dg1 hw5 Solutions
dg1 hw5 Solutions
HOMEWORK ASSIGNMENT 5
MICHAEL EICHMAIR
1. Bookkeeping questions
If you would like feedback on your work on some or all of these questions, please hand your written solutions
to your course assistant or put them into their mailbox in HG F 28 by Friday November 2.
Exercise 1.1 (Potpourri). (i) Let v ∈ Rm . Explain why the linear map Rm → Rm given by w 7→
w + (v · w)v is represented by the matrix
[1] + [v][v]t
with respect to the standard basis of Rm . Here, [1] is the identity matrix and [v] is the column matrix
representing v. Show that the eigenvalues are 1 with multiplicity (m − 1) and 1 + |v|2 with multiplicity
1. What are the corresponding eigenspaces?
(ii) Let v1 , . . . , vm ∈ Rn be linearly independent and let gij = vi · vj . Let g ij denote the components of
the inverse of the matrix [g] with components [g]ij = gij . Show that the orthonormal projection v > of
a vector v ∈ Rn onto the span of v1 , . . . , vm is given by g ij (v · vi )vj . What is the matrix representing
the linear map Rn → Rn given by v 7→ v > in terms of the n × m matrix V with components Vai = via ?
Hint: Make the ansatz v > = ak vk where a1 , . . . , am ∈ R. Take the inner product of both sides of the
equation with vj . Then multiply with g ij . Remark: This formula is known as “least squares” in the
applied math literature (I think).
(iii) Let n > m ≥ 1 be integers. Let U ⊂ Rm be open and u ∈ C ∞ (U, Rn−m ). Show with all bells and
whistles that the graph ϕ : U → Rn given by x 7→ (x, u(x)) is a submanifold chart and that the induced
Riemannian inner product has components gij = δij + ∂i u · ∂j u.
Solution:
(i) Let w ∈ Rm be arbitrary, then ([1] + [v][v]t )[w] = [1][w] + [v]([v]t [w]) = [w] + (v · w)[v] = [w + (v · w)v].
We see that the matrix represents the linear map w 7→ w + (v · w)v. If w · v = 0 then w is mapped
to itself. Clearly, v is mapped to (1 + |v|2 )v. It follows that if v 6= 0 the eigenvalues are 1 (with
multiplicity m − 1) and 1 + |v|2 (with multiplicity 1). If v = 0 the eigenvalues are 1 with multiplicity
m.
(ii) As indicated in the hint we make the Ansatz v > = ak vk . Taking the inner product of both sides with
vj we get
v · vj = v > · vj = ak vk · vj = ak gkj ,
1
HW5 Differential Geometry I, Fall 2013
where we used that v ⊥ · vj = 0. Multiplying with g ij and summing (Einstein summation convention)
we obtain that
g ij (v · vj ) = ak gkj g ij = ak δki = ai .
Note that [g] = V t V . The matrix representing the linear map v 7→ v > is given by
V [g]−1 V t = V (V t V )−1 V t .
(iii) The map ϕ is smooth as its components are smooth. To show that ϕ is an immersion, we compute its
differential and see " #
1
[∂ϕ] = .
∂u
This shows that the first m columns of ∂ϕ already have full rank. Let x, x1 , x2 , . . . ∈ U be such that
ϕ(xi ) = (xi , u(xi )) → ϕ(x) = (x, u(x)). Considering only the first m components we obtain that
xi → x. Thus ϕ is a submanifold chart. For the induced inner product, we find that
Exercise 1.2 (Helicoid). Show that ϕ : R × R → R3 given by (t, θ) 7→ t(cos θ, sin θ, 0) + (0, 0, θ) is a
submanifold chart. What are the Christoffel symbols of the induced Riemannian inner product on R2 ?
Show that the mean curvature of the corresponding simple submanifold vanishes.
Solution: It is clear that ϕ is smooth. To show that it is a homeomorphism and that its differential is
injective, we construct a smooth map ψ : R3 → R2 which restricts to an inverse of ϕ on its image. We define
ψ(x, y, z) = (x cos(z) + y sin(z), z). Note how we reconstruct the parameter t by taking the component of
the vector (x, y, 0) with respect to the unit vector (cos(z), sin(z), 0). This implies directly that ψ ◦ ϕ = idR2 ,
which shows that ψ|ϕ(R2 ) is the inverse of ϕ. From this one sees that ϕ is a homeomorphism. Also, by the
chain rule, we have
v = ∂(ψ ◦ ϕ)|x v = ∂ψ|ϕ(x) ∂ϕ|x v
for all (x, v) ∈ R2 × R2 . It follows that ∂ϕ|x : R2 → R3 is injective for every x ∈ R2 . We compute the
components of the induced Riemannian inner product.
∂t ϕ = (cos(θ), sin(θ), 0)
∂θ ϕ = (−t sin(θ), t cos(θ), 1)
gtt = (∂t ϕ) · (∂t ϕ) = 1
gtθ = gθt = (∂t ϕ) · (∂θ ϕ) = 0
gθθ = (∂θ ϕ) · (∂θ ϕ) = 1 + t2
2
HW5 Differential Geometry I, Fall 2013
Exercise 1.3 (Tangent maps). Let ϕ : U → Rn be an m dimensional submanifold chart. The tangent map
ϕ∗ : U × Rm → Rn × Rn of ϕ given by (x, v) 7→ (ϕ(x), ∂ϕ|x v) is a 2m dimensional submanifold chart.
Solution:
• As U is open, so is U × Rm .
• ϕ∗ is smooth, as ϕ is.
• The differential of ϕ∗ has the form:
" #
∂ϕ 0
[∂ϕ∗ ] =
? ∂ϕ
Using that the matrix [∂ϕ|x ] has rank m for every x ∈ U we see that the matrix [∂ϕ∗ |(x,v) ] has rank
2m for all (x, v) ∈ U × Rm . It follows that ϕ∗ is an immersion.
• Now we need to show that ϕ∗ is also a homeomorphism with respect to its image ϕ∗ (U × Rm ).
• To see that ϕ∗ is injective, note that (ϕ(x), ∂ϕ|x v) = (ϕ(y), ∂ϕ|y w) implies that x = y because ϕ is
injective. Then ∂ϕ|x v = ∂ϕ|x w implies v = w because ∂ϕ|x is injective. We proceed to show that
ϕ∗ is a homeomorphism with its image. Let {(xn , vn )}∞ m m
n=1 ⊂ U × R and (x, v) ∈ U × R be such
that ϕ∗ (xn , vn ) → ϕ∗ (x, v). The goal is to show that (xn , vn ) → (x, v). Clearly, xn → x because ϕ is
a homeomorphism with its image. Since ∂ϕ|x has rank m, there are indices 1 ≤ i1 < . . . < im ≤ m
such that ∂ϕI |x : Rm → Rm is invertible where ϕI = (ϕi1 , . . . , ϕim ). By continuity, the linear maps
∂ϕI |xn : Rm → Rm are also invertible for all n ≥ n0 sufficiently large. Moreover, ∂ϕI |−1 xn → ∂ϕ |x
I −1
as linear maps as n → ∞.1 By assumption, ∂ϕI |xn vn → ∂ϕI |x0 v and thus vn = ∂ϕI |−1 I
xn ∂ϕ |xn vn →
∂ϕI |−1 I
xn ∂ϕ |x v = v, as desired.
Exercise 1.4. Let (U, g) be an m dimensional simple Riemannian manifold. Recall the following funda-
mental definitions /defining properties.
3
HW5 Differential Geometry I, Fall 2013
Rmijk` = − Rmjik`
Rmk`ij = Rmijk`
Rmij`k = − Rmijk`
Rmijk` + Rmjki` + Rmkij` = 0.
Hint: The identity ∂k g ij = −g is g jr ∂k grs from a previous homework set will be useful.
Show that (∂k gij )(0) = 0 for all i, j, k if and only if Γkij (0) = 0 for all i, j, k.
Note that
g ij (0) = δ ij
1
Γkij (0) = (∂j gik + ∂i gkj − ∂k gij ) (0)
2
for all i, j, k.
g ij (0) = δ ij
(∂k g ij )(0) = 0
Γkij (0) = 0
1
Rmijk` (0) = (∂i ∂k gj` − ∂j ∂k gi` + ∂j ∂` gik − ∂i ∂` gjk )(0)
2
for all i, j, k.
Solution:
We compute that
4
HW5 Differential Geometry I, Fall 2013
To see that Rmijk` = Rmk`ij we expand the Christoffel symbols. For clarity, we underline terms to highlight
the changes from one line to the next.
Thus
1
(1) Rmijk` = (∂i ∂k g`j − ∂i ∂` gkj + ∂j ∂k g`i − ∂j ∂` gki )
2
1
− g ♥∗ (−∂i g`♥ + ∂♥ gi` − ∂` gi♥ )(−∂k g∗j − ∂j g∗k + ∂∗ gjk )
4
1
− g ♥∗ (−∂j g`♥ + ∂♥ gj` − ∂` gj♥ )(∂k g∗i + ∂i g∗k − ∂∗ gik )
4
It follows that Rmijk` = Rmk`ij . (Note that the second and the third term on the right hand side of (1)
swap places.)
5
HW5 Differential Geometry I, Fall 2013
In conjunction with Rmijk` = − Rmjik` shown above we obtain that Rmijk` = − Rmik`k .
If (∂k gij )(0) = 0, then directly from the definition of Christoffel symbols, we get Γkij (0) = 0 as all the terms
in the sum are zero.
Assume now that gij (0) = δij . Then g ij (0) = δ ij . Calculating the Christoffel symbol, we have that
1 1
Γkij (0) = δ k∗ (∂j gi∗ (0) + ∂i g∗j (0) − ∂∗ gij (0)) = (∂j gik (0) + ∂i gkj (0) − ∂k gij (0)) .
2 2
Assume now that also ∂k gij (0) = 0. Then (∂k g ij )(0) = −g i∗ (0)g j• (0)∂k g∗• (0) = 0. We calculate that
1 1
∂d Γabc = ∂d g a∗ (∂c gb∗ + ∂b g∗c − ∂∗ gbc ) + g a∗ (∂d ∂c gb∗ + ∂d ∂b g∗c − ∂d ∂∗ gbc )
2 2
Evaluating at 0:
1 1
∂` Γkij (0) = δ k∗ (∂` ∂j gi∗ (0) + ∂` ∂i g∗j (0) − ∂` ∂∗ gij (0)) = (∂` ∂j gik (0) + ∂` ∂i gkj (0) − ∂` ∂k gij (0))
2 2
Finally,
2. Review
There will be no solution for review exercises. Mind you, it would be convenient for us to pretend that you
learnt all these things in that perfect introductory class you took your first year at university. Since we
didn’t take that class, we tirelessly created these exercises to help you get up to speed with the foundational
material. I will eventually assume that you are comfortable with these topics. For the exam, I expect you
to be able to apply these results correctly in context. I would not ask you for their proof.
Review 2.1 (Quadratic forms over R). Let E be a finite dimensional real vector space of dimension d. The
vector spaces
6
HW5 Differential Geometry I, Fall 2013
Let g : E × E → R be a positive definite inner product and let l : E → E be a linear map that is symmetric
with respect to g. Show that there is a basis b1 , . . . , bd of E and real numbers λ1 , . . . , λd ∈ R such that
g(bi , bj ) = δij and such that lbi = λi bi for all i, j = 1, . . . , d. Hint: The following steps may be helpful.
7
HW5 Differential Geometry I, Fall 2013
Let g : E × E → R be a positive inner product and let h : E × E → R be a symmetric bilinear map. Show
that there exists a basis b1 , . . . , bd of E and real numbers λ1 , . . . , λd ∈ R such that g(bi , bj ) = δij and such
that h(bi , bj ) = λi δij for all i, j = 1, . . . , d. Hint: Consider the linear map G−1 H : E → E.
• Let G, H : E → E ∗ be the linear maps associated with g, h. Explain why it is enough to show that
the linear map G−1 H : E → E is diagonalizable.
• Let F be a G−1 H invariant subspace of E. Then F ⊥ = {v ∈ E : g(v, v) = 0 for all v ∈ F } is also
G−1 H invariant.
• Show that G−1 H has an eigenvector. Hint: Explain why the infimum
Let h : E × E → R be a symmetric bilinear map on E. Show that there exist integers d− , d0 , d+ ≥ 0 with
d− + d0 + d+ = d and a basis b1 , . . . , bd of E such that
0 if i 6= j,
−1 if i = j ∈ {1, . . . , d },
−
h(bi , bj ) =
0 if i = j ∈ {d− + 1, . . . , d− + d0 },
1 if i = j ∈ {d− + d0 + 1, . . . , d}.
Show that d0 , d− , d+ are uniquely determined by h. (This is Sylvester’s law of inertia.) Hint: To find
a basis, let g : E × E → R be a positive definite inner product on E. Show that d0 is the dimension of
{v ∈ E : h(v, v) = 0 for all v ∈ E}. Show that any two subspaces of E that are maximal so that h is positive
definite on them have the same dimension.
Let g : E × E → R be a positive definite inner product and let l : E → E be a linear map that is skew-
symmetric with respect to g. There exist k invariant and mutually g perpendicular subspaces E0 , E1 , . . . , Ek
of E such that
E = E0 ⊕ E1 ⊕ . . . ⊕ Ek ,
such that l vanishes on E0 , and such that each Ei is two dimensional when i = 1, . . . , k. For every i = 1, . . . , k
there exist a positive real λi and a g orthonormal basis e2i , e2i+1 of Ei such that le2i = −λi e2i+1 and
le2i+1 = λi e2i . Hint: Show that the linear map l2 = l ◦ l : E → E is symmetric with respect to g and that
its eigenvalues are all non-positive. If v ∈ E is an eigenvector corresponding to a non-zero eigenvalue, show
that spanR {v, lv} is an l invariant subspace.
E admits a non-degenerate skew-symmetric bilinear map if and only if it is even dimensional. Let h1 , h2 :
E × E → R be two non-degenerate skew symmetric bilinear forms. Show that there is an isomorphism
i : E → E such that i∗ h2 = h1 , i.e. such that h1 (v, v) = h2 (iv, iv) for all v, v ∈ E. Hint: Let g : E × E → R
be a positive definite inner product and consider G−1 H1 and G−1 H2 where G, H1 , H2 : E → E ∗ are the
linear maps associated with g, h1 , h2 .
8
HW5 Differential Geometry I, Fall 2013
3. Project
The goal of this project is to follow Bernhard Riemann in his discovery of the curvature tensor.
Let (U, g) be an m dimensional simple Riemannian manifold. We think of U as a map (a coordinate chart)
of an actual physical region. The purpose of the Riemannian inner product
g = gij ∂ i ⊗ ∂ j
is to measure lengths of curves (and, further, distances between points) in the physical region. Indeed, if
c : I → U is a curve, we let Z q
length(c) = (gij ◦ c)ċi ċj .
I
You should check that this definition of length is invariant under reparametrizations of the curve.
The components gij of the Riemannian inner product g on U can be recovered from the knowledge of the
lengths of all curves, i.e. from physical experimentation. Indeed, let x ∈ U and v ∈ Rm . For all ε > 0
sufficiently small the curves cε : [0, ε] → U given by t 7→ x + t v are well-defined and
d
q
length(cε ) = gij (x)v i v j .
dε ε=0
It remains to note that gij (x) is known when gij (x)v i v j is known for all v ∈ Rm . (Why?)
The classical problem of cartography is the design of maps of a given region (e.g. a part of the surface
of the earth) where distances between points can be read off as their Euclidean distance, i.e. maps where
gij = δij . Carl Friedrich Gauss through his Theorema Egregium and further, in the abstract, Bernhard
Riemann recognized that curvature is an obstruction to the design of such maps. You will follow Riemann’s
footsteps below.2
Let (Uα , g α ) be another m dimensional simple Riemannian manifold and let ψαβ : Uα → Uβ be a diffeomor-
phism. Using ψαβ we may think of Uα as a map of the same physical region as Uβ . We say that ψαβ is an
isometry if
∗
g α = ψαβ gβ .
Recall that Uα × Rm 3 (xα , vα ) ∼ (xβ , vβ ) ∈ Uβ × Rm if and only if xβ = ψαβ (xα ) and vβ = ∂ψαβ |xα vα .
2That is to say, this is how we believe Riemann arrived at the abstract definition of curvature according to M. Spivak in the
second volume of his wonderful series [1].
9
HW5 Differential Geometry I, Fall 2013
∗ gβ .
(i) g α = ψαβ
α = (g β ◦ ψ )∂ ψ k ∂ ψ ` for all i, j = 1, . . . , m.
(ii) gij k` αβ i αβ j αβ
α (x )v i w j = g β (x )v k w ` whenever (x , v ) ∼ (x , v ) and (x , w ) ∼ (x , w ).
(iii) gij α α α k` β β β α α β β α α β β
α (x )v i v j = g β (x )v k v ` whenever (x , v ) ∼ (x , v ).
(iv) gij α α α k` β β β α α β β
(v) length(cα ) = length(cβ ) whenever cα : I → Uα and cβ : I → Uβ are related curves, i.e. cβ = ψαβ ◦ cα .
∗ gβ .
We may always force ψαβ to be an isometry by putting g α = ψαβ
From now on we will assume that U = Rm . This assumption is merely for notational convenience and has
no other bearings.
ψ
Show that there is a diffeomorphism ψ : Rm → Rm with ψ(0) = 0 and gij (0) = δij . Hint: Linear algebra.
Assume that gij (0) = δij for all i, j. Let ψ : Rm → Rm be a diffeomorphism such that ψ(0) = 0. Show that
ψ
gij (0) = δij for all i, j if and only if (∂i ψ j )(0) = Rij are the components of an orthogonal transformation of
Rm .
Assume that gij (0) = δij for all i, j. Show that there is a diffeomorphism ψ : Rm → Rm with ψ(0) = 0 and
ψ ψ
(∂i ψ j )(0) = δij for all i, j such that gij (0) = δij and (∂k gij )(0) = 0 for all i, j, k. Hint: Use Proposition 3.2
below.
Assume that gij (0) = δij and (∂k gij )(0) = 0 for all i, j, k. Let ψ : Rm → Rm be a diffeomorphism such that
ψ
ψ(0) = 0 and (∂i ψ j )(0) = δij for all i, j. Show that (∂k gij )(0) = 0 for all i, j, k if and only if (∂i ∂j ψ k )(0) = 0
for all i, j, k.
Assume now that gij (0) = δij and that (∂k gij )(0) = 0 for all i, j, k. Let ψ : Rm → Rm be a diffeomorphism
such that ψ(0) = 0 and (∂i ψ j )(0) = δij and (∂i ∂j ψ k )(0) = 0 for all i, j, k. Show that
ψ
gij (0) = δij
ψ
(∂k gij )(0) = 0
ψ
(∂p ∂q gij )(0) = (∂p ∂q gij )(0) + (∂i ∂p ∂q ψ j )(0) + (∂j ∂p ∂q ψ i )(0)
10
HW5 Differential Geometry I, Fall 2013
for all i, j, k, `. (There is an almost elegant way of doing this.) Thus, no matter how we choose the
diffeomorphism ψ : Rm → Rm (observing the constraints above), if the components of the curvature tensor
at a point are non-zero in one map, then they are non-zero in any other map.
The results below are elementary consequences of the inverse function theorem. Their proofs are non-
examinable.
Lemma 3.1. Let id : Rm → Rm be the identity map x 7→ x. Let ψ : Rm → Rm be a smooth map. There is
ε > 0 such that if
|ψ − id|C 1 = sup |(ψ − id)(x)| + |∂x (ψ − id)| < ε
x∈Rm | {z }
operator norm
then ψ is a diffeomorphism.
Proposition 3.2 (Diffeomorphism with prescribed Taylor expansion). Let n ≥ 2 be an integer. For every
multi-index α = (α1 , . . . , αm ) ∈ Nm with |α| = α1 + . . . + αm ∈ [2, n] and every k ∈ {1, . . . , m} let akα ∈ R be
given. There is a diffeomorphism ψ = (ψ 1 , . . . , ψ m ) : Rm → Rm such that
X
ψ k = xk + akα (x1 )α1 . . . (xm )αm near the origin
2≤|(α1 ,...,αm )|≤n
Proof. Let χ : Rm → R be a smooth function with χ = 1 on B1 (0) and χ = 0 outside of B2 (0). For a given
ε > 0 we let χε : Rm → R be given by x 7→ χ(x/ε). Note that supx∈Rm supi∈{1,...,m} |∂i χε (x)| = O(ε−1 ).
Consider the maps ψε : Rm → Rm with kth component
X
x 7→ xk + χε (x) akα (x1 )α1 . . . (xm )αm .
2≤|(α1 ,...,αm )|≤n
Then |ψε − id|C 1 = O(ε). In particular, for ε > 0 sufficiently small, ψ := ψε is a diffeomorphism by Lemma
3.1.
Exercise 3.1 (Borel’s lemma, not examinable). Let a0 , a1 , a2 , . . . ∈ R be an arbitrary sequence of real
numbers. Show that there is a smooth function f ∈ C ∞ (R) such that f (k) (0)/k! = ak for all k = 0, 1, . . ..
References
1. Michael Spivak, A comprehensive introduction to differential geometry. Vol. II, second ed., Publish or Perish Inc., Wilmington,
Del., 1979. MR 532831 (82g:53003b)
11