B3.

2 GEOMETRY OF SURFACES
Dictionary of some terminology from topology and analysis
Comments and corrections are welcome: [email protected]
Prof. Alexander F. Ritter, Mathematical Institute, Oxford.

1. Topology: a dictionary
⋄ A topological space is a set X and a collection of subsets of X called open sets such that:
(1) the empty set is open,
(2) the whole set is open,
(3) a finite intersection of open sets is open,
(4) an arbitrary union of open sets is open.
Example. A metric1 space (X, d) is a topological space: the open sets are any union of balls
Br (x) = {y ∈ X : d(x, y) < r} (for centres x ∈ X, radii r > 0).
⋄ Convention: our spaces are always understood to be topological spaces.
⋄ A subset is called closed if it is the complement of an open set.
⋄ A neighbourhood2 of x ∈ X is a subset which contains an open set U with x ∈ U .
⋄ A map f : X → Y is continuous if f −1 (open set) is always open.
(1) A composition of continuous functions is continuous,
(2) f continuous ⇒ f (compact subset) is compact,
(3) f continuous ⇒ f (connected subset) is connected.
(4) Continuous bijection from a compact space to a Hausdorff space ⇒ homeomorphism.
(5) Continuous surjection from a compact space to a Hausdorff space ⇒ quotient map.
⋄ f : X → Y is a quotient map if U ⊂ Y open ⇔ f −1 (U ) ⊂ X is open.
⋄ X is Hausdorff if any two points can be separated by open sets.3
⋄ X is compact if every open cover by open sets has a finite subcover.4
⋄ Heine-Borel theorem: subsets of Rn are compact ⇔ they are closed and bounded.
⋄ Example. For metric spaces X, Y :
(1) A subset S ⊂ X is closed ⇔ S ∋ xn → x implies x ∈ S.
(2) A map f : X → Y is continuous ⇔ f (xn ) → f (x) whenever xn → x.
(3) X is automatically Hausdorff.
(4) X is compact ⇔ any sequence has a convergent subsequence.5
⋄ X is connected if every continuous function f : X → Z is constant.
⋄ X is path-connected if any two points are joined by a continuous path.6
⋄ X path-connected ⇒ X connected, but the converse is false in general.7
⋄ X is simply-connected if it is connected and any loop in X is contractible.8
⋄ A continuous deformation of f : X → Y is a continuous map F : X × [0, 1] → Y with

Date: This version of the notes was created on September 22, 2016.
1So a function d : X × X → R with d(x, y) = d(y, x) ≥ 0 with equality if and only if x = y, and such that
the triangle inequality holds: d(x, y) ≤ d(x, z) + d(z, y).
2We do not require the neighbourhood to be an open set. We say open neighbourhood in that case.
3For any x, y ∈ X, there are open sets U , U containing x, y respectively, with U ∩ U = ∅.
x y x y
4So if X = ∪U for some open sets U , then X = U ∪ · · · ∪ U
i i i1 im for some indices i1 , . . . , im .
5So x ∈ X implies x
n nj → x ∈ X for some n1 < n2 < · · ·
6For any x, y ∈ X there is a continuous map f : [0, 1] → X with f (0) = x, f (1) = y.
7The two notions become equivalent if you assume the space is locally path-connected. This means:
for any x ∈ X and any open U containing x, there is an open V ⊂ U which is path-connected, with x ∈ V .
8So for any continuous f : S 1 → X there is a continuous F : D → X with F | 1
S 1 = f . Here S = {z ∈ C :
|z| = 1} is a circle, D = {z ∈ C : |z| ≤ 1} is a disc. By parametrizing D by sz with z = eit ∈ S 1 , s ∈ [0, 1], you
can view F as a family of loops F : S 1 × [0, 1] → X from the constant loop F0 = F (·, 0) to F1 = F (·, 1) = f .
1
2 B3.2 GEOMETRY OF SURFACES, PROF. ALEXANDER F. RITTER

F (x, 0) = f (x). So Fs (x) = F (x, s) is a family of maps, F0 = f , and F1 is the deformed map.
⋄ A map f : X → Y is bijective if there exists a map g : Y → X such that f ◦ g = idY and
g ◦ f = idX are the identity maps. Such a g is unique and called the inverse g = f −1 .
⋄ A homeomorphism f : X → Y is a continuous bijection, with continuous inverse f −1 .
⋄ X, Y are homeomorphic if there exists a homeomorphism f : X → Y .

2. Analysis
n m
⋄ f : R → R is continuously differentiable if all first order partial derivatives exist and
are continuous.1
Explicitly: in coordinates: x = (x1 , . . . , xn ) ∈ Rn maps to f (x) = (f1 (x), . . . , fm (x)) ∈ Rm
for some functions fi : Rn → R, called the components of f .
∂fi ∂fi
So we require that ∂x j
exist and are continuous for all i, j. We abbreviate ∂xj fi = ∂x j
.
⋄ The Jacobian matrix of f : Rn → Rm is the matrix A(x) = (Aij (x)) = (∂xi fj ) of partial
derivatives:  
∂x1 f1 ∂x2 f1 · · · ∂xn f1
 ∂x1 f2 ∂x2 f2 · · · ∂xn f2 
A(x) =  ···


∂x1 fm ∂x2 fm · · · ∂xn fm
The linear map given by “multiplication by A(x)” is the derivative map
Df : Rn → Rm , v 7→ Dx f · v = A(x)v.
Example. For f : R → R, A(x) = (f ′ (x)), Df : R → R is multiplication by f ′ (x).
⋄ Chain rule: Compositions of differentiable maps are differentiable and D(g ◦ f ) = Dg ◦ Df :
Dx (g ◦ f ) = Df (x) g ◦ Dx f.
Example. For f, g : R → R recall (g ◦ f )′ (x) = g ′ (f (x)) · f ′ (x).
⋄ Convention: the vector ∂xj f denotes the j-th column of that matrix.
⋄ Example. Linear maps L : Rn → Rm are differentiable with derivative map L. The whole
point of the derivative map is to find the best linear approximation to a map: f (x) = f (p) +
error
Dp f · (x − p) + error, where kx−pk → 0 as x → p.
⋄ f : Rn → Rm is smooth if it has partial derivatives of all orders (they are automatically
continuous).2
Fact: for smooth functions, partial derivatives commute, e.g. ∂x1 ∂x2 f = ∂x2 ∂x1 f .
For open U, V ⊂ Rn , f : U → V is a diffeomorphism if f is a homeomorphism, and f, f −1
are smooth.
⋄ Integration by substitution (change of variables): If f : V → U is a diffeomorphism,
for open subsets U, V ⊂ Rn , and G = G(x1 , . . . , xn ) : U → R is a smooth function, then
Z Z
G(x) dx1 · · · dxn = G(f (y)) | det Dy f | dy1 · · · dyn
U V
Examples.

1The reason for requiring that the partial derivatives are also continuous is necessary to ensure that the
error
derivative map exists, in the sense that f (x + h) − f (x) = Dx f · h + error, where khk
→ 0 as h → 0.
2For example, for the second order, it means: A : Rn → Rnm , x 7→ A(x) is differentiable. As you increase
the order, this becomes complicated since you choose the succession of which partial derivatives to take.
 B3.2 GEOMETRY OF SURFACES, PROF. ALEXANDER F. RITTER 3

2
(1) Let f be the change of variables from polar
coordinates r, θ to (x, y) inR R . So f (r, θ) =
cos θ −r sin θ
R(r cos θ, r sin θ), so Df = sin θ r cos θ , so | det Df | = r, hence G(x, y) dx dy =
G(r cos θ, r sin θ) r dr dθ.
(2) If γ = γ(t) : [0, 1] → R2 is a smooth curve, and f = f (s) : [a, b] → [0, 1] reparametrizes
R
time (so any strictly increasing smooth function), then the length of the curve, |speed| d(time),
R1 Rb
is well defined independently of the way we parametrize time: 0 kγ ′ (t)k dt = a kγ ′ (f (s))k f ′ (s) ds.
⋄ f : Rn → Rn is a local diffeomorphism near p, if there are open neighbourhoods U, V of
p, f (p) respectively such that the restriction f |U : U → V is a diffeomorphism.
⋄ Convention: we say f : Rn → Rm is defined near p to mean: there is an open set U ⊂ Rn
containing p such that f : U → Rm is defined. We say “for x, y close enough to p, f (p)” to
mean: there are open neighbourhoods U, V of p, f (p) and the statement holds for x ∈ U, y ∈ V .
⋄ Inverse function theorem: Let f : Rn → Rn be a smooth map defined near p ∈ Rn .
If Dp f is invertible, then f is a local diffeomorphism near p.
Explicitly: the theorem hands us a unique smooth map g : Rn → Rn defined near f (p) such
that f (g(y)) = y and g(f (x)) = x (for all x, y close enough to p, f (p) respectively).
Arguably the most important theorem in analysis. It says simple linear algebra (the non-
vanishing of the determinant of a matrix) ensures the smooth invertibility of the map, locally.

⋄ Let f : Rn → Rm be smooth, and n ≥ m. We want to describe the solutions of f (x) = c

near a given solution f (a) = c, where x, a ∈ Rn and c ∈ Rm .
Implicit function theorem: If m columns of Da f are linearly independent, then the vari-
ables xi1 , . . . , xim corresponding to those columns are redundant. Namely, they can be re-
placed by unique smooth functions gi1 , . . . , gim , depending only on the remaining variables,
defined near x = a and satisfying gi1 (a) = ai1 , . . . , gim (a) = aim , so that
f (x)|(xi =c
1 =gi1 , ..., xim =gim )

describes all solutions x near a.

Examples. Below, we seek solutions of f = 0 near x = (0, . . . , 0).
(1) f (x, y) = y: ∂y f = 1 6= 0, so f (x, g(x)) = 0 (indeed g(x) = 0).
(2) f (x, y) = x2 − y: ∂y f = −1 6= 0, so f (x, g(x)) = 0 (indeed g(x) = x2 ).
(3) f (x, y)p= (x + 1)2 − 1 + y 2 : ∂x f |x=0,y=0 = 2 6= 0, so f (g(y), y) = 0 (indeed g(y) =
−1 + 1 − y 2 , which is defined near y = 0, and notice g(0) = 0).
Proof of the implicit function theorem: by relabeling coordinates, we may assume the
last m columns of Da f are linearly independent. Abbreviate k = n − m. Consider F :
Rn → Rn , F (x1 , . . . , xn ) = (x1 , . . . , xk , f (x1 , . . . , xn )). Notice that Da F is invertible (try
writing the matrix). Apply the inverse function theorem. Then F −1 (x1 , . . . , xk , c1 , . . . , cm ) =
(x1 , . . . , xk , gk+1 , . . . , gn ) for unique functions gk+1 , . . . , gn of x1 , . . . , xk , c. 
Smooth dependence on c in the implicit function theorem: Notice above gi1 , . . . , gim
depend smoothly on c. So there are unique smooth functions Gi1 , . . . , Gim : Rn−m × Rm → R
depending only on non-redundant xj variables and y ∈ Rm , defined near x = a, y = c so that
f (x)|(xi =y
1 =Gi1 , ..., xim =Gim )

describes all solutions of f (x) = y for x near a, and y near c.

⋄ A change of coordinates near x = a means a local diffeomorphism ϕ : Rn → Rn near x = a.
A map f : Rn → Rm becomes fe = f ◦ ϕ in the new coordinates. So fe(z) = f (x) for x = ϕ(z).
4 B3.2 GEOMETRY OF SURFACES, PROF. ALEXANDER F. RITTER

⋄ Nonlinear coordinates in the implicit function theorem: There is a change of coor-

dinates of Rn near x = a, and we call the new coordinates z1 , . . . , zn non-linear coordinates,
so that solutions of fe(z) = y near z = ϕ−1 (a) are precisely described by the vanishing
z1 = 0, . . . , zm = 0 of m coordinates (and the other zj coordinates are free).
Proof. First permute coordinates of Rn so that we may assume the i1 , . . . , im above are
1, . . . , m. Then put z1 = x1 − g1 , . . ., zm = xm − gm , and the other zj = xj . 

3. Complex analysis
⋄ A function f : C → C is holomorphic if it is complex differentiable.1
⋄ Fact: f : C → C is holomorphic if and only if F : R2 → R2 , F (x, y) = (f1 (x + iy), f2 (x + iy))
is differentiable with continuous partial derivatives and satisfies
DF ◦ J = J ◦ DF
0 −1


where J = (the matrix which rotates by 90◦ ) corresponds to multiplication by i when
1 0
2
we identify R ≡ C, (x, y) ≡ x + iy.
Remark. DF ◦ J = J ◦ DF ⇔ Cauchy-Riemann equations ∂x f1 = ∂y f2 , ∂y f1 = −∂x f2 hold.
     
∂x f1 ∂y f1 ∂x f1 −∂x f2 cos θ − sin θ
DF = = =r
∂x f2 ∂y f2 ∂x f2 ∂x f1 sin θ cos θ
where r, θ are determined by f ′ (z) = reiθ . Notice Det DF = |f ′ (z)|2 = r2 .
⋄ Fact: f holomorphic ⇒ the above F : R2 → R2 is smooth.
⋄ Fact: f holomorphic near p ⇒ f has an absolutely convergent Taylor series2 at p and f is
equal to its Taylor series near p.
⋄ Identity theorem. If f, g : C → C are holomorphic near p, and there is a sequence
p 6= zn → p with f (zn ) = g(zn ), then f = g near p.
⋄ f : C → C is a biholomorphism if it is bijective and f, f −1 are both holomorphic.
Remark. Since the derivative map is a composition of scaling and rotation, it preserves angles
between vectors. So biholomorphisms are conformal maps, meaning they preserve angles.
⋄ Inverse function theorem. For a holomorphic function f : C → C defined near p, if
f ′ (p) 6= 0 then f is a local biholomorphism near p.
Explicitly: the theorem hands us a unique holomorphic g : C → C defined near f (p) such that
f (g(w)) = w and g(f (z)) = z (for all z, w close enough to p, f (p) respectively).
⋄ Riemann mapping theorem. If U 6= ∅, C is a simply connected open subset of C then
there is a biholomorphism f : U → D onto the open unit disc D = {z ∈ C : |z| < 1}.

4. Differential equations
⋄ For smooth smooth V : Rn → Rn , a flowline γ : [a, b] → Rn is a solution of γ ′ (t) = V (γ(t)).
Idea: V is a vector field (a vector at each point of Rn ), γ is a curve running in the V -direction.
⋄ Theorem. For each point p ∈ Rn there is a flowline γ : (−ε, ε) → Rn of V with γ(0) = p,
for small enough ε > 0. Moreover, γ is smooth, unique and depends smoothly3 on p.

1Meaning f ′ (z) = lim f (z+h)−f (z)

h→0 h
exists. Take h = t ∈ R, let t → 0: then f ′ (z) = ∂x f = ∂x f1 + i∂x f2 .
Take h = it ∈ iR, let t → 0: f ′ (z) = −i∂y f = ∂y f2 − i∂y f1 . Equating gives the Cauchy-Riemann equations.
2P∞ a (z − p)n with a = f (n) (p)/n!
n=0 n n
3Meaning: there is a smooth map F : (−ε, ε) × U → Rn , called flow, defined on a small enough neigh-
bourhood U of p (and ε > 0 depends on U ), such that t 7→ F (t, q) is the flowline of V through q = F (0, q).