Notes for Math 760 – Topics in Differential Geometry

January 2015

0 Introduction: Problems in Riemannian Geometry

Undefined terms: Differentiable manifold, Riemannian manifold, homogeneous manifold,

incompressible submanifold, stabilizer, conformal class

Obviously differential geometry is a sub-study of differentiable manifold theory, and

while geometry has its own internal motivations and problems, to a substantial degree
geometry exists to serve topology. In some sense the central project of differential topology is
to establish a set of numerical or algebraic invariants that can perfectly distinguish manifolds
from one another, as the algebraic invariant a2 + b2 = c2 distinguishes right triangles in
Euclidean space from all other triangles (or, more pedantically, the numerical invariants a,
b, c (side lengths) distinguish all triangles up to similarity).

Nineteenth century mathematics determined that compact 2-manifolds could be classi-

fied by a single numerical invariant: the Euler number (and when χ = 0 also orientability).
Part of the cluster of results surrounding this was the geometrization (also known as uni-
formization) of compact 2-manifolds: any compact 2-manifold possesses a metric of constant
curvature +1, 0, or -1.

Later, the ideas of geometrization leached into other areas of manifold theory; of special
note is the case of 3-manifolds. As opposed to the 3 homogeneous geometries of 2-manifolds,
Thurston was noticed that 3-manifolds could admit just 8 homogeneous geometries (with
compact stabilizers), called the model geometries. He noticed that, although incompressible
spheres and tori obstructed the existence of a model geometry, incompressible 2-manifolds
of other kinds did not. This, along with other evidence, led Thurston conjecture that any
3-manifold can be cut along incompressible tori (in an essentially unique way) such that the
resulting connected components admit a complete metric with one of the 8 geometries.

Since the topological structure of any manifold with a model geometry is completely
understood, if any compact 3-manifold admits such a decomposition, then the topology of
all compact 3-manifolds would be understood.

1
 Starting in the early 1980’s, Richard Hamilton began a program of using the Ricci
Flow to prove the geometrization conjecture for 3-manifolds. The Ricci flow is an evolution
equation for the metric:
dgij
= −2Ricij . (1)
dt
This is a non-linear 2nd order PDE, which resembles a heat equation, or more accurately
a reaction-diffusion equation. Hamilton proved that if singularities do not develop, the
resulting manifold in the limit carries a model geometry. Singularities proved difficult to
control, but in 2003 Perelman classified the singularities, proved that there are at most
finitely many of them, and thereby completed the geometrization program.

The great success of differential-geometric methods in the solution of Thurston’s os-

tensibly topological program has spurred development of such methods in other dimensions.
To simplify (almost to absurdity!), the program has three parts

1) In dimension n, determine the “canonical” geometries, and the topological restrictions

associated to such a geometry
2) Determine a canonical way to break a manifold into pieces, each of which can be given
a canonical geometry
3) Determine the topology of the original manifold from the topology of the pieces with
their gluing pattern

By-and-large, we the mathematical community is stuck on (1), and we expect to be stuck

there for the rest of our lives! (Except in some important special cases.) Surely Einstein
metrics should be canonical; then as an example of the difficulty of (1), you may ask which
manifolds admit an Einstein metric? If you answer this, even in the case n = 4, you’ll
get a fields medal. However we now completely understand which Kähler manifolds admit
Kähler-Einstein metrics.

Many of the problems we’ll explore are involved in the search for canonical metrics.
Our list of topics (we’ll surely not get to all of them) is

• The Yamabe Problem: does a conformal class contain a constant-scalr curvature met-
ric? Is it unique?
• (prescribed scalar curvature) Given a function f : M n → R, is there a metric on M n
so that scal = f ?
• The Ricci flow: can we canonically smooth-out a metric using a heat-like flow in
curvature, like the heat flow “smooths out” any distribution of heat on a compact
body? (Possibly related topic: the Yamabe flow.)
• Complex geometry, especially Kähler geometry: if we add additional structure, do our
problems become more solvable? (Answer: Yes, but they’re still hard.)

2
 • The Calabi conjectures, and Yau’s and Aubin’s partial solutions (possible related topic:
the Calabi flow)
• The Yang-Mills equation and the Yang-Mills flow: vector bundles are of decisive sig-
nificance in Riemannian geometry. Do bundles have canonical geometries? Can we
establish a heat-type flow to attempt to find them? (Yes and yes.)

1 Topic 1 – Bridging Analysis and Geometry

1.1 Stokes’ Theorem

The primary nexus of analysis and differential topology is Stokes Theorem: if ω is a k-form
on M n and N k+1 ⊂ M n is a submanifold, then
Z Z
dω = ω. (2)
N ∂N

Stokes theorem is FTCII plus diffeomorphism invariance of integration.

1.2 The Sobolev Inequalities

The primary nexus of analysis and differential geometry is the cluster of results including and
surrounding the Sobolev embedding theorems. These are familiar in the Euclidean setting,
where they are usually proven (and understood) as consequences of the Poincare inequality,
which is normally proven via iterated applications of FTCII (which is Stokes Theorem in
dimension 1). But careful study of the proof shows that it is dependent on the Euclidean
structure of Rn ; the proof cannot be imitated on manifolds. What’s more, the statement of
the Poincare inequality is very distinctly not diffeomorphism invariant.

The proof of the Sobolev inequalities of Federer-Fleming, dating to the late 1950’s,
showed us their fundamental connections with geometry; it is their proof that we’ll give
here. Before giving the proof, we introduce two constants. If (Ω, g) is a domain with metric
g, then the ν-isometric inequality is
|∂Ωo |
Iν (Ω) = inf ν−1 (3)
Ωo ⊂⊂Ω |Ωo | ν

where |∂Ωo | indicates the (n−1) Hausdorff measure of ∂Ωo and |Ωo | indicates the n-Hausdorff
measure of Ωo . The ν-Sobolev constant is
R
|∇f |
Cν (Ω) = inf ν−1 (4)
f ∈C ∞ (Ω) R ν

Supp(f )⊂Ω f ν−1 ν

Clearly both are less than ∞.

3
Theorem 1.1 (Federer-Fleming)
Iν (Ω) = Cν (Ω)

Proof. Step I: Iν (Ω) ≥ Cν (Ω)

Given any domain Ωo ⊂⊂ Ω, define the function

 
1
f (x) = max 1 − dist(x, Ω0 ), 0 (5)

R ν−1
It is easy to prove that lim→0 fR ν = |Ω|. It is somewhat more complicated, but still
not difficult to prove that lim→0 ∇f = |Ωo |. Therefore
Z Z  ν−1
ν
ν ν−1
|∂Ωo | = lim |∇f | ≥ Cν (Ω) f ν−1 ≥ Cν (Ω) |Ωo | ν . (6)
→0

Step II: Cν (Ω) ≥ Iν (Ω)

This is the more interesting case. Assume f is smooth, with isolated critical points.
On any non-critical point p, let dA indicate the area measure on the level-set of f passing
through p. The co-area formula is
1
dV olp = df ∧ dA (7)
|∇f |
Then using Ωt = {|f | ≥ t} to denote the superlevel sets, we have
Z Z Z sup |f | Z Z ∞
|∇f | dV ol = df ∧ dA = dA df = |∂Ωt | dt
inf |f | {f =0} 0
Z ∞
(8)
ν−1
≥ Iν (Ω) |Ωt | ν

and using the “layer-cake” representation and then a change of variables (a lá multivariable
calculus), we have
Z Z Z |f (p)| Z ∞Z
ν ν 1 ν 1
f ν−1 = t ν−1 dt dV ol(p) = t ν−1 dV ol dt
ν−1 Ω 0 ν−1 0 Ωt
Z ∞ (9)
ν 1
= t ν−1 |Ωt | dt
ν−1 0
Lastly we have to connect these. We have

Lemma 1.2 If g(t) is non-negative and decreasing, and if s ≥ 1, then

 Z ∞  1s Z ∞
1
s ts−1 g(t) dt ≤ g(t) s dt (10)
0 0

4
Proof. First taking a derivative
! 1s ! 1−s
s
Z T Z T
d s−1 s−1 s−1
s t g(t) dt = T g(T ) s t g(t) dt
dT 0 0
! 1−s (11)
Z T s
1 1
≤ T s−1 g(T ) s s ts−1 = g(T ) s
0

and then taking an integral, we get

 Z ∞  1s Z ∞
1
s−1
s t g(t) ≤ g(t) s dt. (12)
0 0


ν
Applying this lemma with s = ν−1 we get
Z ν
 ν−1  Z ∞  ν−1 Z ∞ Z
ν−1 ν 1
ν ν−1 1 (13)
f ν = t ν−1 |Ωt | dt ≤ |Ωt | ν ≤ |∇f |
ν−1 0 0 Iν (Ω)

giving the inequality Iν (Ω) ≤ Cν (Ω). 

pν
Theorem 1.3 For any p < ν we obtain embeddings L ν−p ⊂ W01,p .

The critical value of ν is ν = n. The Sobolev constant on standard Euclidean space

has Cn (Rn ) nonzero, but for any other value of ν, Cν (Rn ) = 0.

If f has more than one weak derivative, then the embedding theorem also improves.

pν
Theorem 1.4 If 1 ≤ p < ∞, k ∈ N, and kp < ν then we obtain embeddings L ν−pk ⊂ W0k,p .

pν
So what happens if pk ≥ ν? As pk % ν, then ν−pk % ∞, so perhaps the embedding
∞
is into L . This naive supposition fails because the constant in the Sobolev inequality also
degenerates (see the exercises).

However if pk > n, then we obtain strong consequences indeed.

n
Theorem 1.5 If l + α = k − p and Cn (Ω) > 0, then

W0k,p (Ω) ⊂ C0l,α (Ω) (14)

Our last word is the Kondrachov-Rellich embedding theorems:

5
Theorem 1.6 The embeddings above are all compact embeddings.

The proof is not geometrically interesting, so we’ll skip it.

As a final note, notice if Ω = M n is compact, obviously Cν (Ω) = Iν (Ω) = 0 for all ν.

This is not useful, so we make the following modification:

|∂Ωo |
Iν (Ω) = inf ν−1
Ωo ⊂⊂Ω
|Ωo |≤ 21 |Ω|
|Ωo | ν

R
|∇f | (15)
Cν (Ω) = inf ν−1
f ∈C ∞ (Ω)
R ν

ν
Supp(f )⊂Ω f ν−1
| supp(f )|≤ 12 |Ω|

1.3 Suggested Problems for 1/21

1) Prove that, if ν 6= n, then Iν (Rn ) = 0.
2) If Ω is any pre-compact domain in Rn and if ν < n, show that Iν (Ω) = 0. Harder: If
ν > n, show that Iν (Ω) 6= 0.
3) Prove Theorem 1.3 by considering the function f γ for some γ, and showing that for

1 R pν  ν−p
pν
ν−p
|∇f |p p ≥ Cν (Ω) p(ν−1)
R
any 1 ≤ p < ν we have f ν−p .

4) Prove Theorem 1.4 by iterating Theorem 1.3.

1.4 Laplacians

1.4.1 Covariant Derivatives and the Rough Laplacian

If T is a tensor, say
O O
p q ∗
T : TM ⊗ T M → R
T (X, Xi1 , . . . , Xip , η j1 , . . . , η jq ) ∈ R (16)
∂ ∂
T = Ti1 ...ip j1 ...jq dxi1 ⊗ · · · ⊗ dxip ⊗ ⊗ · · · ⊗ jq
∂xji ∂x
then we can define its covariant derivative, defined via a Leibniz rule:

(∇T )(X, Xi1 , . . . , Xip , η j1 , . . . , η jq ) = X T (Xi1 , . . . , Xip , η j1 , . . . , η jq )

(17)
− T (∇X Xi1 , . . . , Xip , η j1 , . . . , η jq ) − . . . − T (Xi1 , . . . , Xip , η j1 , . . . , ∇X η jq )

6
The tensor ∇T has an expression in coordinates, where, by convention, the derivative is
indicated with a comma:
∂ ∂
∇T = Ti1 ...ip j1 ...jq ,k dxi1 ⊗ . . . dxip ⊗ ⊗ · · · ⊗ j1 ⊗ dxk . (18)
∂xj1 ∂x
One can compute the following expression for the component functions:
d
Ti1 ...ip j1 ...jq ,k = Ti ...i j1 ...jq + Γski1 Ts...ip j1 ...jq + . . . + Γskip Ti1 ...s j1 ...jq
dxk 1 p (19)
j
+ Γjks1 Ti1 ...s s...jq + . . . + Γksp Ti1 ...s j1 ...s .

Obviously this can be iterated: ∇T , ∇2 T , ∇3 T etc. The so-called rough Laplacian of a

tensor T is the trace of the Hessian:
∂ ∂
4T = g kl Ti1 ...ip j1 ...jq ,kl dxi1 ⊗ . . . dxip ⊗ ⊗ · · · ⊗ j1 . (20)
∂xj1 ∂x
In particular, if f is a function, then

∂2f df
4f = g ij − Γk k , Γk , g ij Γkij . (21)
∂xi ∂xj dx

1.4.2 Forms, the Hodge Duality Operator, and the Hodge Laplacian
V 
k
We recall a few definitions. We have the sections of k-forms, denoted Γ M n . By abuse
Vk n Vk
of notation, we denote this by simply M or just . The Hodge duality operator or just
Hodge star is term given to the tensoral forms of the volume form:
√
dV ol = gij dx1 ∧ · · · ∧ dxn ,
^
k
^
n−k
(22)
∗: −→

The ∗ operator is an isomorphism, as ∗∗ = (−1)k(n−k) . It is also positive definite in the sense

that ∗ (η ∧ ∗η) ≥ 0 with equality if and only if η = 0. Polarizing this quadratic operator, we
define an inner product

hη, γi = ∗ (η ∧ ∗γ) . (23)

Vk Vn
If η, γ ∈ then η ∧ ∗γ ∈ , so we can integrate it. This gives rise to an L2 inner product
Z
(η, γ) , η ∧ ∗γ. (24)

Vk Vk+1
Now let’s look at the exterior derivative d : → . We would like to find its L2
adjoint. Now d is a linear operator, but is only densely defined. If we restrict the L2 space to
just the C ∞ k-forms, then d is defined everywhere, but it is not a bounded (or continuous)

7
linear operator, and the inner peoduct space is not a Hilbert space (due to non-completeness
of C ∞ under the L2 norm). Despite all this, we can find an adjoint for d, by making use of
Stoke’s theorem. We find that d∗ , defined implicitly by

(dη, γ) = (η, d∗ γ) . (25)

Vk Vk−1
One use of Stokes theorem along with the fact that ∗∗ = (−1)k(n−k) gives us d∗ : →
is d∗ η = (−1)nk+k+1 ∗ (d (∗η)).

The first order operators d, d∗ can be combined give a second order operator, the Hodge
Laplacian, by
^ ^
k k
4H : −→ ,
(26)
∗ ∗
4H η = d dη + dd η.

This is not the same as the rough Laplacian, although at the first and second order levels
they are essentially the same. Notice that 4 is negative definite and 4H is positive definite:

(4η, η) = −|∇η|2
(27)
(4H η, η) = |dη|2 + |d∗ η|2 .

It is possible to prove that we have Bochner formulas for the Laplacians

4H η = −4η + F (g, η) (28)

where F is a zero-order differential operator (a linear operator) in η, and a second-order

non-linear operator in g. The following commutation relations are also often very useful:

[d, 4H ] = d4H − 4H d = 0
[d∗ , 4H ] = d∗ 4H − 4H d∗ = 0 (29)
[∗, 4H ] = ∗4H − 4H ∗ = 0.

There are no such simple commutation relation for the rough Laplacian, although later we
shall talk about the important commutator [4, ∇].

1.5 Suggested Problems for 1/26

5) Prove formulae (19) and (21).
√ 
√ 1 ∂ ∂f
6) If f is a function, prove the classic formula 4f = det g ∂xi
det g g ij ∂x j .
Vk Vk
7) Prove ∗∗ : → is the operator (−1)k(n−k) .
Vk Vk−1
8) Prove the codifferential operator is d∗ = (−1)nk+k+1 ∗ d∗ : → .
9) Prove that the inner product hη, γi is symmetric (hint: work on a basis).

8
 10) Prove that a rough-harmonic form is constant, while a Hodge-harmonic form need not
be constant, but must be both closed and co-closed.
11) In addition to the inner product (24), there is another inner product

hη, γi = g i1 j1 . . . g ik jk ηi1 ...ik γj1 ...jk .

Show that the two are proportional, and find the constant of proportionality.

1.6 Sobolev Constants in Action: The Moser Iteration Method

Let η : M n → R≥0 be a test function, meaning η ∈ Cc∞ (M n ), and let f be any weakly
differentiable function that weakly satisfies

4f + uf ≥ 0 (30)

(more general elliptic equations can be considered, but for simplicity we’ll stick to (30) as
the method in the general case is essentially the same.) A standing assumption will be
f ≥ 0.

1.6.1 Initial Apriori Estimates

n
1,p
In addition to (30) assume f ∈ Wloc for some p > 1 and assume u ∈ Lloc 2
. Then an
application of the divergence theorem along with the Cauchy-Schwartz inequality (and some
propitious re-arrangement) gives
Z Z
(p − 1) η 2 f p−2 |∇f |2 = η 2 ∇f p−1 , ∇f
Z
Div η 2 ∇f p−1 ∇f


=
Z p  Z (31)
p
−1 2 p
2 p−2
− p − 1ηf 2 ∇f, √ f ∇η −
2 η f 4f
p−1
p−1
Z Z Z
2
≤ η 2 f p−2 |∇f |2 + |∇η|2 f p + η2 f p u
2 p−1
which gives us an equation that could be regarded as a form of reverse Sobolev inequality:
Z  2 Z Z
2 p−2 2 2 2 p 2
η f |∇f | ≤ |∇η| f + η2 f p u (32)
p−1 p−1
n
Set γ = n−2 and notice γ > 1 (this is convenient for applications of Hölder’s inequality as
1 2
γ + n = 1). One form of the Sobolev inequality is

Z 1
 2γ Z  12
2γ 2
Cn (Ω) g ≤ |∇g| (33)

9
where g ∈ W01,2 (Ω). Using our “reverse Sobolev inequality” along with the actual Sobolev
inequality gives
Z  γ1 Z
p
2 2γ pγ
Cn (supp η) η f ≤ |∇(ηf 2 )|2

p2
Z Z
≤ 2 |∇η|2 f p + η 2 f p−2 |∇f |2 (34)
2
2 Z!
p2
 Z
p 2 p
≤ 2 1+ |∇η| f + η2 f p u
p−1 p−1

With Hölder’s inequality we obtain

Z  γ1
2 2γ pγ
Cn (supp η) η f
2 ! Z  γ1 Z  n2
p2
 Z
p n
≤ 2 1+ |∇η|2 f p + η 2γ f pγ u2
p−1 p−1 supp η (35)

"  n2 # Z  γ1 2 ! Z
p2
Z 
2 n
2γ pγ p
Cn (supp η) − u2 η f ≤ 2 1+ |∇η|2 f p
p−1 supp η p−1
R n
Thus, provided supp η U 2 is sufficiently small with respect to both p and the Sobolev
constant, then the local Lγp -norm of f is controlled in terms of the local Lp -norm of f .
Lifting apriori from lower Lp -norms to higher Lp -norms is called bootstrapping.

Lemma 1.7 (First Bootstrapping Inequality) Assume 4f + uf ≥ 0, f ≥ 0, f ∈ Lploc

n
for some p > 1, u ∈ Lloc
2
, and η ∈ Cc∞ . Then
p−1
Z
n
u2 ≤ Cn (supp η)2 (36)
supp η 2p2
implies
1  p1 Z  p1
4p2
Z  pγ 
2γ pγ 1
η f ≤ |∇η|2 f p . (37)
p − 1 Cn (supp η)2

Iterating this estimate as many times as necessary, and choosing appropriate test func-
tions at each stage (see below), gives us the following theorem

Theorem 1.8 (First -Regularity Theorem) Assume 4f + f u ≥ 0, f ≥ 0, f ∈ Lploc

n
for some p > 1, u ∈ Lloc 2
, and q is any number q ∈ (p, ∞). Then there exist constants
 = (p, q, Cn (Br )) > 0 and C = C(p, q, Cn (Br ), r) < ∞ so that
Z
n
u2 ≤  (38)
Br

10
implies
Z ! q1 Z  p1
q p
f ≤ C f (39)
Br/2 Br

n
In other words, f ∈ Lploc , p > 1 and u ∈ Lloc2
implies f ∈ Lqloc for all q ∈ (p, ∞). However
both  = (p, q, Cn ) and C = C(p, q, Cn , r) deteriorate as q → ∞ (see the exercises), so we
do not obtain apriori L∞ bounds.

1.6.2 Cutoff Functions

It is worthwhile to formalize the choice of test functions necessary to the proof of Theorem
1.8. The test functions we construct below are called cutoff functions.

Let m ∈ M n be any point, and let r = dist(m, ·) be the distance function from m.
Distance functions are not smooth, but are Lipschitz and therefore weakly differentiable.
Since we encounter at worst gradients of η in our integral estimates, the fact that our “test
functions” are just C 0,1 and not C ∞ is no bother to us.

Given some radius r0 ∈ (0, ∞), we may wish to have a test function that is zero
outside Br0 (m), is 1 inside Br0 /2 (m) and has bounded gradient in the intermediate annulus
Br0 (m) \ Br0 /2 (m). Then define

1
 , r ∈ [0, r0 /2]
2
η(r) = 2 − r0 r , r ∈ (r0 /2, r0 ) (40)

0 , r ∈ [r0 , ∞)


We have ∇η = η 0 ∇r and so this function is Lipschitz, and |∇η| = |η 0 | ≤ 2

r0 (a.e.), as the
gradient of a distance functions has norm 1 (a.e.).

It is also sometimes necessary to have a variety of cutoff functions, with the collection
of cutoff functions operating on progressively smaller nested balls, say. To this end consider
the functions

1
 , r ∈ [0, (2−1 + 2−i−2 )r0 ]
2i+1 −1 −i−2
, r ∈ (2−1 + 2−i−2 )r0 , (2−1 + 2−i−1 )r0 (41)



ηi (r) = 1 − r0 r − (2 − 2 )r0
, r ∈ [(2−1 + 2−i−1 )r0 , ∞).

0


The gradients are


0
 , r ∈ [0, (2−1 + 2−i−2 )r0 ]
2i+1
, r ∈ (2−1 + 2−i−2 )r0 , (2−1 + 2−i−1 )r0


|∇ηi | = (42)
 r0
0 , r ∈ [(2−1 + 2−i−1 )r0 , ∞).


Notice that the supports supp|∇i | are non-overlapping.

11
1.6.3 Nash-Moser Iteration and the L∞ estimates

In addition to f ∈ Lp for some p > 1, now assume u is in a strictly larger Lp space than
p = n2 . We shall assume that u ∈ Lpγ . (Aside: u ∈ Lp+δ for any δ > 0 is sufficient to obtain
all our conclusions; we refer you to the sources (eg. Gilbarg-Trudinger) for the details.)
1 2 2 1
Beginning with (34), we observe γ2 + n + nγ = 1, which allows us to use Hölder’s
inequality with greater subtlety:
Z  γ1
2 2γ pγ
Cn (supp η) η f
2 ! Z
p2
 Z 
p 2 1
  4 2
≤ 2 1+ 2 p
|∇η| f + η γ fpγ · η n fpn · u (43)
p−1 p−1
2 ! Z 1  n2 Z  n2 γ1
p2
 Z  Z
p γ2 n
≤ 2 1+ |∇η|2 f p + η 2γ f pγ η2 f p u2γ
p−1 p−1 supp η

An application of Young’s inequality followed by a mild rearrangement of terms gives

" Z  n2 # Z  γ1
2 1 n
γ 2γ pγ
Cn (supp η) − u 2 η f
γ supp η
 2 ! Z   n2 Z (44)
p 2 p n p 2 2 p
≤ 2 1+ |∇η| f + p 2 η f .
p−1 p−1 n
R n
The benefit is that u 2 γ must now be small only compared to Cn2 , not p. Thus, even
n
though the right side deteriorates (goes to infinity) like p 2 , at least there is hope for getting
p
something of value as p → ∞. From now on, assume p ≥ 2, so p−1 ≤ 2. We obtain our
second, more powerful, bootstrap inequality:

Lemma 1.9 (Second BootstrappingR Inequality) Suppose f satisfies 4f +uf ≥ 0 where

n n 1
f ∈ Lp for any p ≥ 2, and u ∈ L 2 γ . If supp η u 2 γ ≤ 2γ Cn (supp η)2 then
Z 1
 pγ Z  p1
1 2  n 1
η 2γ f pγ ≤ C(n) p Cn (supp η) p sup |∇η|2 + p 2 sup η 2 p fp . (45)
supp η

n
Notice that p 2p is uniformly bounded. This gives us the possibility of iteration without the
deterioration of the estimate that we saw above.

Let ηi , i ∈ {0, 1, 2, . . . } be the ith cutoff function constructed above, and set pi = pγ i .
Also, abbreviate Cn = Cn (Br0 ). Then Lemma 1.9 gives us
!p 1 ! p1
Z i+1
1 2 h n
i p1 Z i

f pi+1 (2i r0 )2 + pi f pi
pi i
≤ C(n) Cn
pi 2
. (46)
B(2−1 +2−i−2 )r B(2−1 +2−i−1 )r
0 0

12
This can be iterated all the way down, to obtain
!p 1 " i # Z ! p1
Z i+1 Y 1 2 h n
i p1
pi+1 pk k 2 k p
f ≤ C(n) Cn (2 r0 ) + pk
pk 2
f . (47)
B(2−1 +2−i−2 )r k=0 B r0
0

Theorem 1.10 (Epsilon-Regularity) Suppose f ≥ 0 weakly satisfies 4f + uf ≥ 0,

n
2γ
and assume f ∈ L2loc and u ∈ Lloc . Given any radius r0 > 0 and point m ∈ M n , and
abbreviating Cn (Br0 (m)) by Cn , there exist constants  = (n, Cn ) and C = C(n, Cn , r0 ) so
that if
Z
n
u2γ ≤  (48)
Br0 (m)

then
Z ! 12
sup |f | ≤ C f2 (49)
Br0 /2 (m) Br0 (m)

Proof. One simply checks that

" i
# " i
#
Y 1 2 h n
i p1 Y 1 2
n n 1
k 2 pγ k
 k 2
(2 r0 ) + p 2 γ k 2 pγ k
pk k
C(n) pk
Cn (2 r0 ) + pk2
= C(n) pγ k Cn (50)
k=0 k=0

is uniformly bounded independently of i (see exercises). Setting p = 2 and taking i → ∞

on both sides of (47) gives the result. 

This method of finding apriori L∞ control after assuming just L2 control is called Moser
iteration or Nash-Moser iteration, depending on whether you’re an analyst or a geometric
analyst.

1.7 Suggested Problems for 1/28

12) Supply a proof for Theorem 1.8. Give explicit estimates for  and C in terms of p, q,
Cn (Br ), and r.
−l
13) By considering functions f : Rn → R of the kind f = r−k (log r) , show that there
exist functions f , u that satisfy the hypotheses of Theorem 1.8 but so that f 6∈ L∞
loc .
Conclude that Theorem 1.8 is optimal. (Hint: computing Laplacians should be easy
after you compute 4r and verify the identity 4f (r) = f 00 (r) + f 0 (r)4r.)
2
Q∞ 1
k  n 1
14) Explicitly estimate k=0 C(n) γ k Cnγ (2k r0 )2 + γ k 2 γ k in terms of n, Cn and r0 ,
thereby completing the proof of Theorem 1.10. You could also try leaving p > 1 and
estimating the product in terms of n, Cn , r0 , and p, thereby bounding |f | in terms of
R p
p1 R 2
12
any f , p > 1, instead of just f .

13
1.8 Variational Problems and Euler-Lagrange Equations

1.8.1 Basic Quadratic Functionals and Linear Euler-Lagrange Equations

Let (M n , g) be a Riemannian manifold, which needn’t be compact. If T is any W 1,2 tensor,

consider the quadratic functional
Z
F(T ) = |∇T |2 dV ol. (51)
Mn

One is sometimes interested in finding minimizers (or more generally extremizers) of such
a functional. To do this, we consider variations Tt = T + tH of the tensor T and compute
the first order effect on F (when M n is non-compact it is normal to put a restriction on H,
namely that H ∈ W01,2 or H ∈ Cc∞ even, in order that we can use Stokes’ theorem). If first
derivatives of F(Tt ) vanish regardless of what variation H is chosen, then the tensor T is
said to be a critical point for F. Then if T is an extremizer, then we compute
Z
d d
0 = F(Tt ) = |∇Tt |2
dt t=0 dt
Z Z (52)
= 2 h∇T, ∇Hi = −2 h4T, Hi .

Since H is arbitrary, we can choose H = η4T , where η ≥ 0 is any test function, to obtain the
requirement that 4T = 0. One easily sees that indeed 4T = 0 if and only if T extremizes F.
In general, the pointwise equations one obtains as a requirement for a quantity to extremize
a functional are called the Euler-Lagrange equations of that functional.

There is a more abstract way to look at this process. Let T be the vector space
of all W 1,2 sections of tensors on M n ; this is a Banach space. Then F : T −→ R is a
map from one Banach space to another, and more specifically, it is just a function on the
(infinite dimensional) space T . As such, we should be able to linearize it (that is, find its
“gradient”). This linearization at T ∈ T is called its Frechet derivative of F at T , and is
sometimes denoted

DF T
: W01,2 −→ R. (53)

The first derivative of F at T in the “direction” of H is

d
DF T
(H) = F(T + tH). (54)
dt t=0

We are looking for those points T ∈ T at which the linearized operator DF|T vanishes.
Unfortunately we won’t have the time to explore this abstract point of view much further
than this.

14
1.8.2 Spaces of Riemannian Metrics, and the Riemannian Moduli Space

We will be applying functionals to the Riemannian structure itself, so we will have to consider
the appropriate space of Riemannian structures. So consider the space of Riemannian
metrics
n O o
2 ∗
M et(M n ) = g ∈ T M g is symmetric and positive definite (55)

Obviously if ϕ : M n → M n is any diffeomorphism the metric pulls back. Thus the diffeomor-
phism group Dif f (M n ) acts on M et(M n ), giving the so-called moduli space of Riemannian
metrics

M od(M n ) = M et(M n ) / Dif f (M n ). (56)

The space M et(M n ) is a cone, not a vector space, and is infinte dimensional. The group
Dif f (M n ) is an extremely complicated infinite dimensional Lie group. It is non-compact,
and otherwise very difficult to understand. Thus M od(M n ) is also difficult to directly
understand. Yet is is our fundamental object, as the curvature functionals are invariant
under diffeomorphism and therefore pass from maps on M et(M n ) to maps on M od(M n ).
J2 V2 V2
Indeed we may regard Rm as an operator Rm : → that is diffeomorphism-
invariant:

Rm(ϕ∗ g) = ϕ∗ (Rm(g)) . (57)

1.8.3 Riemannian Functionals

“Optimal” metrics in Riemannian geometry can also be understood as minimizers of certain

“energy” functionals. The first is the Hilbert functional (aka the total scalar curvature
functional):
Z
H(M n , g) = R dV ol. (58)
Mn

This is scale invariant only in dimension n = 2, in which case it is related to the Euler
constant. To fix this we can normalize by volume
Z
n 1
H(M , g) = n−2 R dV ol (59)
V ol(M n ) n M n

or we could take R itself to some other power

Z
n
R(M n , g) = |R| 2 dV ol (60)
Mn

15
Obviously functionals for other curvature quantities also exists:
Z
n
n
RM(M , g) = | Rm | 2 dV ol
n
ZM
n
n
RIC(M , g) = | Ric | 2 dV ol (61)
n
Z M
n
n
W(M , g) = |W | 2 dV ol.
Mn

In dimension 4 these all take on special interest, as n2 = 2. We obtain scale-invariant

quadratic curvature functionals
Z Z Z
| Rm |2 dV ol, | Ric |2 dV ol, | Rı◦c |2 dV ol
M n n n
Z ZM ZM (62)
2
R dV ol, 2
|W | dV ol, |W ± |2 dV ol.
Mn Mn Mn

1.8.4 Riemannian Variational Formulas and non-linear Euler-Lagrange Equa-

tions

To compute variations of these functionals, we have to compute variations of the usual

Riemannian quantities. Suppose we vary the metric

gt = g + th (63)
dg
where h = hij dxi ⊗ dxj is any
 symmetric 2-tensor. We have dt = h. With the usual formula
1 ∂gis ∂gjs ∂gij
k
Γij = 2 ∂xj + ∂xi − ∂xs g sk we compute

dΓkij
   
1 ∂his ∂hjs ∂hij 1 ∂gis ∂gjs ∂gij
= j
+ i
− g sk − + − g su g vk hsv . (64)
dt 2 ∂x ∂x ∂xs 2 ∂x j ∂x i ∂xs
t=0

For theoretical reasons we know that Γ̇kij is tensorial, so we apply the following standard
trick: express Γ̇kij in a coordinate system in which all Christoffel symbols vanish, then convert
all partial derivatives to covariant derivatives. In geodesic normal coordinates, at any single
∂g
chosen point p we have gij = δij + O(2) so the ∂x terms vanish. Thus we have

dΓkij
 
1 ∂his ∂hjs ∂hij
= + − g sk
dt 2 ∂xj ∂xi ∂xs
t=0
(65)
dΓkij 1
= (his,j + hjs,i + hij,s ) g sk
dt 2
t=0

16
Using the standard formulas for Rmijk l in terms of the Christoffel symbols, we obtain

d 1
Rm ijk l (gt ) = (hjs,ki + hks,ji − hjk,si − his,kj + hks,ij − hik,sj ) g sl
dt t=0 2
d 1 s
Ric ij (gt ) = (h i,js + hs j,is − hjk,s s − (T r h),ij ) (66)
dt t=0 2
d
R(gt ) = hst ,ts − 4(T r h) − Ricij hij
dt t=0

d
From this, along with ol = 12 (T r h) dV ol we can easily compute
dt dV
Z Z  
d 1
R dV ol = Ric − R g , h dV ol (67)
dt t=0 2
so that a stable point for the Hilbert functional occurs when the gravitational tensor van-
ishes:
1
Gij , Ric ij − Rgij
2 (68)
Gij = 0.

1.9 Suggested Problems for 2/2

15) (Other Elliptic Equations) If (M n , g) is a compact Riemannian manifold and f, h
are integrable functions, determine the Euler-Lagrange equations for the quadratic
functional
Z
F(u) = f |∇u|2 + h|u|2 . (69)

16) (The p-Laplacian) Determine the Euler-Lagrange equations for the non-quadratic func-
tionals
Z
F(u) = |∇u|p (70)

where p > 1 and, as usual, we take (M n , g) to be compact. Find the Euler-Lagrange

equations for this functionals; these are known as the p-Laplace equations. Obviously
the 2-Laplacian is just the Laplacian. In the limiting cases p = 1 and p = ∞, what
are reasonable notions of the 1- and ∞-Laplacians?
17) Consider the Riemannian metric on R2
4
gij = δij . (71)
(1 + r2 )2

This pulls back to S2 , under standard streographic projection, to the metric of constant
curvature +1. Given any constant Λ > 0, we have diffeomorphisms R2 → R2 given

17
 by (x, y) 7→ (Λx, Λy). Prove that this lifts to a diffeomorphism on S2 (obviously this
map is differentiable away from the north pole; you have to verify it is differentiable
at the north pole as well). The metric on S2 therefore transforms via diffeomorphism
by following this proceedure: pushforward under standard stereographic projection,
pullback under coordinate multiplication by Λ, and then pullback under standard
stereographic projection. Gaussian curvature is a diffeomorphism invariant, so the
sectional curvatures of these metrics remains +1 (this can also be verified by direct
computation). By letting Λ → 0 or Λ → ∞, show that the set of round metrics of
constant curvature +1 is not compact the C ∞ (or C 0 ) topology. Conclude that the
diffeomorphism group of S2 is non-compact.
18) Show that any linear fractional transformation on R2 ∪ {∗} lifts to a diffeomorphism
on S2 , so that P SL(2, R) ⊂ Dif f (S2 ). With this, find a sequence of diffeomorphisms
ϕi : S2 → S2 such that the sequence of pullbacks ϕ∗i (g) of a round metric g converges
pointwise to zero at every point of S2 .
19) Show that the functionals Rk dV ol are diffeomorphism invariant, and therefore in-
R

deed pass to functionals M od(M n ) → R.

20) Prove formally that each functional is a map M od(M n ) → R.
21) Verify the formulas in (66).
22) Show that the Euler-Lagrange equations for the normalized total scalar curvature
functional are Ricij − R
n gij = 0.

23) Find the Euler-Lagrange equations for the quadratic scalar curvature functional. In
dimension other than 4, normalize this functional, and re-compute the Euler-Lagrange
equations.
24) (Einstein’s Cosmological Constant) If RΛ : M n → R is a function, we have the Hilbert
functional with a cosmological term: (R − Λ)dV ol. Find the Euler-Lagrange equa-
tions. Show that there exist non-Ricci flat solutions, and that for these solutions Λ
must be constant. Such manifolds are called Einstein manifolds.

1.10 Elliptic Equations in the Riemann Tensor and Geometric Apri-

ori Estimates

We have seen that the half-dimension Lp norms play two important roles that appear, at
first to be independent. First, the apriori estimates for 4f ≥ uf require u ∈ Lploc for any
p > n2 , with p = n2 being the critical case where a weak conclusion holds (in the case p < n2
n
no conclusions at all are possible). Second, the L 2 Riemannian norms are scale invariant,
and therefore their values carry intrinsic geometric meaning, as opposed to Lp norms for
other values of p, which can be made to take any value whatever simply by scaling the
metric.

Here we shall see how these two roles combine to produce a remarkable regularity
theory for critical metrics (for simplicity we restrict ourselves to Einstein metrics).

18
1.10.1 Derivative Commutator Formulas

From basic Riemannian geometry we have the definition of the Riemann tensor as a com-
mutator: if X is any vector field then
X l ,ij − X l ,ji = Rmijs l X s . (72)
By computation this can be extended to other tensors:
T l1 ...ls ,ij − T l1 ...ls ,ji = Rmijs k1 T sl2 ...ln + Rmijs l2 T l1 s...ln + . . . + Rmijs ln T l1 ...ln−1 s (73)
and to covectors η = ηi dxi
ηk,ij − ηk,ji = −Rmijk s ηs
(74)
Tk1 ...kn ,ij − ηk,ji = − Rmijk1 s Tsk2 ...kn − Rmijk2 s Tk1 s...kn − . . . − Rmijkn s Tk1 ...kn−1 s .
This can be extended to tensors of any type. For instance
Ta bc ,ij − Ta bc ,ji = Rmija s Ts bc − Rmijs b Ta sc − Rmijs c Ta bs (75)

1.10.2 The Second Bianchi Identity and the Riemann Tensor

Both the first and second Bianchi identities can be derived from the diffeomorphism invari-
ance of the Riemann tensor (Reference: J. Kazdan, 1981). If we regard Rm = Rm(g) as a
second order differential operator on the metric g, then diffeomorphism invariance takes the
form
Rm(ϕ∗ g) = ϕ∗ (Rm(g)) (76)
n n ∗
where ϕ : M → M is any diffeomorphism and ϕ is the pullback. Letting ϕt be a time-
dependent family of diffeomorphisms with a well-chosen variational field, then taking a Lie
derivative on both sides of (76) gives the second Bianchi identity (see exercises). A similar
process gives the first Bianchi identity. In this light the Bianchi identities are understood
as Leibniz rules.

Alternatively, the second Bianchi identity identity can be computed directly from the
definition of Rm, where it ultimately follows from the Jacobi idenity, which itself is both an
expression of diffeomorphism invariance (of the bracket), and also a Leibniz rule.

The 2nd Bianchi identity can be expressed in either of the following two ways:
Rmijkl,m + Rmijlm,k + Rmijmk,l = 0
(77)
Rmijkl,m + Rmjmkl,i + Rmmikl,j = 0.
Tracing once gives
Ricij,k − Ricik,j = Rmsijk ,s (78)
Tracing again gives
R,k = 2Ricsk ,s . (79)

19
1.10.3 The Laplacian of the Riemann Tensor

Combining the 2nd Bianchi identity with the commutator formula, we can obtain and ex-
pression for the Laplacian of the full Riemann tensor. First using the 2nd Bianchi idenitity
we get
(4 Rm)ijkl , g st Rmijkl,st
(80)
= −g st Rmijls,kt − g st Rmijsk,lt
Using the commutator formula on both terms gives
(4 Rm)ijkl = −g st Rmijls,tk − g st Rmijsk,tl
(81)
+ g st Rmkti u Rmujls + . . .
where “. . . ” indicates a further seven quadratic monomials in the Riemann tensor. Whenever
the exact expression doesn’t matter, we shall abbreviate any linear combination of tensor
products and traces of a tensor T with a tensor S simply by S ∗ T . To (81), apply the 2nd
Bianchi identity again to get
(4 Rm)ijkl = g st Rmjtls,ik + g st Rmtils,jk + g st Rmjtsk,il + g st Rmtisk,jl + Rm ∗ Rm
(82)
= −Ricjl,ik + Ricil,jk + Ricjk,il − Ricik,jl + Rm ∗ Rm
The second derivative terms are now entirely on various copies of the Ricci tensor. Schemat-
ically expressed, we have
4 Rm = ∇2 Ric + Rm ∗ Rm . (83)

In the n ≥ 3 Einstein case we have Ric = λg where λ = const, and so we obtain the
tensoral elliptic equation 4 Rm = Rm ∗ Rm. This is difficult to deal with using the above
elliptic theory, however, because it was developed for functional Laplace equations of the
sort 4f ≥ uf . We therefore resort to the so-called Kato inequality: for any tensor T we
have
|∇|T || ≤ |∇T |. (84)
Then the two identities
1
4|T |2 = h4T, T i + |∇T |2
2 (85)
1
4|T |2 = |T |4|T | + |∇|T ||2
2
along with the Kato inequality give
|T |4|T | ≥ h4T, T i . (86)
Using T = Rm and | hRm ∗ Rm, Rmi | ≥ −C| Rm |3 (from Cauchy-Schwartz) we get the
equation
4| Rm | ≥ C(n)| Rm |2 (87)
which holds strongly away from | Rm | = 0 and holds weakly everywhere.

20
1.10.4 Consequences of the Einstein Condition

We have seen that the Einstein condition gives the weak elliptic inequality

4u ≥ −u2 (88)

where u = C| Rm |. Using the Lp theory above, where f = u, gives the statement that there
are constants  = (n, Cs ), = C(n, Cs , r) so that if

Z ! n2
n
| Rm | 2 ≤  (89)
Br (m)

then
  n2 γ1 ! n2
Z Z
n n
γ
 | Rm | 2  ≤ C | Rm | 2 (90)
B 3 r (m) Br (m)
4

4
(this is from Theorem 1.8). It can be shown that the r–dependency of C is r− n (see
n
exercises). But then with a uniform local estimate of the L 2 γ -norm of u we can actually
use the more powerful -regularity theorem (Theorem 1.10) to obtain

Z ! n2 Z ! n2
n n
| Rm | 2 ≤  =⇒ sup | Rm | ≤ C | Rm | 2 (91)
Br (m) B 1 r (m) Br (m)
2

In the exercises, you’ll show that the r-dependence in C is actually r−2 .

This remarkable theorem, due independently to Anderson and Bando-Kasue-Nakajima,

gives local apriori control over the pointwise norm of the Riemann tensor in terms of local
n
L 2 norms of | Rm |.

1.11 Problems for 2/4

25) In this exercise we’ll work through how the diffeomorphism invariance of the various
curvature tensors imply the doubly-traced Second Bianci identity (79). Let ϕt be a
family of diffeomorphisms with variation field X. Consider the expression R(ϕ∗t g) =
ϕ∗t (R(g)), and take the derivative dt
d
t=0
on both sides. On the right side, obtain
simply X(R), and on the left side use the variation formulas for scalar curvature from
(66) along with the identity (LX g)ij = (Xi,j − Xj,i ). Finally, apply a commutator
formula to obtain (66).

26) (Harder, but not terrible) Use the process from Problem 25 and the variation formula
for the Riemann tensor to derive the traced 2nd Bianci and finally the full 2nd Bianchi
identity.

21
 27) Derive the exact expression for Rm ∗ Rm in (83). Then derive an explicit estimate for
C = C(n) in the weak inequality 4| Rm | ≥ −C| Rm |2 .
28) Prove the Kato inequality.
29) Run through the “First Bootstrapping Inequality” (Lemma 1.7) with a proper choice
of η to obtain
  n2 γ1 ! n2
Z Z
γn 4
−n n
 | Rm | 2  ≤ Cr | Rm | 2 (92)
B 3 r (m) Br (m)
4

R n
 n2
for C = C(n, Cn ), under the condition that Br (m)
| Rm | 2 ≤ .

30) Use your result from Problem 29, along with a slightly more careful Moser iteration
process from Theorem 1.10 to obtain

Z ! n2
n
sup | Rm | ≤ Cr−2 | Rm | 2 (93)
B 1 r (m) Br (m)
2

R n
 n2
for C = C(n, Cn ), under the condition that Br (m)
| Rm | 2 ≤ .

1.12 Harmonic Coordinates

The material in this section is due to DeTurck-Kazdan.

1.12.1 Definition and First Uses

One interested in what local coordinates are “best.” Obviously what “best” means will
vary from situation to situation , and geodesic normal coordinates are particularly useful in
certain cases, as we already saw when we computed variation formulas.

However, for investigating regularity issues, geodesic normal coordinates are not good.
If a metric is only C k,α , for instance, then the exponential map causes a derivative loss
when constructing the coordinate frame, so, in fact, a C k,α metric will not appear even to
be C k,α in its own exponential normal coordinate chart (see the DeTurck-Kazdan paper for
details). So if the metric is C k,α , what sort of coordinate chart will show this?

An harmonic coordinate chart is simply a coordinate chart {x1 , . . . , xn } in which each

coordinate is harmonic:

4 xi = 0. (94)

22
 2
Using 4 f = g ij ∂x∂i ∂x
f k ∂f 1 n
j − Γ ∂xk , we easily see that {x , . . . , x } is an harmonic chart if

and only if each of the traced Christoffel symbols Γk = g ij Γkij vanishes. A theorem of
DeTurck-Kazdan states that, for regularity issues, this coordinate frame is the best:

Theorem 1.11 (DeTurck-Kazdan) Let (M n , g) be a differentiable manifold along with

a metric of class C k,α . Assume T is a tensor of class C l,β . If {x1 , . . . , xn } is an harmonic
coordinate chart, then the component functions of T will be differentiable of class C m,γ where
m + γ = min{k + α, l + γ}.

In particular, a C k,α metric will appear to be C k,α in its own harmonic coordinate chart.

1.12.2 The Ricci Operator

The Ricci curvature tensor can be considered a non-linear differential operator that takes
the metric g as its input and produces a symmetric 2-tensor as its output. In harmonic
coordinates this transformation is explicitly a semilinear elliptic differential operator, as we
shall see.

We have
∂Γsij ∂Γssj
Ricij = − + Γstj Γtis − Γsij Γtts . (95)
∂xs ∂xi
We shall be interested exclusively in the second order part, so we shall denote all lower order
terms with the expression Q = Q(∂g, g), which will be quadratic separately in ∂g and in
g, and a polynomial of 4th degree in both together. Expanding out the Christoffel symbols
and making a cancellation we obtain
1 st ∂ 2 gij
 2
∂ 2 gjs ∂ 2 gst

1 ts ∂ git
Ricij = − g + g + − + Q. (96)
2 ∂xs ∂xt 2 ∂xs ∂xj ∂xi ∂xt ∂xi ∂xj
j
∂Γ
Next consider the expression g ui ∂xu . We have

r
 2
∂ 2 gvs ∂ 2 guv

∂Γ 1 ∂ gus
gri j = gri g uv g sr + − + Q
∂x 2 ∂xj ∂xv ∂xj ∂xu ∂xj ∂xs
(97)
∂ 2 gui
 2 
1 ∂ guv
= g uv v j − g uv + Q.
∂x ∂x 2 ∂xi ∂xj
Comparing this with (98) we see that, in general,
∂ 2 gij ∂Γr ∂Γr
 
1 1
Ricij = − g st s t + gri j + grj i + Q. (98)
2 ∂x ∂x 2 ∂x ∂x
Therefore in an harmonic coordinate system this simplifies to
1
Ric ij = − 4c gij + Q (99)
2
2
where 4c indicates the coordinate Laplacian 4c , g st ∂x∂s ∂xt .

23
1.12.3 Bootstrapping Regularity for Einstein Manifolds

Recall that in the case of Einstein metrics in dimension n ≥ 3, we derived apriori local
estimates on the Riemann curvature tensor, obtaining local L∞ bounds. Via the Jacobi
equation, this puts apriori C 2 control on the metric tensor. Now consider, in an harmonic
coordinate system, the equation
1
λgij = − 4c gij + Q, (100)
2
where λ = const, which comes from (99) and the Einstein condition. Because g ∈ C 2
we have Q ∈ C 1 , and that the coefficients on 4c are C 2 . The Schauder apriori estimates
therefore imply that the components gij must be C 2,α functions. But then Q ∈ C 1,α , so
the Schauder theory gives gij ∈ C 3,α . Now Q ∈ C 2,α , so Schauder implies gij ∈ C 4,α .
Continuing in this way, we obtain apriori C ∞ estimates for gij .

Indeed, as g satisfies an elliptic equation whose coefficients are themselves determined

by g, it is possible to prove that the gij functions must be C ω (real-analytic). Thus an
Einstein metric is necessarily real-analytic.

This concludes our investigation of Einstein manifolds from an elliptic-analytic point

of view.

1.13 Problems for 2/9

k
31) In an arbitrary coordinate system, show that
√ the traced Christoffel symbols Γ are
logarithmic derivatives of the volume forms det g.
j
32) (Modified harmonic coordinates) Suppose 4xi + bij ∂x i j i
k ∂xk + cj x + f = 0. Write down
ij i i
some conditions on the functions bk , cj , f so that the Ricci operator is still elliptic
semilinear in the resulting gauge.

1.14 Introduction to Basic Complex Geometry

1.14.1 Almost Complex Structures and Initial Constructions

Many approaches to complex geometry exist; in these notes we’ll choose an approach that
should give differential geometers the gentlest possible introduction. We will regard com-
plex geometry and/or topology to simply be differential geometry and/or topology with an
additional differentiable structure, the so-called almost complex structure J, which is simply
any tensor

J : T M −→ T M such that J ◦ J = −Id. (101)

24
Note that J can be dualized: if η is any covector then we define J(η) to be the covector
η ◦ J (beware: sometimes the convention J(η) = −η ◦ J is used instead).
√
Because J squares to −1 its eigenvalues are ± −1, and its eigenvectors come in com-
plex conjugate pairs. If we define the complexified tangent space to be

TC M , C ⊗ T M (102)

then we have an eigenspace decomposition of TC M into two complex vector bundles

TC M = TC0 M ⊕ TC00 M (103)

√
√ the holomorphic and antiholomorphic tangent bundles (respectively the + −1 and
called
− −1 eigenspaces of J). We often denote these TC0 M = T(1,0) M and TC00 M = T(0,1) M . Sim-
ilarly the complexified cotangent bundle TC∗ M = T (1,0) M ⊕ T (0,1) M splits into eigenspaces.
We have projectors
1 √

π(1,0) : TC M −→ T(1,0) M, π(1,0) = Id − −1J
2
1 √

π(0,1) : TC M −→ T(0,1) M, π(0,1) = Id + −1J
2 (104)
(1,0) (1,0) 1 √
π (1,0) =


π : TC M −→ T M, Id − −1J
2
1 √
π (0,1) : TC M −→ T (0,1) M, π (0,1) =


Id + −1J .
2
The projectors are complimentary in the sense that π(1,0) ◦ π(0,1) = π(0,1) ◦ π(1,0) = 0 and
Id = π(1,0) + π(0,1) , and similarly for the cotangent bundle projectors.

A vector field X is a section of the holomorphic vector bundle if X = π(1,0) (X), or

√
equivalently if π(0,1) (X) = 0. Such a vector field X must have the form X = Y − −1JY
for some real
√ vector field Y . Likewise X is a section of the antiholomorphic vector bundle
if X = Y + −1JY for some real field Y .

1.14.2 Integrability and the Nijenhuis Tensor

Because T(0,1) M is a complex vector bundle, it does not make sense to talk about its
“integrability” in the sense that in spans any kind of submanifold. Yet it makes complete
sense that we can talk about the Frobenius criterion, so we say that T(1,0) M is integrable
√ √
iff [T(1,0) M, T(1,0) M ] ⊆ T(1,0) M . Letting Y − −1JY and Z − −1JZ be sections of the
holomorphic vector bundle, linearity of the bracket gives that
√ √
[Y − −1JY, Z − −1JZ] ∈ T(1,0) M (105)

if and only if

[Y, Z] − [JY, JZ] + J ([Y, JZ] − [JY, Z]) = 0. (106)

25
It is convenient to define the Nijenhuis tensor NJ to be
NJ (Y, Z) = [Y, Z] − [JY, JZ] + J ([Y, JZ] − [JY, Z]) . (107)
Although it may not be obvious, the Nijenhuis tensor is actually linear with respect to
multiplying Y or Z by a function (see exercises). Thus the condition for integrability is that
NJ ≡ 0. An almost complex structure that is integrable is called a complex structure.

1.14.3 Example: Orientable Riemannin Manifolds of Dimension 2

If (Σ2 , g) is an oriented manifold with a Riemannian metric, then a Hodge-∗ exists. On

Vk Vk
any n-manifold we have ∗∗ : → , ∗∗ = (−1)k(n−k) . Therefore on Σ2 we have
∗ ∗
∗∗ : T Σ → T Σ, ∗∗ = −1, is an almost complex structure. Further, since the manifold
is Riemannian, the brackets that constitute the Nijenhuis tensor can be evaluated using
covariant derivatives. This, combined with the covariant-constancy of the Hodge-∗, gives in
fact N∗ ≡ 0. Thus J = ∗ is a complex structure on any oriented Riemannian 2-manifold.
One can verify that ∗ : T ∗ M → T ∗ M is a conformal invariant, so therefore the natural
complex structure obtained from g is also conformally invariant.

Notice that in dimension 2 (and only dimension 2) the Laplace equation 4f = 0 is also
conformally invariant, meaning that if a function f is harmonic with respect to some metric,
it remains harmonic in any conformally related metric. This can be verifies as such: 4f =-
∗d(∗df ) so therefore 4f = 0 is the same as d(∗df ) = 0. But the 1-form ∗df is conformally
invariant (because d is independent of the metric and ∗ is conformally invariant), and so the
equation d(∗df ) = 0 is conformally invariant.

Given any harmonic function f defined on a simply connected domain Ω ⊆ Σ2 , there

is a conjugate harmonic function g, defined implicitly by dg = ∗df (since ∗df is closed, this
follows from the Poincare lemma).

Theorem 1.12 If f is√any harmonic function and g is its conjugate in the sense that
dg = ∗df , then z = g + −1f is a holomorphic function.

Proof. Compute π (0,1) (dz) and verify that it is zero. 

1.14.4 Holomorphic functions, holomorphic coordinate systems, and integra-

bility

A function is called holomorphic if its differential df is a section of the holomorphic cotangent

bundle; that is, if df = π (1,0) (df ).

An excellent question is whether holomorphic functions exist. More broadly, if (M 2n , J)

is an almost complex manifold, one wishes to know when there exist complex functions
√ √
z 1 = x1 + −1y 1 , . . . , z n = xn + −1y n (108)

26
so that the 2n many functions x1 , y 1 , . . . , xn , y n form a coordinate system, and so that the
function {z i } are all holomorphic. When this happens, we say that {z 1 , . . . , z n } form a
holomorphic coordinate system.

This question was answered by Newlander-Nirenberg:

Theorem 1.13 (Newlander-Nirenberg) Let (M 2n , J) be an almost complex manifold.

Then exists a holomorphic coordinate system {z 1 , . . . , z n } near any point p ∈ M if and only
if NJ ≡ 0.

We shall not attempt any sort of proof of this difficult theorem.

One final word. If {z 1 , . . . , z n } and {w1 , . . . , wn } are two different holomorphic coordi-
∂z i
nates systems, then the transition functions ∂w j are in fact automatically holomorphic. If

NJ ≡ 0, the Newlander-Nirenber theorem states that we can cover M n with domains that
each have holomorphic coordinate systems. Since the transitions on the overlaps are auto-
matically holomorphic, this means we have an equivalence amongst three different notions
of a complex manifold. Namely (M 2n , J) is a complex manifold iff the holomorphic tangent
bundle T(1,0) M is integrable iff the Nijenhuis tensor NJ vanishes iff M 2n can be covered by
J-holomorphic coordinate charts with holomorphic coordinate transitions.

1.15 Problems for 2/11

33) Formally show that if a manifold M n admits an almost complex structure, then n is
even.
√
34) Verify that if X is a section of the holomorphic vector bundle iff X = Y − −1JY for
some real vector field Y .
35) Prove that the Nijenhuis tensor is indeed a tensor. In particular, if f is any function,
show NJ (f Y, Z) = f NJ (Y, Z) = NJ (Y, f Z).

36) Show that T(1,0) M is integrable if and only if T(0,1) M is integrable.

37) Verify that on oriented Riemann 2-manifolds (Σ2 , g) we have that ∗ is covariant-
constant, and that the Nijenhuis tensor N∗ actually vanishes.
V1 V1
38) Prove that ∗ : → is a conformal invariant on any oriented Riemannian 2-
manifold.
39) If f is any function, show that df = π (1,0) df are precisely the Cauchy-Riemann equa-
tions. (Hint: do this first for the 2-dimensional case.)

27
1.16 The Bigraded Exterior Algebra, and the ∂ and ∂¯ Operators

1.16.1 The bigraded exterior algebra

The decomposition of TC∗ M into eigenspaces of the J operator implies a refinement of the
usual grading of the exterior algebra.

If J is any almost complex structure, we have separate (complex) vector bundles

T (1,0) M and T (0,1) M , and so we have separate exterior algebras built over them:
^∗,0 n ^ ∞ ^ n ^
M k M k M
= T (1,0) M = T (1,0) M = k,0

k=0 k=1 k=1

n ^ ∞ ^ n ^
(109)
^0,∗ M k M k M 0,k
(0,1) (0,1)
= T M = T M =
k=0 k=1 k=1
V∗ V∗,∗
Of course the full exterior algebra T M , usually denotes has the following decom-
position
^∗,∗ ^∗  
, T (1,0) M ⊕ T (0,1) M
M M k ^
i ^k−i ! (110)
(1,0) (0,1)
= T M⊗ T M .
k i=0
Vi,j Vi Vj Vk Lk Vi,k−i V∗ V∗,∗
We denote = T (1,0) M ⊗ T (0,1) M , so that = i=1 and = =
Ln Vi,j
i,j=0 .

Define operators
^k,l ^k+1,l
∂ : −→ , ∂η = π (k+1,l) (dη)
^k,l ^k,l+1 (111)
∂¯ : −→ , ∂η = π (k,l+1) (dη) .
V0 V1 ¯ However, this does not hold in
The operator d : → obviously decomposes d = ∂ + ∂.
general. All we can say is that
^k,l ^k+l+1
d : → , (112)

¯ For general J, there is no straightforward

but there is no information that d = ∂ + ∂.
decomposition for d. However if J is integrable, the interaction is straightforward,
as we see presently.

Consider a holomorphic coordinate system {z 1 , . . . , z n }. By Newlander-Nirenberg, this

exists if and only if J is integrable. The Cauchy-Riemann equations are
√
¯ = 0,
∂f or Jdf = −1df (113)

28
and therefore dz i = ∂z i . Newlander-Nirenberg says that T(0,1) M is integrable, and a Frobe-
V1,0 V2,0 V1,1
nius theorem says therefore d : → ⊕ (see exercises) and, but taking complex
V0,1 V0,2 V1,1
conjugates, also d : → ⊕ . This means, in particular, that for any function f
we have
¯ )
0 = d2 f = d(∂f ) + d(∂f (114)

But the integrability of J gives that

^2,0 ^1,1
d∂f ∈ ⊕ , so that ¯
d∂f = ∂∂f + ∂∂f
^0,2 ^1,1 (115)
¯ ∈
d∂f ⊕ , so that ¯ = ∂¯∂f
d∂f ¯ + ∂ ∂f
¯

Matching by type, we conclude that

^0,0 ^20
∂∂ : −→ , ∂∂ ≡ 0
^0,0 ^0,2
∂¯∂¯ : −→ , ∂¯∂¯ ≡ 0 (116)
^0,0 ^1,1
¯ + ∂ ∂¯ :
∂∂ −→ , ¯ + ∂ ∂¯ ≡ 0.
∂∂

Vk,l
Theorem 1.14 If J is an integrable complex structure, there is a unique operator ∂ : →
Vk+1,l
that satisfies

1) (Constant linearity) ∂(nη + mγ) = n∂η + m∂γ

2) (Leibniz Rule) ∂(η ∧ γ) = ∂η ∧ γ + (−1)deg η η ∧ ∂γ
V0,0 V2,0
3) (Flatness) The operator ∂∂ : → is precisely zero
4) (Action on functions) ∂f = π 1,0 df .

Proof. See exercises. 

Vk,l
Theorem 1.15 If J is an integrable complex structure, there is a unique operator ∂¯ : →
Vk,l+1
that satisfies

¯
1) (Constant linearity) ∂(nη ¯ + m∂γ
+ mγ) = n∂η ¯
¯ ∧ γ) = ∂η
2) (Leibniz Rule) ∂(η ¯ ∧ γ + (−1)deg η η ∧ ∂γ
¯
V0,0 V2,0
3) (Flatness) The operator ∂¯∂¯ : → is precisely zero
¯ = π 0,1 df .
4) (Action on functions) ∂f

29
Proof. See exercises. 

Vk,l Vk+1,l Vk,l+1

Theorem 1.16 If J is an integrable complex structure, then in fact d : → ⊕
¯
and d = ∂ + ∂.

Proof. See exercises. 

1.17 Holomorphic Vector Bundles and the ∂¯ Operator

1.17.1 Definition of Holomorphic Vector Bundles

A rank k real vector bundle E over M n can be described via charts {Uα }α that cover
M 2n (countably many with finite intersections) and “trivializations” Uα × Rk , along with a
collection of “transition functions”

ρβα : Uα ∩ Uβ −→ GL(R, k) (117)

that satisfy the cocycle conditions that ραβ = ρ−1βα and that on triple intersection ργα =
ργβ ρβα . Then we can form equivalence classes: given (x, v) ∈ Uα × Ck and (y, w) ∈ Uβ × Ck
we say

(x, v) ∼ (y, w) iff x = y and w = ρβα v. (118)

A rank k complex vector bundle is the same thing, except now the transitions are

ρβα : Uα ∩ Uβ −→ GL(R, k) (119)

A complex vector bundle is a holomorphic vector bundle provided the base space M n
is itself a complex manifold and provided, after choosing a basis for Ck and therefore being
able to describe the ρβ,α as matrix-values maps ρβα : Uα ∩ Uβ → Mn×n C
, we have that the
transitions are themselves holomorphic:

∂¯ ρβα = 0. (120)

1.17.2 The ∂¯ operator on holomorphic vector bundles

In the real case, the exterior derivative exists exclusively on the base manifold. By contrast,
in the case of a holomorphic vector bundle E, there does exist a uniquely-defined exterior
derivative

∂¯ : E 7−→ T (0,1) M ⊗ E (121)

30
(however it will be impossible to iterate this operator, so for instance ∂¯∂¯ will not be defined).
If {e1 , . . . , ek } is a basis for Ck , then any section e can be expressed over trivializations
Uα × Ck by e = ai ei and over Uβ × Ck by e = bi fi where the transition is fi = (ρβα )ji ej .

However the transition component functions (ρβα )ij are holomorphic, so by defining
¯ = ∂(a
∂e ¯ i ) ⊗ ei (122)
over a trivialization, we see that this definition is invariant over which trivialization is chosen.
To wit:
¯ = ∂(b
∂e ¯ i ) ⊗ fi
¯ j (ρβα )i ) ⊗ fi
= ∂(a j
(123)
¯
= ∂(a ) ⊗ (ρβα )i fi
j
j
¯ j ) ⊗ ei .
= ∂(a
Thus we see that the definition of ∂¯ is trivialization-independent, so ∂¯ exists as an operator
on E. A section e of E that has ∂e¯ = 0 is called a holomorphic section.

The holomorphic tangent bundle T(1,0) M of course has a ∂¯ operator. A vector field X
¯ = 0.
on M that is a section of T(1,0) M is called an holomorphic vector field provided ∂X

Obviously there are no holomophic functions on a compact complex manifold except

the constants (due to the maximum principle). But there are sometimes holomorphic vector
fields, as you’ll see in the exercises.

1.18 Problems for 2/16

40) Verify, using a Frobenius theorem and some trivial identities, that [T(1,0) M, T(1,0) M ] ⊂
V1,0 V2,0 V1,1
T(1,0) M if and only if d : → ⊕ .
41) Prove Theorems 1.14, 1.15, and 1.16. Hint for Theorem 1.16: compute the sum
∂ i ∂ i 1 1 n n
∂z i dz + ∂ z̄ i dz̄ in terms of the real coframe basis {dx , dy , . . . , dx , dy }.

42) Suppose (M n , J) is a complex manifold, and E is a complex k-vector bundle over

M n . Assume that E has an exterior derivative ∂¯ : E → T (0,1) M ⊗ E with ∂(f
¯ 4) =
¯ ¯
∂f ⊗ e + f ∂e for any section e of E. Prove that, in fact, E is an holomorphic vector
bundle.
43) Consider the 2-sphere S2 , with a complex structure J give by the right-hand rule,
oriented via the inward pointing normal. Prove that the stereographic projections
d
preserve J. Prove that the vector field z dz on C lifts to a smooth field on S2 ; this is
called the “Euler field.” Prove that it is a non-constant holomorphic vector field on
(S, J).
44) Suppose X is a smooth field on any complex√manifold (M n , J) that preserves J in
the sense that LX J = 0. Prove that X − −1JX is a holomorphic vector field.
Conversely, if X is any holomorphic vector field, prove that LRe(X ) J = LIm(X ) J = 0.

31
1.19 Covariant Exterior Differentiation and Curvature of Vector
Bundles

For a moment we retreat from complex geometry as such and develop some ideas in dif-
ferential geometry that will be of use to us, even in the complex case. A connection on a
vector bundle E (whether real or complex) is a map
∇ : E −→ T ∗ M ⊗ E (124)
that obeys a Leibniz rule: if σ is any section of E and f is any function, then
∇(f σ) = df ⊗ E + f ∇σ. (125)
We can combine any connection ∇ on E with the differential operator d on the base manifold
to create covariant exterior differentiation, which is often denoted D. We do this by simply
requiring that a Leibnz rule hold:
^k ^k+1
D : ⊗ E −→ ⊗E
(126)
D(η ⊗ σ) = dη ⊗ σ + (−1)deg(η) η ∧ Dσ.

Suppose {σ1 , . . . , σk } is some local frame for E. Then we define the 1-forms θij by

∇σi = θij ⊗ σj . (127)

The symbols θji are called the connection 1-forms.

Of course iterating the exterior derivative d gives zero: d2 = 0. What happens when
we iterate the covariant exterior derivative D?
 
DD(η ⊗ σ) = D dη ⊗ σ + (−1)deg(η) η ∧ Dσ
= ddη ⊗ σ + (−1)deg(η)+1 dη ∧ Dσ + dη ⊗ σ + (−1)2deg(η) η ∧ DDσ (128)
= η ∧ DDσ
V0
Therefore D2 , while not necessarily zero, is tensorial. In particular, if η ∈ is any function
f , then the computation above shows D2 (f σ) = f D2 (σ). It is usual to denote this tensor
by Θ, and call it the curvature tensor:
^k ^k+2
Θ : ⊗ E −→ ⊗E
(129)
Θ(η ⊗ σ) = D (η ⊗ σ) = η ∧ D2 (σ).
2

This is identical to the more familiar notion of the curvature of line bundles in Riemnnian
geometry (see exercises). We can express this tensor in terms of the connection 1-forms by
 
D2 (σi ) = D θij ⊗ σj
= dθij ⊗ σj − θij ∧ θjk ⊗ σk (130)
 
= dθij + θis ∧ θsj σj

32
Therefore

Θji = dθij + θis ∧ θsi (131)

This is often abbreviated Θ = dθ + θ ∧ θ.

1.20 Holomorphic Vector Bundles and the Chern Connection

The theory of covariant exterior differentiation dovetails nicely with the holomorphic vector
bundle theory, in the sense that holomorphic vector bundles automatically have a natural
connection, the ∂¯ operator, as we have already discussed. A connection ∇ : E → T M ⊗ E
is said to be compatible with the holomorphic structure of E provided ∇0,1 = ∂. ¯ Then if E
has an hermitian inner product h·, ·i, we obtain the following uniqueness theorem:

Theorem 1.17 (Fundamental Theorem of Hermitian Holomorphic Vector Bundles)

If E is a holomorphic vector bundle over E with Hermitian inner product E, then there is
a unique connection ∇ on E that satisfies

• ∇ : E −→ T ∗ M ⊗ E
• (Leibniz rule) ∇(f σ) = df ⊗ σ + f ∇σ
• (Compatibility with the Hermitian inner product) d hσ1 , σ2 i = h∇σ1 , σ2 i + hσ1 , ∇σ2 i
¯
• (Compatibility with the holomorphic structure) ∇0,1 σ = ∂σ.

Before moving on, we should note that the curvature of such

V1,1 an Hermitian holomorphic
vector bundle has a spacial bidegree decomposition: Θij ∈ . Further, in a holomorphic
frame Θij is purely imaginary and antihermitian in the sense that Θi̄ = −Θjı̄ .

This can be seen in two steps. Let {σi } be a holomorphic frame, meaning ∂σ ¯ = 0.
1,0 i
V2,0 V1,1 V2,0 V2,0 i V1,1
Therefore ∇σi = ∇ σi implies dθj ∈ ⊕ and θ ∧ θ ∈ so that Θ ∈ ⊕ .
Then we note
0 = d2 hσi , σj i
= d hDσi , σj i + d hσi , Dσj i
(132)
= hΘ(σi ), σj i − hDσi , Dσj i + hDσi , Dσj i + hσi , Θ(σj )i
= hΘ(σi ), σj i + hσi , Θ(σj )i

so that 0 = Θsi hs̄ + Θsj his̄ = Θsi hs̄ + Θsj hsı̄ .

Now we specialize to the case that (M, J) is a complex manifold, so T1,0 M is a holo-
morphic vector bundle and has a ∂¯ operator. Suppose (M, J, h) is a complex manifold
with Hermitian inner product h. Then M has a natural connection, ∇C , called the Chern
connection—in general this is different from the Levi-Civita connection.

33
 We would like to establish the conditions under which the ∇C = ∇L−C . That is, the
conditions under which the holomprhic theory of curvature and the real (Riemnnian) theory
coincide. This question leads us to the topic of Kähler geometry.

1.21 Problems for 2/18

∂
45) Suppose ∇ is a connection on T M , and { ∂x i } is a coordinate frame. Show that the

connection 1-forms are just the Christoffel symbols: θjk = Γkij dxi .

46) In the Riemannian setting where E = T M , show that Θlk is indeed the Riemannian
curvature operator.
47) Consider the 2-sphere, with metric given by gij = (1 + r2 )−2 δij . Determine the
connection 1-forms θji and the curvature tensor Θij .
48) In the real setting, what can you say about the traced curvature 2-form ρ = Θii ?
49) Prove the second Bianchi identity: dΘ + θ ∧ Θ + Θ ∧ θ = 0. Hint: start by showing
D (Θ(σ)) = Θ(D(σ)), and work out what this means in terms of the θ-symbols.
50) Starting with dhi̄ = d hei , ej i = θis hs̄ + θjs hs̄ and pairing off by bi-degree, show that
θji = his̄ ∂hj s̄ and that Θij = ∂¯ his̄ ∂hj s̄

1.22 Chern Classes and Kähler Manifolds

1.22.1 The Ricci form and the First Chern Class

We shall not attempt to develop the theory of characteristic classes or even Chern classes
of complex vector bundles, but shall content ourselves with describing the first Chern class
of a holomorphic vector bundle.

Given an Hermitian holomorphic vector bundle (E, ∂) ¯ over a holomorphic manifold

n
M of complex dimension n, we have the curvature operator
^2
Θ ∈ ⊗ E∗ ⊗ E
(133)
Θ = Θij ⊗ ej ⊗ ei .
where {ei } is any local holomorohic basis for E and {ei } is the corresponding dual basis for
E ∗ . If h = hij̄ ei ⊗ ej is the Hermitian form, then we have the array of 2-forms Θij given by

Θij = ∂¯ his̄ ∂hj s̄

(134)
(see previous exercises). But we can trace Θ in its E ∗ ⊗ E factor to obtain (a complex
multiple of) the Hermitian holomorphic vector bundle’s Ricci 2-form, which we define to be
√ ^2
ρ = −1Θii ∈ (135)

34
Recall the Jacobi formula for matrices: if A = (Aij ) is a matrix dependent on some variable
d dA
x, then dx log A = Aji dxij . Using this, we obtain
√ √
−1∂¯ his̄ ∂his̄ = − −1∂ ∂¯ log det(h),


ρ = (136)

which is clearly a (1, 1)-form. We shall show below that it is a real (1,1)-form, but for now
¯
we will take this for granted. Because ∂∂(f ) = d∂(f ) (as a result of ∂ 2 = 0, we see that
ρ is locally exact, and therefore it is globally closed. Therefore [ρ] represents a deRham
cohomology class.

However, it is not yet clear that this class represents anything fundamental, because
within the definition of ρ is a choice of Hermitian form h on E. So suppose h and k are two
Hermitian forms on E. Then we can connect them through a path:

ht , (1 − t)h + tk. (137)

Note that ht is an Hermitian metric for all t ∈ [0, 1], and that h0 = h and h1 = k. For each
ht we have a new ρt , but we see
d √ d √
ρ = − −1∂ ∂¯ log det ht = − −1∂ ∂¯ (ht )i̄ (ki̄ − hi̄ )


dt √ dt (138)
= d − −1∂¯ T rht (k − h)

d
But T rht (k − h) is a globally defined function. Therefore dt ρt is an exact 2-form, which
means that the class [ρt ] is actually invariant. The class [ρ] is called the first Chern class of
¯ and denoted c1 (E). This gives the following remarkable theorem:
(E, ∂),

Theorem 1.18 (The First Chern Class) Assume (E, ∂) ¯ is a holomorphic vector bun-
dle. Given any Hermitian form h on E, the resulting Ricci 2-form ρ is real and closed. The
2
resulting deRham class [ρ] ∈ HdR (M n , R) is invariant under choice of Hermitian form h,
and is therefore a constant of the holomorphic structure ∂¯ of the bundle E.

1.22.2 Hermitian, Metric, and Kähler forms on Complex Manifolds

The theory of Chern curvature obviously exists when we choose E = TC0 M , the holomorphic
tangent bundle. The resulting first Chern class c1 (T M ) = c1 (M ) is referred to as the first
Chern class of the manifold itself.

Let h = hi̄ dz i ⊗ dz j be an Hermitian metric on the complex n-manifold M n . We can

separate h into its symmetric and anti-symmetric parts:

h = hi dz i ⊗ dz̄ j
1 1 (139)
= hi dz i z̄ j + hi dz i ∧ dz̄ j
2 2

35
We make the definitions
1
g = hi̄ dz i z̄ j
2
√ (140)
−1
ω = hi̄ dz i z̄ j
2
√
so that h = g − −1ω. Notice that, since hi̄ = hjı̄ , we have
1 1
g =hi̄ dz i dz̄ j = hjı̄ dz̄ i dz j
2 2 (141)
1 j i
= hjı̄ dz dz̄ = g.
2
Likewise we can compute ω = ω, so ω is also real. The real symmetric tensor g is called
the Riemannian metric associated to h, and the real antisymmetric 2-tensor ω is called the
Kähler form associated to h. We can prove that ω(·, ·) = g(J·, ·) (see exercises).

Further, it is a simple matter of algebra to prove that

1
dV olg = ω ∧ ··· ∧ ω (142)
n!
(n-fold wedge product), and that
1
∗ω = ω ∧ ··· ∧ ω (143)
(n − 1)!
((n − 1)-form wedge product), so that |ω|2 = ∗ (ω ∧ ∗ω) = n.

1.22.3 The Kähler condition and Initial Considerations

Every complex manifold has at least one Hermitian inner product; this can be obtained, for
instance, by using creating locally defined Hermitian forms, and gluing them together with
a partition of unity. But this is too “flexible” to be of further use.

If the Kähler 2-form ω satisfies the Kähler condition that

dω = 0, (144)
then we say that (M n , J, ω) is a Kähler manifold, and that the associated metric g is a
Kähler metric.

This is actually a restrictive condition, and not every complex manifold admits a Kähler
metric. For instance, if M n is compact, note that
Z
n 1
Vol(M ) = ω ∧ · · · ∧ ω. (145)
n! M n
However,
R if ω = dη is actually exact, then ω∧· · ·∧ω = d(η∧ω∧· · ·∧ω) is also exact, meaning
ω ∧ · · · ∧ ω = 0, an impossibility. Therefore ω is closed but not exact, and therefore defines
a non-zero cohomology class [ω], called the Kähler class of the manifold.

36
1.22.4 Various Forms of the Kähler Condition, and the Equivalence of Real
and Complex Geometry on Kähler manifolds

√
Setting ω = ωi̄ dz i ∧ dz̄ j where of course ωij̄ = −1hi̄ , the Kähler condition is

dωi̄ k dωi̄ k
0 = dω = dz ∧ dz i ∧ dz̄ j + dz̄ ∧ dz i ∧ dz̄ j . (146)
dz k dz̄ k
for this to be zero, it is necessary and sufficient that
dωi̄ dωk̄
k
= (147)
dz dz i
for all i, j, k. But this has an expression in terms of Christoffel symbols: because ωi̄ =
√
−1hi̄ , we have
   
d d d d d
hi̄ = ∇ d , + , ∇ d/dz k dz j
dz k dz k dz i dz j dz i
= Γski hs̄ (148)
d
hk̄ = Γsik hs̄ .
dz i
Therefore the Kähler condition is equivalent to Γtik = Γtki . However this is precisely the
torsion-free condition of Riemannian geometry.

Theorem 1.19 The Chern connection on an Hermitian complex manifold is the Levi-Civita
connection if and only if the metric is Kähler.

Theorem 1.20 An Hermitian complex manifold (M n , J, h) is Kähler if and only if J is

covariant-constant with respect to the Chern connection: ∇Y (JX) = J∇Y X.

Proof. Exercise. 

1.23 Problems for 2/23

51) Prove that ω = ω, so that ω is a real 2-form.
52) Defining g, ω from h as above, prove that g(J·, J·) = g(·, ·) and ω(J·, J·) = ω(·, ·).
53) Defining g, ω from h as above, prove that ω(·, ·) = g(J·, ·).

ω(·, ·) =
54) Prove the converse of (52) and (53), namely if g is J-invariant, then setting √
g(·, ·), show that ω is a J-invariant, antisymmetric 2-form, and that h = g − −1ω is
an Hermitian metric.

37
 55) If (M n , J, g) is a complex manifold with metric g, show that this is a Kähler manifold
provided the two compatibility conditions g(J·, J·) = g(·, ·) and dg(J·, ·) = 0 both
hold.
dωi̄ dωk̄
56) Verify the assertion before (147). In particular, show that dz k
= dz k
is equivalent
dω
to dz̄i̄
k = dω ik̄
dz̄ j

57) Prove Theorem 1.20.

1.23.1 The Hodge Stars ∗ and ¯∗, and the Bidegree Decomposition

1.23.2 ¯
The ∂ ∂-operator

1.23.3 Signed Forms. Pluriharmonic and Plurisubharmonic functions

1.23.4 Signed Cohomology Classes

1.23.5 The Calabi Conjecture for Kähler-Einstein Metrics

1.24 Exercises for 2/25

38