A FEW APPLICATIONS OF DIFFERENTIAL FORMS

MATTHEW CORREIA

Abstract. This paper introduces the concept of differential forms by defin-

ing the tangent space of Rn at point p with equivalence classes of curves and
introducing the cotangent space as the dual of the tangent space. The first ap-
plication presented is a formalization of the separation of variables technique
for solving differential equations which is used in many introductory calculus
classes. The next application is a proof of the Fundamental Theorem of Alge-
bra by using a 1-form to detect the winding of the image under a polynomial
of a large and small circle in the complex plane. Finally, the Gauss-Bonnet
theorem is presented using more intuitive definitions rather than by formal
proof.

Contents
1. Introduction to Differential Forms 1
2. Using Differential Forms to Solve Differential Equations 4
3. A Proof of the Fundamental Theorem of Algebra 6
4. The Gauss-Bonnet Theorem 8
Acknowledgments 10
References 10

1. Introduction to Differential Forms

We begin our discussion by introducing the concept of differential forms.
Definition 1.1. A curve in Rn centered on p is a smooth map γ : R → Rn such
that γ(0) = p.
Suppose we have a curve in Rn centered on p, γ, such that γ(t) = (γ1 (t), · · · , γn (t)).
Then we can define the tangent vector to γ at p in the following way:
 
dγ1 (t)

 dt 0  
Definition 1.2. The tangent vector to γ at p is the vector  ..
 with
 
 .  
 dγn (t) 
dt 0
origin at p.
In order to define a tangent space with some generality, we will first introduce a
relation ∼.

Date: August 28, 2015.

1
2 MATTHEW CORREIA

n
Definition 1.3. If α and β are  both curves
  in R  centered on p, we say α ∼ β if
dαi (t) dβi (t)
for all i ∈ {1, · · · , n} we have = .
dt 0 dt 0
We can easily check that ∼ is reflexive, symmetric, and transitive. Thus, ∼ is
an equivalence relation and we will now refer to curves in Rn centered on p within
their equivalence classes under ∼. With this, we are ready to define the tangent
space of Rn at p.
Definition 1.4. The tangent space of Rn at p is the set of all equivalence classes
of curves in Rn at p and is denoted by Tp Rn .
Remark 1.5. Tp Rn is a vector space. If [γ], [δ] ∈ Tp Rn , then we define
[γ] + [δ] = [γ + δ],
where
(γ + δ)(t) = (γ1 (t) + δ1 (t), · · · , γn (t) + δn (t)).
It is a simple matter to check that for any α ∈ [γ] and β ∈ [δ] we will have
α + β ∈ [γ + δ]. Similarly, scalar multiplication is defined coordinate-wise, such
that for any c ∈ R we have c[γ] = [cγ] where cγ(t) = (cγ1 (t), · · · , cγn (t)).
Now, we introduce a bilinear map:
Definition 1.6. We define D(·) (·) : Tp Rn × C ∞ (Rn ) → R such that
D[γ] (f ) = (f ◦ γ)0 (0).
Remark 1.7. Notice that for fixed [γ] and for f, g ∈ C ∞ (Rn ) we have
D[γ] (f ∗ g) = f (p) ∗ D[γ] (g) + g(p) ∗ D[γ] (f ).
Definition 1.8. For any fixed p = (p1 , · · · , pn ) ∈ Rn , ei will refer to the path
ei : R → Rn such that ei (t) = (p1 , · · · , pi−1 , pi + t, pi+1 , · · · , pn ).
 ∂f 
Remark 1.9. Notice that for any f ∈ C ∞ (Rn ) we have D[ei ] (f ) = . Fur-
∂xi p
n
thermore, the set {[e1 ], · · · , [en ]} forms a basis of Tp R . Often, this basis will be
 ∂ 
referenced with the notation for [ei ].
∂xi p
Definition 1.10. The cotangent space at p is defined to be the dual space of Tp Rn ,
that is the set of all linear functions from Tp Rn to R. We denote the cotangent
space at p by (Tp Rn )∗
Remark 1.11. We will write {(dx1 )p , · · · , (dxn )p } as a basis of (Tp Rn )∗ where
 ∂ 
(dxi )p ( ) = δij where δij is the Kronecker Delta.
∂xj p
Definition 1.12. A differential 1-form ω on Rn is an assignment of some ωp ∈
(Tp Rn )∗ to each p ∈ Rn that is smooth in the following sense: if
ωp = f1 (p)(dx1 )p + · · · + fn (p)(dxn )p
then f1 , · · · , fn are smooth.
Now, we define a map from C ∞ (Rn ) to the set of 1-forms on Rn .
 A FEW APPLICATIONS OF DIFFERENTIAL FORMS 3

Definition 1.13. We define d : C ∞ (Rn ) → {1-forms on Rn } such that for all

∂f i
f ∈ C ∞ (Rn ) we have d(f ) = dx using Einstein summation convention to sum
∂xi
over indices.
Remark 1.14. Suppose we have f ∈ C ∞ (Rn ) and a curve in Rn centered on p, γ,
 ∂f   ∂γ 
i
such that γ = (γ1 (t), · · · , γn (t)). Then D[γ] (f ) = . Notice that if
∂xi p ∂t 0
we take d(f ) and evaluate it at (p, [γ]), we get
 ∂f   ∂f   ∂γ 
i i
(d(f ))p ([γ]) = (dx ) p ([γ]) = = D[γ] (f ).
∂xi p ∂xi p ∂t 0
n i n
Remark 1.15.   also note for coordinate functions on R , i.e. x : R → R
We will
α1 !
i  .. 
such that x  .  = αi , we have d(xi ) is the 1-form that evaluates to (dxi )p
αn
at all p ∈ Rn .
Definition 1.16. A differential k-form on Rn is an assignment of some
ωp ∈ Λk ((Tp Rn )∗ )
to each p ∈ Rn that is smooth in the following sense: if
X
ωp = fi1 ···ik (p)dxi1 ∧ · · · ∧ dxik
i1 <···<ik

then each fi1 ···ik is a smooth function on Rn .

Remark 1.17. Notice that {dxi1 ∧ · · · ∧ dxik | i1 < · · · < ik ; i1 , · · · ik ∈ {1, · · · , n}}
forms a basis of Λk ((Tp Rn )∗ )
Now, we will extend the d operator in the following way, keeping with our original
definition by referring to functions in C ∞ (Rn ) as 0-forms:
Definition 1.18. Now, we define d : {k-forms on Rn } → {(k + 1)-forms on Rn }
such that
 X  X n X ∂fi ···i
d fi1 ···ik dxi1 ∧ · · · ∧ dxik = 1
j
k
dxj ∧ dxi1 ∧ · · · ∧ dxik .
i <···<i j=1 i <···<i
∂x
1 k 1 k

Finally, we will establish how these forms transition between vector spaces. We
will assume we have a smooth map F : Rn → Rm .
Definition 1.19. The pushforward of vectors under F at a fixed point p ∈ Rn is
a map F∗ : Tp Rn → TF (p) Rm such that F∗ ([γ]) = [F ◦ γ].
The pullback of a differential form on Rm under F is a differential form on Rn .
We will define the pullback of a 1-form, but it can be extended to k-forms.
Definition 1.20. Let ω = f1 (y 1 , · · · y m )dy 1 + · · · + fm (y 1 , · · · y m )dy m be a dif-
ferential 1-form on Rm . Then the pullback of ω under F is a 1-form on Rn ,
m n
X X ∂y i j
F ∗ω = (fi ◦ F ) ∗ ( dx ).
i=1 j=1
∂xj
4 MATTHEW CORREIA

Remark 1.21. For fixed p ∈ Rn and a fixed curve in Rn centered on p, γ, and a

fixed 1-form on Rm , ω = f1 (y 1 , · · · y m )dy 1 + · · · + fm (y 1 , · · · y m )dy m , we have
m n
X X ∂y i
(F ∗ ω)p ([γ]) = (fi ◦ F )p ∗ ( j
(dxj )p ([γ]))
i=1 j=1
∂x
which simplifies to
m
X
(F ∗ ω)p ([γ]) = (fi ◦ F )p ∗ (dy i )p (F∗ [γ])
i=1
and further simplifies to
(F ∗ ω)p ([γ]) = ωF (p) (F∗ [γ]).

2. Using Differential Forms to Solve Differential Equations

First, we will introduce a few classifications of differential forms.
Definition 2.1. A differential 1-form ω is exact if there exists f ∈ C ∞ (Rn ) such
that df = ω.
Definition 2.2. A differential 1-form ω is closed if dω = 0.
Now, a few useful facts:
Lemma 2.3. If a differential 1-form is exact, it is closed.
Proof. Suppose ω is an exact differential 1-form. Then there exists f ∈ C ∞ (Rn )
such that df = ω. Therefore,
n
X ∂f i
ω= dx
i=1
∂xi
Then, we have
n
n X
X ∂2f
dω = dxj ∧ dxi
j=1 i=1
∂xj ∂xi
However,
∂2f j i ∂2f
dx ∧ dx = − dxi ∧ dxj
∂xj ∂xi ∂xi ∂xj
It follows that dω = 0 and ω is closed. 
The following lemma is presented without proof as it is beyond the scope of this
paper.
Lemma 2.4 (Poincaré’s Lemma for 1-forms). A 1-form ω defined on an open set
U ⊂ Rn is closed if and only if for all p ∈ U there exists a neighborhood V ⊂ U
containing p with a differentiable function f : V → R such that df = ω|V .
Now, we will consider differential equations.
Definition 2.5. A first order differential equation is an equation of the form
dy
φ0 (x) = f (x, φ(x)), which is often written in the form y 0 = f (x, y) or = f (x, y).
dx
Definition 2.6. A first order separable differential equation is an equation of the
form φ0 (x) = g(x)h(φ(x)), which again may be written as y 0 = g(x)h(y) or
dy
= g(x)h(y).
dx
 A FEW APPLICATIONS OF DIFFERENTIAL FORMS 5

Suppose we have a first order separable differential equation, y 0 = g(x)h(y). We

choose a differential 1-form ω = dy − g(x)h(y)dx and we need to find all paths φ
such that φ∗ ω = 0.
ω
If for some function f (x, y) we have is closed, then there exists a function
f (x, y)
ω
F (x, y) such that dF = . Curves of the form φ(t) such that F (t, φ(t)) = c
f (x, y)
ω
where c is some constant are desired. The reason for this is if = dF then
f (x, y)
1
Fy =
f (x, y)
and
g(x)h(y)
Fx = − .
f (x, y)
If (t, φ(t)) satisfies F (t, φ(t)) = c for all t, then
dF (t, φ(t)) = 0.
It follows that
Fx dt + Fy ∗ φ0 (t)dt = 0.
Therefore,
φ0 (t) g(t)h(φ(t))
dt − dt = 0.
f (t, φ(t)) f (t, φ(t))
It follows that φ0 (t) = g(t)h(φ(t)), so φ(t) is a solution to the differential equation.
Remark 2.7. The integration of 1-forms is equivalent to a line integral in multivari-
ate calculus. There are more formal ways to specify this, but it is not necessary for
this application.
ω dy − g(x)h(y)dx
We first assume dF = = . For simplicity, we take
f (x, y) f (x, y)
f (x, y) = h(y). If we integrate dF from some initial condition p = (b, a) along an
arbitrary path to (x, y) we will find a solution to the differential equation. Thus,
Z F Z y Z x
dy
F = dF = − g(x)dx.
0 a h(y) b

If these expressions are integrable, we can set it equal to an arbitrary constant and
solve for y.
Let us consider an example of a separable differential equation, y 0 = xy. Our
first step is to set ω = dy − xydy. In this case, we choose f (x, y) = y so
ω dy
= − xdx
f (x, y) y
leaves us with separated variables. Thus, we have
Z F Z y Z x
dy
dF = − xdx
0 a y b
which evaluates to
x2 b2
F (x, y) = ln(y) − ln(a) − +
2 2
6 MATTHEW CORREIA

We now set it equal to an arbitrary constant and combine constants to find

x2
ln(y) − =c
2
which we can solve for y and find
2
y = kex /2

with k = ec .

3. A Proof of the Fundamental Theorem of Algebra

In order to prove the Fundamental Theorem of Algebra, we first present another
result concerning closed 1-forms.
Remark 3.1. We stated earlier that integrals of 1-forms could be considered to be
regular line integrals. This makes use of a more general fact: if f is a map from A
to B and ω is a 1-form on B and we let γ be a path in A, the integral of ω over the
path f ◦ γ is equivalent to the integral of f ∗ ω over γ.
Lemma 3.2. If ω is a closed 1-form on Rn and γ1 , γ2 are closed paths in Rn such
that there exists a smooth map F : S 1 × [0, 1] → n
R that F |S 1 ×{0} = γ1 and
R R such
1
F |S 1 ×{1} = γ2 (where S is the unit disk) then γ1 ω = γ2 ω.
Proof. Observe that Z Z
ω= F ∗ω
γ1 S 1 ×{0}
Similarly, Z Z
ω= F ∗ω
γ2 S 1 ×{1}
Now, we notice that
Z Z Z
F ∗ω − F ∗ω = F ∗ω
S 1 ×{0} S 1 ×{1} ∂(S 1 ×{0})

By Stoke’s Theorem, we have

Z Z
F ∗ω = d(F ∗ ω)
∂(S 1 ×{0}) S 1 ×{0}

This can be simplified, as

Z Z Z
∗ ∗
d(F ω) = F dω = F ∗0 = 0
S 1 ×{0} S 1 ×{0} S 1 ×{0}

Therefore, we have Z Z
∗
F ω− F ∗ω = 0
S 1 ×{0} S 1 ×{1}
and we have the desired result
Z Z
ω= ω
γ1 γ2

Theorem 3.3 (Fundamental Theorem of Algebra). Every complex polynomial has
at least one complex root.
 A FEW APPLICATIONS OF DIFFERENTIAL FORMS 7

Proof. Consider a polynomial of the form an z n + · · · + a1 z + a0 . We assume a0 6= 0

else z = 0 is a trivial root. Furthermore, we assume an = 1 because we can divide
through by a constant without changing the roots of the polynomial. We are left
with the polynomial z n + · · · + a1 z + a0 .
dz
Consider the 1-form ω = on C \ {0}. It is easy to verify that dω = 0 so ω
z
is closed. We suppose that z n + · · · + a1 z + a0 has no complex roots, and thus we
can define a continuous map f : C → C \ {0} such that f (z) = z n + · · · + a1 z + a0 .
The pullback of ω is thus a closed form on C. It can be easily verified that
(nz n−1 + an−1 (n − 1)z n−2 + · · · + a2 (2)z + a1 )dz
f ∗ω =
z n + · · · + a1 z + a0

Consider γ1 : [0, 2π] → C such that γ1 (t) = reit and γ2 : [0, 2π] → C such that
γ2 (t) = Reit . These paths satisfy the conditions of Lemma 3.2. However, when we
integrate, we will notice a discrepancy.
Z Z 2π
f ∗ω = (reit )∗ f ∗ ω
γ1 0

To take this pullback, we recognize z = reit and make the necessary substitutions
from the formulas above
Z 2π n−1 n−2
(n(rei t) + an−1 (n − 1)(rei t) + · · · + a2 (2)(rei t) + a1 )dt
Z
f ∗ω = ireit n
γ1 0 (rei t) + · · · + a1 (rei t) + a0
Z 2π n n−1 2
(n(rei t) + an−1 (n − 1)(rei t) + · · · + a2 (2)(rei t) + a1 (rei t))dt
= i n
0 (rei t) + · · · + a1 (rei t) + a0
We are specifically interested in the case of a small circle around the origin, so we
take the limit as r approaches 0:
Z Z 2π
lim f ∗ω = 0dt = 0
r→0 γ1 0

Now we consider our other, larger curve.

Z Z 2π
f ∗ω = (Reit )∗ f ∗ ω
γ2 0

Pulling back in the same manner, we obtain

Z 2π n−1 n−2
(n(Rei t) + an−1 (n − 1)(Rei t) + · · · + a2 (2)(Rei t) + a1 )dt
Z
= iReit n
γ2 0 (Re t) + · · · + a1 (Rei t) + a0
i

Z 2π n n−1 2
(n(Rei t) + an−1 (n − 1)(Rei t) + · · · + a2 (2)(Rei t) + a1 (Rei t))dt
= i n
0 (Rei t) + · · · + a1 (Rei t) + a0
This time, in the interest of examining a large curve, we let R approach ∞:
Z Z 2π
lim = indt = 2inπ
R→∞ γ2 0

This therefore results in a contradiction, and z n + · · · + a1 z + a0 must have at least

one complex root. 
8 MATTHEW CORREIA

4. The Gauss-Bonnet Theorem

First, we will develop the concept of the Euler Characteristic for closed surfaces.
It is easiest to think of in terms of polyhedra.
Definition 4.1. The Euler Characteristic χ for a polyhedron is given by
χ=V −E+F
where V is the number of vertices, E is the number of edges, and F is the number
of faces.
The Euler Characteristic is also defined for more general closed surfaces; the
simplest way to understand the definition is to note that if a surface can be deformed
into another without breaking the surface, making new connections, or intersecting
itself, then the two surfaces share the same Euler Characteristic. For example, a
cube has 8 vertices, 12 edges, and 6 faces. χ = 8 − 12 + 6 = 2. A cube can be
deformed in a sphere without breaking, gluing, or self intersecting, so for a sphere,
χ = 2.
Next, we develop a seemingly unrelated property of surfaces, the Gaussian cur-
vature. First, we recall the curvature of a curve in space.
Definition 4.2. Let γ be a curve in R3 with arclength parameterization given by
γ(s). The curvature κγ of γ at point p is given by κγ (p) = |(γ 00 (s))p |.
Definition 4.3. The Gaussian curvature K of a surface M ⊂ R3 at point p is given
by κ1 κ2 where κ1 is the maximum curvature of a geodesic in M at p and κ2 is the
minimum curvature of a geodesic in M at p.
Now, we consider a special map from a surface in R3 to the unit sphere, S 2 .
Definition 4.4. For a surface M ⊂ R3 , the Gauss map, G : M → S 2 is defined
such that every p ∈ M maps to its unit normal vector, that is a point on S 2 .
We consider the differential 2-form
xdy ∧ dz + ydz ∧ dx + zdx ∧ dy
ω=
(x2 + y 2 + z 2 )3/2
defined on R3 \ {0}. This is related to the form we used to investigate the Funda-
mental Theorem of algebra. It shares a property of counting a winding number -
the previous form counted how many radians around the origin a curve wrapped
when
R integrated, and this form counts a similar wrapping for surfaces. For instance,
S 2 ω = 4π.
An important property of ω is that G∗ ω = Kdσ where Rdσ is the areaR element of
M . In fact, there is also an important relation between M G∗ ω and S 2 ω which
allows us to quickly integrate the Guassian curvature. To quantify this relation, we
must first examine the idea of the degree of G. An informal explanation of deg(G)
would be the number of preimages of the majority of points in S 2 .
To formally examine deg(G), we start by considering the height function h for M
such that h(x1 , x2 , x3 ) returns the height (x3 coordinate) of M at that location. We
find the zeroes of −∇h, the negative gradient of the surface (this can be visualized
if you think of pouring water over the surface - the points with no net flow are
zeroes of this vector field). We can divide these zeroes into two classes: those with
upward unit normal vectors and those with downward unit normal vectors. We
 A FEW APPLICATIONS OF DIFFERENTIAL FORMS 9

now further split the zeroes into two more classes: those with positive K (extrema)
and those with negative K (saddle points). The degree of G can be given by the
number of zeroes with upward (or downward) normal vectors and positive K less
the number of zeroes with upward (or downward) normal vector and negative K.
Note that the calculation can use either upward or downward normal vector zeroes,
but not a mix of the two.
In Figure 1, we show an example
Figure 1. Negative of the gradient sketched on a torus.
Gradient of Height The orange dots are zeroes with upward
Function on Torus pointing normal vectors. The purple
dots are zeroes with downward point-
ing normal vectors. For both, there is
on extremum and one saddle point, so
the degree of the Gauss map on a torus
is 0.
With a firm way to calculate deg(G),
we turn backR to our application,R and
present that M G∗ ω = deg(G) S 2 ω.
This relation is not unexpected, be-
cause both ω and the degree of G are
related to the way our surface wraps
around the origin.
Another important result is the Poincare-Hopf theorem, which allows us to relate
the degree of this map back to the Euler characteristic. Using this theorem, we
obtain the relation χ(M ) = 2deg(G). This final piece allows us to string together
Z Z Z
Kdσ = G∗ ω = deg(G) ω = 4πdeg(G) = 2πχ(M )
M M S2
This final result, Z
Kdσ = 2πχ(M )
M
is known as the Gauss-Bonnet theorem.
Now, we present a brief application of the Gauss-Bonnet theorem.
Corollary 4.5. Let f be a smooth function f : R2 → R such that for some compact
B ⊂ R2 f is identically zero outside of B. Then the integral of the gaussian
curvature of the graph of f in R3 is 0.
Let A0 be the graph of f in R3 . We note that the Gauss map G : A0 → S 2 can
be expressed such that
!
−fu −fv 1
G(u, v, f (u, v)) = p ,p ,p .
fu2 + fv2 + 1 fu2 + fv2 + 1 fu2 + fv2 + 1
This can be verified by directly calculating the unit normal vectors to this surface.
Further, we present that
fuu fvv − fuv fvu
Kdσ = G∗ ω = 3/2
du ∧ dv.
(fu2 + fv2 + 1)
This too can be verified through directly pulling back ω under the above map,
however the calculation is tedious. We observe that it should be possible to directly
10 MATTHEW CORREIA

integrate this expression for any function of the specified type. However, in practice
the integration will usually not be possible without numerical methods. Instead,
we present a much simpler proof using the Gauss-Bonnet theorem.
Proof. Let A be the graph of f |B in R3 . It is possible to construct a smooth surface
S ⊂ R3 such S ∪ A is a smooth surface in R3 with no holes or self intersections,
i.e. S ∪ A can be smoothly deformed into a sphere. Such a surface S is hard to
express explicitly, but an example would be the lower hemisphere of a sphere with
the edges tapered to smoothly connect with A. Because A must smoothly join with
the identically zero region of the graph of f outside of B, the same S should be
usable for any choice of f .
We recognize that since S ∪ A can be smoothly deformed into a sphere,
χ(S ∪ A) = 2.
We consider the Gauss-Bonnet theorem in regards to S ∪ A:
Z
Kdσ = 2πχ(S ∪ A) = 4π.
S∪A
We observe that we can split the integral into the two subsections of the surface:
Z Z Z
Kdσ = Kdσ + Kdσ = 4π.
S∪A S A
2
Now, we make the observation that f : R → R such that f (x, y) = 0 satisfies
these conditions. We note that the curvature of the graph of this function is iden-
tically zero. This can be verified by calculation with the formula presented above.
When we put this result into our calculation with the Gauss-Bonnet theorem, we
see that Z
Kdσ = 4π.
S
Therefore, for any f that satisfies the above conditions, we see that
Z
Kdσ = 0
A
. 
Acknowledgments. It is a pleasure to thank my mentor, Yiwen Zhou, for sharing
his time and knowledge, along with his endless patience and kindness. I would
also like to thank the Mathematics Department at the University of Chicago, and
particularly Peter May, for running this REU.

References
[1] Manfredo P. do Carmo. Differential Forms and Applications. Springer-Verlag Berlin Heidelberg
GmbH. 1994.
[2] S. S. Chern, W. H. Chen and K. S. Lam. Lectures on Differential Geometry. World Scientific
Publishing Co. Pte. Ltd. 1999.