Solution Manual For Manifolds, Tensor and Forms ( P. Rentein) PDF

Solution Manual
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
1 Linear algebra page 1

2 Multilinear algebra 20
3 Differentiation on manifolds 33
4 Homotopy and de Rham cohomology 65
5 Elementary homology theory 77
6 Integration on manifolds 84
7 Vector bundles 90
8 Geometric manifolds 97
9 The degree of a smooth map 151
Appendix D Riemann normal coordinates 154
Appendix F Frobenius’ theorem 156
Appendix G The topology of electrical circuits 157
Appendix H Intrinsic and extrinsic curvature 158
iii
1
Linear algebra
1.1 We have
0 = c1 (1, 1) + c2 (2, 1) = (c1 + 2c2 , c1 + c2 )

⇒ c2 = −c1 ⇒ c1 − 2c1 = 0 ⇒ c1 = 0 ⇒ c2 = 0,
so (1, 1) and (2, 1) are linearly independent. On the other hand,
0 = c1 (1, 1) + c2 (2, 2) = (c1 + 2c2 , c1 + 2c2 )
can be solved by choosing c1 = 2 and c1 = −1, so (1, 1) and (2, 2) are

linearly dependent (because c1 and c2 are not necessarily zero).
1.2 Subtracting gives

0= vi ei − vi ei = (vi − vi )ei .
i i i
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
where si ∈ S. Then some of the ci ’s are nonzero. We cannot have c = 0,

because S is linearly independent. Therefore v lies in the span of S, which
says that dim V = |S| < n, a contradiction.
1.5 Let S, T : V → W be two linear maps, and let {ei } be a basis for V .

Assume Sei = T ei for all i, and that v = i ai ei . Then Sv = i ai Sei =

i ai T ei = T v.
1.6 Let v1 , v2 ∈ ker T . Then T (av1 + bv2 ) = aT v1 + bT v2 = 0, so ker T is
closed under linear combinations. Moreover ker T contains the zero vector of
V . All the other vector space properties are easily seen to follow, so ker T is a
subspace of V . Similarly, let w1 , w2 ∈ im T and consider aw1 + bw2 . There
exist v1 , v2 ∈ V such that T v1 = w1 and T v2 = w2 , so T (av1 + bv2 ) =
aT v1 + bT v2 = aw1 + bw2 , which shows that im T is closed under linear
combinations. Moreover, im T contains the zero vector, so im T is a subspace
of W .
1.7 For any two vectors v1 and v2 we have
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
suppose x ∈ ker π ∩ im π. Then there is a y ∈ V such that x = π(y). But

then 0 = π(x) = π 2 (y) = π(y) = x.
b. Let { f i } be a basis for W , and complete it to a basis of V by adding a lin-
early independent set of vectors {g j }. Let U be the subspace of V spanned
by the gi ’s. With these choices, any vector v ∈ V can be written uniquely
as v = w + u, where w ∈ W and u ∈ U . Define a linear map π : V → V
by π(v) = w. Obviously π(w) = w, so π 2 = π .
1.10 Clearly, T 0 = 0, so T −1 0 = 0. Let T v1 = v1 and T v2 = v2 . Then
aT −1 v1 + bT −1 v2 = av1 + bv2 = (T −1 T )(av1 + bv2 ) = T −1 (av1 + bv1 ),
which shows that T −1 is linear.

1.11 The identity map I : V → V is clearly an automorphism. If S ∈ Aut V
then S −1 S = SS −1 = I . Finally, if S, T ∈ Aut V , then ST is invertible,
with inverse (ST )−1 = T −1 S −1 . (Check.) This implies that ST ∈ Aut V .
(Associativity is automatic.)
1.12 By exactness, the kernel of ϕ1 is the image of ϕ0 . But the image of ϕ0 consists
only of the zero vector (as its domain consists only of the zero vector). Hence
the kernel of ϕ1 is trivial, so by Exercise 1.7, ϕ1 must be injective. Again by
exactness, the kernel of ϕ3 is the image of ϕ2 . But ϕ3 maps everything to zero,
so V3 = ker ϕ1 , and hence V3 = im V2 , which says that ϕ2 is surjective. The
converse follows by reversing the preceding steps. As for the last assertion, ϕ
is both injective and surjective, so it is an isomorphism.
1.13 If T is injective then ker T = 0, so by the rank/nullity theorem rk T =
dim V = dim W , which shows that T is surjective as well.
1.14 The rank of a linear map is the dimension of its image. There is no way
that the image of ST can be larger than that of either S or T individually,
because the dimension of the image of a map cannot exceed the dimension of
its domain.
1.15 If v ∈ [v] then v = v + u for some u ∈ U . By linearity ϕ(v ) = ϕ(v) + w
for some w ∈ W , so [ϕ(v )] = [ϕ(v) + w] = [ϕ(v)].
1.16 Pick a basis {ei } for V . Then,

(ST )i j ei = (ST )e j = S( Tk j ek ) = Tk j Sek = Tk j Sik ei .
i k k ik
Hence

(ST )i j = Sik Tk j = (ST )i j ,
k
which shows that ST → ST .

4 Linear algebra
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
It follows that T e j is represented by the j th column of T , so the maxi-

mum number of linearly dependent vectors in the image of T is precisely
the maximum number of linearly independent columns of T .

1.19 Suppose i ci θi = 0. By linearity of the dual pairing,

0 = ej, ci θi = ci e j , θi = ci δi j = c j ,
i i i
so the θ j ’s are linearly independent.

Now let f ∈ V ∗ . Define f (e j ) =: a j and introduce a linear functional

g := i ai θi . Then

g(e j ) = g, e j = ai δi j = a j ,
i
so f = g (two linear functionals that agree on a basis agree everywhere).

Hence the θ j ’s span.

1.20 Suppose f (v) = 0 for all v. Let f = i f i θi and v = e j . Then f (v) =
f (e j ) = f j = 0. This is true for all j, so f = 0. The other proof is similar.
1.21 Let w ∈ W and θ1 , θ2 ∈ Ann W . Then
(aθ1 + bθ2 )(w) = aθ1 (w) + bθ2 (w) = 0,
so Ann W is closed under linear combinations. Moreover, the zero func-
tional (which sends every vector to zero) is clearly in Ann W , so Ann W is a
subspace of V ∗ .
Conversely, let U ∗ ⊆ V ∗ be a subspace of V ∗ , and define
W := {v ∈ V : f (v) = 0, for all f ∈ U ∗ }.
If f ∈ U ∗ then f (v) = 0 for all v ∈ W , so f ∈ Ann W . It therefore suffices
to prove that dim U ∗ = dim Ann W . Let { f i } be a basis for U ∗ , and let {ei }
be its dual basis, satisfying f i (e j ) = δi j . Obviously, ei ∈ W . Thus dim W =
dim V − dim U ∗ . On the other hand, let {wi } be a basis for W and complete
Linear algebra 5
it to a basis for V : {w1 , . . . , wdim W , edim W +1 , . . . , edim V }. Let {u i } be a basis

for Ann W . Then u i (e j ) = 0, else e j ∈ W . So dim Ann W = dim V −dim W .
1.22 a. The map is well defined, because if [v ] = [v] then v = v + w for some
w ∈ W , so ϕ( f )([v ]) = f (v ) = f (v + w) = f (v) + f (w) = f (v) =
ϕ( f )([v]). Moreover, if ϕ( f ) = ϕ(g) then for any v ∈ V , 0 = ϕ( f −
g)([v]) = ( f − g)(v), so f = g. But the proof of Exercise 1.21 shows that
dim Ann W = dim(V /W ) = dim(V /W )∗ , so ϕ is an isomorphism.
b. Suppose [g] = [ f ] in V ∗ / Ann W . Then g = f + h for some h ∈
Ann W . So π ∗ ([g])(v) = g(π(v)) = f (π(v)) + h(π(v)) = f (π(v)) =
π ∗ ([ f ])(v). Moreover, if π ∗ ([ f ]) = π ∗ ([g]) then f (π(v)) = g(π(v)) or
( f − g)(π(v)) = 0, so f = g when restricted to W . Dimension counting
shows that π ∗ is an isomorphism.
1.23 Let g be the standard inner product on Cn and let u = (u 1 , . . . , u n ), v =
(v1 , . . . , vn ) and w = (w1 , . . . , wn ). Then

g(u, av + bw) = u i (avi + bwi )
i

=a u i vi + b u i wi
i i
= ag(u, v) + bg(u, w).
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
because the modulus squared of a complex number is always nonnegative, so

g is nonnegative definite. Moreover, the only way we could have g(u, u) = 0
is if each u i were zero, in which case we would have u = 0. Thus g is
positive definite. The same proof applies in the Euclidean case, but fails in
the Lorentzian case because then
6 Linear algebra

n−1
g(u, u) = −u 20 + u i2 ,
i=1
and it could happen that g(u, u) = 0 but u = 0. (For example, let u =

(1, 1, 0, . . . , 0).)
1.24 We have
(A∗ (a f + bg))(v) = (a f + bg)(Av) = a f (Av) + bg(Av)
= a(A∗ f )(v) + b(A∗ g)(v) = (a A∗ f + b A∗ g)(v),
so A∗ is linear. (The other axioms are just as straightforward.)

1.25 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
∗
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
dim Vi = dim ker ϕi + dim im ϕi .
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,
because the sum is telescoping.

1.31 An arbitrary term of the expansion of det A is of the form
(−1)σ A1σ (1) A2σ (2) . . . Anσ (n) . (1)
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
(−1)σ Aσ −1 (1)1 Aσ −1 (2)2 . . . Aσ −1 (n)n , (2)
where σ −1 is the inverse permutation to σ . For example, suppose σ (5) = 1.

Then there would be a term in (1) of the form A5σ (5) = A51 . This term appears
first in (2), as σ −1 (1) = 5. Since a permutation and its inverse both have the
8 Linear algebra
−1
same sign (because σ σ −1 = e implies (−1)σ (−1)σ = 1), Equation (2) may
be written
−1
(−1)σ Aσ −1 (1)1 Aσ −1 (2)2 . . . Aσ −1 (n)n . (3)
Hence
−1
det A = (−1)σ Aσ −1 (1)1 Aσ −1 (2)2 . . . Aσ −1 (n)n . (4)
σ ∈Sn
As σ runs over all the elements of Sn , so does σ −1 , so (4) may be written

−1
det A = (−1)σ Aσ −1 (1)1 Aσ −1 (2)2 . . . Aσ −1 (n)n . (5)
σ −1 ∈Sn
But this is just det A T .

1.32 By (1.46) the coefficient of A11 in det A is

(−1)σ A2σ (2) . . . Anσ (n) , (1)
σ ∈Sn
where σ means a general permutation in Sn that fixes σ (1) = 1. But

this means the sum in (1) extends over all permutations of the numbers
{2, 3, . . . , n}, of which there are (n − 1)!. A moment’s reflection reveals that
(1) is nothing more than the determinant of the matrix obtained from A by
removing the first row and first column, namely A(1|1).
Now consider a general element Ai j . What is its coefficient in det A?
Well, consider the matrix A obtained from A by moving the i th row up
to the first row. To get A we must execute i − 1 adjacent row flips, so
det A = (−1)i−1 det A. Now consider the matrix A obtained from A by
moving the j th column left to the first column. Again we have det A =
(−1) j−1 det A . So det A = (−1)i+ j det A. The element Ai j appears in the
(11) position in A , so by the reasoning used above, its coefficient in det A
is just det A (1|1) = det A(i| j). Hence, the coefficient of Ai j in det A is
(−1)i+ j det A(i| j) = A i j .
Next consider the expression
11 + A12 A
A11 A 12 + · · · + A1n A
1n , (2)
which is (1.57) with i = 1. Thinking of the Ai j as independent variables, each

term in (2) is distinct (because, for example, only the first term contains A11 ,
etc.). Moreover, each term appears in (2) precisely as it appears in det A (with
the correct sign and correct products of elements of A). Finally, (2) contains
n(n −1)! = n! terms, which is the number that appear in det A. So (2) must be
Linear algebra 9
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)
Of course, we could have substituted a new j th column instead, with similar

results.
If we were to let the Bi j be the elements of any row of A other than the i th ,
then the expression in Equation (1) would vanish, as the determinant of any
matrix with two identical rows is zero. This gives us the following result:
i1 + Ak2 A
Ak1 A i2 + · · · + Akn A
in = 0, k = i. (2)
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
If i = k this is an expansion in terms of alien cofactors and vanishes. If i = k

then this is just the determinant of A. Hence [A(adj A)]ik = (det A)δik . This
proves the first half. To prove the second half, note that (adj A)T = (adj A T ).
That is, the transpose of the adjugate is the adjugate of the transpose. (Just
trace back the definitions.) Hence, using the result (whose easy proof is left
to the reader) that (AB)T = B T A T for any matrices A and B,
[(adj A)A)]T = A T (adj A)T = A T adj A T = (det A T )I = (det A)I. (3)
1.34 By (1.59),
A(adj A) = (adj A) A = (det A)I,
so if A is nonsingular, then the inverse of A is adj A/ det A, and if A is

invertible, then multiplying both sides of this equation by A−1 gives adj A =
(det A) A−1 , which implies A is nonsingular (because the adjugate cannot
10 Linear algebra
vanish identically). Next, suppose Av = 0. If A were invertible, then multi-

plying both sides of this equation by A−1 would give v = 0. So v is nontrivial
if and only if A is not invertible, which holds if and only if det A = 0.
1.35 A is nonsingular, so
1
x = A−1 b = (adj A)b.
det A
But expanding by the i th column gives

det A(i) = ji =
bj A (adj A)i j b j ,
j j
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
b. By orthogonality of R, for any u, v ∈ V ,

g(v, w) = g(Rv, Rw) = g(R † Rv, w).
It follows that R † R = I , where I is the identity map. (Just let v and w run
through all the basis elements.) By the discussion following Exercise 1.26,
R † is represented by R T , so R T R = I. As a left inverse must also be a
right inverse, R R T = I. Tracing the steps backwards yields the converse.
c. We have I = R T R, so by Exercise 1.31 and Equation (2.54), 1 =
det R T det R = (det R)2 .
d. Let R be orthogonal so that R T R = I. In components, Rik R jk = δi j . A
priori this looks like n 2 conditions (the number of entries in the identity
matrix), but δi j is symmetric, so the independent conditions arise from
those pairs (i, j) for which i ≤ j. To
count these we observe that there
are n pairs (i, j) with i = j, and n2 = n(n − 1)/2 pairs with i < j.
Adding these together gives n(n + 1)/2 constraints. Therefore the number
of independent parameters is n − n(n + 1)/2 = n(n − 1)/2.
1.39 From (2.54) we get
1 = det I = det A A−1 = (det A)(det A−1 ),
so
det A−1 = (det A)−1 .
1.40 In our shorthand notation we can write

Ae j = ei Ai j ⇒ Ae = e A, (1)
i
and similarly,

Aej = ei Ai j ⇒ Ae = e A . (2)
i
Substituting into (2) we get

AeS = eS A ⇒ Ae = eS A S−1 ,
so comparing with (1) (and using the fact that e is a basis) gives
A = S A S−1 or A = S−1 AS.
1.41 Assume A has n linearly independent eigenvectors {v1 , v2 , . . . , vn } with cor-
responding eigenvalues {λ1 , λ2 , . . . , λn }, and let S be a matrix whose columns
are the vectors vi , i = 1, . . . , n. Then S is clearly nonsingular (because its
rank is maximal), and multiplication reveals that AS = S, where is
the diagonal matrix diag(λ1 , . . . , λn ) with the eigenvalues of A along the
12 Linear algebra
diagonal. It follows that S−1 AS = . Conversely, if there exists a nonsin-

gular matrix S such that S−1 AS = , then as AS = S, the columns of
S are the eigenvectors of A (which are linearly independent because S is
nonsingular), and the diagonal elements of are the eigenvalues of A.
1.42 The equation Av = λv holds if and only if ( A − λI)v = 0, which has a
nontrivial solution for v if and only if A − λI is singular, and this holds if and
only if det( A − λI) = 0. So the roots of the characteristic polynomial are the
eigenvalues of A.
1.43 Let p A (λ) = det( A − λI) be the characteristic polynomial of A. Then
p S−1 AS = det(S−1 AS − λI)

= det(S−1 ( A − λI)S)
= (det S)−1 p A (λ) det S
= p A.
It follows that the eigenvalues of A are similarity invariants.

1.44 Let p A (λ) be the characteristic polynomial of A. Then we can write
p A (λ) = (−1)n (λ − μ1 )(λ − μ2 ) · · · (λ − μn ),
where the roots (eigenvalues) μi are not necessarily distinct. By expanding

out the product we see that the constant term in this polynomial is the product
of the eigenvalues, but the constant term is also p A (0) = det A.
Again by expanding, we see that the coefficient of the term of order λn−1
is the sum of the eigenvalues times (−1)n−1 . Now consider det( A − λI). Of
all the terms in the Laplace expansion, only one contains n − 1 powers of λ,
namely the product of all the diagonal elements. (In order to contain n − 1
powers of λ the term must contain at least n − 1 diagonal elements, which
forces it to contain the last diagonal element as well.) But the product of all
the diagonal elements is

(A11 − λ)(A22 − λ) · · · (Ann − λ) = (−1)n λn + (−1)n−1 λn−1 Aii + · · · ,
i
where the missing terms are of lower order in λ.

1.45 For any two matrices A andB,

tr AB = Ai j B ji = B ji Ai j = tr B A.
ij ij
The general case follows by setting A = A1 A2 · · · An−1 and B = An .

Linear algebra 13
1.46 a.
∞

(1 + t x j ) = (1 + t x1 )(1 + t x2 ) · · ·
j=1
= 1+t (x1 + x2 + · · · )+t 2 (x1 x2 + x1 x3 + · · · + x2 x3 + · · · )

+ · · ·+ t n (x1 x2 · · · )
∞

= t jej.
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
From d E/dt = E(t)P(−t) we get

kek t k−1 = ( ei t i )( p j (−t) j−1 )
k i j

= (−1) j−1 ei p j t i+ j−1 .
j i
Equating powers of t on both sides gives

k
kek = (−1) j−1 ek− j p j .
j=1
14 Linear algebra
d. Write down the first n Newton identities in matrix form to get

⎛ ⎞⎛ ⎞ ⎛ ⎞
1 0 ··· ··· e1 s1
⎜ s1 2 0 · · · · · ·⎟ ⎜e2 ⎟ ⎜ s2 ⎟
⎜ ⎟⎜ ⎟ ⎜ ⎟
⎜ s2 · · ·⎟ ⎜e3 ⎟ ⎜ ⎟
⎜ s 1 3 0 ⎟ ⎜ ⎟ = − ⎜ s3 ⎟ ,
⎜ . .. .. .. ⎟ ⎜ . ⎟ ⎜.⎟
⎝ .. . . . ⎠ ⎝ .. ⎠ ⎝ .. ⎠
sn−1 sn−2 · · · s1 n en sn
where si := (−1)i pi . Then Cramer’s rule gives en = det A/n!, where
⎛ ⎞
1 0 ··· ··· −s1
⎜ s1 2 0 ··· ··· −s2 ⎟
⎜ ⎟
⎜ s2 s1 3 0 ··· −s3 ⎟
⎜ ⎟
A=⎜ ⎜ . .
. .
. . . . . . . .
.
⎟
⎜ . . . . . ⎟⎟
⎜ .. ⎟
⎝sn−2 sn−3 · · · s1 n − 1 . ⎠
sn−1 sn−2 · · · s1 −sn
Taking the determinant of this, commuting the last column to the front,
and pulling out a sign gives

s1 1 0 · · · · · ·

s2 s1 2 0 · · ·

n s3 · · · · · · .
det A = (−1) s 2 s 1
. . .. ..
.. .. . .

s s ··· s
n n−1 sn−2 1
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
e. If Av = λv then Ak v = λk v. As the trace is the sum of the eigenval-

ues, tr Ak = pk (λ1 , . . . , λn ), where λ1 , . . . , λn are the n eigenvalues of A.
The determinant is the product of the eigenvalues, which is en (λ1 , . . . , λn ).
Thus,

p1 1 0 0

p2 p1 2 0
4!e4 = = p14 − 6 p12 p2 + 3 p22 + 8 p1 p3 − 6 p4 ,
p p p 3
3 2 1
p p3 p2 p 1
4
Linear algebra 15
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
1.49 By positive definiteness,
0 ≤ g(u + αv, u + αv) = g(u, u) + 2αg(u, v) + α 2 g(v, v).
Minimizing the right side with respect to α gives

g(u, v)
α=− ,
g(v, v)
and plugging back in gives
g(u, v)2 g(u, v)2
0 ≤ g(u, u) − 2 +
g(v, v) g(v, v)
or
0 ≤ g(u, u)g(v, v) − g(u, v)2 .
Moreover, equality holds if and only if u + αv = 0, or u = −αv.

1.50 Symmetry is obvious, and bilinearity follows from the linearity of integration.
For example,
∞
( f, ag + bh) = f (x)(ag(x) + bh(x)) d x
−∞
∞
= (a f (x)g(x) + b f (x)h(x)) d x
−∞
= a( f, g) + b( f, h).
Finally,
∞
( f, f ) = f 2 (x) d x ≥ 0
−∞
because f 2 ≥ 0. The integral vanishes if and only if f = 0, so the map (·, ·)

is positive definite.
1.51 Suppose v1 , . . . , vn are linearly dependent. Then there exist constants c j , not
all zero, such that
c1 v1 + c2 v2 + . . . cn vn = 0. (1)
Take the inner product of (1) with each of the vectors vi to get
g(v1 , v1 )c1 + g(v1 , v2 )c2 ··· + g(v1 , vn )cn = 0,

g(v2 , v1 )c1 + g(v2 , v2 )c2 ··· + g(v2 , vn )cn = 0,
.. .. .. (2)
. . .
g(vn , v1 )c1 + g(vn , v2 )c2 ··· + g(vn , vn )cn = 0.
Linear algebra 17
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
e2 = x 2 − e0 g(e0 , x 2 ) − e1 g(e1 , x 2 ).
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.
g(e2 , e2 ) = g(x 2 −1/3, x 2 −1/3) = g(x 2 , x 2 )−(2/3)g(x 2 , 1)+(1/9)g(1, 1).
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
e3 = x 3 − e0 g(e0 , x 3 ) − e1 g(e1 , x 3 ) − e2 g(e2 , x 3 ).
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
For the second, if f ∈ im T ∗ then f = T ∗ g for some g ∈ W ∗ . So, if v ∈

ker T ,
f (v) = T ∗ g(v) = gT (v) = g(0) = 0,
so f ∈ Ann ker T . Conversely, let f ∈ Ann ker T . By Theorem 1.3 we can
write
V = ker T ⊕ S(im T ),
where S is a section of T . Define an element g ∈ W ∗ by

f (S(w)), if w ∈ im T ,
g(w) :=
0, otherwise.
For any v ∈ V we have
g(T v) = f (v), (1)
because if v ∈ ker T then both sides of (1) are zero, and if v ∈ ker T then
both sides of (1) are equal by virtue of the definition of g (and the fact that
T ◦ S = 1). We conclude that f = T ∗ g ∈ im T ∗ .

1.54 Suppose that v = k vk ek . Then, as the basis is orthonormal,

g(v, ei ) = vk g(ek , ei ) = g(ei , ei )vi .
k
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
so f v f = f (two linear maps that agree on a basis agree everywhere).

Therefore, the two maps v → f v and f → v f are indeed inverses of one
another.
2
Multilinear algebra
2.1 Let e and e be two bases related by the change of basis matrix A, so that

ei = ei Ai i .
i
Then the components of T in the two bases are related by

Ti j = Ai i A j j Ti j .
ij
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
In the first equality we are allowed to switch i and j in Ai j because A is

symmetric. In the second equality we swap i and j in B i j at the cost of a
minus sign, because B is antisymmetric. In the third equality we change the
names of the dummy indices from i and j to j and i, respectively. The last
equality follows because the only number equal to its negative is zero.
2.3 Just repeat what was done in the text, without the signs.
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
Hence (cf., Exercise 2.3),

1 1 1 −1
alt(T τ ) = (−1)σ (T τ )σ = (−1)σ T σ τ = (−1)π τ T π
p! p! p!
σ ∈Sp σ ∈Sp π τ −1 ∈Sp
1 1
= (−1)τ (−1)π T π = (−1)τ (−1)π T π = (−1)τ alt(T ).
p! −1 p!
πτ ∈Sp π∈Sp
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
This proves (i).

For simplicity, we prove (ii) for two rank one tensors, as the general case
follows by similar considerations. Suppose S = v1 ⊗ · · · ⊗ v p and T =
v p+1 ⊗ · · · ⊗ v p+q . Then
1
alt(S ⊗ T ) = (−1)σ vσ (1) ⊗ vσ (2) ⊗ · · · ⊗ vσ ( p+q) . (1)
( p + q)!
σ ∈Sp+q
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.
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,
(R ∧ S) ∧ T − alt(R ⊗ S ⊗ T ) = alt((R ∧ S) ⊗ T ) − alt(R ⊗ S ⊗ T )

= alt((R ∧ S − R ⊗ S) ⊗ T )
= (R ∧ S − R ⊗ S) ∧ T.
But this vanishes by (ii), because by (i), alt(R ∧ S) = alt(alt(R ⊗ S)) =

alt(R ⊗ S), so
alt(R ∧ S − R ⊗ S) = 0.
Similar reasoning shows that
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 .
On the other hand, the matrix representing T is

⎛ ⎞
1 0 1
⎝2 3 0⎠
0 2 1
and its determinant is—you guessed it—7.

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,
eσ (1) ∧ · · · ∧ eσ (n) = (−1)σ e1 ∧ · · · ∧ en .
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
columns equal kills the determinant, (3) adding a multiple of a column to

another column leaves the determinant unchanged because, by multilinearity,
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
η ∧ λ = e I ∧ g(e J , e J )e J = g(e J , e J )σ = (−1)d g(e I , e I )σ

= (−1)d g(η, λ)σ.
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
for some coefficients ai j . Hence

k
n
k
n
0= vi ⊗ ai j v j = ai j vi ⊗ v j .
i=1 j=1 i=1 j=1
But vi ⊗ v j , 1 ≤ i, j, ≤ n is a basis for the tensor product space V ⊗ V .

Therefore all the ai j must vanish.
2.13 By definition,

(A ⊗ B)(ei ⊗ f j ) = (A ⊗ B)k,i j (ek ⊗ f ),
k
where (A ⊗ B)k,i j is the (k, i j)th component of the matrix representing

A ⊗ B in the basis {ei ⊗ f j }. But also,

(A ⊗ B)(ei ⊗ f j ) = Aei ⊗ B f j = Aki Bj (ek ⊗ f ).
k
It follows that
(A ⊗ B)k,i j = Aki Bj .
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
for some matrices A and B. Thus,

⎛ ⎞

p

p
p

n
0= vi ∧ wi = vi ∧ ⎝ Ai j v j + Bi j v j ⎠
i=1 i=1 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
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
2.18 From (2.59) and (2.61) we get

(−1)d g(λ, μ)σ = λ ∧ μ = g(λ, μ)σ,
from which the result follows.
2.19 Let {ei }, 1 ≤ i ≤ m be a basis for V and { f j }, 1 ≤ j ≤ n a basis for W . Then
{(ei , 0), (0, f j )}, 1 ≤ i ≤ m, 1 ≤ j ≤ n is a basis for V ⊕ W . The vector

space n (V ⊕ W ) is thus generated (as a direct sum) by all basis elements of
the form
(ei1 , 0) ∧ · · · ∧ (eik , 0) ∧ (0, f j1 ) ∧ · · · ∧ (0, f jn−k ),
where k runs from 0 to n (and all the indices are in increasing—but not
necessarily sequential—order). Under the linear map defined in the problem,
(ei1 ∧· · · eik )⊗( f j1 ∧· · ·∧ f jn−k ) →(ei1 , 0)∧· · ·∧(eik ,0)∧(0, f j1 )∧· · ·∧(0, f jn−k),
so
n

k n−k
V⊗ W ∼
=
n
(V ⊕ W )
k=0
(because the induced map on the direct sum is clearly bijective).

2.20 a. If A is diagonalizable with eigenvalues {λi } then I +z A is diagonalizable

with eigenvalues {1 + λi z}. So (denoting the kth elementary symmetric
function by E k )

det(I + z A) = (1 + λk z) = E k (λ1 , . . . , λn )z k .
k k
On the other hand, if {ei } is a basis for V ,

( k A)(ei1 ∧ · · · ∧ eik ) = Aei1 ∧ · · · ∧ Aeik
= (λi1 · · · λik )(ei1 ∧ · · · ∧ eik ),
so
k
tr A= λi1 · · · λik = E k (λ1 , . . . , λn ).
i 1 <i 2 <···<i k
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 ]
+ z 3 [Ae1 ∧ Ae2 ∧ Ae3 ]

= e1 ∧ e2 ∧ e3

+ z[ 1 Ae1 ∧ e2 ∧ e3 + e1 ∧ 1 Ae2 ∧ e3 + e1 ∧ e2 ∧ 1 Ae3 ]

+ z 2 [ 2 A(e1 ∧ e2 ) ∧ e3 + 2 A(e1 ∧ |e2 | ∧ e3 )

+ e1 ∧ 2 A(e2 ∧ e3 )]

+ z 3 [ 3 A(e1 ∧ e2 ∧ e3 )]

= (1 + z tr 1 A + z 2 tr 2 A + z 3 tr 3 A)(e1 ∧ e2 ∧ e3 ),
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:
∗ ∗ ∗| ∗ ∗||∗,
where the number of stars in each “compartment” determined by the bars

equals the number of times a particular element appears in the multiset.
For a multiset of size p we must have p stars, while n is the number of
compartments, so there must be n −1 bars. Altogether there are p +n −1
symbols, and a multiset is a choice of which n − 1 of those symbols

will
be bars. By definition, the number of such choices is n+n−1 p−1
.
b. The natural map ψ : Sym V → F[e1 , e2 , . . . , en ] given by e1 · · ·
e p → e1 e2 · · · e p and extended by linearity, provides the natural isomor-
phism we seek. The tedious but entirely straightforward details are left
to the reader.
c. Just follow the proof of Exercise 2.19.
d. If A is diagonalizable with eigenvalues {λi } then I −z A is diagonalizable

with eigenvalues {1 − λi z}. So
1
= (1 − λk z)−1 = Hk (λ1 , . . . , λn )z k ,
det(I − z A) k k
where Hk is the k th homogeneous symmetric function. On the other hand,

if {ei } is a basis for V ,
(Symk A)(ei1 · · · eik ) = Aei1 · · · Aeik

= (λi1 · · · λik )(ei1 · · · eik ),
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
This expression is symmetric in u and v, so [

a (u),
a (v)]− = 0.
(ii).

a † (u)
a † (v)v1 · · · v p =
a † (u)v v1 · · · v p
= u v v1 · · · v p .
This is manifestly symmetric in u and v, so [

a † (u),
a † (v)]− = 0.
(iii).

a (u)
a † (v)v1 · · · v p =
a (u)v v1 · · · v p

= (u, vi )v v1 · · ·
vi · · · v p ,
i
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.
(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
3.5 Let X be Hausdorff, and suppose x ∈ X . If y ∈ X , then there is an open set

U containing y but not x. Therefore X is open (being the union of open sets),
so x is closed.
3.6 Let X be compact, and let Y ⊆ X be closed. Let U be an open cover of Y .
Then U ∪ Y is an open cover of X and so admits a finite subcover U of X .
But Y is not covered by Y , so it must be covered by all the sets in U distinct
from Y , namely by a finite subcollection of the sets in U .
3.7 Let K be closed in Y . Then Y \K is open. Now, f −1 (Y \K ) = X \ f −1 (K ),
because the inverse image of Y is X itself, so everything not in f −1 (K ) maps
to Y \K . By hypothesis, f −1 (K ) is closed, so X \ f −1 (K ) is open. As every
open set is the complement of some closed set, the inverse image of every
open set is open, so f is continuous.
3.8 Label the two types of continuity C1 (for the inverse image definition) and
C2 (for the epsilon-delta definition). Assume f is C1, and let p ∈ R. The
interval I := ( f ( p) − , f ( p) + ) is an open subset of R, so f −1 (I ) is open
as well. But f −1 (I ) contains p, so it must contain an open interval (a, b)
containing p (because every open set in R is a union of open intervals). Now
let δ = min{ p − a, b − p}. Then the condition |x − p| < δ guarantees that
x ∈ (a, b), and hence that f (x) ∈ I , which means that | f (x) − f ( p)| < .
Hence f is C2.
To prove the converse it suffices to show that the inverse image of a single
open interval is open, because the inverse image of a union of sets is the union
of the inverse images. So let I = (c, d), and let p ∈ f −1 (I ) so that f ( p) ∈ I .
Choose = min{ f ( p) − c, d − f ( p)}. As f is C2, there exists a δ such that
the open interval ( p − δ, p + δ) is in f −1 (I ). Therefore f −1 (I ) is the union
of open sets, which shows that f is C1.
3.9 Let U be an open cover of Y . By continuity and surjectivity, { f −1 (U ) : U ∈
U} is an open cover of X , which therefore has a finite subcover U . By con-
struction the images of the open sets in the finite set U are elements of U,
and by surjectivity they cover Y .
3.10 Let U be open in X . Then ι−1 (U ) = Y ∩ U , which is clearly open in τY , so ι
is continuous in the subspace topology. On the other hand, if ι is continuous
in some topology σ on Y , then for any open set U in X , ι−1 (U ) = Y ∩ U is
open in σ , so if Y ∩ U were not in σ then ι would not be continuous.
3.11 Let X be Hausdorff and let Y be a subspace of X . If p and q are distinct points
of Y then they are distinct in X , so there exist subsets U and V with U ∩V = ∅
and p ∈ U and q ∈ V . But then (Y ∩ U ) ∩ (Y ∩ V ) = Y ∩ (U ∩ V ) = ∅ as
well, so Y is Hausdorff.
3.12 By Exercise 3.5, y is closed in Y , so by Exercise 3.7, f −1 (y) is closed. But
by Exercise 3.6, the closed subset of a compact set is compact.
Differentiation on manifolds 35
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 .
For the second, note that id(x) = x, so in components (id)i (x) = x i .

Applying the definition of the derivative gives
∂(id)i ∂xi
(D id)(x) = = = δ ij .
∂x j ∂x j
[Incidentally, this is why it is a bit dangerous to write ‘1’ for the identity,
because one might think that the derivative of ‘1’ should be 0!]
3.16 We consider the nine coordinate charts suggested in the problem hint. First
we observe that these patches do indeed cover the torus. Next, we must
show that the transition maps between patches are diffeomorphisms. We
do this for two overlaps, leaving the rest to your imagination. On U1 ∩ U2
we have (ϕ1 ◦ ϕ2−1 )(x, y) = (x, y) for the same reason as in the Möbius
case. On U2 ∩ U8 (where U8 = {[x, y] : 2/3 < x < 4/3, 1/3 < y < 1})
we have, for example, ϕ2−1 (1/6, 1/2) = [1/6, 1/2] = [7/6, 1/2], so
(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
We construct an inverse map from Rn to Rn+1 for 0 ≤ j ≤ n by
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 ).
For the other direction, we have

x0 x1 x j−1 x j+1 xn
(ϕ −1
j ◦ ϕ j )( p) = ϕ −1
j , , . . . , , . . . ,
xj xj xj xj xj
0 1 j−1 j+1 n
x x x x x
= j
, j , . . . , j , 1, j . . . , j
x x x x x
= p,
because p represents the equivalence class of [x 0 , x 1 , . . . , x n ].

To show that ϕi ◦ ϕ −1j is a diffeomorphism, we must show it is a smooth
bijection. If p is as above, then
n
x0 x1 x j−1 x j+1 x
(ϕi ◦ ϕ −1
j ) j
, j
, . . . , j
, j ..., j
x x x x x
0 1
x x x j−1 x j+1 xn
= ϕi , , . . . , , 1, . . . ,
xj xj xj xj xj
⎛ ⎞
1 x x0 1 %
x i x n
= i j ⎝ j , j , . . . , j , . . . , !"#$ 1 ,..., j ⎠
x /x x x x x
j th place
0 1
x x x i−1 x i+1 xn
= , , . . . , , . . . , .
xi xi xi xi xi
(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
where y1 = 0. This map is infinitely differentiable (smooth), so we are done.

The Jacobian of this map is
⎛ ⎞
−1/(y 1 )2 0 0 ··· 0
⎜−y 2 /(y 1 )2 1/y 1 0 ··· 0 ⎟
⎜ ⎟
⎜−y 3 /(y 1 )2 1/y 1 ··· ⎟
⎜ 0 0 ⎟.
⎜ .. .. .. .. .. ⎟
⎝ . . . . . ⎠
−y n /(y 1 )2 0 0 · · · 1/y 1
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.
3.18 Define f : Rn+1 → R by

f (x) = (x i )2 .
i
As R and Rn+1 are obviously manifolds and as S n is f −1 (1), we need only

show that 1 is a regular value of f . For this, we compute the Jacobian matrix,
which is a 1 by n + 1 matrix whose j th entry is ∂ f /∂ x j = 2x j . This matrix is
rank deficient (i.e., less than one) only at x = 0, but fortunately for us, x = 0
is not on the sphere, so all is well.
3.19 Applying the chain rule gives

∂ ∂x ∂ ∂y ∂ ∂z ∂
= + +
∂r ∂r ∂ x ∂r ∂ y ∂r ∂z

∂ ∂x ∂ ∂y ∂ ∂z ∂
= + +
∂θ ∂θ ∂ x ∂θ ∂ y ∂θ ∂z

∂ ∂x ∂ ∂y ∂ ∂z ∂
= + + .
∂φ ∂φ ∂ x ∂φ ∂ y ∂φ ∂z
Substituting into these equations using
∂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
Now we have to normalize the vectors by dividing by their lengths. ∂/∂r is

already a unit vector, so we just identify er = ∂/∂r . ∂/∂θ has length r , so
eθ = (1/r )∂/∂θ. Finally, ∂/∂/φ has length r sin θ, so eφ = (1/r sin θ)∂/∂φ.
3.20 a. The product of linear maps is linear, as is the sum, so [X, Y ] is linear.
Also
[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,
so [X, Y ] is also a derivation.

b. In the given local coordinates,

∂ j ∂f
(X Y ) f = X i
Y
∂xi ∂x j
j
∂Y ∂ f j ∂ f
2
=X i
+Y .
∂xi ∂x j ∂xi ∂x j
Subtracting a term just like this with X and Y interchanged gives

j
i ∂Y i ∂X ∂f
j
[X, Y ] f = X −Y ,
∂x i ∂x i ∂x j
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.
[X, [Y, Z ]] + [Y, [Z , X ]] + [Z , [X, Y ]]

= [X, Y Z − Z Y ] + [Y, Z X − X Z ] + [Z , X Y − Y X ],
= [X, Y Z ] − [X, Z Y ] + [Y, Z X ] − [Y, X Z ] + [Z , X Y ] − [Z , Y X ],
= XY Z − Y Z X − X ZY + ZY X + Y Z X − Z XY
− Y X Z + X ZY + Z XY − XY Z − ZY X + Y X Z,
= 0.
3.21 Plug (3.37) into (3.38) to get
∂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.
Next, using X = X i (x)∂/∂ x i = X j (y)∂/∂ y j , α = ai (x) d x i = b j (y) dy j

and (3.24) gives
∂y j
X i (x)ai (x) = X, α = Y j (y)b j (y) = b j (y)X i (x) .
∂xi
Stripping off X i from both sides we get (3.38).

3.22 Under a change of basis from x to y, X = X i (∂/∂ y i ), where

i ∂ yi
X =X . i
∂xi
It follows that

∂ Xi ∂ X i ∂ x j ∂ yi i ∂ y
2 i
∂xk
= + X ,
∂y j ∂x j ∂y j ∂xi ∂ xk∂ xi ∂ y j
which differs from (3.53) by the second term.
3.23 Set
d x U2 = d x U1 = d x 1
on the overlap U1 ∩ U2 . But on the overlap,
∂x1 2
dx1 =d x = d x 2,
∂x2
because the Jacobian ‘matrix’ is just unity. So, if we choose d xU2 = d x 2
everywhere on U2 , then it agrees with d xU1 on the overlap. Similar reason-
ing shows that, if we choose d xUi = d x i on the i th patch, the 1-forms all
(smoothly) agree on the overlaps, so together they define a global 1-form d x.
The same argument works for dy.
3.24

∂f ∂f ∂f ∂f ∂f ∂f
df ∧df = dx + dy + dz ∧ dx + dy + dz
∂x ∂y ∂z ∂x ∂y ∂z
∂f ∂f ∂f ∂f
= (d x ∧ dy + dy ∧ d x) + (dy ∧ dz + dz ∧ dy)
∂x ∂y ∂ y ∂z
∂f ∂f
+ (dz ∧ d x + d x ∧ dz)
∂z ∂ x
= 0.
3.25
dα = 6x y 3 z d x ∧ d x + 9x 2 y 2 z dy ∧ d x + 3x 2 y 3 dz ∧ d x
+ y 3 z 2 d x ∧ dy + 3x y 2 dy ∧ dy + 2x y 3 z dz ∧ dy
+ 2z 2 dy ∧ dz + 4yz dz ∧ dz
= (y 3 z 2 − 9x 2 y 2 z) d x ∧ dy + (2z 2 − 2x y 3 z) dy ∧ dz + 3x 2 y 3 dz ∧ d x,
and
d 2 α = −18x y 2 z d x ∧ d x ∧ dy + (3y 2 z 2 − 18x 2 yz) dy ∧ d x ∧ dy

+ (2y 3 z − 9x 2 y 2 ) dz ∧ d x ∧ dy − 2y 3 z d x ∧ dy ∧ dz
− 6x y 2 z dy ∧ dy ∧ dz + (4z − 2x y 3 ) dz ∧ dy ∧ dz
+ 6x y 3 d x ∧ dz ∧ d x + 9x 2 y 2 dy ∧ dz ∧ d x
= 0.
3.26
dω = (2x 2 y − 3yz 3 ) d x ∧ dy ∧ dz,
and d 2 ω = 0 because all the 4-forms vanish.

3.27
∂a ∂ 2a
d 2 (a d x I ) = d d x k
∧ d x I
= d x ∧ d x k ∧ d x I = 0,
k
∂ x k
k
∂ x ∂ x k
because mixed partial derivatives commute, but wedge products anticom-

mute. (See Exercise 2.2.)
3.28 Starting again from
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,
Applying the exterior derivative operator and collecting terms we get

∂ Ey ∂ Ex ∂ Bz
dF = − + (dt ∧ d x ∧ dy)
∂x ∂y ∂t

∂ Ez ∂ Ey ∂ Bx
+ − + (dt ∧ dy ∧ dz)
∂y ∂z ∂t

∂ Ex ∂ Ez ∂ By
+ − + (dt ∧ dy ∧ dz)
∂z ∂x ∂t

∂ Bx ∂ By ∂ Bz
+ + + (d x ∧ dy ∧ dz).
∂x ∂y ∂z
If d F = 0 then each term vanishes individually, whereupon we obtain
Maxwell’s equations without sources.
3.29 Starting from
J = −ρ d x ∧ dy ∧ dz + Jx dt ∧ dy ∧ dz
+ Jy dt ∧ dz ∧ d x + Jz dt ∧ d x ∧ dy.
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
antisymmetric under the interchange of X and Y , we have shown that, on

p forms,
i X i Y = −i Y i X ,
which proves the claim.
d. Using the properties of the interior product we get
ω(X 1 , X 2 , X 3 ) = i X 3 i X 2 i X 1 ω
= i X 3 i X 2 i X 1 (λ ∧ μ ∧ ν)
= i X 3 i X 2 (i X 1 λ ∧ μ ∧ ν − λ ∧ i X 1 (μ ∧ ν))
= i X 3 i X 2 (i X 1 λ ∧ μ ∧ ν − λ ∧ i X 1 μ ∧ ν + λ ∧ μ ∧ i X 1 ν)
= i X 3 i X 2 (λ(X 1 )μ ∧ ν − μ(X 1 )λ ∧ ν + ν(X 1 )λ ∧ μ)
= i X 3 (λ(X 1 )(i X 2 μ ∧ ν − μ ∧ i X 2 ν)
− μ(X 1 )(i X 2 λ ∧ ν − λ ∧ i X 2 ν)
+ ν(X 1 )(i X 2 λ ∧ μ − λ ∧ i X 2 μ))
= i X 3 (λ(X 1 )(μ(X 2 )ν − ν(X 2 )μ)
− μ(X 1 )(λ(X 2 )ν − ν(X 2 )λ)
+ ν(X 1 )(λ(X 2 )μ − μ(X 2 )λ))
= λ(X 1 )μ(X 2 )ν(X 3 ) − λ(X 1 )μ(X 3 )ν(X 2 )
− λ(X 2 )μ(X 1 )ν(X 3 ) + λ(X 3 )μ(X 1 )ν(X 2 )
+ λ(X 2 )μ(X 3 )ν(X 1 ) − λ(X 3 )μ(X 2 )ν(X 1 ).
e. We see immediately that c3 = 1, and in general c p = 1.
3.31 We choose the easy way. If {X 1 , X 2 , . . . , X n } are linearly dependent, then we
may write

n−1
Xn = ai X i
i=1
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
But, by antisymmetry, each summand vanishes, because in each term at least

two of the vectors are the same.
3.32 We compute
φ ∗ ω = 3(t 2 )(t)(2t dt) − 7(2t − 1)2 (t)(2 dt) + 2(t 2 )2 (2t − 1)(dt)
= (6t 4 − 14t (2t − 1)2 + 2t 4 (2t − 1)) dt
= (4t 5 + 4t 4 − 56t 3 + 56t 2 − 14t) dt.
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.
3.36 Suppose h : M → R is a function. Then if f is the identity map, ( f ∗ h)(x) =

h( f (x)) = h(x), which shows that f ∗ is just the identity. Similarly, f (x) = x
implies y i = f i (x) = x i , and the Jacobian matrix (∂ y i /∂ x j ) representing f ∗
is just the identity matrix.
If f : U → V and g : V → W and h : W → R, then
(g ◦ f )∗ h = h ◦ (g ◦ f ) = (h ◦ g) ◦ f = f ∗ (h ◦ g) = f ∗ g ∗ h,
so (g ◦ f )∗ = f ∗ ◦ g ∗ . Lastly, by Exercise 3.35, f ∗ is represented locally by
the Jacobian of the transformation, so by the chain rule, (g ◦ f )∗ = g∗ f ∗ .
3.37 4. We have

φt∗ ω, Y p = ω, Y |q = ωq , Yq = ω(Y )q .
But also

φt∗ ω, φt∗ Y p = (φt∗ ω) p , (φt∗ Y ) p = (φt∗ ω) p , (φ−t∗ Y ) p
= ω(φt∗ φ−t∗ Y )|q = ω(Y )q ,
so,

φt∗ ω, Y = φt∗ ω, φt∗ Y ,
i.e., pullback commutes with contraction. Applying this result gives
φt∗ ω, Y − ω, Y
L X ω, Y = lim
t→0
∗ t
∗
φ ω, φt Y − ω, Y
= lim t
t→0
∗ t
φt ω, φt∗ Y − φt∗ ω, Y + φt∗ ω, Y − ω, Y
= lim
t→0
& ' t & '
∗
∗ φt Y − Y φt∗ ω − ω
= lim φt ω, + lim ,Y
t→0 t t→0 t
= ω, L X Y + L X ω, Y .
5. Taking the hint, we observe that L X d = dL X because pullback com-
mutes with d. So, invoking 4 we get
X (Y ( f )) = L X (Y ( f )) = L X d f, Y
= L X d f, Y + d f, L X Y
= dL X f, Y + (L X Y )( f )
= d X ( f ), Y + (L X Y )( f )
= Y (X ( f )) + (L X Y )( f ),
whereupon we conclude that
L X Y = X Y − Y X = [X, Y ].
6. By 5 and the Jacobi identity,
[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
(x P , y P ) = (0, 1) + t (u, −1) = (tu, 1 − t).
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.
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)).
On the overlap (Ui × V j ) ∩ (Uk × V ),

(ϕi × ϕ j ) ◦ (ϕk × ϕ )−1 (x, y) = (ϕi ◦ ϕk−1 (x), ϕ j ◦ ϕ−1 (y)).
This map is a diffeomorphism, as each component of the map is a diffeomor-
phism. To show orientability, it suffices to show that the sign of the Jacobian
of the composite map f × g is the product of the signs of the Jacobians of the
component maps f and g. For this, we have, symbolically,
∂ f ∂ f
( )( )
∂f ∂g
∂x ∂y
∂g ∂g = ,
∂x ∂y

∂x ∂y
whence the assertion follows.
3.40 a. In spherical polar coordinates
r = (x 2 + y 2 + z 2 ),
θ = cos−1 (z/r ),
φ = tan−1 (y/x).
For future use we observe that
∂r 1 x
= (x 2 + y 2 + z 2 )−1/2 · 2x = ,
∂x 2 r
and similarly for y and z. Also,
∂ z z x xz
=− 2 · =− 3,
∂x r r r r
∂ z z y yz
=− 2 · =− 3,
∂y r r r r
∂ z r − z(z/r ) r 2 − z2 x 2 + y2
= = = .
∂z r r2 r3 r3
Therefore

1 ∂(z/r ) ∂(z/r ) ∂(z/r )
dθ = − * dx + dy + dz
1 − (z/r )2 ∂x ∂y ∂z

1 xz yz x 2 + y2
= −* − 3 d x − 3 dy + dz .
1 − (z/r )2 r r r3
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 .
Putting this all together gives

∗ 1 1
f σ =− [(−x 2 z − y 2 z)(d x ∧ dy)
x + y r3
2 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
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
of the entries of each individual matrix, which is clearly a smooth map.

Finally, the inverse of a matrix is the ratio of two polynomials in the
matrix entries, namely the adjugate and the determinant, which is also
smooth whenever the determinant does not vanish.
b. An element of Mn+ (R) is determined by n(n + 1)/2 numbers: the n diag-
onal elements and the n(n − 1)/2 elements above the diagonal. Thus, it is
naturally a submanifold of Mn (R) diffeomorphic to Rn(n+1)/2 . The map
ϕ : Mn (R) → Mn+ (R) is smooth (because each entry of the image is a
quadratic polynomial in the matrix entries), and O(n) = ϕ −1 (I ), where
I is the identity matrix. To show I is a regular value we must examine
the derivative map ϕ∗ , which maps the tangent space of Mn (R), which we
identify with Mn (R) (because it is a vector space), to the tangent space to
Mn+ (R), which is identified with Mn+ (R). Here is one way to do it. (For
another, see [33], pp. 22-23.) We view ϕ as a map on Rn 2 by considering
its effect on individual matrix entries. Thus,
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 .
To check that I is a regular value of ϕ we must check that, for every

A ∈ ϕ −1 (I ) = O(n) and for any symmetric matrix C there exists a
matrix B such that ϕ∗A B = C. Every symmetric matrix is of the form
1
C = (C + C T ),
2
so it suffices to solve the equation B A T = 12 C. But A is orthogonal, so
multiplying both sides on the right by A gives B = 12 C A. Therefore, by
Theorem 3.2, O(n) is a smooth manifold. As it inherits the smooth group
operations from G L(n, R), it is a Lie group. Again by Theorem 3.2, the
dimension of O(n) is n 2 − n(n + 1)/2 = n(n − 1)/2.
c. The determinant is a smooth map from O(n) to R, and S O(n) is the

inverse image of +1. From the result of Exercise 1.36, we have
∂
(det A) = (det A)(A−1 ) ji (1)
∂ Ai j
whenever det A = 0, which is the case for A ∈ O(n). As this never
vanishes, the derivative map is surjective everywhere, so by Theorem 3.2
S O(n) is a smooth submanifold of O(n). The other component of O(n)
is not a Lie group for the simple reason that it is not a group, because
det AB = det A det B, so if A and B both have determinant −1 then AB
has determinant +1.
d. Start with a real matrix

a b
A= .
c d
We have

a b a c a 2 + b2 ac + bd
AA =
T
= ,
c d b d ac + bd c2 + d 2
so the condition A A T = I gives
1 = a 2 + b2 = c2 + d 2 ,
0 = ac + bd.
From det A = 1 we get
ad − bc = 1.
The second, third and fourth equations yield

c d a 0 a −c −d 0 d
= ⇒ =− = ,
d −c b 1 b −d c 1 −c
so a general element of S O(2) can be written

a b
,
−b a
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
is obviously smooth, and it restricts to a bijection from S 1 to S O(2).

Hence, by the fact supplied in the hint, ϕ restricts to a diffeomorphism
from S 1 to S O(2).
3.44 a. L g is a diffeomorphism, so by (3.87),
L g∗ [X, Y ] = [L g∗ X, L g∗ Y ] = [X, Y ].
b. If X is left invariant, then for all g ∈ G, X g = L g∗ X e . Conversely, given

Y we define a vector field X by X g = L g∗ Y . Then L g∗ X h = L g∗ L h∗ Y =
L gh ∗ Y = X gh , so X is left invariant.
c. Given a one parameter subgroup γ , its derivative γ (0) is an element
of Te G = g. Conversely, given a vector X e ∈ Te G, consider the left
invariant vector field X it determines. Define γ to be the unique integral
curve of X through e satisfying γ (0) = X e , so that γ (t) = X γ (t) =
L γ (t)∗ X e for |t| < . Fix s with |s| < and |t + s| < and define two
new curves: η1 (t) := γ (s + t) and η2 (t) := γ (s)γ (t). We have
η1 (t) = γ (s + t) = X s+t = X η1 (t) .
Now view η2 (t) as a composition: (L γ (s) ◦γ )(t). Then as the pushforward

behaves nicely with respect to composition (chain rule), we have
d
η2 (t) = η2 (t) = η2∗ (d/dt) = L γ (s)∗ γ∗ (d/dt) = L γ (s)∗ γ (t)
dt
= L γ (s)∗ X γ (t) = X γ (s)γ (t) = X η2 (t) .
In particular, η1 and η2 are both integral curves of X and they satisfy the
same initial condition
η1 (0) = γ (s) = η2 (0),
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,
(L ∗A x)(B) = x(L A B) = x(AB) = AB,
so
L ∗A x = Ax.
On the other hand,
(L ∗A x −1 )(B) = x −1 (L A B) = (AB)−1 = B −1 A−1 ,

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.
3.46 S O(n) is the set of all n by n real matrices satisfying A A T = I , and by

the previous exercise we may identify so(n) with a subset of Mn (R). Con-
sider a curve A : (a, b) → S O(n) given by t → A(t), with A(0) = e.
Differentiating I = A A T with respect to t at t = 0 gives

dA T d A T d A d A T d A d A T
0= A +A = + = + ,
dt dt 0 dt 0 dt 0 dt 0 dt 0
because A(0) = e. Elements of so(n) are tangent vectors to the curve at
the identity and therefore may be identified with the derivatives d A/dt at
t = 0. (We are using the fact that Mn (R) is a vector space, so it coincides
with its tangent space.) Therefore, so(n) consists of skew symmetric matrices.
Moreover, if A and B are skew symmetric then
[A, B]T = (AB − B A)T = (AB)T − (B A)T = B T A T − A T B T

= (−B)(−A) − (−A)(−B) = [B, A] = −[A, B],
so so(n) is indeed a subalgebra of Mn (R).

3.47 Mn (R) is a vector space, so the differential of a map just equals the map
itself. Hence L A∗ = L A . The differential equation defining the integral curve
therefore becomes
dγ (t)
= γ (t)A,
dt
where A = γ (0). To find the solution to the differential equation, rewrite it
as an integral equation,
t
γ (t) = I + γ (t)A dt.
0
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))† ,
which shows that elements of su(n) must be anti-Hermitian. (The overdot

means time derivative.) In the real case the condition that Ȧ(0) be traceless
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),
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
As et B e−(t /2)[A,B] |t=0 = 1, the proof is complete.

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,
(ad Z (X ), Y ) = ([Z , X ], Y ) = tr(ad[Z , X ] ◦ ad Y )

= tr([ad Z , ad X ] ◦ ad Y )
= tr(ad Z ◦ ad X ◦ ad Y ) − tr(ad X ◦ ad Z ◦ ad Y )
= tr(ad Y ◦ ad Z ◦ ad X ) − tr(ad Z ◦ ad Y ◦ ad X )
= −(ad Z (Y ), X ).
3.53 a. We have
⎛ ⎞⎛ ⎞ ⎛ ⎞
0 1 0 0 0 −1 0 0 0
E 1 E 2 = ⎝−1 0 0⎠ ⎝0 0 0 ⎠ = ⎝0 0 1⎠
0 0 0 1 0 0 0 0 0
and
⎛ ⎞⎛ ⎞ ⎛ ⎞
0 0 −1 0 1 0 0 0 0
E 2 E 1 = ⎝0 0 0 ⎠ ⎝−1 0 0⎠ = ⎝0 0 0⎠ ,
1 0 0 0 0 0 0 1 0
so
⎛ ⎞
0 0 0
E 1 E 2 − E 2 E 1 = ⎝0 0 1⎠ = E 3 .
0 −1 0
Similarly we find
[E 2 , E 3 ] = E 1 and [E 3 , E 1 ] = E 2 ,
from which the result follows.
b. The jk th element of the matrix representing ad E i in the given basis is
−i jk , because (using the summation convention),
(ad E i )(E j ) = (ad E i )k j E k = i jk E k .
Therefore,
(E i , E j ) = tr(ad E i ◦ ad E j ) = (−imn )(− jnm ) = −2δi j .
The metric is negative definite, so it is nondegenerate and its signature is
(−, −, −).
3.54 Elements of su(2) consist of 2×2 complex traceless skew-Hermitian matrices
with determinant one. But A is anti-Hermitian if and only if i A is Hermitian,
so we may as well find all 2×2 traceless Hermitian matrices with determinant
one. By Hermiticity

a b ā c̄
= ,
c d b̄ d̄
which gives
a = ā, d = d̄, and b = c̄.
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.
This proves (1).

b. We want to show that, for ω a p-form,
L X i Y ω − i Y L X ω = i L X Y ω. (2)
Let p = 1. By property (iii) of the Lie derivative,
L X Y, ω − Y, L X ω = L X Y, ω .
But this is just
L X i Y ω − i Y L X ω = i L X Y ω,
so (2) holds for p = 1. Now assume (2) holds for p −1, and let ω = η ∧λ
with deg η = r . Then, using the induction hypothesis we get
L X i Y (η ∧ λ) − i Y L X (η ∧ λ) = L X [i Y η ∧ λ + (−1)r η ∧ i Y λ]
− i Y [L X η ∧ λ + η ∧ L X λ]
Lλ
= L X iY η ∧ λ + η∧
iY X
(( ((
+( ((L(X η ∧ i Y λ + (−1) η ∧ L X i Y λ
(−1)r r
(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
= (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
d. By the inductive hypothesis and the antisymmetry of the contraction

mapping,
(di X 0 ω)(X 1 , . . . , X p )

p
= (−1)i−1 L X i ((i X 0 ω)(X 1 , . . . ,
X i , . . . , X p ))
i=1

+ (−1)i+ j (i X 0 ω)(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 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
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 .
Upon changing dummy indices and stripping off

∂i1 ⊗ · · · ⊗ ∂ir ⊗ d x j1 ⊗ · · · ⊗ d x js
from each term we arrive at (3.124).
3.57 a. We have
ω = dq 1 ∧ dp1 + · · · + dq n ∧ dpn ,
so
ωn = (dq 1 ∧ dp1 + · · · + dq n ∧ dpn )n
= (dq 1 ∧ dp1 ) ∧ · · · ∧ (dq n ∧ dpn )
= (−1)n(n−1)/2 dq 1 ∧ · · · ∧ dq n ∧ dp1 ∧ · · · ∧ dpn ,
where the sign is obtained by permuting all the dq i ’s to the left in succes-
n−1
sion, starting with dq 2 . This gives i=1 i = n(n − 1)/2 sign flips. Thus,
ω is manifestly nonvanishing. As ω is expressible in the same way in
n n
every coordinate patch, it extends globally to all of M.

b. As the pullback map is a ring homomorphism, f ∗ ν n = μn . In particular,
the Jacobian determinant of f is unity, which means that the differential
f ∗ is an isomorphism of tangent spaces. Hence, by the inverse function
theorem, f is a local diffeomorphism.
c. Let Y = a i (∂/∂q i ) + bi (∂/∂ pi ) be an arbitrary vector field in local
coordinates. Then by definition,
∂f ∂f
ω(X f , Y ) = i Y i X f ω = i Y d f = d f (Y ) = Y f = a i + bi .
∂q i ∂ pi
But also
ω(X f , Y ) = (dq i ∧ dpi )(X f , Y )
= dq i (X f )dpi (Y ) − dq i (Y )dpi (X f )
= Y ( pi )X f (q i ) − Y (q i )X f ( pi )
= bi X f (q i ) − a i X f ( pi ).
As Y was arbitrary,
∂f ∂f
X f ( pi ) = − and X f (q i ) = ,
∂q i ∂ pi
whence the conclusion follows.
d. From Part (c),
∂H ∂H
X H (q i ) = and X H ( pi ) = − .
∂ pi ∂q i
But the tangent vector to the integral curve γ can be written

∂ ∂ i ∂ ∂
γ̇ = q̇ i + ṗi = X H = X H (q ) + X H ( pi ) ,
∂q i ∂ pi ∂q i ∂ pi
whence Hamilton’s equations follow.
e. By definition,

d H d
= γ ∗ H = γ∗ (d/dt)H = X H H
dt γ (t) dt
∂H ∂H ∂H ∂H
= − i = 0.
∂ pi ∂q i ∂q ∂ pi
In old fashioned notation,
dH ∂ H dq i ∂ H dpi
= + ,
dt ∂q dt
i ∂ pi dt
which gives the same thing.
f. Following the hint, we have
L X H ω = i X H dω + di X H ω = 0,
because ω is closed, and i X H ω = d H . From the definition of the Lie
derivative, it follows that
d ∗
ϕ ω = 0,
dt t
which shows that ϕt∗ ω is a constant, independent of t. But ϕ0 = id,
whence we obtain ϕt∗ ω = ω.
g. Using Part (c) we get
∂g ∂ f ∂g ∂ f
ω(X f , X g ) = d f (X g ) = X g ( f ) = − i .
∂ pi ∂q i ∂q ∂ pi
h. From (3.122) we have
dω(X f , X g , X h )
= L X f (ω(X g , X h )) − L X g (ω(X f , X h )) + L X h (ω(X f , X g ))
− ω([X f , X g ], X h ) − ω([X g , X h ], X f ) + ω([X f , X h ], X g ). (1)
The symplectic form ω is closed, so the left side vanishes. We also have
{ f, g} = ω(X f , X g ) = −ω(X g , X f ) = −dg(X f ) = −X f (g) = −L X f g,
so the second line in (1) is
− { f, {g, h}} + {g, { f, h}} − {h, { f, g}}. (2)
Cartan’s second formula says that

[L X , i Y ]ω = i [X,Y ] ω.
Thus, using Cartan’s first formula, the fact that ω is closed, and the
definition of X f we get
ω([X f , X g ], X h ) = i X h i [X f ,X g ] ω
= (i X h L X f i X g − i X h i X g L X f )ω
= (i X h (i X f d + di X f )i X g − i X h i X g (i X f d + di X f )ω
= i X h i X f d(dg) + i X h d{g, f } − i X h i X g d(d f )
= d{g, f }(X h )
= −{h, {g, f }}.
It follows that the third line in (1) is
{h, {g, f }} + { f, {h, g}} − {g, {h, f }}. (3)
Adding (2) and (3) gives
0 = −2({ f, {g, h}} + {g, {h, f }} + {h, { f, g}}).
4
Homotopy and de Rham cohomology
4.1 This is just a matter of unpacking the definitions. If X is contractible then

there exist maps f : X → { p} and g : { p} → X satisfying f ◦ g = id p and
g ◦ f ∼ id X . This, in turn, means that there is a homotopy F : I × X → X
such that F(0, x) = g( f (x)) = x0 and F(1, x) = id X (x) = x for some
x0 ∈ X and all x ∈ X . But this just says that id X is null homotopic, because
the existence of F shows that id X is homotopic to the one point map g ◦ f :
X → x0 .
Conversely, if id X is null homotopic, then there exists a continuous map
F : I × X → X such that F(0, x) = x0 and F(1, x) = id X (x). Pick a point
p ∈ X and define maps f : X → { p} and g : { p} → X by g( p) = x0 . (Note
that both maps are automatically continuous, because the inverse image of
every closed set is closed.) Then f ◦ g = id p , F(0, x) = x0 = g( f (x)), and
F(1, x) = id X , so X is contractible.
4.2 We show that idC X is homotopic to the one point map f : C X → { p} sending
every point to the apex of the cone. Thus, consider the continuous map F :
I ×C X → C X given by F(t, (s, x)) = ((1−s)t +s, x). Then F(0, (s, x)) =
(s, x) = idC S (s, x) and F(1, (s, x)) = (1, x) = p = f (s, x).
4.3 a. We have
F ∗ ω = A(t x, t y, t z) d(t y) ∧ d(t z) + B(t x, t y, t z) d(t z) ∧ d(t x)
+ C(t x, t y, t z) d(t x) ∧ d(t y)
= A(t x, t y, t z)(y dt + t dy) ∧ (z dt + t dz)
+ B(t x, t y, t z)(z dt + t dz) ∧ (x dt + t d x)
+ C(t x, t y, t z)(x dt + t d x) ∧ (y dt + t dy)
= A(t x, t y, t z)(t y dt ∧ dz − t z dt ∧ dy)
+ B(t x, t y, t z)(t z dt ∧ d x − t x dt ∧ dz)
+ C(t x, t y, t z)(t x dt ∧ dy − t y dt ∧ d x)
+ terms without dt.
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
b. For ease of writing, define

1
e(x, y, z) := A(t x, t y, t z) t dt (1)
0
1
f (x, y, z) := B(t x, t y, t z) t dt (2)
0
1
g(x, y, z) := C(t x, t y, t z) t dt, (3)
0
so that
α := h F ∗ ω = e(y dz − z dy) + f (z d x − x dz) + g(x dy − y d x).
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.
(ker αi ⊆ im δi−1 .) If αi [a] = [ϕi a] = 0 then ϕi a = di−1 b for some

b ∈ B i−1 . Define c := ψi−1 b. Then δi−1 [c] = [a].
2. (im αi ⊆ ker βi .) By exactness,
βi αi [a] = βi [ϕi a] = [ψi ϕi a] = 0.
(ker βi ⊆ im αi .) If βi [b] = [ψi b] = 0 then ψi b = di−1 c for some

c ∈ C i−1 . By exactness, there is a b ∈ B i−1 such that ψi−1 b = c. But
then ψi b = di−1 c = di−1 ψi−1 b = ψi di−1 b , so ψi (b − di−1 b ) = 0. By
exactness, there exists an a ∈ Ai such that ϕi a = b − di−1 b . But then
αi [a] = [ϕi a] = [b].
3. (im βi ⊆ ker δi .)
δi βi [b] = δi [ψi b] = [a],
where ϕi+1 a = di b for any b with ψi b = ψi b. So choose b = b. But

di b = 0, so ϕi+1 a = 0 and therefore by exactness a = 0.
(ker δi ⊆ im βi .) If δi [c] = [a] = 0 then a = di u for some u ∈ Ai .

So di b = ϕi+1 a = ϕi+1 di u = di ϕi u, which gives di (b − ϕi u) = 0. As
b − ϕi u is closed it determines a cohomology class in B i , and by exactness,
βi [b − ϕi u] = [ψi b] = [c].
4.7 By stereographic projection we can cover S n by two patches. U covers every-
thing but the south pole while V covers everything but the north pole, and
each is homeomorphic to Rn . The question is what happens on the overlap.
In the case of the circle, the two patches overlap in two disconnected pieces,
each of which is homotopic to a point—in other words, the intersection is
homotopic to a zero-dimensional sphere. In the case of the two sphere, the
two patches overlap in a fat band around the equator, which is homotopic to a
one dimensional sphere. We claim that the same thing happens in any dimen-
sion: U ∩ V is homotopic to an n − 1 dimensional sphere. There are many
ways to see this. Here is one.
Define S n = {x ∈ Rn+1 : (x 1 )2 + · · · + (x n )2 = 1}. Let the north pole be
(0, . . . , 1) and the south pole (0, . . . , −1). Using homotopy, shrink U and V
to sets that just overlap along the equator x n = 0. Then U and V overlap in
an open set consisting of all points on the n-sphere satisfying x n ∈ (−, ).
This set is homotopic to the equator itself, which is clearly an n − 1-sphere.
We proceed by induction on n. The base case n = 1 is the previous exam-
ple. Assume it is true for n −1 for n > 1. We know that HdR 0
(U ) = HdR
0
(V ) =
R and HdR (U ) = HdR (V ) = 0 for k = 1, and by hypothesis HdR (S ) = R
k k 0 k
for k < n. Therefore, the bottom two rows of the Mayer-Vietoris sequence
look like this:
As in the case of the circle, rk ψ = dim ker ψ = 1, so by exactness we

conclude that HdR0
(S n ) = R (which we already knew) and HdR
1
(S n ) = 0.
Now, by hypothesis HdR (S ) = 0 and HdR
n−2 n−1
(S ) = R, so the top three
n−1 n−1
rows look like this:
By exactness we therefore conclude HdR (S ) = 0 and HdR

n−1 n n
(S n ) = R.
Finally, using the induction hypothesis again on the middle range of the
Mayer-Vietoris diagram we get HdR k
(S n ) = 0 if k = 0 and k = n.
4.8 a. f ∼ f via the homotopy F(t, x) = f (x) for all t.
b. f ∼ g implies there exists a continuous map F : I × X → Y such that
F(0, x) = f (x) and F(1, x) = g(x). Define G(t, x) = F(1 − t, x). This
is clearly continuous, and provides the required homotopy from g to f .
c. f ∼ g and g ∼ h implies there exist continuous maps F : I × X → Y and
G : I × X → Y such that F(0, x) = f (x), F(1, x) = g(x), G(0, x) =
g(x), and G(1, x) = h(x). Define

F(2t, x), if 0 ≤ t ≤ 1/2, and
H (t, x) =
G(2t − 1, x) if 1/2 ≤ t ≤ 1.
Then H (t, x) is a homotopy from f to h. It is continuous for all x and for

all t ∈ [0, 1] because F(1, x) = G(0, x).
4.9 If f ∼ g there exists a continuous F : I × X → Y such that F(0, x) =
f (x) and F(1, x) = g(x). Define a continuous map G : I × X → Z by
G(t, x) = h ◦ F(t, x). Then G(0, x) = h( f (x)) and G(1, x) = h(g(x)), so
G is a homotopy between h ◦ f and h ◦ g.
If g ∼ h there exists a continuous F : I × Y → Z such that F(0, y) =

g(y) and F(1, y) = h(y). Define a continuous map G : I × X → Z by
G(t, x) = F(t, f (x)). Then G(0, x) = g( f (x)) and G(1, x) = h( f (x)), so
G is a homotopy between g ◦ f and h ◦ f .
4.10 Recall that X ∼ Y if there exist continuous maps f : X → Y and g : Y → X
such that g ◦ f ∼ id X and f ◦ g ∼ idY .
a. Set f = g = id X to show that X ∼ X .
b. If X ∼ Y then clearly Y ∼ X .
c. If X ∼ Y and Y ∼ Z then, in addition to f and g above, there exist
continuous maps h : Y → Z and k : Z → Y such that k ◦ h ∼ idY and
h ◦ k ∼ id Z . Define u : X → Z and v : Z → X by u = h ◦ f and
v = g ◦ k. Then by the results of Exercise 4.9, v ◦ u = g ◦ k ◦ h ◦ f ∼ id X
and u ◦ v = h ◦ f ◦ g ◦ k ∼ id Z . Thus X ∼ Z .
4.11 a. This follows from the results of Exercise 4.8, because path-homotopy
is a more restrictive kind of homotopy, namely one that preserves the
endpoints.
b. Suppose αβ is defined, so that α(1) = β(0). Let α ∈ [α] and β ∈ [β].
Then α (1) = β (0) as well, so α β makes sense. There exist homotopies
F : I × I → X and G : I × I → X such that F(0, t) = α(t), F(1, t) =
α (t), F(s, 0) = α(0) = α (0), F(s, 1) = α(1) = α (1), G(0, t) = β(t),
G(1, t) = β (t), G(s, 0) = β(0) = β (0), and G(s, 1) = β(1) = β (1).
Define a continuous map H : I × I → X by

F(s, 2t), if 0 ≤ t ≤ 1/2,
H (s, t) =
G(s, 2t − 1), if 1/2 ≤ t ≤ 1.
Then H (0, t) = αβ(t), H (1, t) = α β (t), H (s, 0) = αβ(0) = α β (0),

and H (s, 1) = αβ(1) = α β (1), so α β is path-homotopic to αβ via the
homotopy H .
c. Let α(0) = p and α(1) = q, and suppose α ∈ [α]. Then there exists
a path-homotopy F : I × I → X such that F(0, t) = α(t), F(1, t) =
α (t), F(s, 0) = p, and F(s, 1) = q. Define G(s, t) = F(s, 1 − t). Then
G(0, t) = α(1 − t) = α −1 (t), G(1, t) = α (1 − t) = α −1 (t), G(s, 0) = q,
and G(s, 1) = p. So α −1 is path-homotopic to α −1 , which shows that
[α]−1 is independent of class representative.
d. i). If α and β are two loops based at p then αβ is a loop based at p. So
[α][β] = [αβ] exists and is a path-homotopy equivalence class of loops
based at p.
ii). Let α, β, and γ be loops based at p. To show multiplication is associative
we must show that [(αβ)γ ] = [α(βγ )]. Now
⎧
⎨α(4t),
⎪ if 0 ≤ t ≤ 1/4,
((αβ)γ )(t) = β(4t − 1), if 1/4 ≤ t ≤ 1/2,
⎪
⎩
γ (2t − 1), if 1/2 ≤ t ≤ 1,
and
⎧
⎨α(2t),
⎪ if 0 ≤ t ≤ 1/2,
(α(βγ ))(t) = β(4t − 2), if 1/2 ≤ t ≤ 3/4,
⎪
⎩
γ (4t − 3), if 3/4 ≤ t ≤ 1
Define
⎧
⎪
⎨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
Then F(0, t) = ((αβ)γ )(t), F(1, t) = (α(βγ ))(t), and F(s, 0) =

F(s, 1) = p, so F is the homotopy we desire.
iii). [id p ] is the identity element of π1 (X, p), for [α][id p ] = [α · id p ] = [α]
and [id p ][α] = [id p · α] = [α].
iv). If α is a loop based at p then so is α −1 . We must show that [α][α −1 ] =
[id p ]. For that, define a continuous family γs (t) of loops based at p
indexed by s ∈ [0, 1] according to

α(2st), if 0 ≤ t ≤ 1/2,
γs (t) :=
α(2s(1 − t)), if 1/2 ≤ t ≤ 1.
Define a map F : I ×I → X by F(s, t) = γs (t). Then F(0, t) = α(0) =
p = id p (t), F(1, t) = (αα −1 )(t), and F(s, 0) = F(s, 1) = α(0) = p.
Hence F is a homotopy from αα −1 to id p .
e. Let γ be a path from p to q. If β is a loop at q then α := γ −1 βγ is a loop
at p. So we get a map γ : π1 (X, q) → π1 (X, p) given by [β] → [α]. The
claim is that γ is a group isomorphism. (It is well defined, because if β is
path homotopic to β (fixing q), then γ −1 βγ is path homotopic to γ −1 β γ
(by a homotopy that fixes p and q).)
i).
γ is a homomorphism.
γ ([β1 β2 ]) = [γ −1 β1 β2 γ ] = [γ −1 β1 γ γ −1 β2 γ ]
γ ([β1 ][β2 ]) =

= [γ −1 β1 γ ][γ −1 β2 γ ] = (
γ [β1 ])(
γ [β2 ]).
ii).
γ is bijective. If γ [β2 ] then there is a homotopy from γ −1 β1 γ
γ [β1 ] =
−1
to γ β2 γ that fixes p. By conjugating with γ (and, if necessary, chang-
ing the parameterization of the curves), we get a homotopy from β1
to β2 that fixes q, so [β1 ] = [β2 ]. Also, by shrinking the tail γ −1 γ ,

γ [γ αγ −1 ] = [α], so
γ is surjective.
f. See, e.g., [55], p. 81 or [77], p. 61.
4.12 Obviously f = idY ◦ f . But Y is contractible, so idY ∼ idq for some q ∈ Y .
So, by the results of Exercise 4.9, f ∼ idq ◦ f = idq .
If f : X → S n is not surjective then we may view it as a map to S n minus
a point. But using stereographic projection we see that S n minus a point is
homotopic to Rn , which, in turn, is contractible.
4.13 a. S 0 is a two point space, and B 1 is an interval of the real line. If f is
0-connected and f (−1) = p and f (1) = q, then f can be extended to a
map from the interval [−1, 1] into X whose endpoints are p and q. In other
words, X is path connected. Conversely, if X is path connected, then there
is a path f : [−1, 1] → X between any two points, say p = f (−1) and
q = f (1), which means that any map f : {−1, 1} → X can be extended
to a map from B 1 ∼ = [−1, 1] into X , so X is 0-connected.
b. Let f : S i → X be given, where S i = {x ∈ Rn+1 : x = 1} is the
standard sphere. Suppose f extends to a map fˆ on the (i + 1)-ball B i+1 =
{x ∈ Rn+1 : x ≤ 1}. Consider the map F : I × B i+1 → X given by
F(t, x) = fˆ(t x).
Then for any x ∈ S i , F(0, x) = fˆ(0) and F(1, x) = fˆ(x) = f (x), so f
is null homotopic.
Conversely, if f is null homotopic there exists a continuous map F : I ×
S i → X with F(0, x) = q for some q ∈ X and F(1, x) = f (x). Define
fˆ : B i+1 → X by

F(x, x/x), if x = 0, and
fˆ(x) =
q, if x = 0.
Clearly, fˆ restricted to the sphere is F(1, x) = f (x), so it remains to show

that fˆ is continuous at x = 0. This is actually a bit subtle. Although it is
true that x/x does not approach a well-defined limit as x → 0, it does
not matter. Set x = tv for some unit vector v. Then in the limit that t → 0,
x/x = v, so in that case
lim F(x, x/x) = lim F(t, v) = q.
x→0 t→0
As this holds independent of v, fˆ is continuous.

c. If X ∼ Y there exist f : X → Y and g : Y → X , both continuous, such
that g◦ f ∼ id X and f ◦g ∼ idY . By virtue of the result of Part (b), we need
only show that if every continuous map from S i to X is null homotopic
(1 ≤ i ≤ k) then every continuous map from S i to Y is null homotopic. So

let h : S i → Y be given, and define u := g ◦ h : S i → X . By hypothesis,
u is null homotopic. But by Exercise 4.9, f ◦ u = f ◦ g ◦ h ∼ idY ◦ h = h.
As u is null homotopic, so is f ◦ u, therefore h is null homotopic.
d. If S n were n-connected, then every continuous map f : S n → S n would
be null homotopic. In particular, the identity map would be null homotopic,
which would say that the sphere is contractible. But it’s not.
4.14 Let U and V be two overlapping cylinders covering the torus. They intersect
in two disjoint cylinders. A cylinder is homotopic to a circle, so the Mayer-
Vietoris sequence looks like this:
The torus is connected, so HdR 0

(T 2 ) = R. As in the case of the circle, the
difference map ψ0∗ has rank one, so dim ker δ0 = rk ψ0∗ = 1. By count-
ing dimensions we get rk δ0 = 1, so dim ker ϕ1∗ = 1. Assume for the
moment that rk ψ1∗ = 1. Then dim ker ψ1∗ = 1, so rk ϕ1∗ = 1. Hence,
rk ϕ1∗ + dim ker ϕ1∗ = 2, so HdR1
(T 2 ) = R ⊕ R. Also, dim ker δ1 = rk ψ1∗ = 1
so rk δ1 = 1. By exactness, δ1 is surjective, so HdR2
(T 2 ) = R.
∗
It remains to show that rk ψ1 = 1. Write U ∩ V = X ∪ Y , where X
and Y are disjoint cylinders. X is a deformation retract of U (and of V ), and
similarly for Y , so the inclusion maps iU∗ : HdR•
(U ) → HdR •
(X ∪ Y ) and i V∗ :
• •
HdR (V ) → HdR (X ∪ Y ) are isomorphisms in cohomology. Thus ψ1∗ (ω, ν) =
∗ ∗
(i V − iU )(ω, ν) = (ν − ω, ν − ω) is a cohomology class on X ∪ Y , which
shows that rk ψ1∗ = 1.
Natural generators of cohomology in the various degrees are as follows.
(1) Degree 0: the unit function; (2) Degree 1: the one forms d x and dy from
Exercise 3.23; (3) Degree 2: the two form d x ∧ dy. (1) is immediate. As
for (2), d x and dy are obviously closed, but neither is exact, because nei-
ther x nor y is single valued on the torus. Lastly, (3) holds because if we
had d x ∧ dy = dθ for some θ then we would have to have θ = x dy or
θ = −y d x or some linear combination, none of which are single-valued.
(A slightly more rigorous argument based on integration can be found in [84],
Proposition 28.2.)
4.15 We have T = U ∪ V and U ∩ V ∼

= S 1 , so the Mayer-Vietoris sequence
becomes
T is connected, so H 0 (T ) = R. dim ker δ0 = rk ψ0∗ = 1, so ϕ1∗ is injective,

which means that H 1 (T ) is at least two dimensional. But ψ1∗ is the zero
map, so H 1 (T ) = R ⊕ R and δ1 is injective. By counting dimensions δ1
is also surjective, so dim ker ϕ2∗ = 1, which means that rk ϕ2∗ = 0. Hence
H 2 (T ) = 0.
4.16 We denote the genus 2 surface by T2 . (The subscript denotes the genus, not
the dimension; the latter is always two in this exercise.) Let U = V = T
be the punctured torus with the hole stretched to an open disk, so that U ∩ V
is the circle. Then using the previous exercise the Mayer-Vietoris sequence
looks like this:
T2 is connected, so H 0 (T2 ) = R. As before, dim ker δ0 = rk ψ0∗ = 1 and

ψ1∗ is the zero map, so by exactness we conclude that HdR1
(T2 ) = R4 and
HdR (T2 ) = R.
2
The obvious guess is that

⎧
⎪
⎪ R, k = 0,
⎪
⎪
⎨R2g , k = 1,
HdRk
(Tg ) =
⎪
⎪ R, k = 2, and
⎪
⎪
⎩
0, k ≥ 2.
This turns out to be true, and follows by a simple inductive argument along
the lines given above.
4.17 A suspension is just a double cone over the base. If we remove one of the
endpoints we are left with a cone whose base is attached to a cylinder over
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
0 −−−→ H k (M) −−−→ H k+1 ( M) −−−→ 0,
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)
(That’s because M is connected even if M is not.) Also, dim ker δ0 =

rk ψ0∗ = 1, so
H 0 (M) = H 1 ( M) ⊕ R. (3)
It follows that
H k (S r ) = H k+1 ( S r ) = H k+1 (S r +1 ) (k ≥ 1). (4)
Now we proceed in steps. 1) H 0 (S 0 ) = R ⊕ R because S 0 is just two points.

2) H 0 (S n ) = R for any n ≥ 1 by (2). 3) H 1 (S 1 ) = R by (1) and (3). 4)
H n (S n ) = R for all n > 1 by (3) and (4). 5) H 1 (Sr ) = 0 for r ≥ 2 by (2) and
(3). 6) H k (S r ) = 0 if r − k ≥ 1 and k > 1 by (5) and (4).
4.18 By construction U and V are both open in M ∨ N (because they are the
complements of closed sets). Moreover, U is homotopic to M (just shrink
the ball), V is homotopic to N (ditto), and U ∩ V is contractible to p = q.

The Mayer-Vietoris sequence therefore looks like this:
Suppose M has r connected pieces and N has s. Then obviously M ∨ N has

r + s − 1 pieces. So H 0 (M ∨ N ) ⊕ R = H 0 (M) ⊕ H 0 (N ). It follows that
dim ker ψ0∗ = rk ϕ0∗ = r + s − 1, so rk ψ0∗ = 1. By exactness, δ0 is the zero
map. It follows immediately that H k (M ∨ N ) ∼
= H k (M) ⊕ H k (N ) for k ≥ 1.
5
Elementary homology theory
5.1 Let K be the set of convex combinations of the points of S. Let p, q ∈ K .

Then p = i si pi and q = i ti pi with si , ti > 0 and i si = i ti = 1.

Any point r on the line segment pq is r = t p + (1 − t)q = i ri pi where

0 ≤ t ≤ 1 and ri := tsi + (1 − t)ti . As ri ≥ 0 and i ri = 1, r ∈ K , so K is
convex.

If k = 2 and p ∈ K then p = t1 p1 + t2 p2 with ti ≥ 0 and i ti = 1, which
shows that p is in every convex set that contains p1 and p2 —i.e., p ∈ P. If
|S| = n, then p ∈ K means

p= ti pi = (1 − tk )(s1 p1 + · · · sk−1 pk−1 ) + tk pk ,
i
k k−1
where ti ≥ 0, i=1 ti = 1, and si = ti /(1 − tk ). But i=1 si = 1, so by
induction p is on a segment joining two points of p, which means p ∈ P.
Hence K ⊆ P.
But K contains the points p1 , . . . , pk and is convex, so K ⊇ P, which
shows that P = K .
5.2 By Exercise 5.1, every point in [S] can be written as a convex combination of
its vertices. So we need only prove uniqueness. For that, suppose (s0 , . . . , sd )
and (t0 , . . . , td ) represent the same point of [S]. Define ci := ti − si . Then

d
d
0= ci pi = (c0 + c1 + . . . cd ) p0 + ci ( pi − p0 ).
i=0 i=1

But the first term vanishes because i ci = i ti − i si = 1 − 1 = 0. By

linear independence ci = 0 for all i > 0, and when combined with i ci = 0
we get c0 = 0 as well.
5.3 The real homology groups of a space are of the form Hk = Rβk where βk is
the k th Betti number. As Z k = ker ∂k , Bk = im ∂k+1 , and Hk = Z k /Bk , we
have
βk = dim Hk = dim ker ∂k − dim im ∂k+1 .
77
78 Elementary homology theory
Thus, to find the homology groups of the tetrahedron K , we just need to

compute the ranks and nullities of the boundary maps ∂.
Let K = (P0 , P1 , P2 , P3 ). Every 0-chain on K is of the form
c0 = a0 P0 + a1 P1 + a2 P2 + a3 P3 .
The boundary of a point is zero, so ∂0 c0 = 0. That is, ∂0 is the zero map. In
particular, dim ker ∂0 = dim C0 = 4.
The most general 1-chain in K is
c1 = b01 (P0 , P1 ) + b02 (P0 , P2 ) + b03 (P0 , P3 )
+ b12 (P1 , P2 ) + b13 (P1 , P3 ) + b23 (P2 , P3 ). (1)
Its boundary is
∂1 c1 = b01 (P1 − P0 ) + b02 (P2 − P0 ) + b03 (P3 − P0 )
+ b12 (P2 − P1 ) + b13 (P3 − P1 ) + b23 (P3 − P2 ).
Thus, ∂1 c1 = c0 provided
⎛ ⎞
b
⎛ ⎞ ⎜ 01 ⎟ ⎛ ⎞
−1 −1 −1 0 0 0 ⎜b02 ⎟ a0
⎜ ⎜ ⎟ ⎜ ⎟
⎜ 1 0 0 −1 −1 0⎟ ⎜ ⎟
⎟ ⎜ 03 ⎟ = ⎜a1 ⎟ .
b
⎝ (2)
0 1 0 1 0 −1 ⎠ ⎜b12 ⎟ ⎝a2 ⎠
⎜ ⎟
0 0 1 0 1 1 ⎝b13 ⎠ a3
b23
The matrix in (2) represents the boundary map ∂1 . The maximum rank of the
matrix is 4, because it has 4 rows, but it turns out that its rank is only 3. (This
follows because the rows are not all linearly independent: their sum is zero.)
Therefore
β0 = dim ker ∂0 − dim im ∂1 = 4 − 3 = 1.
For use below, observe that dim ker ∂1 = 3, which follows from the
rank/nullity theorem.
Next we find β1 . The most general 2-cycle is of the form
c2 = d0 (P1 , P2 , P3 ) + d1 (P0 , P2 , P3 ) + d2 (P0 , P1 , P3 ) + d3 (P0 , P1 , P2 ).
Its boundary is
∂2 c2 = d0 [(P1 , P2 ) − (P1 , P3 ) + (P2 , P3 )]
+ d1 [(P0 , P2 ) − (P0 , P3 ) + (P2 , P3 )]
+ d2 [(P0 , P1 ) − (P0 , P3 ) + (P1 , P3 )]
+ d3 [(P0 , P1 ) − (P0 , P2 ) + (P1 , P2 )].
Elementary homology theory 79
We have ∂2 c2 = c1 for c1 as in (1) if

⎛ ⎞ ⎛
⎞
0 0 1 1 b01
⎜ 0 −1 ⎟ ⎛d ⎞ ⎜ b ⎟
⎜ 1 0 ⎟ 0 ⎜ 02 ⎟
⎜ ⎟⎜ ⎟ ⎜ ⎟
⎜ 0 −1 −1 0 ⎟ ⎜d1 ⎟ ⎜b03 ⎟
⎜ ⎟⎝ ⎠ = ⎜ ⎟. (3)
⎜ 1 0 0 1 ⎟ d2 ⎜b12 ⎟
⎜ ⎟ ⎜ ⎟
⎝ −1 0 1 0 ⎠ d3 ⎝b13 ⎠
1 1 0 0 b23
The maximum rank of ∂2 is 4, because it has 4 columns, but the actual rank is
3, because columns one and three add to the same vector as columns two and
four. It follows that
β1 = dim ker ∂1 − dim im ∂2 = 3 − 3 = 0.
Note that dim ker ∂2 = 1.
Lastly, consider ∂3 , which maps from three chains to two chains. There are
no three chains in K (which comprises only the surface of a tetrahedron), so
the image of ∂3 is only the zero vector, which has dimension 0. Hence
β2 = dim ker ∂2 − dim im ∂3 = 1 − 0 = 1.
The analogous answer for the n-sphere is that it has trivial homology in all
degrees except 0 and n:

R, p = 0 or n,
H p (S ) =
n
0, otherwise.
5.4 As x and A both have positive real entries, so does Ax. Moreover, by con-

struction i f i = 1, so f maps the simplex σ n−1 to itself. As σ n−1 is
homeomorphic to the (n − 1)-ball, Brouwer’s theorem guarantees the exis-
tence of a fixed point, namely a vector satisfying f (x) = x. But then
Ax = |Ax|x and the conclusion follows.
5.5 Let P denote the crosspolytope, and let F = P∩H be a proper face. If F were
to contain both ei and −ei it would contain the origin 0 (as the intersection
of convex sets is convex), so we would have 0 ∈ H as well. But F is proper,
so there are vertices of P not in H , say e j , j = i. We cannot have −e j ∈ H
either, otherwise the segment (−e j , 0) would lie in H , and by linear extension
so would the segment (0, e j ). Hence there are points of P on both sides of
H , a contradiction.
Every face of a polytope is the convex hull of some subcollection of the
vertices of P, so we examine all such subcollections. Let S = {v1 , . . . , vn } be
any maximal subset of the vertex set of P that does not contain both ei and
−ei for any i. S is linearly independent, so {v2 − v1 , . . . , vn − v1 } are linearly
independent. In other words, S is affinely independent, so it determines a

simplex F in the affine subspace H spanned by S. We claim that F is a facet
of P. To show this, assume that S = {e1 , . . . , en } (the general case is similar).
Suppose H were not a supporting hyperplane of P. Then there would exist
a point p ∈ P on the opposite side of H from the origin. Let q be the point
of intersection of the ray (0 p) with H , so that p = tq for some t > 1. As

q = i>0 si ei with i si = 1 and si ≥ 0, we would have p = i>0 tsi ei

with tsi ≥ 0. But p ∈ P, so we must have i tsi = 1, a contradiction. A
similar argument shows that every maximal subcollection of the vertex set of
P not containing both ei and −ei is a facet of P. All the other faces are then
determined by subsets of these subcollections.
5.6 a. By simple counting we obtain the following table.
Platonic Solid V E F
tetrahedron 4 6 4
cube 8 12 6
octahedron 6 12 8
dodecahedron 20 30 12
icosahedron 12 30 20
The face numbers of the cube and octahedron are the reverse of one
another, as are the face numbers of the dodecahedron and the icosahede-
dron. This is because those figures are related by duality: putting a vertex in
the middle of each face and taking the convex hull gives you the dual figure,
reversing the face numbers. The tetrahedron is in its own class because it
is self-dual.
b. We use the classic method of double counting. Let |(e, f )| be the number
of pairs with the edge e contained in the 2-face f . There are F 2-faces, and
each 2-face contains p edges, so |(e, f )| = p F. But there are E edges, and
each edge is contained in two 2-faces, so |(e, f )| = 2E. Hence 2E = p F.
Similarly, by counting pairs (v, e) of vertices v contained in edges e in two
ways, we get q V = 2E. It follows that p F = q V = 2E. Euler’s equation
thus gives

2E 2E 1 1 1 −1
2=V −E+F = −E+ ⇒ E= + − .
q p p q 2
As E > 0,
1 1 1
+ > .
p q 2
Now start plugging in numbers. We must have p ≥ 3, because the triangle
is the smallest regular polygon. Similarly, q ≥ 3, otherwise the figure is
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
χ (U ∪ V ) = χ ((U ∪ V )\V ) + χ(V )

= χ (U \(U ∩ V )) + χ(V )
= χ (U ) − χ(U ∩ V ) + χ(V ).
b. The only question is what happens to the Euler characteristic of a surface

if you remove an open disk D o . Using the additivity of χ , we have
χ (open disk) = χ (closed disk) − χ(circle) = 1 − 0 = 1.
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
χ (T #k ) = χ (T #(k−1) \D o ) + χ(T 2 \D o ) − χ(S 1 )

= (2 − 2(k − 1) − 1) + (0 − 1) − 0
= 2 − 2k.
In the solutions to Exercise 4.16 we asserted that the cohomology of the

g-holed torus was R in dimensions 0 and 2 and R2g in dimension 1. By
definition, the Euler characteristic is the alternating sum of Betti numbers,
so we ought to have χ = 1 − 2g + 1 = 2 − 2g, which agrees with our
answer above, as it must.
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 ).
As we showed in Exercise 5.10, the map h i : Hi → Hi is defined as follows.

Let αi be a cycle in Ci . It defines a cohomology class [αi ] in Hi , and we set
h i ([αi ]) = [ψi (αi )]. In other words, the map ψi , restricted to Hi , is precisely
the map h i . Hence tr(ψi , Hi ) = tr(h i , Hi ). Thus

n
n
(−1) tr(ψi , Ci ) =
i
(−1)i [tr(ψi , Z i ) + tr(ψi−1 , Bi−1 )]
i=0 i=0

n
= (−1)i [tr(ψi , Bi ) + tr(ψi , Hi ) + tr(ψi−1 , Bi−1 )]
i=0

n
= (−1)i [tr(ψi , Bi ) + tr(h i , Hi ) + tr(ψi−1 , Bi−1 )].
i=0
But the sum over the boundary spaces telescopes to zero:

n
(−1)i [tr(ψi , Bi ) + tr(ψi−1 , Bi−1 )]
i=0

n
n−1
= (−1)i tr(ψi , Bi ) + (−1)i+1 tr(ψi , Bi )
i=0 i=−1
n n
= (−1)i tr(ψi , Bi ) + (−1)i+1 tr(ψi , Bi )
i=0 i=0
= 0,
because B−1 = Bn = 0. Hence the claim is proved.
6
Integration on manifolds
6.1 We have
γ ∗ ω = (t 2 )(t 3 ) dt + (t 3 )(t)(2t) dt + (t)(t 2 )(3t 2 ) dt = 6t 5 dt,
so
1
∗
1
6
ω= γ ω=6 t dt = t = 1.
5
γ I 0 0
6.2 We have
ϕ ∗ (dy ∧ dz) = ϕ ∗ (dy) ∧ ϕ ∗ (dz)

= d(ϕ ∗ y) ∧ d(ϕ ∗ z)
= d(y ◦ ϕ) ∧ d(z ◦ ϕ)
= dϕ 2 ∧ dϕ 3
2 3
∂ϕ ∂ϕ 2 ∂ϕ ∂ϕ 3
= du + dv ∧ du + dv
∂u ∂v ∂u ∂v
2 3
∂ϕ ∂ϕ ∂ϕ 2 ∂ϕ 3
= − du ∧ dv
∂u ∂v ∂v ∂u

∂ϕ ∂ϕ 1
= × du ∧ dv
∂u ∂v
= n 1 du ∧ dv
and
ϕ ∗ β = (Bx ◦ ϕ)ϕ ∗ (dy ∧ dz) + cyclic.
Thus
ϕ ∗ β = B(ϕ(u, v)) · n(ϕ(u, v)) du ∧ dv,
84
Integration on manifolds 85
and

∗
β= ϕ β= B(ϕ(u, v)) · n(ϕ(u, v)) du ∧ dv = B · d S.
c c
6.3 By the change of variables theorem,

dω = ϕ ∗ dω.
U
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
On the other hand,

ω= ϕ ∗ ω.
∂ ∂U
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
as expected. Hence Stokes’ theorem is confirmed in this case.

6.4 Both results are direct consequences of Stokes’ theorem. If ω is closed and c
is any chain, then

ω = dω = 0.
∂c c
86 Integration on manifolds
If ω = dμ is exact and the chain c has no boundary, then

ω = dμ = μ = 0.
c c ∂c
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
It follows that d 2 = 0 implies ∂ 2 = 0 and vice versa.

6.6 Let {V j } be another coordinate cover of M, and let {λi } be a partition of unity
subordinate to {Vi }. Note that

ρi λ j ω = ρi λ j ω
Ui Vj
because ρi λ j ω has support in Ui ∩ V j . Therefore

ρi ω = ρi λ j ω = ρi λ j ω = λ j ω.
i Ui i, j Ui ij Vj j Vj

because i ρi = j λ j = 1.
6.7 Stokes’ theorem gives

fη = d( f η) = df ∧η + f ∧ dη.
∂M M M M
6.8 For any -chain c,

ω (c) = dτ (c) = dτ = τ= τ (∂c) = ∂ ∗ τ (c).
c ∂c
6.9 Let f ∈ Z (M) be a cocycle. Then f (∂c) = (∂ ∗ f )(c) = 0, so f vanishes

on all boundaries. By de Rham’s 0 second theorem, there exists
0 a closed form
ω such that, for every cycle z, z ω = f (z). This shows that ω and f agree
on cycles, and0hence define the same cocycle. In other words, every cocycle
0
is of the form ω for some closed ω, which means that the map [ω] → [ ω]
is surjective. 0 0

Next, suppose ω is zero0 in H 0 (M). Then ω ∈ B (M), which means
there exists a τ such that ω = ∂ ∗ τ . Hence, if z is a cycle,

∗
ω= ω (z) = ∂ τ (z) = τ (∂z) = 0.
z
By de Rham’s
0 first theorem, ω is exact, so [ω] = 0, whence we see that
[ω] → [ ω] is injective.
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
where U0 is the region of (u 1 , u 2 )-space corresponding to the southern hemi-

sphere and V0 is the region of (v1 , v2 )-space corresponding to the northern
hemisphere. In both cases, this region is the interior of the unit circle in the
plane. The first integral is

du 2 ∧ du 1
ωU =
U0 (1 + η)
2
U0
2π 1
r dr dθ
= dθ dr
0 0 (1 + r 2 )2
1
1/2 π
= −2π · = .
1+r 0 2 2
A similar computation gives

π
ωV = ,
V0 2
and adding them together gives π , as before.

6.11 As in the previous exercise, it is easiest just to integrate d x ∧ dy over U1 and
ignore the missing parts, which are sets of measure zero. But then the integral
is about as simple as it gets:
1 1
d x ∧ dy = dx dy = 1.
U1 0 0
88 Integration on manifolds
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
On the other hand, the disk can be parameterized by

√
σ (u, v) = (u cos v, u sin v, 3a/2), 0 ≤ u ≤ a/2, 0 ≤ v ≤ 2π.
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
Again, Stokes’ theorem is verified.

6.14 Let D be the closed disk in M containing p whose boundary is C, and let D 0
be its interior. Then M −{ p} is homotopic to M − D 0 , and as homotopic maps
induce the same map in cohomology, it suffices to show that ι∗ : HdR 1
(M −
D ) → H (C) is the zero map, where now ι is the inclusion ι : C → M − D 0 .
0 1
But M − D 0 is a compact oriented surface with boundary C, so by Stokes’

theorem, for any ω ∈ HdR 1
(M − D 0 ),

∗ ∗
ι ω= ι ω= dω = 0,
C ∂(M−D 0 ) M−D 0
because ω is closed. By de Rham’s theorem, integration gives an isomorphism

between HdR 1
(C) and R, so ι∗ ω must be zero.
6.15 Let (Ui , ϕi ) be an atlas for a manifold M with boundary. We must show that,
given p ∈ ∂ M there is an open neighborhood U ⊂ ∂ M containing p that is
homeomorphic to Rn−1 . Let p ∈ Ui . By definition, X := ϕi (Ui ∩ ∂ M) ⊆
∂Hn ∼= Rn−1 . Pick q ∈ X . If ϕi−1 (q) ∈ ∂ M, then ϕi−1 (q) has a neighborhood
in M homeomorphic to Rn . But ϕi is a homeomorphism, so q has a neigh-
borhood in X homeomorphic to Rn , a contradiction. We may therefore take
U = ϕi−1 (X ) = Ui ∩ ∂ M, which is clearly open in ∂ M.
7
Vector bundles
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
7.3 Let X be a vector field on G. The map L g∗ : Te G → Tg G is an isomorphism

of vector spaces, so the bundle isomorphism we seek is afforded by the map
X g → (g, L g−1 ∗ X g ).
7.4 By definition of the curvature operator,
R( f X, gY )h Z = ∇ f X ∇gY h Z − ∇gY ∇ f X h Z − ∇[ f X,gY ] h Z . (1)
Expanding each term on the right side of (1) in succession gives the following
set of equations.
∇ f X ∇gY h Z = f ∇ X (g∇Y (h Z ))
= f ∇ X (gY (h)Z + gh∇Y Z )
= f X (gY (h))Z + f gY (h)∇ X Z
+ f X (gh)∇Y Z + f gh∇ X ∇Y Z . (2)
∇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)
∇[ f X,gY ] h Z = ∇ f X (g)Y + f g X Y −gY ( f )X − f gY X h Z

= f X (g)∇Y h Z − gY ( f )∇ X h Z + f g∇[X,Y ] h Z
= f X (g)Y (h)Z + f X (g)h∇Y Z
− gY ( f )X (h)Z − gY ( f )h∇ X Z
+ f g([X, Y ]h)Z + f gh∇[X,Y ] Z . (4)
Combining (1), (2), (3), and (4) gives
R( f X, gY )h Z = f X (gY (h))Z + f gY (h)∇ X Z
! "# $ ! "# $
a b
+ f X (gh)∇Y Z + f gh∇ X ∇Y Z
! "# $ ! "# $
c d
− gY ( f X (h))Z − f g X (h)∇Y Z
! "# $ ! "# $
a c
− gY ( f h)∇ X Z − f gh∇Y ∇ X Z
! "# $ ! "# $
b d
− f X (g)Y (h)Z − f X (g)h∇Y Z
! "# $ ! "# $
a c
92 Vector bundles
+ 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
(ϕ, ψ)U (q) = ( p, (v, w)),
where π(q) = p, ϕU (q) = ( p, v), and ψU (q) = ( p, w). We have
[(ϕ, ψ)V ◦ (ϕ, ψ)U−1 ]( p, (x, y)) = ( p, (x , y )),
where x = g E,U V x and y = g F,U V y. In other words, the transition functions

for the direct sum are g E⊕F,U V = g E,U V ⊕ g F,U V , where (A ⊕ B)(x, y) :=
(Ax, By).
7.8 To show that Dt is a connection, we need only show that it obeys the two
axioms (because it clearly maps sections of E to one form valued sections
of E, since each piece of it does). The two axioms are (i) linearity (which
is obvious, because a convex combination is a special kind of linear com-
bination), and (ii) the Leibniz rule. To show (ii), let s be a section and f a
function. Then, by linearity,
Vector bundles 93
f)
Dt (s f ) = (1 − t)D(s f ) + t D(s
· f +s ⊗df}
= (1 − t){Ds · f + s ⊗ d f } + t{ Ds
= Dt s · f + s ⊗ d f.
(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,
tr k = tr(A−1 k A) = tr(k A A−1 ) = tr k .
b. We have
d tr k = tr dk = tr d( ∧ · · · ∧ ) (1)

= tr(d ∧ ∧ · · · ∧ + ∧ d ∧ · · · ∧
+ · · · + ∧ · · · ∧ d) (2)
= k tr(d ∧ k−1
) (3)
= k tr({ ∧ ω − ω ∧ } ∧ k−1
) (4)
= k tr(ω ∧ − ω ∧ )
k k
(5)
= 0. (6)
Equation (1) follows because d is linear, so it commutes with the trace.

Equation (2) holds by virtue of the fact that is a matrix of two forms: d
is an antiderivation, so d(λ ∧ η) = dλ ∧ η + (−1)(deg λ)(deg η) λ ∧ dη. In our
case, the degree of each entry of is two, so the sign is always positive.
Equation (3) is a bit subtle. Consider, for example, the second term in (2).
We have (with the usual implicit summation)
tr( ∧ d ∧ · · · ∧ ) = i1 i2 ∧ di2 i3 ∧ i3 i4 ∧ · · · ∧ ik i1

= di2 i3 ∧ i3 i4 ∧ · · · ∧ ik i1 ∧ i1 i2
= tr(d ∧ k−1 ),
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
c. By an argument similar to that given in Part (b), we have

d dt
tr t = k tr
k
∧ t k−1
.
dt dt
But
t = dωt + ωt ∧ ωt
= d(ω + tη) + (ω + tη) ∧ (ω + tη)
= + t (dη + η ∧ ω + ω ∧ η) + t 2 η ∧ η,
so
dt
= dη + η ∧ ω + ω ∧ η + 2tη ∧ η,
dt
and
d
tr kt = k tr {dη + η ∧ ω + ω ∧ η + 2tη ∧ η} ∧ k−1

t .
dt
Under a change of frame field, the connection matrix transforms
according to
ω = A−1 ω A + A−1 d A
so
η =
ω − ω = A−1
ω A − A−1 ω A = A−1 η A,
showing that
α = tr(η ∧ k−1
t )
is a global form on M.
We have
dα = d tr(η ∧ k−1
t )
= tr(dη ∧ k−1
t − η ∧ dk−1
t ).
The second term in this expression is
tr(η ∧ dk−1
t ) = tr(η ∧ {dt ∧ t ∧ · · · ∧ t
+ t ∧ dt ∧ t ∧ · · · ∧ t
+ t ∧ · · · ∧ t ∧ dt }).
Using the Bianchi identity
dt = t ∧ ωt − ωt ∧ t
Vector bundles 95
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
Integrating both sides with respect to t gives

1 1 1
d
k − tr 2 =
tr tr kt dt = k dα dt = d k α dt ,
0 dt 0 0
thereby establishing the desired result.

7.10 a.
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.
b. Starting from A = Ai d x i we get
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.
8.6 We have g(eb̂ , eĉ ) = ±δb̂ĉ , so by metric compatibility,
0 = eâ g(eb̂ , eĉ ) = ∇eâ g(eb̂ , eĉ )

= g(∇eâ eb̂ , eĉ ) + g(eb̂ , ∇eâ eĉ )
= g( d̂ â b̂ ed̂ , eĉ ) + g(eb̂ , d̂ â ĉ ed̂ )
= d̂ â b̂ g(ed̂ , eĉ ) + d̂ â ĉ g(eb̂ , ed̂ )
= ωd̂ b̂ (eâ )gd̂ ĉ + ωd̂ ĉ (eâ )gb̂d̂
= (ωĉb̂ + ωb̂ĉ )(eâ ).
As eâ is a basis, the conclusion follows.

The curvature matrix components are
ˆ ˆ ˆ
n̂ = dω n̂ + ω â ∧ ωâ n̂ .
ˆ
The downstairs components are m̂ n̂ = gˆm̂ n̂ , so
m̂ n̂ = dωm̂ n̂ + ωm̂ â ∧ ωâ n̂

= dωm̂ n̂ + ωm̂ â ∧ ωb̂n̂ g â b̂ .
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.
8.10 Start with the following four equations, each of which follows from (8.49):
g(W, R(X, Y )Z ) + g(W, R(Y, Z )X ) + g(W, R(Z , X )Y ) = 0,

g(X, R(Y, Z )W ) + g(X, R(Z , W )Y ) + g(X, R(W, Y )Z ) = 0,
g(Y, R(Z , W )X ) + g(Y, R(W, X )Z ) + g(Y, R(X, Z )W ) = 0,
g(Z , R(W, X )Y ) + g(Z , R(X, Y )W ) + g(Z , R(Y, W )X ) = 0.
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
0 = 2g(W, R(X, Y )Z ) − 2g(X, R(W, Z )Y ),
as desired.
8.11 Begin with (8.49) and take inner products to get
g(W, R(X, Y )Z + R(Y, Z )X + R(Z , X )Y ) = 0.
According to equation (8.44)
g(∂i , R(∂k , ∂ )∂ j ) = Ri jk ,
so (8.49) is equivalent to the condition
Ri jk + Rikj + Rijk = 0. (1)
On the other hand, if we expand out the condition
Ri[ jk] = 0 (2)
we get
Ri jk + Rikj + Rijk − (Ri jk + Rik j + Rik j ) = 0,
or
Ri jk + Rikj + Rijk = Ri jk + Rik j + Rik j . (3)

But by antisymmetry in the last two indices, we must have

Ri jk + Rik j + Rik j = −(Ri jk + Rikj + Rijk ) (4)
as well. Equations (3) and (4) together imply (1). It is obvious that (1) implies
(2), so (1) and (2) are equivalent constraints.
8.12 Property 3 by itself implies that the last three indices
of
the Riemann
n
tensor
must be distinct. Naively, then, Property 3 imposes n n−1 3
= 4 4
constraints.
But in fact it imposes fewer than that, because by Properties 1 and 4, the first
index cannot equal any of the other three. For example, if we were to choose
‘1’ for the first index and ‘234’ for the last three indices, Property 3 would
give
R1123 + R1231 + R1312 = 0.
But by Property 1, the first term vanishes, and the last equals −R3112 , so the
equation would become
R1231 − R3112 = 0,
which is already true by Property 4. It follows that all four
n
indices must be
distinct, which explains why Property 3 imposes only 4 constraints. (All
these constraints are independent of the ones imposed by the other Properties,
because each one reduces to a condition on the cyclic permutation of the last
three indices, none of which follow from conditions on pairs of indices (or
pairs of pairs).)
A symmetric N by N matrix has N (N + 1)/2 independent components
(where N = (n 2 − n)/2), so altogether there are

1 2 n 1 1 2 1 2
(N + N ) − = (n − n) + (n − n)
2
2 4 2 4 2
n(n − 1)(n − 2)(n − 3)
−
4!
1 4
= (n − n 2 )
12
independent components.
8.13 By antisymmetry, the last two components of the Riemann tensor cannot be
equal, so the only independent components are of the form R i jθφ . We have
R θ θθφ = g θθ Rθθθφ + g θφ Rφθθφ = 0,
where the first term vanishes by virtue of the antisymmetry of the first two
components, and the second term vanishes because there are no off-diagonal
terms in the inverse metric. By similar reasoning we get
R θ φθφ = g θθ Rθφθφ + g θφ Rφφθφ = Rθφθφ ,

R φ θφθ = g φθ Rθθφθ + g φφ Rφθφθ = csc2 θ Rθφθφ ,
and
R φ φφθ = g φθ Rθφφθ + g φφ Rφφφθ = 0.

Therefore the only nonvanishing components of the Riemann tensor are
R θ φθφ and R φ θφθ , as advertised. Using Equation (8.46) and the Christoffel
symbols given in Exercise 8.5 we get
R φ θφθ = φ θθ,φ − φ φθ,θ + φ φm m θθ − φ θ m m φθ
= csc2 θ − cot2 θ = 1.
It follows that
R θ φθφ = Rθφθφ = sin2 θ R φ θφθ = sin2 θ.
8.14 By symmetry Property 4,
R ji = g k Rjki = g k Rkij = R ij = Ri j .
8.15
Rθθ = R k θ kθ = R θ θθθ + R φ θφθ = 1,
Rθφ = Rφθ = R k θ kφ = R θ θθφ + R φ θφφ = 0,
Rφφ = R k φkφ = R θ φθφ + R φ φφφ = sin2 θ.
8.16 Fix X = ∂i and Y = ∂ j . The linear map T , given by
Z → R(Z , ∂ j )∂i
is then represented by the matrix T , where
T ∂k = T m k ∂m = R(∂k , ∂ j )∂i = R m ik j ∂m ,
whence we conclude that the trace of T is
T m m = R m im j = Ri j .
8.17
R = g i j Ri j = g θθ Rθθ + g φφ Rφφ = 1 · 1 + csc2 θ · sin2 θ = 2.
8.18 If s = s(t) is another parameterization of γ , we have

dγ i ∂Y k dt dγ i ∂Y k
+ ijY
k j
= + ijY
k j
= 0.
ds ∂ x i ds dt ∂xi
8.19 If Y and Z are both parallel translated along X , then ∇ X Y = ∇ X Z = 0. So

the metric compatibility equation yields
∇ X g(Y, Z ) = g(∇ X Y, Z ) + g(Y, ∇ X Z ) = 0,
which says that g(Y, Z ) is constant along the curve. Substituting Y for Z or
Z for Y shows that the lengths of vectors are also preserved, so the angle
between Y and Z is preserved.
8.20 Let s = s(t) be a reparameterization. Then
dxk d x k dt
=
ds dt ds
and
2
d2xk d2xk dt d x k d 2t
= + .
ds 2 dt 2 ds dt ds 2
Thus,
2
d2xk k dx dx
i j
d2xk k dx dx
i j
dt d x k d 2t
+ i j = + i j + .
ds 2 ds ds dt 2 dt dt ds dt ds 2
In order for the curve to be a geodesic with respect to both the s and t parame-
ters simultaneously we must have d 2 t/ds 2 = 0, which means that t is a linear
function of s (and vice versa).
8.21 We basically solved this one already. If X = γ∗ (d/dt) is the tangent vector
along the geodesic, then from Exercise 8.19 and the fact that X is parallel
translated along itself, ∇ X g(X, X ) = 0, which is the equation above.
8.22 This is best illustrated by example, say n = 3. The domain of integration in
the path ordered integral is a cube of side t that can be broken up into six (3!)
chambers satisfying t1 > t2 > t3 , t1 > t3 > t2 , t2 > t1 > t3 , t2 > t3 > t1 ,
t3 > t1 > t2 , and t3 > t2 > t1 . (We can disregard the cases of equality, which
are sets of measure zero.) Thus we get
t t t
dt1 dt2 dt3 P(A(t1 )A(t2 )A(t3 ))
0 0 0
t t1 t2
= dt1 dt2 dt3 A(t1 )A(t2 )A(t3 )
0 0 0
t t1 t3
+ dt1 dt3 dt2 A(t1 )A(t3 )A(t2 )
0 0 0
t t2 t1
+ dt2 dt1 dt3 A(t2 )A(t1 )A(t3 )
0 0 0
t t2 t3
+ dt2 dt3 dt1 A(t2 )A(t3 )A(t1 )
0 0 0
t t3 t1
+ dt3 dt1 dt2 A(t3 )A(t1 )A(t2 )
0 0 0
t t3 t2
+ dt3 dt2 dt1 A(t3 )A(t2 )A(t1 )
0 0 0
t t1 t2
= 3! dt1 dt2 dt3 A(t1 )A(t2 )A(t3 ).
0 0 0
The general case is similar.
8.23 From (8.84) we have
ϑ(γ , ∇)ϑ( p , ∇) = ϑ( p , ∇)ϑ(γ , ∇) = ϑ(γ , ∇)
so ϑ( p , ∇) = id.
For the second claim, if we reverse the direction of the curve 0we reverse
t
the tangent
0t
vector, so A → −A. We need to show that Pe 0 A(t) dt =
(Pe− 0 A(t) dt )−1 . It suffices to do this for an infinitessimal parameter distance,
because every curve can be built from such pieces. So, let t = δ be a small
quantity. Starting from the equation
t
Y (t) = Y (0) − dt1 A(t1 )Y (t1 )
0
and using Taylor’s theorem and keeping only terms of first order in δ we get
Y (δ) ≈ (1 − δ A(0))Y (0).
Again by Taylor, (1 − δ A(0))−1 ≈ 1 + δ A(0), so
Y (0) ≈ (1 + δ A(0))Y (δ),
which shows that reversing the sign of A carries us backwards from Y (δ) to
Y (0).
8.24 Let be a loop based at q. Then γ γ −1 is a loop based at p, so
ϑ(γ −1 , ∇)ϑ(, ∇)ϑ(γ , ∇) ∈ H (∇; p),
and therefore,
ϑ(, ∇) ∈ ϑ(γ , ∇)H (∇; p)ϑ(γ −1 , ∇).
A similar argument yields the reverse inclusion.
8.25 Let y 1 , . . . , y n be local coordinates on a neighborhood V that meets U . On
the overlap we have
*
σ (x) = G(x) d x 1 ∧ · · · ∧ d x n
*
= J 2 G (y)J −1 dy 1 ∧ · · · ∧ dy n
*
= G (y) dy 1 ∧ · · · ∧ dy n ,
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
= η∧λ+ d((−1)k η ∧ ν + μ ∧ λ + μ ∧ dν)

M M
= ([η], [λ]),
where the second integral in the penultimate line vanishes by Stokes’
theorem, because M has no boundary.
8.30 By Poincaré duality, βn = β0 = 1, so M cannot be contractible.
8.31 In a coordinate basis, we have
T = T i j ∂i ⊗ d x j .
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

− k j1 T i1 ...ir j2 ... js − k j2 T i1 ...ir j1 ... js + · · · − k js T i1 ...ir j1 ... js−1 X k

− T ki2 ...ir j1 ... js i1 k − T i1 k...ir j1 ... js i2 k − · · · − T i1 ...ir−1 k j1 ... js ir k

+ T i1 ...ir k j2 ... js k j1 + T i1 ...ir j1 k... js k j2 + · · · + T i1 ...ir j1 ... js−1 k k js X

= (L X T )i1 ...ir j1 ... js .
In this computation we used the fact that the Christoffel symbols are sym-
metric in the two lower indices so all the terms in parentheses cancel. For
example, the term
i1 k T i2 ...ir j1 ... js X k
cancels with the term
T ki2 ...ir j1 ... js i1 k X
because changing dummy indices then switching indices in the Christoffel
symbol gives
T ki2 ...ir j1 ... js i1 k X = T i2 ...ir j1 ... js i1 k X k = T i2 ...ir j1 ... js i1 k X k .
8.33 From equation (8.103) we get
gi j;k = gi j,k − ki gj − k j gi .
But the right hand side vanishes by virtue of the metric compatibility equation
Xg(Y, Z ) = g(∇ X Y, Z ) + g(Y, ∇ X Z )
with X = ∂k , Y = ∂i and Z = ∂ j :
gi j,k = g( ki ∂ , ∂ j ) + g(∂i , k j ∂ ) = ki gj + k j gi .
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 .
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 .
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.
8.38 a. Watch the birdie:
ξi, jk = ξi,k j (mixed partials commute)

= −ξk,i j (Killing’s equation)
= −ξk, ji (mixed partials again)
= ξ j,ki (Killing’s equation)
= ξ j,ik (mixed partials again)
= −ξi, jk (Killing’s equation)
=0 (x = −x ⇒ x = 0).
Integrating ξi, jk = 0 gives
ξi, j = αi j
for some constant tensor which, by virtue of Killing’s equation must be

antisymmetric. Integrating again gives
ξi = αi j x j + ai
for some constants ai , as advertised.

b. With ai = 0 the integral curve equation reads ẋ i = ξ i = δ i j α jk x k . Setting
α12 = −α21 = 1 and all other components to zero we get
ẋ 1 = x 2
ẋ 2 = −x 1
ẋ 3 = 0.
The latter equation shows that x 3 = constant, so the curve is restricted

to the x y-plane. The other two equations can be solved by diagonaliz-
ing a matrix (or appealing to complex numbers), but it is simpler just to
differentiate the first and substitute the second to get
ẍ 1 = ẋ 2 = −x 1 x 1 = A cos(t + δ).
Plugging this into the second equation and integrating gives
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
x = r sin θ cos φ r = (x 2 + y 2 + z 2 )1/2

y = r sin θ sin φ θ = cos−1 (z/r )
z = r cos θ φ = tan−1 (y/x).
Applying the chain rule gives

∂ ∂r ∂ ∂θ ∂ ∂φ ∂
= + +
∂x ∂ x ∂r ∂ x ∂θ ∂ x ∂φ

∂ ∂r ∂ ∂θ ∂ ∂φ ∂
= + +
∂y ∂ y ∂r ∂ y ∂θ ∂ y ∂φ

∂ ∂r ∂ ∂θ ∂ ∂φ ∂
= + +
∂z ∂z ∂r ∂z ∂θ ∂z ∂φ
Next we have to compute all those partial derivatives, then convert back
to spherical polar coordinates. (We could also invert the Jacobian matrix
of the inverse transformation, but that’s just as irritating.) For example, we
have
∂r 1 x
= (x 2 + y 2 + z 2 )−1/2 (2x) = = sin θ cos φ,
∂x 2 r
∂θ 2 −1/2 xz cos θ cos φ
= −(1 − (z/r ) ) (−z/r )(x/r ) = 2 2
2
= ,
∂x r (x + y 2 )1/2 r
∂φ −y/x 2 y sin φ
= =− 2 =− .
∂x 1 + (y/x) 2 x +y 2 r sin θ
Continuing in this way gives
∂ ∂ cos θ cos φ ∂ sin φ ∂
= sin θ cos φ + − ,
∂x ∂r r ∂θ r sin θ ∂φ
∂ ∂ cos θ sin φ ∂ cos φ ∂
= sin θ sin φ + + ,
∂y ∂r r ∂θ r sin θ ∂φ
∂ ∂ sin θ ∂
= cos θ − .
∂z ∂r r ∂θ
From the previous parts we have
∂ ∂
ξ (1) = y −z ,
∂z ∂y
∂ ∂
ξ (2) =z −x ,
∂x ∂z
∂ ∂
ξ (3) =x −y ,
∂y ∂x
whereupon the desired result follows by substituting using the above

computations. For example,

(1) ∂ sin θ ∂
ξ = (r sin θ sin φ) cos θ −
∂r r ∂θ

∂ cos θ sin φ ∂ cos φ ∂
− (r cos θ) sin θ sin φ + +
∂r r ∂θ r sin θ ∂φ
∂ ∂
= − sin φ − cot θ cos φ .
∂θ ∂φ
The other two vector fields follow similarly.
8.39 a. From Equation (3.124) we get

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):
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)
Combining (2)-(7) with (8)-(12) we have the following constraints:
Fτρ,θ = 0 ((2) and (8)), (14)

Fτ θ,θ = Fτ φ = 0 ((3) and (9)), (15)
Fτ θ = 0 ((4) and (10) and (15)), (16)
Fρθ,θ = Fρφ = 0 ((5) and (11)), (17)
Fρθ = 0 ((6) and (12) and (17)), (18)
Fθφ = B(τ, ρ) sin θ ((7) and (13)), (19)
where B(τ, ρ) is some arbitrary function. By (1) and (14) Fτρ = A(τ, ρ)
for some arbitrary function A. Thus
F = A(τ, ρ) dτ ∧ dρ + B(τ, ρ) sin θ dθ ∧ dφ,
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.
As the determinant of a diagonal matrix is just the product of the diagonal

elements,
*
|G| = abr 2 sin θ.
Also, inverting the metric gives

g = −a −2 ∂τ2 + b−2 ∂ρ2 + r −2 ∂θ2 + sin−2 ∂φ2 ,
where ∂τ2 := ∂τ ⊗ ∂τ etc., so


g(dθ, dθ) g(dθ, dφ)

g(dθ ∧ dφ, dθ ∧ dφ) =
g(dφ, dθ) g(dφ, dφ)
−2
r 0
=
−2
−2
0 r sin θ
= r −4 sin−2 θ.
Hence
r 2 sin θ
(dτ ∧ dρ) = dθ ∧ dφ.
ab
Using 2 = (−1)k(n−k)+d on k forms in n dimensions with d = 1 for a
Lorentzian metric gives
ab
(dθ ∧ dφ) = − dτ ∧ dρ.
r 2 sin θ
c. The first Maxwell equation gives
∂B ∂B
0 = dF = sin θ dτ ∧ dθ ∧ dφ + sin θ dρ ∧ dθ ∧ dφ,
∂τ ∂ρ
so B must be constant. The second Maxwell equation gives
∂(Ar 2 /ab) ∂(Ar 2 /ab)
0 = dF = sin θ dτ ∧dθ ∧dφ+ sin θ dρ∧dθ ∧dφ,
∂τ ∂ρ
so Ar 2 /ab is a constant.
The metric is already diagonal, so by inspection we have
θ 0̂ = a dτ,
θ 1̂ = b dρ,
θ 2̂ = r dθ,
θ 3̂ = r sin θ dφ.
(For instance, g(θ 1̂ , θ 1̂ ) = a 2 g(dτ, dτ ) = a 2 (−a −2 ) = −1, etc..)

Therefore
1
F = Fî ĵ θ î ∧ θ ĵ
2
= abF0̂1̂ dτ ∧ dρ + r 2 sin θ F2̂3̂ dθ ∧ dφ + · · · .
We conclude that F0̂1̂ = A/ab and F2̂3̂ = B/r 2 , with all other components
vanishing. It follows that F0̂1̂ = constant/r 2 and F2̂3̂ = constant/r 2 . The
magnitudes and signs of the constants follow from the fact that the electric
and magnetic field components are related to the field strength tensor in a
local orthonormal frame as they are related in flat spacetime, namely by

ˆ
E ĵ = F 0̂ ĵ and F ĵ k̂ = ε ĵ k̂ Bˆ.
8.40 a. Simply observe that the metric tensor can be written
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
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
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 .
Integrating both sides with respect to t gives

ln φ̇ = −2 ln sin θ + c
and exponentiating yields
(sin2 θ)φ̇ = J,
where J = ec . Substituting this into (8.113) gives
cos θ
θ̈ = J 2 .
sin3 θ
Multiply both sides by θ̇ and integrate with respect to t to get
1 2 1
θ̇ = − J 2 sin−2 θ + c,
2 2
which gives (8.116).
Next we observe that (8.116) is separable:
(sin2 θ)θ̇ 2 = a 2 sin2 θ − J 2
*
⇒ (sin θ)θ̇ = a 2 sin2 θ − J 2

sin θ dθ
⇒ * = dt
a 2 sin2 θ − J 2

d(cos θ)
⇒ − * = at + c
1 − cos2 θ − (J/a)2
3 4
cos θ
⇒ − arcsin * = at + c
1 − (J/a)2
*
⇒ cos θ = − 1 − (J/a)2 sin(at + c),
which is (8.117).
For (8.115) we have
J
φ̇ =
sin2 θ
J
=
1 − cos2 θ
J
=
1 − (1 − (J/a)2 ) sin2 (at + c)
J
=
cos (at + c) + (J/a)2 sin2 (at + c)
2
J sec2 (at + c)
= ,
1 + (J/a)2 tan2 (at + c)
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
tan(φ − φ0 ) = (J/a) tan(at + c),
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
cos θ = C sin θ[sin φ cos φ0 − cos φ sin φ0 ].
Transforming to Cartesian coordinates gives
z = Ay − Bx,
where A := C cos φ0 and B := C sin φ0 . This is the equation of a plane

passing through the origin. The geodesics are the intersections of this plane
with the sphere, namely great circles.
8.43 a. Let X = a X + bY and Y = cX + dY . Then using the fact that R(X, Y )Z

is antisymmetric in X and Y we have
R(X , Y )Y = R(a X + bY, cX + dY )(cX + dY )
= acd R(X, Y )X + ad 2 R(X, Y )Y
+ bc2 R(Y, X )X + bcd R(Y, X )Y
= (ad − bc)(c R(X, Y )X + d R(X, Y )Y ).
Also, g(W, R(X, Y )Z ) is antisymmetric in W and Z , so
g(X , R(X , Y )Y ) = (ad −bc)(bcg(Y, R(X, Y )X )+adg(X, R(X, Y )Y ))
= (ad − bc)2 g(X, R(X, Y )Y ).
Lastly,
g(a X + bY, a X + bY )g(cX + dY, cX + dY ) − g(a X + bY, cX + dY )2
= (a 2 g(X, X ) + 2abg(X, Y ) + b2 g(Y, Y ))
× (c2 g(X, X ) + 2cdg(X, Y ) + d 2 g(Y, Y ))
− (acg(X, X ) + (ad + bc)g(X, Y ) + bdg(Y, Y ))2
= (ad − bc)2 (g(X, X )g(Y, Y ) − g(X, Y )2 ).
Thus,
g(X , R(X , Y )Y )
g(X , X )g(Y , Y ) − g(X , Y )2
(ad − bc)2 g(X, R(X, Y )Y )
=
(ad − bc)2 (g(X, X )g(Y, Y ) − g(X, Y )2 )
= K (").
b. By hypothesis
g(X, R(X, Y )Y ) = g(X, R (X, Y )Y ), (1)
so
g(X + W, R(X + W, Y )Y ) = g(X + W, R (X + W, Y )Y ).
Expanding both sides and using (1) gives
g(X, R(W, Y )Y ) + g(W, R(X, Y )Y )
= g(X, R (W, Y )Y ) + g(W, R (X, Y )Y ).
Now apply (8.50) to the first term on each side to get
2g(W, R(X, Y )Y ) = 2g(W, R (X, Y )Y ). (2)
In (2) send Y → Y + Z to get, after the dust has cleared,
g(W, R(X, Y )Z ) + g(W, R(X, Z )Y )

= g(W, R (X, Y )Z ) + g(W, R (X, Z )Y ).
Equivalently,
g(W, R(X, Y )Z ) − g(W, R (X, Y )Z )

= g(W, R(Z , X )Y ) − g(W, R (Z , X )Y ).
This says that
g(W, R(X, Y )Z ) − g(W, R (X, Y )Z )
is invariant under cyclic permutations of X , Y , and Z . So using (8.50) we

get
3(g(W, R(X, Y )Z ) − g(W, R (X, Y )Z )) = 0,
which yields the desired result.

c. Assume
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,
g(X, R (X, Y )Y ) = g(X, X )g(Y, Y ) − g(X, Y )2 ,
so (8.120) gives, for any X and Y ,
g(X, R(X, Y )Y ) = K (")g(X, R (X, Y )Y ) = K p g(X, R (X, Y )Y ).
But R and R both satisfy the symmetry properties (8.47)-(8.50), so by the

proof of Part (b) we must have g(W, R(X, Y )Z ) = K p g(W, R (X, Y )Z ).
Taking X = ∂k , Y = ∂ , Z = ∂ j and W = ∂i yields the result.
d. From Part (c) we get, by isotropy,
Râ b̂ĉd̂ = K (δâ ĉ δb̂d̂ − δâ d̂ δb̂ĉ )
in an orthonormal basis. Raising indices with the Kronecker delta gives
R â b̂ ĉd̂ = δ â m̂ δ b̂n̂ Rm̂ n̂ ĉd̂ = K (δĉâ δd̂b̂ − δd̂â δĉb̂ ).
Therefore from (8.45) we get

1
â b̂ = K (δĉâ δd̂b̂ − δd̂â δĉb̂ )θ ĉ ∧ θ d̂ = K θ â ∧ θ b̂ .
2
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 .
Applying the Bianchi identity gives
d K ∧ θ â ∧ θ b̂ = 0.
Now d K = αâ θ â for some αâ ’s, so by linear independence of θ â ∧ θ b̂ ∧ θ ĉ

(this is where we need the dimension to be at least three) we must have
αâ = 0, whence we conclude that d K = 0 and K = constant.
8.44 In two dimensions there is only one linearly independent component of the
Riemann tensor, which we can take to be R1212 in some coordinate basis. The
Ricci curvature scalar is therefore
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
so R = 2K (a basis independent result).

8.45 a. The metric components in the coordinate basis (x, y) are gi j = δi j /y 2 .
Consider the parameterized curve γ : I → H+ 2 from (0, 0) to (0, 1), say,
given by t → (0, t). Then
1 1 1
dxi dx j dt
ds = gi j (t) dt = = ln t = ∞.
γ 0 dt dt 0 t 0
b. In a coordinate basis (8.39) reads

1 1
ki j = (∂i g jk + ∂ j gik − ∂k gi j ) = (g jk,i + gik, j − gi j,k ).
2 2
Then, for example,
1
x x y = gx x,y = −1/y 3 .
2
Similarly we obtain
x x x = x yy = yx y = yyx = 0
and
x x y = x yx = − yx x = yyy = −1/y 3 .
Next we raise indices with the inverse metric (g i j = y 2 δ i j ) to get, for
example,
x x x = g x x x x x + g x y yx x = 0.
Continuing in this way we get
x x x = x yy = y x y = y yx = 0
and
x x y = x yx = − y x x = y yy = −1/y.
c. The geodesic equations read
d2xi j
i dx dx
k
+ jk = 0.
dt 2 dt dt
Substituting from Part (b) we get
2
ẍ −ẋ ẏ = 0,
y
1 1
ÿ + ẋ 2 − ẏ 2 = 0,
y y
and simplifying gives
y ẍ − 2ẋ ẏ = 0,
y ÿ + ẋ 2 − ẏ 2 = 0.
From the first equation we get
ẍ ẏ
=2 .
ẋ y
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.
Plugging back into ẋ = cy 2 gives

c
cy 2 =
1 − (ct/r + h)2
or
y = (1 − (ct/r + h)2 )−1/2 .
d. The Gaussian curvature is the sectional curvature of any two plane at a

point. At any point we can choose the basis ∂x and ∂ y . Thus
Rx yx y
K = .
gx x g yy − gx2y
We have
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)
which shows that Möbius transformations are closed

under composition.
1 0
Composition is associative. Also, if I = then TI (z) = z, the
0 1
identity map on C. This is the identity element of M because
T A ◦ TI (z) = TA (z) and TI ◦ TA (z) = T A (z).
Lastly, from (1) we get
TA ◦ TA−1 = TA−1 ◦ TA = TI ,
which implies that
TA−1 (z) = TA−1 (z).
(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,
and this holds for all z only if c = b = 0 and a = d = λ = 0.

This shows that the kernel of the homomorphism consists of all nonzero
complex multiples of the identity matrix.
c. The existence of the inverse map TA−1 = TA−1 shows that the map is bijec-
tive. Moreover, because the entries of A are real, multiplying the numerator
and denominator by the complex conjugate of the denominator gives

az + b az + b cz̄ + d ac|z|2 + adz + bcz̄ + bd
= = ,
cz + d cz + d cz̄ + d |cz + d|2
where as usual |z|2 = z z̄. Thus, as a map from R2 to itself TA is given by
ac(x 2 + y 2 ) + (ad + bc)x + bd
x →
(cx + d)2 + (cy)2
and
y
y → .
(cx + d)2 + (cy)2
(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
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,
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).
The metric tensor components are therefore
E = g(σu , σu ) = (b + a cos v)2 ,

F = g(σu , σv ) = 0,
G = g(σv , σv ) = a 2 .
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
ω1̂ 2̂ = sin v du.
(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

= cos v dv ∧ du
cos v
=− θ 1̂ ∧ θ 2̂ ,
a(b + a cos v)
whereupon we conclude that K = − cos v/a(b + a cos v).
8.50 a. By definition, Ad g = (Rg−1 ◦ L g )∗,e = Rg−1 ∗,g ◦ L g∗,e . Thus
(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).
b. Let X , Y , and Z be left invariant vector fields. The Koszul formula reads
2(∇ X Y, Z ) = X (Y, Z ) + Y (X, Z ) − Z (X, Y )

+ ([X, Y ], Z ) − ([X, Z ], Y ) − ([Y, Z ], X ).
By Exercise 3.52c the last two terms cancel, because
(ad Z (X ), Y ) + (ad Z (Y ), X ) = 0.
Also, by left invariance of X and Y ,
(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 ].
c. From Part (b) and the Jacobi identity,
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
Ric(X, Y ) = tr(Z → R(Z , Y )X ),
so using the result of Part (c) gives

1
Ric(X, Y ) = − tr(Z → [[Z , Y ], X ])
4
1
= − tr(Z → ad X ad Y ◦ Z )
4
1
= − tr(ad X ad Y )
4
1
= − (X, Y ).
4
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
From dθ â = −ωâ b̂ ∧ θ b and the properties of the connection matrix in an

orthonormal frame we guess and verify that
f 2̂
ω1̂ 2̂ = −ω2̂ 1̂ = − θ = − f dθ,
f
f
ω1̂ 3̂ = −ω3̂ 1̂ = − θ 3̂ = − f sin θ dφ,
f
1
ω2̂ 3̂ = −ω3̂ 2̂ = − cot θ θ 3̂ = − cos θ dφ.
f
b.
1̂ 2̂ = dω1̂ 2̂ + ω1̂ 3̂ ∧ ω3̂ 2̂

f 1̂
= − f dr ∧ dθ = − θ ∧ θ 2̂ ,
f
1̂ 3̂ = dω1̂ 3̂ + ω1̂ 2̂ ∧ ω2̂ 3̂
= − f sin θ dr ∧ dφ − f cos θ dθ ∧ dφ + f cos θ dθ ∧ dφ
f 1̂
=− θ ∧ θ 3̂ ,
f
2̂ 3̂ = dω2̂ 3̂ + ω2̂ 1̂ ∧ ω1̂ 3̂
= sin θ dθ ∧ dφ − ( f )2 sin θ dθ ∧ dφ
1 − f 2 2̂
= θ ∧ θ 3̂ .
f2
c. By Schur’s theorem we must have
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
f = ±1, so a = ±1 and f = ±r . Lastly, if K = 1 a similar argument

gives f = ± sin r .
8.52 a. Laplace expansion and the definition of the Levi-Civita alternating symbol
gives

εi1 ...in Ai1 1 · · · Ain n = (−1)σ Aσ (1) 1 · · · Aσ (n) n = det A.
σ ∈Sn
Now the claim is that
εi1 ...in Ai1 j1 · · · Ain jn
is totally antisymmetric under the interchange of the j’s. We see this as

follows:
εi1 ...ik ...i ...in Ai1 j1 · · · Aik jk · · · Ai j · · · Ain jn

= εi1 ...i ...ik ...in Ai1 j1 · · · Ai jk · · · Aik j · · · Ain jn (flip dummy indices)
= εi1 ...i ...ik ...in A j1 · · · A j · · · A jk · · · A jn
i1 ik i in
(commute components)
= −εi1 ...ik ...i ...in Ai1 j1 · · · Aik j · · · Ai jk · · · Ain jn (antisymmetry of ε).
Moreover, when ( j1 , . . . , jn ) = (1, . . . , n) it gives det A, so
εi1 ...in Ai1 j1 · · · Ain jn = ε j1 ... jn det A.
b. By the result of Part (a),

∂ x i1 ∂ x in −1
··· εi ...i = J εi1 ...in ,
∂ y i1 ∂ y in 1 n
so ξ = J .
c. Under a coordinate transformation,
∂xi ∂x j
gi j = gi j .
∂ yi ∂ y j
Taking the determinant of both sides gives G = J −2 G.
d. Under a coordinate transformation we get
*
ii ...in = |G |εi1 ...in
* ∂ x i1 ∂ x in
= (|J |−1 |G|)J i · · · i εi1 ...in
∂y 1 ∂y n
∂x i1
∂x in
= (sgn J ) i · · · i i1 ...in .
∂y 1 ∂y n
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 .
Therefore we must choose +1 in the Euclidean case and −1 in the

Lorentzian case.
h. The covariant derivative of the Levi-Civita tensor vanishes because it is
built from the metric, and the Levi-Civita connection is metric compatible.
For instance, we have
√ √
∇k a1 ...an = ∇k ( Gεa1 ...an ) = ∇k ( G)εa1 ...an = 0,
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
.
This gives, by Laplace expansion on the last column,

1
εi1 ...ik ik+1 ...in ε j1 ... jk ik+1 ...in
(n − k − 1)!
⎛ i ⎞
δ j11 ··· δ ij1k δiik+1
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
εi1 ...ik ik+1 ...in ε j1 ... jk ik+1 ...in = δ ij11...i

... jk .
k
(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
i1 ...ik ik+1 ...in j1 ... jk ik+1 ...in = ±(n − k)!δ ij11...i

... jk ,
k
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,
where σ is the canonical volume element. Let β := d x ik+1 ∧ · · · ∧ d x in . Then

1
α∧β = ai ...i d x i1 ∧ · · · ∧ d x in
k! 1 k
1
= ai1 ...ik εi1 ...in d x 1 ∧ · · · ∧ d x n
k!
1 1
= √ ai1 ...ik εi1 ...in σ
k! |G|
1
= i1 ...in ai1 ...ik σ.
k!
On the other hand,
1
g(α, β)σ = a∗ g(β, β)σ
(n − k)! ik+1 ...in
1
= a∗ g(β, β)σ
(n − k)! ik+1 ...in
1
= (a ∗ ) jk+1 ... jn g jk+1 ik+1 · · · g jn in g(β, β)σ.
(n − k)!
But
⎛ ⎞
g(d x ik+1 , d x ik+1 ) · · · g(d x ik+1 , d x in )
⎜ .. .. .. ⎟
g(β, β) = det ⎝ . . . ⎠
g(d x , d x ) · · · g(d x , d x )
in i k+1 in in
⎛ i i ⎞
g k+1 k+1 · · · g ik+1 in
⎜ .. ⎟ ,
= det ⎝ ... ..
. . ⎠
g in ik+1 ··· gin in
so
i ···i
g jk+1 ik+1 · · · g jn in g(β, β) = δ jk+1 n
k+1 ··· jn
.
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
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
div X = ∇i X i = ∂i X i + i ji X j = G −1/2 (G 1/2 X j ), j .
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 ).
For the first term, we have
L X (G 1/2 ) = X (G 1/2 ) = X j (G 1/2 ), j .
For the second, we note that

L X (d x j ) = d(X (x j )) = d X j = X j ,k d x k ,
k
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
L X σ = (X j (G 1/2 ), j + G 1/2 X j , j ) d x 1 ∧ · · · ∧ d x n = (div X )σ.
c. From the Ricci identity,

1
div ◦ curl X = ∇k ( i jk ∇i X j ) = i jk ∇[k ∇i] X j = − i jk R q jki X q = 0,
2
by virtue of the symmetry properties of the Riemann tensor. Also,
(curl ◦ grad f )k = i jk ∇i ∇ j f = i jk ∇i ∂ j f = i jk (∂i ∂ j f − i j ∂ f ) = 0,
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.
b. We have, for example,

1 1
g(eφ̂ , eφ̂ ) = g(∂φ , ∂φ ) = gφφ = 1.
r 2 sin θ
2
r 2 sin2 θ
In a similar fashion we can check that
g(eî , e ĵ ) = δî ĵ ,
so the basis is indeed orthonormal.
c. For future use we observe that
G 1/2 = r 2 sin θ.
Also,
∂ ∂ ∂
X = Xr + Xθ + Xφ = X r̂ er̂ + X θ̂ eθ̂ + X φ̂ eφ̂ ,
∂r ∂θ ∂φ
so
1 θ̂ 1
X r = X r̂ , Xθ = X , and Xφ = X φ̂ .
r r sin θ
Also,
X r = grr X r = X r̂ ,
X θ = gθθ X θ = r X θ̂ ,
X φ = gφφ X φ = r sin θ X φ̂ .
First, the divergence. We have
div X = G −1/2 (G 1/2 X j ), j .
Now
(G 1/2 X r ),r = (r 2 X r̂ ),r sin θ,
(G 1/2 X θ ),θ = r (sin θ X θ̂ ),θ ,
(G 1/2 X φ ),φ = X φ̂ ,φ ,
and therefore
1 ∂ 2 r̂ 1 ∂ θ̂
1 ∂ φ̂
div X = r X + sin θ X + X .
r ∂r
2 r sin θ ∂θ r sin θ ∂φ
Next, the curl. We have
curl X = i jk (∇i X j )∂k = i jk (∂i X j − m i j X m )∂k
= i jk (∂i X j )∂k = G −1/2 εi jk (∂i X j )∂k ,
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 .
Thus we have the following.

(curl X )r = G −1/2 (∂θ X φ − ∂φ X θ )
1 6 7
= 2 r sin θ X φ̂ ,θ − r X θ̂ ,φ
r sin θ
1 6 7
= sin θ X φ̂ ,θ − X θ̂ ,φ
r sin θ
= (curl X )r̂ .
Also,
(curl X )θ = G −1/2 (∂φ X r − ∂r X φ )
1 6 r̂ 7
= 2 X ,φ − sin θ r X φ̂ ,r
r sin θ
1
= (curl X )θ̂ ,
r
which implies
1 6 r̂ 7
(curl X )θ̂ = X ,φ − sin θ r X φ̂ ,r .
r sin θ
Lastly,
(curl X )φ = G −1/2 (∂r X θ − ∂θ X r )
1 6 θ̂ 7
= 2 r X ,r − X r̂ ,θ
r sin θ
1
= (curl X )φ̂ ,
r sin θ
so
1 6 θ̂ 7
(curl X )φ̂ = r X ,r − X r̂ ,θ .
r
Putting everything together, we obtain

8 9
1 ∂ φ̂ ∂ X θ̂
curl X = (sin θ X ) − er
r sin θ ∂θ ∂φ
8 9
1 1 ∂ X r̂ ∂ φ̂
+ − (r X ) eθ
r sin θ ∂φ ∂r
8 9
1 ∂ θ̂ ∂ X r̂
+ (r X ) − eφ .
r ∂r ∂θ
For the gradient, we have
grad f = g i j (∇i f )∂ j = g i j (∂i f )∂ j
1 1
= (∂r f )er̂ + 2 (∂θ f )r eθ̂ + (∂φ f )(r sin θ)eφ̂
r r sin2 θ
2
∂f 1 ∂f 1 ∂f
= er̂ + eθ̂ + e .
∂r r ∂θ r sin θ ∂φ φ̂
Finally, the Laplacian may be read off immediately by taking the diver-
gence of the gradient of f .
8.58 a. We start with
1
α = ai1 ...ik d x i1 ∧ · · · ∧ d x ik
k!
and apply the exterior differential to get
1
dα = ai ...i , j d x j ∧ d x i1 ∧ · · · ∧ d x ik
k! 1 k
1
= ai1 ...ik ; j + ji1 ai2 ...ik + · · · + jik ai1 ...ik−1

k!
× d x j ∧ d x i1 ∧ · · · ∧ d x ik
1
= ai ...i ; j d x j ∧ d x i1 ∧ · · · ∧ d x ik
k! 1 k
1
= ∇ j ai1 ...ik d x j ∧ d x i1 ∧ · · · ∧ d x ik ,
k!
where we used the symmetry of the Christoffel symbols in the last two
indices.
The second expression for dα given in the problem holds by virtue of the
antisymmetry of the wedge product, as discussed in Exercise 2.4.
To get the last expression, we start from
1
∇[i1 ai2 ...ik+1 ] = (−1)σ ∇iσ (1) aiσ (2) ...iσ (k+1)
(k + 1)!
σ ∈Sk+1
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
b. By definition, δα = (−1)nk+n+1 d α. We write

1
α= ai ...i d x i1 ∧ · · · ∧ d x ik ,
k! 1 k
1
α = a i1 ...ik i1 ...in d x ik+1 ∧ · · · ∧ d x in ,
k!(n − k)!
1 i ...i
d α = a 1 k i1 ...in ; j d x j ∧ d x ik+1 ∧ · · · ∧ d x in

k!(n − k)!
1
= ∇ j1 a i1 ...ik i1 ...ik j2 ... jn−k+1 d x j1 ∧ · · · ∧ d x jn−k+1

k!(n − k)!
= b j1 ... jn−k+1 d x j1 ∧ · · · ∧ d x jn−k+1 ,
where
1
b j1 ... jn−k+1 := ∇ j1 a i1 ...ik i1 ...ik j2 ... jn−k+1 .

k!(n − k)!
Thus we have
1
d α = b j1 ... jn−k+1 j1 ... jn d x jn−k+2 ∧ · · · ∧ d x jn
[n − (n − k + 1)]!
1
= b p ... p j ... j
(k − 1)! 1 n−k+1 1 n
× g p1 j1 · · · g pn−k+1 jn−k+1 d x jn−k+2 ∧ · · · ∧ d x jn
1
= ∇ p1 a i1 ...ik i1 ...ik p2 ... pn−k+1 j1 ... jn

k!(n − k)!(k − 1)!
× g p1 j1 · · · g pn−k+1 jn−k+1 d x jn−k+2 ∧ · · · ∧ d x jn
1
= g p1 j1 ∇ p1 ai1 ...ik i1 ...ik j2 ... jn−k+1 j1 ... jn

k!(n − k)!(k − 1)!
× d x jn−k+2 ∧ · · · ∧ d x jn ,
where we used metric compatibility. Applying the Leibniz rule we have

∇ p1 ai1 ...ik i1 ...ik j2 ... jn−k+1 j1 ... jn

= ∇ p1 (ai1 ...ik ) i1 ...ik j2 ... jn−k+1 j1 ... jn

+ ai1 ...ik ∇ p1 i1 ...ik j2 ... jn−k+1 j1 ... jn .
The second term vanishes, while the first term becomes, after permuting
some indices,
∇ p1 (ai1 ...ik ) i1 ...ik j2 ... jn−k+1 j1 ... jn

= (−1)(n−k)(k−1) ∇ p1 (ai1 ...ik ) i1 ...ik j2 ... jn−k+1 j1 jn−k+2 ... jn j2 ... jn−k+1
= (−1)(n−k)(k−1) (n − k)!∇ p1 (ai1 ...ik )δ ij11...i k
jn−k+2 ... jn
= (−1)(n−k)(k−1) k!(n − k)!∇ p1 (a j1 jn−k+2 ... jn ).
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
βi2 ...ik := −∇ j a ji2 ...ik .
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
where we coalesced the signs.

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
Adding the two expressions together we get

k

(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
By swapping indices we have

R p j j ir = −R j pj ir = −R j p jir = −R p ir .
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
Putting everything all together yields

(a)i1 ...ik = −∇ 2 ai1 ...ik

k
+ ai1 ...ir −1 pir +1 ...ik R p ir
r =1
1
− ai ...i ji ...i pi ...i R j p ir is ,
2 r =1...k 1 r −1 r+1 s−1 s+1 k
s=1...k
r =s
which is the Weitzenböck formula.

8.59 a. α is harmonic, so

0 = k!(α, α) = a i1 ...ik (a)i1 ...ik σ

M
= a i1 ...ik − ∇ 2 ai1 ...ik

M

k
+ ai1 ...ir−1 iir +1 ...ik R i ir
r =1
1
− ai1 ...ir −1 iir+1 ...is−1 jis+1 ...ik R i j ir is σ.
2 r=1...k
s=1...k
r =s
We have, by permuting indices,

k
a i1 ...ik ai1 ...ir−1 iir +1 ...ik R i ir
r =1

k
= a i1 ...ir−1 jir+1 ...ik ai1 ...ir−1 iir +1 ...ik R i j
r=1

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.
But f = δd f because δ f = 0. ( f is a zero form.) Thus,

(∇ 2 f )σ = −(δd f, 1) = −(d f, d1) = 0.
M
Now let f = a i1 ...ik ai1 ...ik and expand to get

0= ∇ 2 (a i1 ...ik ai1 ...ik )σ
M
= ∇ i ∇i (a i1 ...ik ai1 ...ik )σ
M
= ∇ i {(∇i a i1 ...ik )ai1 ...ik + a i1 ...ik ∇i ai1 ...ik }σ
M
= {(∇ 2 a i1 ...ik )ai1 ...ik + (∇i a i1 ...ik )(∇ i ai1 ...ik )
M
+ (∇ i a i1 ...ik )(∇i ai1 ...ik ) + a i1 ...ik ∇ 2 ai1 ...ik }σ

= 2 {(∇ 2 a i1 ...ik )ai1 ...ik + (∇ i a i1 ...ik )(∇i ai1 ...ik )}σ.
M
c. Combining (a) and (b) gives

1 2
(∇i ai1 ...ik )(∇ i a i1 ...ik ) + k F(α) σ = 0. (1)
M
From (8.109) we have
Ri j = g pq R piq j = K g pq (g pq gi j − g pj giq ) = K (n − 1)gi j ,
so
a ji2 ...ik aii2 ......ik R i j = K (n − 1)k!α2 .
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
F(α) = K k!(n − k)α2 ,
as anticipated. By the nondegeneracy of the inner product, α2 is noneg-

ative, as is ∇ j ai1 ...ik 2 , so each term in the brackets in (1) must vanish. In
particular, we must have F(α) = 0, and hence α = 0. Evidently there are
no nonzero harmonic k forms on M for 0 < k < n. By Hodge’s theorem,
the corresponding Betti numbers must vanish.
8.60 a.
1
1
Vn+1 = r n An dr = An .
0 n+1
b. We compute
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.
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
The second integral is the volume of an (n − 1)-ball of radius (1 − (x 1 )2 +

(x 2 )2 )1/2 , so

Vn+1 = Vn−1 (1 − (x 1 )2 + (x 2 )2 )(n−1)/2 d x 1 d x 2
(x 1 )2 +(x 2 )2 ≤1
1
= 2π Vn−1 (1 − r 2 )(n−1)/2 r dr
0
2π
= Vn−1 .
n+1
Equivalently,
Vn 2π
= .
Vn−2 n
Temporarily set
π n/2
f n := .
((n/2) + 1)
Then by the recursive property of the gamma function,
fn 2π
= .
f n−2 n
Moreover, V1 = f 1 = 2 and V2 = f 2 = π. As Vn and f n satisfy the same
recurrence relation with the same initial conditions, we must have Vn = f n .
8.61 a. The Laplacian in spherical polar coordinates in R3 is

1 ∂ ∂f 1 ∂ ∂f 1 ∂2 f
∇2 f = 2 r2 + 2 sin θ + .
r ∂r ∂r r sin θ ∂θ ∂θ r 2 sin2 θ ∂φ 2
Plugging in f = r n h gives

1 ∂ ∂h 1 ∂ 2h
0 = n(n + 1)h + sin θ + 2
sin θ ∂θ ∂θ sin θ ∂φ 2
= n(n + 1)h + ∇ 2 h
= n(n + 1)h − h.
b. Let h n and h m be spherical harmonics of degrees n and m, respectively.

Then
n(n + 1)(h n , h m ) = (h n , h m ) = (h n , h m ) = m(m + 1)(h n , h m ).
If n = m then (h n , h m ) = 0.
8.62 Let η be harmonic. Clearly, if λ = ω then
(λ, η) = (ω, η) = (ω, η) = 0.
Conversely, suppose (λ, η) = 0 for all η. By the Hodge decomposition, there
exist α, β, and γ such that
λ = dα + δβ + γ .
Choosing η = γ we get
0 = (λ, γ ) = (dα + δβ + γ , γ ) = (γ , γ ),
because γ is closed and co-closed. By nondegeneracy, γ = 0.
Again by the Hodge decomposition,
α = dα1 + δβ1 + γ1 ,
so
dα = dδβ1 .
Once again we have
β1 = dα2 + δβ2 + γ2 ,
whereupon we find
dα = dδβ1 = dδdα2 = (dα2 ).
A similar argument shows that δβ is harmonic, and the sum of two harmonic
functions is harmonic.
9
The degree of a smooth map
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
so the degree of f is zero.

9.3 First note that choosing x = 0 gives g = f . Then

∗
f (dϕ) = (Deg f ) dϕ = 2π(Deg f ).
S1 S1
But f ∗ (dϕ) = d f ∗ ϕ = d(ϕ ◦ f ) = k dθ, so Deg f = k. (OK, that’s a little

cavalier, because θ and ϕ are not globally defined functions, but the argument
works because k is the Jacobian of the transformation f at every point, so the
local argument extends to a global one.)
151
152 The degree of a smooth map
9.4 Let σ be the volume form (generator of top dimensional cohomology) on S n .

Flipping the sign of all the x i ’s sends σ → (−1)n+1 σ . So

∗
(−1) n+1
σ = (−id) σ = (Deg(−id)) σ.
Sn Sn Sn
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
(1 − t) f (x) = t x ⇔ (1 − t) f (x) = t x ⇔ 1 − t = t ⇔ t = 1/2,
(because x, f (x) ∈ S n ). This holds provided f (x) = x, but f has no fixed

points by hypothesis, so F is indeed well-behaved. The result now follows
from the homotopy invariance of the degree.
9.6 By Exercise 4.12, f is null homotopic. This means that it is homotopic to a
constant map. But the degree of the constant map is zero. (One way to see this
is that pullback is the inverse of pushforward, and the derivative of a constant
is zero. Alternatively, the degree is the sum of the indices at a regular point,
and each index is a sign of a Jacobian determinant, and therefore zero.)
9.7 Suppose we had g(x) = x for some x. By construction, g sends all the points
on the sphere to the southern hemisphere S, so the fixed point would have to
be in S. But then we would have f (ϕ S (x)) = ϕ S (x), a contradiction, as f has
no fixed points. Hence, by Exercise 9.5 the degree of g is (−1)n+1 . But g is a
nonsurjective map of the sphere to itself, so by Exercise 9.6 its degree must
be zero. This contradiction shows that f must have a fixed point.
9.8 The first part is straightforward:
λn = 2−n Q i1 i2 · · · Q i2n−1 i2n θ i1 ∧ · · · ∧ θ i2n

= 2−n (−1)σ Q σ (1)σ (2) · · · Q σ (2n−1)σ (2n) σ
σ ∈S2n
= n! pf(Q) σ.
For the second part, under a change of basis θ = A−1 θ so θ = Aθ or

θ = A j k θ k . Hence,
j
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

The metric tensor is symmetric, so it has n(n + 1)/2 = n+1 2

components,
and by the symmetry of partial derivatives there are this many independent

2
second derivatives as well, so there are b = n+1 2
independent second
derivatives of the metric.
We have

(n + 1)n 2 (n + 2)(n + 1)n
b−a = −n
2 6
( )
n + 1 n + 2
= n (n + 1)
2
−
4 6
n (n − 1)
2 2
= .
12
154
Riemann normal coordinates 155
According to (8.51), this is precisely the number of independent components

of the Riemann tensor, which shows that the Riemann tensor is completely
determined by the second derivatives of the metric, or equivalently, that all
the curvature information is contained in the second derivatives of the metric.
Yet another way to put it is that, whereas we can always find a coordinate
system in which the first derivatives of the metric vanish, we cannot change
coordinates so as to make any of the second derivatives go away. (Physicists
would say that the second derivatives of the metric are “physical”, as opposed
to “gauge degrees of freedom”.)
Appendix F
Frobenius’ theorem
m
F.1 First we show that (2) and (3) are equivalent. If dθ i = j=1 A j θ then
i j
wedging with any of the θ ’s (1 ≤ j ≤ m) will kill it. Conversely,

j
we can complete the θ i ’s to a coframe field, in which case we can write

dθ i = nj,k=1 a i jk θ j ∧ θ k . Then

n
0 = dθ ∧ (θ ∧ · · · ∧ θ ) =
i 1 m
a i jk θ j ∧ θ k ∧ θ 1 ∧ · · · ∧ θ m
j,k=1
If m = n there is nothing to prove, so suppose m < n. Then by linear inde-

pendence of the θ j ’s we must have a i jk = 0 for k = m − n, . . . , n, so (3)
holds.
Next we show that (1) and (3) are equivalent. Let X, Y ∈ . By (3.123),
dθ i (X, Y ) = X θ i (Y ) − Y θ i (X ) − θ i ([X, Y ]) = −θ i ([X, Y ]).
Completing the θ i ’s to a basis we can write

n
dθ (X, Y ) =
i
a i jk θ j ∧ θ k (X, Y )
j,k=1

n
= a i jk θ j ∧ θ k (X, Y ),
j,k=m−n
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
∇ f X (gY ) = f ∇ X (gY ) = f X (g)Y + f g∇ X Y,
so
f X (gY ) − ∇ f X (gY ) = f g(∇
α( f X, gY ) = ∇ X Y − ∇ X Y ) = f gα(X, Y ).
H.2 From the definition of R(X, Y )Z and Gauss’s formula we obtain

R(X, X ∇
Y )Z = ∇ Y Z − ∇
Y ∇
X Z − ∇[X,Y ] Z
=∇X (∇Y Z + α(Y, Z )) − (X ↔ Y ) − ∇[X,Y ] Z − α([X, Y ], Z )
X α(Y, Z )
= ∇ X ∇Y Z + α(X, ∇Y Z ) + ∇
Y α(X, Z )
− ∇Y ∇ X Z − α(Y, ∇ X Z ) − ∇
− ∇[X,Y ] Z − α([X, Y ], Z ),
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
H.3 The tangent space to M is orthogonal to the normal space to N , so by metric

compatibility and Gauss’s formula,
X (g(Y, ξ )) = g(∇
0 = X (g(Y, ξ )) = ∇ X Y, ξ ) + g(Y, ∇
X ξ )
= g(∇ X Y + α(X, Y ), ξ ) − g(Y, Sξ (X ))
= g(α(X, Y ), ξ ) − g(Sξ (X ), Y ).
H.4 From Gauss’s equation (H.7), the properties of projections, and metric
compatibility,

g( R(X, Y )Z , W )
X α(Y, Z ) − ∇
= g(R(X, Y )Z , W ) + g(∇ Y α(X, Z ), W )
X W ) + g(α(X, Z ), ∇
= g(R(X, Y )Z , W ) − g(α(Y, Z ), ∇ Y W )
= g(R(X, Y )Z , W ) − g(α(Y, Z ), α(X, W )) + g(α(X, Z ), α(Y, W ))
= g(R(X, Y )Z , W ) + IIξ (X, Z )IIξ (Y, W ) − IIξ (Y, Z )IIξ (X, W ).
Again, by metric compatibility,
X α(Y, Z ), ξ ) = −g(α(Y, Z ), ∇
g(∇ X ξ ) = −IIξ (Y, Z )g(ξ, ∇
X ξ ) = 0.
So by the Codazzi-Mainardi equation (H.8) and metric compatibility,

g( R(X, Y )Z , ξ ) = IIξ (X, ∇Y Z ) − IIξ (Y, ∇ X Z ) − IIξ ([X, Y ], Z )
= g(Sξ (X ), ∇Y Z ) − g(Sξ (Y ), ∇ X Z ) − g(Sξ ([X, Y ]), Z )
= g(∇ X Sξ (Y ) − ∇Y Sξ (X ) − Sξ ([X, Y ]), Z ).
H.5 R3 is flat, so its curvature vanishes. By the Gauss equation,
g(R(X, Y )Y, X ) = −IIξ (X, Y )IIξ (Y, X ) + IIξ (Y, Y )IIξ (X, X ).
Choose X = e1 and Y = e2 to be orthornormal at p. Then the left side of
the above equation is just the Gaussian curvature K . As for the right side, we
have
IIξ (e j , ei ) = g(Sξ (e j ), ei ) = Si j ,
where Si j is the matrix representation of the shape operator S in the basis {ei }.
So we get
K = −S12 S21 + S22 S11 = det S.
H.6 If the ambient space is Euclidean then its curvature vanishes, and Gauss’s
equation becomes
g(R(X, Y )Z , W ) = −IIξ (X, Z )IIξ (Y, W ) + IIξ (Y, Z )IIξ (X, W ).
160 Intrinsic and extrinsic curvature
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.
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 ).
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),
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
To save writing we define x := u 2 + v 2 and α := 1 + x. Then the individual

components of η are:
î : (2uv)(−2v) − (2u)(1 − v 2 + u 2 ) = −4uv 2 − 2u + 2uv 2 − 2u 3
= −2u(1 + u 2 + v 2 ) = −2uα
ĵ : (2u)(2uv) − (−2v)(1 − u 2 + v 2 ) = 4u 2 v + 2v − 2vu 2 − 2v 3
= 2v(1 + u 2 + v 2 ) = 2vα
k̂ : (1 − (u 2 − v 2 ))(1 + (u 2 − v 2 )) − (2uv)2 = 1 − (u 2 − v 2 )2 − 4u 2 v 2
= 1 − (u 4 + 2u 2 v 2 + v 4 ) = 1 − x 2 .
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 .
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.

Solution Manual For Manifolds, Tensor and Forms ( P. Rentein) PDF

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Solution Manual For Manifolds, Tensor and Forms ( P. Rentein) PDF

Caricato da

Copyright:

Formati disponibili

Solution Manual

1 Linear algebra page 1

0 = c1 (1, 1) + c2 (2, 1) = (c1 + 2c2 , c1 + c2 )

so (1, 1) and (2, 1) are linearly independent. On the other hand,

0 = c1 (1, 1) + c2 (2, 2) = (c1 + 2c2 , c1 + 2c2 )

can be solved by choosing c1 = 2 and c1 = −1, so (1, 1) and (2, 2) are

where si ∈ S. Then some of the ci ’s are nonzero. We cannot have c = 0,

suppose x ∈ ker π ∩ im π. Then there is a y ∈ V such that x = π(y). But

aT −1 v1 + bT −1 v2 = av1 + bv2 = (T −1 T )(av1 + bv2 ) = T −1 (av1 + bv1 ),

which shows that T −1 is linear.

which shows that ST → ST .

It follows that T e j is represented by the j th column of T , so the maxi-

so the θ j ’s are linearly independent.

so f = g (two linear functionals that agree on a basis agree everywhere).

it to a basis for V : {w1 , . . . , wdim W , edim W +1 , . . . , edim V }. Let {u i } be a basis

because the modulus squared of a complex number is always nonnegative, so

and it could happen that g(u, u) = 0 but u = 0. (For example, let u =

so A∗ is linear. (The other axioms are just as straightforward.)

dim Vi = dim ker ϕi + dim im ϕi .

because the sum is telescoping.

(−1)σ A1σ (1) A2σ (2) . . . Anσ (n) . (1)

(−1)σ Aσ −1 (1)1 Aσ −1 (2)2 . . . Aσ −1 (n)n , (2)

where σ −1 is the inverse permutation to σ . For example, suppose σ (5) = 1.

As σ runs over all the elements of Sn , so does σ −1 , so (4) may be written

But this is just det A T .

where σ means a general permutation in Sn that fixes σ (1) = 1. But

which is (1.57) with i = 1. Thinking of the Ai j as independent variables, each

Of course, we could have substituted a new j th column instead, with similar

If i = k this is an expansion in terms of alien cofactors and vanishes. If i = k

[(adj A)A)]T = A T (adj A)T = A T adj A T = (det A T )I = (det A)I. (3)

A(adj A) = (adj A) A = (det A)I,

so if A is nonsingular, then the inverse of A is adj A/ det A, and if A is

vanish identically). Next, suppose Av = 0. If A were invertible, then multi-

b. By orthogonality of R, for any u, v ∈ V ,

Substituting into (2) we get

diagonal. It follows that S−1 AS = . Conversely, if there exists a nonsin-

p S−1 AS = det(S−1 AS − λI)

It follows that the eigenvalues of A are similarity invariants.

p A (λ) = (−1)n (λ − μ1 )(λ − μ2 ) · · · (λ − μn ),

where the roots (eigenvalues) μi are not necessarily distinct. By expanding

where the missing terms are of lower order in λ.

The general case follows by setting A = A1 A2 · · · An−1 and B = An .

= 1+t (x1 + x2 + · · · )+t 2 (x1 x2 + x1 x3 + · · · + x2 x3 + · · · )

From d E/dt = E(t)P(−t) we get

Equating powers of t on both sides gives

d. Write down the first n Newton identities in matrix form to get

e. If Av = λv then Ak v = λk v. As the trace is the sum of the eigenval-

1.49 By positive definiteness,

0 ≤ g(u + αv, u + αv) = g(u, u) + 2αg(u, v) + α 2 g(v, v).

Minimizing the right side with respect to α gives

0 ≤ g(u, u)g(v, v) − g(u, v)2 .

Moreover, equality holds if and only if u + αv = 0, or u = −αv.

because f 2 ≥ 0. The integral vanishes if and only if f = 0, so the map (·, ·)

g(v1 , v1 )c1 + g(v1 , v2 )c2 ··· + g(v1 , vn )cn = 0,

e2 = x 2 − e0 g(e0 , x 2 ) − e1 g(e1 , x 2 ).

g(e2 , e2 ) = g(x 2 −1/3, x 2 −1/3) = g(x 2 , x 2 )−(2/3)g(x 2 , 1)+(1/9)g(1, 1).

e3 = x 3 − e0 g(e0 , x 3 ) − e1 g(e1 , x 3 ) − e2 g(e2 , x 3 ).

For the second, if f ∈ im T ∗ then f = T ∗ g for some g ∈ W ∗ . So, if v ∈

so f v f = f (two linear maps that agree on a basis agree everywhere).

Then the components of T in the two bases are related by

In the first equality we are allowed to switch i and j in Ai j because A is

Hence (cf., Exercise 2.3),

This proves (i).

η ∧ λ = e I ∧ g(e J , e J )e J = g(e J , e J )σ = (−1)d g(e I , e I )σ

where (A ⊗ B)k,i j is the (k, i j)th component of the matrix representing

(Symk A)(ei1 · · · eik ) = Aei1 · · · Aeik

This expression is symmetric in u and v, so [

This is manifestly symmetric in u and v, so [