Cambridge Part III Maths

Lent 2016

Applications of Differential
Geometry to Physics

based on a course given by written up by

Maciej Dunajski Josh Kirklin

Please send errors and suggestions to [email protected].

Contents
1 Introduction 2
1.1 Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Vector Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2 Lie groups 7
2.1 Geometry of Lie groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2 Metrics on Lie groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

3 Hamiltonian Mechanics/Symplectic Geometry 12

3.1 Symplectic manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.2 Symplectic structure on cotangent bundles . . . . . . . . . . . . . . . . . . . . . . . 14
3.3 Canonical transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.4 Geodesics, Killing vectors and tensors . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.5 Geodesics in non-Riemannian geometries . . . . . . . . . . . . . . . . . . . . . . . . 16
3.6 Null Kaluza-Klein reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.7 Integrable systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

4 Topological charges in field theory 18

4.1 Scalar kinks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.2 Degree of a map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.3 Applications to physics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

5 Gauge Theory 23
5.1 Hodge duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
5.2 Yang-Mills equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
5.3 Yang-Mills instantons and Chern forms . . . . . . . . . . . . . . . . . . . . . . . . 25
5.4 Fibre bundles & connections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
5.5 Back to Yang-Mills instantons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

–1–
 1 Introduction 1 Introduction

1 Introduction
Lecture 1
14/01/16
Geometry has always played a key role in physics. For example, consider Kepler orbits, i.e. paths
r(t) in R3 obeying:
GM
ṙ = 3 r
r
The solutions are conic sections. If we view a conic section as a
path with coordinates (x, y) in the plane it inhabits, then we can
find that all conic sections can be written as the locus of solutions
of an equation of the following form:

ay 2 + bx2 + cxy + dy + ex + f = 0
(x, y)
where a, b, c, d, e, f ∈ R. It was first deduced by Apollonius of Perga
that given five arbitrary points in the plane, there is a unique conic
section through those points. Thus if we observe the position of a
planet at five different points, we can deduce its orbit.
The space of conic sections is RP5 = R6 / ∼ where x ∼ y ⇐⇒ ∃ C ∈ R s. t. x = Cy. We can see
this by first labelling each conic section by its coefficients (a, b, c, d, e, f ) ∈ R6 in the above equation,
and then seeing that we can simply multiply both sides of the equation to get the same conic.
This is geometry, but it is an algebraic sort of geometry. In this course we will be more interested
in differential geometry. There will be three main topics that we will cover in some detail:

The Hamiltonian formalism (19th century), which involves a generalised coordinate q, its conju-
gate momentum p, and a function H(p, q) such that paths in (p, q) space obey:

∂H ∂H
q̇ = , ṗ = −
∂p ∂q

We will see how this leads to a notion of symplectic geometry (20th

R
century), which concerns
symplectomorphisms, which are diffeomorphisms that preserve dp ∧ dq, and are generated
by Hamiltonian vector fields: XH = ∂H ∂ ∂H ∂
∂p ∂q − ∂q ∂p .

General relativity (1915), which leads to Riemannian geometry.

Gauge theory (both Maxwell and Yang-Mills), which will lead to discussions on an object known
as the connection on a principal bundle.

A unifying feature of these topics will be the presence of Lie groups.

This course will involve lots of examples (often instead of proofs), and the emphasis will be on
calculations.

1.1 Manifolds

Definition. An n-dimensional (smooth) manifold is a set M together with a collection of open

sets Uα , α = 1, 2, . . ., such that the Uα cover M, and there exist bijections φα : Uα → Vα ⊂ Rn
(called charts) such that φβ ◦ φ−1 α : φα (Uα ∩ Uβ ) → φβ (Uα ∩ Uβ ) is smooth.

–2–
1 Introduction 1.1 Manifolds

M
Rn
Uα Vα
φα

φβ ◦ φ−1
α

φβ
Rn
Vβ
Uβ

Informally, we can say that M is a topological space with some extra structure that allows us to
introduce a differential calculus.

Example. The trivial manifold is M = Rn with a single chart, for example the identity.

Although it is not obvious, it is in fact usually possible to choose other differential structures on Rn .
For example, it has been shown that there are infinitely many exotic structures on R4 (Donaldson
1984).

Example. The n-dimensional sphere S n = {r ∈ Rn+1 s. t. |r| = 1} ⊂ Rn+1 is an n-manifold.

We choose open sets U = S n \ N, Ũ = S n \ S where N = {0, . . . , 0, 1}, S = {0, . . . , 0, −1}, and
define charts as follows:

N
r1 rn
 
p φ(r) = ,..., = (x1 , . . . , xn ) on U
1 − rn+1 1 − rn+1
r1 rn
 
φ̃(p) φ̃(r) = ,..., = (x̃1 , . . . , x̃n ) on Ũ
1 + rn+1 1 + rn+1
φ(p)
S

Note that:
2
r12 + · · · + rn2 1 − rn+1 1 + rn+1
x21 + · · · + x2n = = =
(1 − rn+1 )2 (1 − rn+1 )2 1 − rn+1
So:
1 − rn+1 xk
x̃k = xk = 2
1 + rn+1 x1 + · · · + x2n
is smooth on U ∩ Ũ .

Example. Let f1 , . . . , fk : RN → R and set M = {r ∈ RN s. t. f1 = · · · = fk = 0}. Then M is

∂f i N
a manifold if rank ∂xa is maximal, and is called a surface in R .

In fact:

–3–
 1 Introduction 1.2 Vector Fields

Theorem 1 (Whitney). Any n-dimensional manifold M can be obtained as a surface in RN , where

N ≤ 2n + 1.

Example. Real projective space RPn is a manifold:

Rn+1 \ 0
RPn = where [X 1 , . . . , X n+1 ] ∼ [cX 1 , . . . , xC n+1 ] for all c ∈ R∗ = R \ 0
∼
We have n + 1 open sets Uα = {p ∈ RPn s. t. X α 6= 0}, and charts on each open set:

X1 X α−1 X α+1 X n+1

x1 = , ... xα−1 = , xα+1 = , ... xn+1 =
Xα Xα Xα Xα

1.2 Vector Fields

Lecture 2
19/01/16
Suppose we have two manifolds M, M̃ with dimensions n, ñ, open sets Uα , Ũβ and charts φα , φβ
respectively, and let f be a map from M to M̃.

Definition. f is smooth if φ̃β ◦ f ◦ φα−1 : Rn → Rñ is smooth for all α, β.

Definition. If M̃ = R, then f : M → R is a function.

Definition. If M = R, then f : R → M̃ is a curve.

Suppose we have a curve γ : R → M such that γ(0) = p ∈ M. Let U be an open neighbourhood of

p, U ' Rn and choose coordinates xa , a = 1, . . . , n on U .

Definition. The tangent vector to γ at p is defined as:


dγ() 
V |p = ∈ Tp M
d =0
where Tp M is the tangent space at p, defined as the set of all tangent vectors to all curves at p.

V |p γ

M p

Definition. The tangent bundle is given by:

[
TM = Tp M
p∈M

–4–
1 Introduction 1.2 Vector Fields

A vector field assigns a tangent vector to each point p ∈ M. Let f : M → R. The rate of change
of f along γ is given by:
n 
d X ∂f 
f (xa ())|=0 = ẋa ()
d a=1
∂xa =0
n 
X
a ∂ 
= V (x) f
a=1
∂xa p
| {z }
=V
n o
V is a vector field. V a (x) are the components of V in the basis ∂x∂ 1 , . . . , ∂x∂n at p.
We have gone from a curve to a vector field. We can also go the other way:

Definition. An integral curve or flow γ() of a vector field V is defined by:

d a
γ̇() = V |γ() or, equivalently x () = V a (x())
d

We have:
xa (, xa (0)) = xa (0) + V a (x(0)) + O(2 )
We say that the vector field V generates the flow.

Definition. An invariant of a vector field V is a function that is constant along the flow of V :

f (xa (0)) = f (xa ()) ∀  or, equivalently V (f ) = 0

∂ ∂
Example. Let M = R2 , xa = (x, y), and V = x ∂x + ∂y . The integral curves of V are given
by ẋ = x, ẏ = 1, and we can solve this to obtain (x(), y()) = (x(0)e , y(0) + ).

Note that on these curves xe−y is constant, so this is an invariant of V .

–5–
1 Introduction 1.2 Vector Fields

Example. Consider the 1-parameter group of rotations on R2 . Under these rotations, (x0 , y0 )
transforms to the point (x(), y()) = (x0 cos  − y0 sin , x0 sin  + y0 cos ). The vector field
that generates these curves is given by:

dy() ∂ dx() ∂
 
V = +
d ∂y d ∂
∂ ∂
=x −y
∂y ∂x
The distance to the origin is an invariant of V :

V (x2 + y 2 ) = −2xy + 2xy = 0

Definition. A Lie bracket of two vector fields V and W is a vector field [V, W ] defined by its
action on functions as:
[V, W ](f ) = V (W (f )) − W (V (f ))

The Lie bracket has some important properties:

Antisymmetry : [V, W ] = −[W, V ]

Jacobi identity : [U, [V, W ]] + [V, [W, U ]] + [W, [U, V ]] = 0

∂ ∂ ∂ ∂
Example. If V = x ∂x + ∂y and W = ∂x , then [V, W ] = − ∂x = −W .

Definition. A Lie algebra is a vector space g equipped with an antisymmetric bilinear operation
[ , ] : g × g → g that satisfies the Jacobi identity.

If g is finite-dimensional and Vα , α = 1, . . . , dim g span g, then g is determined by its structure

γ
constants fαβ , defined as follows:
X γ
[Vα , Vβ ] = fαβ Vγ
γ

Example. There are only two 2D Lie algebras (up to isomorphism), each determined by the
bracket of two basis elements:

[V, W ] = 0 or [V, W ] = −W

Example. gl(n, R) (the set of all n × n real matrices) is a Lie algebra when equipped with the
matrix commutator as a bracket. Its dimension is n2 .

–6–
 2 Lie groups

Example. Vector fields on a manifold M form an infinite dimensional Lie algebra. Consider for
example g = diff(S 1 ) or diff(R), by which we mean the set of all vector fields on the manifold.
We have the following basis of g:
∂
Va = −xα+1 where α ∈ Z
∂x
With this basis, the Lie bracket gives:

[Vα , Vβ ] = (α − β)Vα+β

Lecture 3
21/01/16

Example. The Virasoro algebra is a so-called central extension of diff(S 1 ). This is defined as
Vir = diff(S 1 ) ⊕ R, and we choose c ∈ R to be a basis vector of the R sector (c is sometimes
referred to as the central charge). In addition, we have the following brackets:
c 3
[Vα , c]Vir = 0, [Vα , Vβ ]Vir = (α − β)Vα+β + (α − α)δα+β,0
12
Consider the bracket of two vector fields in diff(S 1 ):

∂ ∂ ∂
 
f (x) , g(x) = (f g 0 − gf 0
∂x ∂x | {z } ∂x
Wronskian

This extends to the Virasoro algebra in a slightly non-trivial manner. According to Witten:
∂ ∂ ∂ ic
  Z
0 0
f (x) , g(x) = (f g − gf ) + (fxxx g − gxxx f ) dx
∂x ∂x Vir ∂x 48π

Theorem 2 (Ado). Every finite-dimensional Lie algebra is isomorphic to some matrix algebra.

2 Lie groups

Definition. A Lie group is a group that is also a manifold such that the group operations
G × G → G, (g1 , g2 ) → g1 g2 and G → G, g → g −1 are smooth maps.

Example. G = GL(n, R) is a Lie group with dimension n2 .

n(n−1)
Example. G = O(n, R) is a Lie group with dimension 2 .

–7–
2 Lie groups 2.1 Geometry of Lie groups

Definition. A group action of a group G on a manifold M is a function G × M → M,

(g, p) → g(p) such that e(p) = p and g1 (g2 (p)) = (g1 g2 )(p). Groups acting on manifolds are
referred to as transformation groups.

Example. Consider M = R2 and G = E(2), the 3-dimensional Euclidean group, consisting of

rotations and translations of the plane. An element g ∈ G with coordinates (θ, a, b) acts on M
in the following way: ! ! ! !
x cos θ − sin θ x a
g = +
y sin θ cos θ y b

We see that the manifold of G is S 1 × R2 . G has three 1-parameter Lie subgroups:

• Gθ , elements with coordinates (θ, 0, 0).

• Ga , elements with coordinates (0, a, 0).

• Gb , elements with coordinates (0, 0, b).

Each 1-parameter subgroup with parameter  is in fact a flow, generated by the vector field:

d 
V |p = (g (p))
d =0

We have:
∂ ∂ ∂ ∂
Vθ = x −y , Va = , Vb =
∂y ∂x ∂x ∂y
These form a basis for a 3D Lie algebra of E(2), in which we have:

[Va , Vθ ] = Vb , [Vb , Vθ ] = −Va , [Va , Vb ] = 0

2.1 Geometry of Lie groups

Let M and M̃ be manifolds, and f : M → M̃.

Definition. A tangent map f∗ : Tp M → Tf (p) M̃ is defined by:


d 
f∗ (V ) = f (γ())
d =0

where γ is an integral curve of V . This definition extends to T M

∂f i
(f∗ (V ))i = V j
∂xj
where i = 1, . . . , dim M̃, j = 1, . . . , dim M.

–8–
 2 Lie groups 2.1 Geometry of Lie groups

Definition. Let V , W be vector fields with V = γ̇. The Lie derivative is defined as follows:

W (p) − γ()∗ W (p())

LV W = lim = [V, W ]
→0 
We also define LV (f ) = V (f ) for functions f , and define Lie derivatives on higher order tensors
by the Leibnitz rule.
Lecture 4
26/01/16
Given Ω an r-form, we can find its Lie derivative using the Cartan formula:

LV Ω = dV yΩ + V y dΩ

Definition. A Lie algebra g of a Lie group G is the tangent space of G at the identity, with
bracket in g given by the commutator of fields on G, defined in the following way. Given a
g ∈ G, its associated left translation function is Lg : G → G, Lg (h) = gh. This induces a map
(Lg )∗ : g → Tg (G). Given a v ∈ g, the vector field g 7→ (Lg )∗ v is left-invariant. The Lie bracket
is then given in terms of commutators of these left-invariant fields:

[(Lg )∗ v, (Lg )∗ w] = (Lg )∗ [v, w]g

If we have a basis of g, then left translation naturally gives a set of dim(G) independent global
non-vanishing vector fields on G. Manifolds that admit such a set are said to be parallelisable.
(Note that not all parallelisable manifolds can be made into Lie groups.)
Let Lα , α = 1, . . . , dim g be a basis of left invariant vector fields. The structure of the commutator
is given by its structure constants
X γ
[Lα , Lβ ] = fαβ Lγ .
γ

Let σ α be a dual basis of left-invariant 1-forms, Lα yσ β = δαβ . For any 1-form Ω and vector fields
v, w, we have the identity

dΩ (v, w) = v(Ω(w)) − w(Ω(v)) − Ω([v, w]),

so in particular we have
1 α β
dσ α + fβγ σ ∧ σ γ = 0.
2

Definition. The Maurer-Cartan 1-form ρ is a Lie algebra valued 1-form defined by ρg (v) =
(Lg−1 )∗ v.

Assume from now on that G is a matrix Lie group, and let g ∈ G. Then it can be shown that
ρg = g −1 dg. This is indeed left-invariant

(g0 g)−1 d(g0 g) = g −1 g0−1 g0 dg = g −1 dg ,

–9–
 2 Lie groups 2.1 Geometry of Lie groups

and it is a member of the Lie algebra, as we can see by considering the curve

−1 −1 dg 
g (s)g(s + ) = e +  g +O(2 ).
ds =0
| {z }
∈g

We can expand the Maurer-Cartan 1-form as

g −1 dg =
X
σ α ⊗ Tα ,
α

where Tα are matrices in a basis of g satisfying

X γ
[Tα , Tβ ] = fαβ Tγ .
γ

We have
dρ = dg −1 ∧ dg = −g −1 dg g −1 ∧ dg = −ρ ∧ ρ.
Lecture 5
28/01/16 In terms of the expansion,
1 1 γ α
dσ α ⊗ Tα = σ α ∧ σ β Tα Tβ = σ α ∧ σ β [Tα , Tβ ] = fαβ σ ∧ σ β Tγ ,
2 2
so σ α are a left-invariant dual basis.

Example. The Heisenberg group is composed of 3 × 3 matrices of the form

 
1 x z
g = 0 1 y  = 1 + xT1 + yT2 + zT3 .
 
0 0 1

This group obeys the Heisenberg algebra

[T1 , T2 ] = T3 , [T2 , T3 ] = [T3 , T1 ] = 0.

In quantum mechanics, we might identify T1 ∼ position, T2 ∼ momentum and T3 ∼ i~×identity.

Let ρ = g −1 dg = T1 σ 1 + T2 σ 2 + T3 σ 3 . Note that
   
1 −x z + xy 0 dx dz
g −1 = 0 1 −y  and dg = 0 0 dy  ,
   
0 0 1 0 0 0

so we obtain ρ = T1 dx + T2 dy + T3 (dz − x dy), i.e.

σ1 = dx , σ2 = dy , σ3 = dz − x dy .

Note that dσ 1 = 0, dσ 2 = 0 and dσ 3 = − dx ∧ dy, so we verify dσ α + 12 fβγ

α σ β σ γ . The vector

fields dual to the σ α form a left-invariant vector field basis

∂ ∂ ∂ d
L1 = , L2 = +x , L3 = ,
∂x ∂y ∂z dz
and note that these fields form another representation for the Heisenberg algebra:
∂
[L1 , L2 ] = = L3 , [L1 , L3 ] = [L2 , L3 ] = 0
∂z

– 10 –
2 Lie groups 2.2 Metrics on Lie groups

We could also define right translations by Rg (h) = hg, and similarly obtain right-invariant
forms, along with right-invariant vector fields Rα :
∂ ∂ ∂ ∂
R1 = +y , R2 = , R3 =
∂x ∂z ∂y ∂z
These obey
γ
[Rα , Rβ ] = −fαβ Rγ and [Rα , Lβ ] = 0 ∀ α, β.

Note a strange terminological artifact: right-invariant vector fields generate left translations and
vice versa.

2.2 Metrics on Lie groups

A left-invariant metric on a Lie group G is given as

h = hαβ σ α ⊗ σ β

where hαβ is a constant symmmetric non-degenerate matrix, and α, β = 1, . . . , dim G. Right-

invariant vector fields are Killing vector fields for (G, h), i.e. LRα h = 0. Thus we have dim G
KVFs.

Example. Return to the Heisenberg group defined above, and let

h = δαβ σ α ⊗ σ β = dx2 + dy 2 + (dz − x dy)2 .

This has three isometries:

∂
z → z +  generated by = R3
∂z
∂
y → y +  generated by = R2
∂y
x→x+ ∂ ∂
generated by +y = R1
z → z + y ∂x ∂z

Now we look at the Kaluza-Klein interpretation, by considering the motion on the space of orbits
∂
of R3 = ∂z . Geodesics of h have Lagrangian

L = ẋ2 + ẏ 2 = (ż − xẏ)2 .

The associated equations of motion give

ẍ = −C ẏ and ÿ = C ẋ,

where c = ż − xẏ is a constant of motion. Compare this to geodesic motion in the presence of a
Riemannian manifold (M, g), where g = gij dxi dxj . The magnetic fields are given by the field
strength
1
F = Fij dxi ∧ dxj with dF = 0,
2
– 11 –
 3 Hamiltonian Mechanics/Symplectic Geometry 3 Hamiltonian Mechanics/Symplectic Geometry

and the geodesic equation is

ẍi + Γijk ẋj ẋk = cF ij ẋj (∗)
where we raise and lower indices with gij . Take M = R2 , gij = δij , xi = (x, y). We have Γijk = 0.
Also assume F = − dx ∧ dy, so Fij = ij (a volume form for M). From (∗) we obtain

ẍi = −cij ẋj .

From this we see that the geodesics of the left-invariant metric h on G project to trajectories of a
charged particle moving in a constant magnetic field on M.
More generally, we might have
h = (dz + A)2 + gij dxi dxj
where A = Ai (x) dxi is a 1-form. i
n o This leads to a conserved charge c = ż + Ai ẋ , and the geodesics
∂
Lecture 6 projected down to M = G/ ∂z are magnetic geodesics with F = dA.
02/02/16
If we calculate the Laplacian of a left-invariant metric h = hab σ a σ b , we find

∇2 = hab La Lb + hab fac

c
Lb .

Example. Using the metric above for the Heisenberg group, we can write down the Schrödinger
equation as
∇2 φ = φxx + φzz + (∂y + x∂z )2 φ = −Eφ.
If we take φ = ψ(x, y)eiez where e is a constant, we obtain

ψxx − e2 ψ + (∂y + ixe)2 ψ = −Eψ.

Compare this with

(∂x − ieAx )2 ψ + (∂y − ieAy )2 ψ = −(E − e2 )ψ,
where A = Ax dx + Ay dy, and dA = F is the magnetic field.

3 Hamiltonian Mechanics/Symplectic Geometry

3.1 Symplectic manifolds
Let M be a 2n-dimensional phase space. Given two functions f, g : M → R, their Poisson bracket
is a third function defined as
n
X ∂f ∂g ∂f ∂g
{f, g} = − ,
k=1
∂qk ∂pk ∂pk ∂qk

where pk , qk , k = 1, . . . , n are local coordinates on M.

Hamiltonian mechanics is the definition of a function H : M → R called the Hamiltonian. Dynamical
paths on M obey Hamilton’s equations
∂H ∂H
ṗj = − and q̇j = .
∂qj ∂pj
Equivalently, paths t → (p(t), q(t)) are integral curves of a Hamiltonian vector fields
X ∂H ∂ ∂H ∂
XH = − = {H, };
j
∂pj ∂qj ∂qj ∂pj

– 12 –
3 Hamiltonian Mechanics/Symplectic Geometry 3.1 Symplectic manifolds

using Hamilton’s equations, we see that

q̇j = XH (qj ) and ṗj = XH (pj ).

A more general framework is that of Poisson structures. Now assume that dim M = m (not
necessarily even), and let ω ij = ω [ij] : M → R. The Poisson structure given by ω is the bracket
m
X ∂f ∂g
{f, g} = ω ij (x) ,
i,j=1
∂xi ∂xj

and we require ω to be such that the Jacobi identity is satisfied. Note that there are no a priori
distinctions between posisitions and momenta. In this framework we have
X ∂H X ∂H ∂
ẋi = ω ij and XH = ω ij .
j
∂xj i,j
∂xj ∂xi

If {f (x), xi } = 0 for some f (x), then we call f (x) a Casimir.

Example. Let M = R3 and ω ij = k ijk xk . This gives rise to a Poisson structure with the
P

following algebra:
{x1 , x2 } = x3 , {x2 , x3 } = x1 , {x3 , x1 } = x2
Let r2 = (x1 )2 + (x2 )2 + (x3 )2 . Then we have {xi , r2 } = 0, so r2 is a Casimir.
Suppose we have the Hamiltonian
!
1 (x1 )2 (x2 )2 (x3 )3
H= + 2 + 2 ,
2 a21 a2 a3

where ai 6= 0 are constants. Then Hamilton’s equations give

a3 − a2 2 3 a1 − a3 3 1 a2 − a1 1 2
ẋ1 = x x , ẋ2 = x x , ẋ3 = x x ,
a2 a3 a3 a1 a1 a2
which are just Euler’s equations for a rigid body.

Example. Now restrict the Poisson structure from the previous example to S 2 ⊂ R3 , and
reparametrise to spherical coordinates

x1 = sin θ cos φ, x2 = sin θ sin φ, x3 = cos θ.

It can be shown that the Poisson structure in these coordinates is given by

{θ, φ} = sin−1 (θ).

Note that this structure has no Casimirs; we say it is non-degenerate. We can define a 2-form
on S 2 by
(ω − 1)ab dxa ∧ dxb = sin θ dθ ∧ dφ .
This 2-form is called a symplectic structure, and arises from the Poisson structure if ω is
invertible.

– 13 –
 3 Hamiltonian Mechanics/Symplectic Geometry 3.2 Symplectic structure on cotangent bundles

Definition. A symplectic manifold is a smooth manifold M of even dimension 2n, equipped

with a closed 2-form ω ∈ Λ2 (M) which is non-degenerate (i.e. ω
| ∧ ·{z
· · ∧ ω} 6= 0).
n

Lecture 7 ω provides an isomorphism between T M and T ∗ M given by v ∈ T M 7→ vy ω ∈ T ∗ M.

04/02/16
Given an f : M → R, we have its gradient df , a 1-form, and define the associated Hamiltonian
vector field Xf by
Xf y ω = − df .
We then define the bracket to be {f, g} = Xg (f ). From this we see

{f, g} = Xg (f ) = df (Xg ) = −(Xf y ω)(Xg ) = ω(Xg , Xf ) = (Xg y ω)(Xf ) = − dg (Xf ) = −{g, f },

so the bracket is antisymmetric. We can also write the bracket in terms of components as
X ∂f ∂g
{f, g} = ω ij .
∂xj ∂xi
The Jacobi identity is true as a consequence of ω being closed.
Exercise: show that [Xf , Xg ] = −X{f,g} is a homomorhpism between functions on M and the Lie
algebra of M.
Hamiltonian vector fields preserve the symplectic structure of M:

LXf ω = Xf y |{z}
dω + d(Xf y ω ) = − d(df ) = 0
| {z }
=0 =−df

Theorem 3 (Darboux). Let (M, ω) be a 2n-dimensional symplectic manifold. Then there exist
local coordinates x1 = q 1 , . . . , xn = q n and xn+1 = p1 , . . . , x2n = pn such that ω = ni=1 dpi ∧ dqi ,
P

and the Poisson bracket takes the “standard” form.

Proof (sketch). We proceed by induction with respect to half of the dimension of M.

1. Choose a function p1 : M → R, and construct q1 : M → R by solving an ODE Xp1 (q1 ) = 1

(really a PDE solvable by a method of characteristics).

2. Let M1 = {x ∈ M s. t. p1 = constant, q1 = constant}. It can be shown that M1 is locally

symplectric with ω1 = ω|p1 ,q1 constant .

3. Repeat: pick p2 : M1 → R, find q 2 , construct M2 , etc.

3.2 Symplectic structure on cotangent bundles

Let Q be an n-dimensional manifold. M = T ∗ Q admits a global symplectic structure. For each
covector in T ∗ M, let π be the map that projects that covector onto its point in Q. Then we
have π −1 (x) ' (Rn )∗ for each x ∈ Q. Let (q 1 , . . . , q n ) be a chart on U ⊂ Q, and (p1 , . . . , pn ) be
coordinates on (Rn )∗ = Tq∗ Q. From π : T ∗ Q → Q, we can construct its pullback π ∗ : T ∗ Q → T ∗ T ∗ Q.
Let p ∈ T ∗ Q, and write θ(p) = π ∗ p. We define ω = dθ, and note that dω = 0. Locally, we have
θ = pi dq i (where we abuse notation to write q i instead of q i ◦ π), and ω = i dpi ∧ dq i . Hence, ω
P

gives a natural symplectic structure on M.

– 14 –
 3 Hamiltonian Mechanics/Symplectic Geometry 3.3 Canonical transformations

3.3 Canonical transformations

Let (M, ω) be a symplectic manifold of dimension 2n. Canonical transformations are one parameter
familties of transformations generated by Hamiltonian vector fields that preserve ω, i.e. f : M → M
such that f ∗ (ω) = ω.
If n = 1, these are area preserving maps in 2d phase space (Liouville theorem).
Given a canonical transformation, we can locally construct a corresponding generating function.
Suppose we have a canonical transformation given in Darboux coordinates by
f
(pi , q i ) → (Pi (p, q), Qi (p, q)).
The canonical condition gives dpi dq i = − dQi dPi , so we have that pi dq i + Qi dPi is a closed and
hence locally exact 1-form. Thus there exists a function S(q, P ) such that
∂S i ∂S
dS = pi dq i + Qi dPi = i
dq + dPi .
∂q ∂Pi
∂S ∂S
Comparing coefficients we find pi = ∂qi
and Qi = ∂P i
. This procedure can be carried out in reverse,
so that if we are given a generating function S, we can solve these equations to find a canonical
transformation Q(q, p), P (q, p).

3.4 Geodesics, Killing vectors and tensors

Consider a (possibly peudo-) Riemannian manifold (M, g), and write g = gij dxi dxj where {xi } is
a chart on M. We have the geodesics equations (for affine parameter)
ẍi + Γijk ẋj ẋk = 0.
This is one way to view geodesics. We will consider another, in which geodesics on M are projections
Lecture 8 from T ∗ M of integral curves of some Hamiltonian vector field XH .
09/02/16
Define ẋi = pi , and pi = gij pj . Then xi , pj are coordinates on T ∗ M, with ω = dpi ∧ dxi . The
geodesic equations then become 2n first order ODEs
ẋi = pi and ṗi = −Γijk pj pk .
The vector field defining this system is
∂ ∂
X = pi i
− Γijk pj pk i .
∂x ∂p
Integral curves of X project down to geodesics. We would like to find some Hamiltonian H so that
we can write X as its Hamiltonian vector field:
∂H ∂ ∂H ∂
X= i
−
∂pi ∂x ∂xi ∂pi
Such a Hamiltonian is the geodesic Hamiltonian H(x, p) = 12 g ij (x)pi pj (to verify this, use ∇i gjk =
∂gjk mg m
∂xi
− γij mk − γik gjm = 0).
Killing vectors are those under which the metric is preserved, Lκ g = 0 ⇐⇒ ∇(i κj) = 0. From
∂
Killing vectors we can form first integrals that are linear in the momenta, κ = κi ∂x i
i → K = κ pi .
We have
∂κi
{K, H} = {κi pi , H} = pi pj − Γijk pj pk κi = (∇i κj )pi pj = 0.
∂xj
Note that we can recover the Killing vector from K by pushing XK forward with π : T ∗ M → M:
κ = π∗ (XK )

– 15 –
 3 Hamiltonian Mechanics/Symplectic Geometry 3.5 Geodesics in non-Riemannian geometries

Definition. Let K = κij...k pi pj . . . pk where κ is a symmetric (r, 0) tensor on M. Then κ is a

| {z }
r
rank r Killing tensor iff
{K, H} = 0 ⇐⇒ ∇(i κj...k) .

i...jk ∂
We have XK = rκi...jk pi . . . pj ∂x∂ k − ∂κ∂xl pi . . . pk ∂p l
, and we see that π∗ (XK ) = 0 for r > 1, i.e.
there are no 1-paramater groups of transformations of M given by higher order Killing tensors.
Killing tensors are for this reason sometimes called hidden symmetries of a geodesic motion.

3.5 Geodesics in non-Riemannian geometries

Suppose now that in the geodesic equation we do not assume that the connection is necessarily
Levi-Civita. This is useful in several instances. For example, when considering the geometry of
unparametrised geodesics (paths), we may with to eliminate t from the geodesic equations, and for
example use xn as a parameter.
Consider trajectories of particles in Newtonian physics, obeying
ẍ = −∇V,
where V : R3 → R is the gravitational potential. We would like to reinterpret this as geodesic
motion of some connection, so that
ẍa + Γabc ẋb ẋc = 0 where xa = (x, t) are points in 4D Newtonian spacetime.
∂V
Comparing the two equations, we see that we must take Γi00 = δ ij ∂x j and other components
zero. The space components of the geodesic equation then replicate Newton’s equation, and the
time component is ẗ = 0, meaning we can take t as a parameter. Note that this is not a metric
connection.

Definition. A Newton-Cartan structure on a 4D manifold M is a triple (∇, h, θ) where ∇ is

a torsion-free connection, h = hab ∂x∂ a ⊗ ∂x∂ b is a degenerate contravariant metric of rank 3, and
θ is a (closed) 1-form such that hab θa = 0 and moreover ∇a θb = 0, ∇a hb c = 0.

Note that dθ = 0 implies that there exists a time coordinate t : M → R such that θ = dt (with
some topological assumptions). In this context we call θ a clock.

3.6 Null Kaluza-Klein reduction

Lecture 9
11/02/16 a
Consider a pseudo-Riemannian metric g on a 5D manifold with coordinates (u, t, x) with line
element
ds2 = 2 dt du − 2V (x, t) dt2 + δij dxi dxj .
∂
Note that ∂u is a null Killing vector of this metric. Consider the equations of motion. The
u-equation gives ẗ = 0, so ṫ2 is a constant, and we will choose ṫ2 = 1. The x-equation is just that of
Newtonian geodesics:
∂V
ẍi = −δ ij j ṫ2 .
∂x
The t-equation is
∂
∂V 2
u̇ − 2V ṫ = − ṫ .
∂τ ∂t
– 16 –
3 Hamiltonian Mechanics/Symplectic Geometry 3.7 Integrable systems

3.7 Integrable systems

In the general case of geodesic motion, integrability is the existence of sufficiently many first
integrals.

Definition. An integrable system is a symplectic manifold (M, ω) of dimension 2n together

with n functions fi : M → R, i = 1, . . . , n such that {fi , fj } = 0 for all i, j (the functions are
said to be in involution), and df1 ∧ df2 ∧ · · · ∧ dfn 6= 0 on M.

Theorem 4 (Arnold-Liouville). Let (M, ω, fi ) be an integrable system with Hamiltonian f1 (say).

Then:

1. If Mf = {x ∈ M, f1 = c1 , . . . , fn = cn }, where the ci are constants, is compact and connected,

then it is diffeomorphic to an n-dimensional torus T n = |S 1 × S 1 × {z
· · · × S 1}.
n

2. There exists a canonical transformation to “action-angle” variables (I1 , . . . , In , φ1 , . . . , φn ),

| {z } | {z }
actions angles
such that the angles φ are periodic with period 2π and are coordinates on Mf , and Ij are first
integrals (in a neighbourhood of Mf in M), such that Hamilton’s equations are solvable by
quadratures:
∂H ∂H
I˙i = − = 0, φ̇i = = ωi (I1 , . . . , In ).
∂φi ∂I i
So, H = H(I1 , . . . , In ) and Ii (t) = Ii (0), φi (t) = φi (0) + ωi (I)t.

Proof (sketch). Mf is a manifold. Each fi gives rise to a Hamiltonian vector field Xfi tangent
to Mf , and we have [Xfi , Xfj ] = −X{fi ,fj } = 0. Hence the Xfi define an n-dimensional Abelian
group Rn of transformations on M. Restrict this group to Mf and use the orbit-stabilizer theorem
to obtain
Rn
Mf ' = T n,
Γ
where Γ is some lattice subgroup. Note that Mf ⊂ M is “Lagrangian” (i.e. ω|Mf = 0) since the
Xfi span Tp Mf at any p ∈ Mf , and ω(Xfi , Xfj ) = 0. Therefore we can write ω = dθ for some
1-form θ.
On T n we have n distinct non-contractible closed paths (cycles). For example there are two of
these on the 2-torus:

Choose coordinates 0 ≤ φ̃i ≤ 2π, 1 ≤ i ≤ n, such that each cycle can be written as

Γk = {0 ≤ φ̃k ≤ 2π, φ̃i = constant, i 6= k}.

– 17 –
 4 Topological charges in field theory 4 Topological charges in field theory

Define n action coordinates by

1
I
Ik = θ.
2π Γk

Note that these don’t depend on the choice of Γk (that is, up to homeomorphism). Thus Ik only
depends on f1 , . . . , fn and so are first integrals with H = f1 .
Choose a point q0 on Mf and define a generating function
Z q
S(qj , Ij ) = θ.
q0

∂S
The angle coordinates are then defined by φj = ∂I j
.
(pi , qj ) → (Ii , φj ) is a canonical transformation, and we can write H = H(I1 , . . . , In ).

Example. All Hamiltonian systems in 2 dimensions are integrable. For example, take H(p, q) =
1 2 2 2 2
2 (p + ω q ) and M = R , ω = dp ∧ dq = d(p dq). Then Mf = {H(p, q) = E}, and we call E
the energy. Mf is an eclipse in the p-q plane; let ε be the region it encloses. We have action
coordinate
1 1 area E
I ZZ
I= p dq = dp dq = = ,
2π Mf 2π  2π ω
∂H
so can write H(p, q) = E = ωI. On paths, I is constant, and we have φ̇(t) = ∂I = ω, so
φ(t) = φ(0) + ωt.

4 Topological charges in field theory

Lecture 10
11/02/16 b 4.1 Scalar kinks
Consider Minkowski space R1,1 with metric ds2 = dt2 − dx2 , and let φ : R1,1 → R be a scalar field
with Lagrangian given by
1 2
Z 
L= φt − φ2x − U (φ) dx = T − V.
R 2
The Euler-Lagrange equations provide the equation of motion
dU
φtt − φxx = − .
dφ

We want the field to habe a stable ground state, so we need U (φ) ≥ U0 for some constant U0 . We
will normalise so that U0 = 0.
Now assume that U −1 (0) = {φ1 , φ2 , . . . } has more than one element.

φ
φ1 φ2

– 18 –
4 Topological charges in field theory 4.1 Scalar kinks

There are different ways in which we could handle this. One way is perturbation theory, leading
to QFT, Feynman diagrams, and uncomfortable infinities. We choose a ground state, φ1 say, and
set φ ≈ φ1 + δφ, where δφ is small and dynamical. Then the Euler-Lagrange equations become
( + m2 )δφ = 0.
Another way is to assume that finite energy solutions must approach an element of U − 1(0) sufficiently
quickly as x → ±∞.

Definition. Solitons are non-singular, static, finite energy solutions to Euler-Lagrange equa-
tions.

For example, we might have a soliton with φ → φ1 as x → −∞, and φ → φ2 as x → +∞ (called a

non-perturbative kink):

φ
φ2

φ1

h i2
Example. Suppose U (φ) = 12 dWdφ(φ) (in such an instance W is called the superpotential).
The energy of the field is then given by
1
Z  
E= dx φ2t + (φx + Wφ )2 ∓ 2 φx Wφ
2 R | {z }
= dW
dx
1
Z 
= dx φ2t + (φx ± Wφ )2 ∓ [W (φ(∞)) − W (φ(−∞))]
2 R
≥ |W (φ(∞)) − W (φ(−∞))| .

This is the Bogomolny bound, and it is saturated if the Bogomolny equations hold:
dφ dW
φt = 0, =∓
dx dφ

Definition. Let φ± = limx→±∞ φ. The topological charge of φ is given by

Z +∞
dφ
N = φ+ − φ− = dx .
−∞ dx

The topological charge only depends on boundary conditions at ±∞, and since E is finite, it is
conserved. We can classify solitons by their topological charges:

– 19 –
 4 Topological charges in field theory 4.2 Degree of a map

φ φ φ

x x x

(a) N > 0: kink (b) N < 0: antikink (c) N = 0: trivial

Lecture 11
23/02/16

4.2 Degree of a map

Definition. Let M and M0 be oriented, compact manifolds without boundary, and let
f : M → M0 be a smooth map. The topological degree of f is given by
Z Z
f ∗ (vol(M0 )) = deg(f ) vol(M0 ),
M M0

where ω = vol(M0 ) ∈ Λn (M0 ) and dim M = dim M0 = n.

Topological degree is independent of the choice of volume form ω. Suppose ω and ω̂ are such that
Z Z
ω= ω̂.
M0 M0
R
Then we have M0 (ω̂ − ω) = 0, so ω̂ = ω + dα for some (n − 1)-form α. Hence
Z Z Z Z Z
∗ ∗ ∗ ∗
f (ω̂) − f (ω) = f (dα) = d(f α) = f ∗α = 0
M M M M ∂M=∅

Also, deg(f ) ∈ Z and counts the number of pre-images of a point in M0 under f .

Theorem 5. deg(f ) is an integer given by

X
deg(f ) = sign(J(x)),
x∈f −1 (y)

∂y i
 
where J(x) = det ∂xj
, and y is regular.

Example. For functions f : S 1 → S 1 , deg(f ) is equivalent to the winding number.

– 20 –
4 Topological charges in field theory 4.2 Degree of a map

f (θ)

2π

f0 − −
+ + +

θ
0 2π

We have
∂f
X  
N= sign ,
θ:f (θ)=f0
∂θ

which is 1 − 1 + 1 + 1 − 1 = 1 in the above example. Equivalently we have

Z 2π
1 1 df
Z
N= df = dθ .
vol(S 1 ) S 1 2π 0 dθ

For example, if we view S 1 as the set of complex numbers with modulus 1, and let f (z) = z k ,
then deg(f ) = k.

Example. Consider f : S 2 → S 2 , and write f a = f a (xi ) ∈ R3 , a = 1, 2, 3, |f | = 1. Using

spherical coordinates

x = r sin θ cos φ, y = r sin θ sin φ, z = r cos θ,

it can be shown that

1 ∂f b ∂f c 2
Z
deg(f ) = ij abc f a d ρ,
8π ∂ρi ∂ρj
where ρi = (θ, φ).

Example. Let f : M → SU (2) ' S 3 , where dim M = 3. Then, using vol(S 3 ) = 2π 2 , it can be
shown that
1
Z h i
− 3
deg(f ) = Tr (f 1 df ) .
24π 2 M

– 21 –
 4 Topological charges in field theory 4.3 Applications to physics

4.3 Applications to physics

Sigma model lumps
Consider a field φ : R2,1 → S N −1 ⊂ RN that is free (i.e. there is no potential, V (φ) = 0). The
target space is non-linear. Treating φa as a point in RN , the Lagrangian is
1 ∂φa ∂φb
L = η µν µ ν δab ,
2 ∂x ∂x
and we have the added constraint
N
X
φa φa = 1.
a=1
Thus we introduce a Lagrange multiplier, making the new Lagrangian
1 
L0 = L − 1 − |φ|2 .
2
The Euler-Lagrange equations give φa − λφa = 0. Dotting with φ, we find that λ = a φa ,
P
aφ
so we obtain a non-linear set of field equations
N
!
X
φa − φb φb φa = 0.
b=1

Another way we could approach this would be to solve |φ|2 = 1 for φN , and then substitute it back
into the Lagrangian. We obtain
1 ∂φp ∂φq φp φq
L = gpq (φ)ηµν µ ν where gpq = δpq + P −1 r r , p, q = 1, . . . , N − 1.
2 ∂x ∂x 1− N
r=1 φ φ

g is the metric induced on S N −1 from δ on RN . Proceeding with the Euler-Lagrange equations

obtains the same equation as above.
φ
In general a field φ is a map between two pseudo-Riemannian manifolds, (Σ, η) → (M, g). Some
examples:
• For a point particle moving on a curved spacetime, we have Σ = R, η = ds2 and (M, g) is
spacetime.
• In the bosonic sector of a superstring, (Σ, g) is a Riemann surface (the string worldsheet),
and dim(M) = 10.
• When dim(Σ) = k, we say that we are considering a “(k − 1)-brane”.
Lecture 12
25/02/16 Return to the more specific case and set N = 3, so that φ : R2,1 → S 2 . For a static field, we have
φ : R2 → S 2 . We want to consider finite energy conditions, i.e. those with finite
1
Z
E= ∂i φa ∂i φa dx dy .
2 R2

If we change to polar coordinates, we see that (schematically) we then require |∇φ|2 r dr dθ → 0

sufficiently quickly as r → ∞, or equivalently r|∇φa | → 0 as r → ∞. Thus we have φ → φ∞ , a
constant, at spatial infinity. Without loss of generality, we can take φ∞ to be the North pole, (0, 0, 1).
Since φ is constant near spatial infinity, φ extends to a one point compactification R2 + {∞} = S 2 .
Therefore static, finite energy configurations are maps φ : S 2 → S 2 . These are classified by their
degree (topological charge). This is only a partial classification, as fields can have different energies
with the same deg(φ).

– 22 –
5 Gauge Theory 5 Gauge Theory

Lemma 1. The energy is bounded from below, with E ≥ 4π| deg(φ)|. Equality is obtained when the
1st order Bogamolny equations hold:
∂φa ∂φc
i
= ±ij abc φb j
∂x ∂x
Proof. Consider the following inequality:
Z   
∂i φa ± ij abc φb ∂j φc ∂i φa ± ik ade φd ∂k φe d2 x ≥ 0

If we use ij ik = δjk and abc ade = δ bd δ ce − δ be δ cd , and φa ∂i φa = 12 ∂i φa φa = 0, then we see
Z h i
0≤ ∂i φa ∂i φa + δjk (δ bd δ ce − δ be δ cd )φb φd ∂j φc ∂k φe ± 2ij ∂i φa abc φb ∂j φc d2 x
= 2E + 2E ± 16πQ ≥ 0,
where Q = deg(φ). Since E ≥ 0, we obtain the result.

5 Gauge Theory
5.1 Hodge duality
Consider Rn , η = dx2 − dt2 , with signature (n − t, t), and xa = (x, t), and assume we have a volume
1
form vol = n! ab...c dxa ∧ dxb ∧ · · · ∧ dxc . We have a natural inner product on p-forms given by
1 a1 ...ap
(α, β) = α βa1 ...ap .
p!
We define the Hodge operator ∗ : Λp → Λn−p by
λ ∧ µ = (∗λ, µ) · vol ∀ µ ∈ Λn−p .
In components, we have
(−1)t
(∗λ)b1 ...bq = a1 . . . ap b1 . . . bq λa1 ...ap ,
p!
and following things through, we find ∗ ∗ λ = (−1)t+p(n−p) λ.

Example. Consider n = 4, t = 1, p = 2. Then ∗ ∗ F = −F (e.g. the Maxwell field). We choose

an orientation such that the volume form is dx ∧ dy ∧ dz ∧ dt. With the Maxwell 2-form we
usually write
F = −Ei dt ∧ dxi + B1 dx2 ∧ dx3 + B2 dx3 ∧ dx1 + B3 dx1 ∧ dx2 .
Under the hodge star, we have ∗(E, B) = (−B, E). Comparing with
1
F = F0i dt ∧ dxi + Fij dxi ∧ dxj ,
2
we see Ei = F0i and Bi = 12 ijk Fjk . The Maxwell equations dF = 0 give F = dA for some
1-form A. If we write A = −φ dt + Ai dxi , then we see
∂φ ∂Ai
 
F = dA = − i
− dxi ∧ dt + ∂[k Ai] dxk ∧ dxi ,
∂x ∂t
or
E = −Ȧ − ∇φ, B = ∇ × A.

– 23 –
5 Gauge Theory 5.2 Yang-Mills equations

Example. Consider n = 4, t = 0 (Euclidean 4-space). For all 2-forms F , we have ∗ ∗ F = F .

If F = dA and ∗F = F , then the Bianchi equations dF = 0 imply the Maxwell equations
d ∗ F = 0. The 2nd order Maxwell equations follow from the 1st order self-duality condition
F = ∗F .
Note that any F can by written
1 1
F = (F + ∗F ) + (F − ∗F ) .
2
| {z } |2 {z }
=F+ =F−

We have ∗F+ = F+ and ∗F− = −F− , i.e. F+ is self-dual and F− is anti-self-dual. In other
words
Λ2 (R4 ) = Λ2+ ⊕ Λ2− ,
where the first vector space is dimension 6, and the latter two are dimension 3.

5.2 Yang-Mills equations

Consider (Rn , η, vol), and G a Lie group with Lie algebra g. A gauge potential A is a G-valued
1-form on Rn . Expanding in components, we write
A = Ab dxb = Aαb Tα dxb , Aαb : Rn → R,
where Tα , α = 1, . . . , dim G is a basis of g, with structure constants given by [Tα , Tβ ] = cαβψ Tγ .
The gauge field corresponding to A is
1
F = dA + A ∧ A = Fab dxa ∧ dxb .
2
Evaluating the components of F , we have
Fab = ∂a Ab − ∂b Aa + [Aa , Ab ] = [Da , Db ],
where D = d + A is the covariant derivative of A.
We identify (A, A0 ) and (F, F 0 ) if
A0 = gAg −1 − dg · g −1 , F 0 = gF g −1
for some g : Rn → G, called a gauge transformation.
F is in the adjoint representation, so we can deduce the Bianchi identity
DF = dF + [A, F ]
= d2 A + dA ∧ A − A ∧ dA + A ∧ dA − dA ∧ A + A ∧ A ∧ A − A ∧ A ∧ A
= 0.
The Yang-Mills equations are D ∗ F = 0. These arise from the action
Z
S=− Tr(F ∧ ∗F ).

If we set G = U (1), we recover Maxwell theory. If we take G = SU (2), then } is the set of
2 × 2 anti-Hermitian matrices, and we have a basis where [Tα , Tβ ] = −αβγ Tγ . In this basis,
Tr(Tα Tβ ) = − 12 δαβ , and hence
1  
− Tr(F ∧ ∗F ) = − Tr F ab Fab vol.
2
– 24 –
 5 Gauge Theory 5.3 Yang-Mills instantons and Chern forms

5.3 Yang-Mills instantons and Chern forms

Definition. Instantons are non-singular equations of motion in Euclidean space (R4 ), whose
action is finite.
Lecture 13
01/03/16
Consider R4 , η = δ, with boundary conditions on A such that
1 1
   
Fab (x) ∼ O 3 , Aa (x) ∼ O 2 − ∂a g · g −1
r r

as r → ∞. The gauge transformation g need only be defined asymptotically as r → ∞ on ∂R4 ' S∞ 3 ,

3 3
so for example g : S∞ → G = SU (2) = S . The topological degree of g will become relevant.
Consider gauge theories on Rn for general n. We define the Chern forms in the following way. Let
i
 
C(F ) = det 1 + F = 1 + C1 (F ) + C2 (F ) + . . . .
2π

Cp (F ) is the 2p-form part of the Chern form and is O(F p ); it is called the pth Chern form. C is
gauge invariant (C(F ) = C(gF g −1 )), and closed (by the Bianchi identity).
We have:
i
• C1 (F ) = 2π Tr(F ). This vanishes for G = SU (2) (but not G = U (1)).
1 1
• C2 (F ) = 8π 2
(Tr(F ∧ F ) − Tr F ∧ Tr F ). For G = SU (2), this is C2 (F ) = 8π 2
Tr(F ∧ F ).

Using the Bianchi identity DF = df + A ∧ F − F ∧ A = 0, we have

1 1
dC2 = 2
Tr(dF ∧ F ) = 2 Tr(−A ∧ F ∧ F + F ∧ A ∧ F ) = 0.
4π 4π
Hence C2 is exact, so we can write C2 = dY3 , where Y3 is the Chern-Simons 3-form. It is an
exercise to verify that
1 2 3
 
Y3 = 2 Tr dA ∧ A + A .
8π 3
Returning to n − 4, we define the Chern number by
Z
c2 = C2 (F ).
R4

We have
Z
c2 = dY3
R4
1 1
Z   
= 2 d Tr F ∧ A − A3
8π R4 3
1
Z  
=− Tr A3
24π 2 S∞3

1
Z h i
−1 3
= Tr (dg · g )
24π 2 S 3
= deg(g) ∈ Z.

This integer gives a classification of instantons on R4 .

– 25 –
 5 Gauge Theory 5.4 Fibre bundles & connections

Theorem
R
6. The Yang-Mills action S = − R4 Tr(F ∧ ∗F ) on a given topological sector c2 ≥ 0
is bounded from below by 8π 2 c2 . The bound is saturated iff the anti-self-dual (ASD) Yang-Mills
equations hold (F = − ∗ F , or F12 = −F34 , F13 = −F42 , F14 = −F23 ).

Proof. First note that F ∧ F = ∗F ∧ ∗F for any 2-form F (this can be seen as the difference of two
squares). Thus we have
1
Z Z
S=− Tr [(F + ∗F ) ∧ (F + ∗F )] + Tr(F ∧ F ) ≥ 8π 2 c2 ,
2 R4 R4
| {z } | {z }
total square, trace negative definite =8π 2 c2

and equality iff F + ∗F = 0.

This is the Bogomolny bound for the Yang-Mills actions. A similar calculation with c2 ≤ 0 gives
SDYM (F = ∗F ).
If ∗F = ±F , then D ∗ F = ± DF = 0, so the full Yang-Mills equations hold.
The instanton number is k = −c2 .

5.4 Fibre bundles & connections

Lecture 14
03/03/16
Yang-Mills theory is equivalent to a theory of connections on principal bundles, with the gauge
group G on fibres. Fibre bundles are “locally product manifolds”. For example, the cylinder and
Möbius band are both examples of S 1 “locally multiplied” by R. More concretely:

Definition. A fibre bundle {E, π, B, F, G} is a structure consisting of:

1. Two manifolds E and B with dim E > dim B, and a smooth map π : E → B. π is the
projection from the total space E onto the base B.

2. A manifold F (the fibre) such that for any point x ∈ B, there exists an open Uα ⊂ B
with x ∈ Uα and an diffeomorphism φα : Uα × F → π −1 (Uα ) satisfying π(φα (x, f )) = x.

3. Transition functions φαβ = φ−1 α ◦ φβ and G 3 φαβ : F → F , f 7→ φαβ (x, f ) with

x ∈ Uα ∩ Uβ . G is a Lie group: the structure group of the bundle, the group of
transformations of F . We require φαα = identity, and φαβ φβγ = φαγ for x ∈ Uα ∩ Uβ ∩ Uγ
(this is the “cocycle relation”).

A principal bundle has F = G, and G acts on F by left translations.

A vector bundle has F = Rn and G ⊂ GL(n, R).
A trivial bundle has E ' B × F . It turns out that a bundle is always trivial if B is contractible.

Example. E = T B = x∈B Tx B, the tangent bundle, is a vector bundle with Fx = π −1 (x) '
S

Rn and G = GL(n, R). Points in E take the form (x, v), where x is a “position”, and v is a
“velocity”.

– 26 –
 5 Gauge Theory 5.4 Fibre bundles & connections

Example. Consider B = S 2 , F = G = U (1) ' S 1 . This is the “magnetic monopole bundle”.

U+

U−

We can write B = U+ ∪ U− as shown in the diagram. We use a fibre coordinate ψ for eiψ ∈ U (1),
with 0 ≤ ψ ≤ 2π. On the base we have local spherical coordinates (θ, φ) with 0 ≤ θ ≤ π
and 0 ≤ φ ≤ 2π. Suppose a point in U+ ∩ U− has coordinates (θ, φ, eiψ+ ) on U+ × U (1) and
coordinates (θ, φ, eiψ− ) on U− × U (1). We have eiψ− = einφ eiψ+ . In order for E to be a manifold
n must be an integer, and it is known as the monopole number. n classifies the U (1) principal
bundles over S 2 . If n = 0, then E = S 2 × S 1 , a trivial bundle. If n = 1 then E = S 3 , and this
is known as the Hopf fibration.
Lecture 15
08/03/16

Example. Consider bundles with B = S 2 , F = G = S 1 , and monopole number n = 1, so

E = S 3 . We can view S 3 as a subset of C2 with

S 3 = {(z1 , z2 ) ∈ C2 s. t. |z1 |2 + |z2 |2 }.

U (1) acts on S 3 by right multiplication (z1 , z2 ) → (z1 eiα , z2 eiα ). Note that this action preserves
the ratio z1 /z2 ∈ S 2 , so π : S 3 → S 2 = S 3 /U (1).
Another description involves taking C2 ' R4 . A complex line A1 z1 + A2 z2 = 0 through the
origin is CP1 ' S 1 , and this intersects S 3 in S 1 . This is a fibre over a point z1 /z2 ∈ CP1 ' S 2 .

Definition. A section of a bundle π : E → B is a map s : B → E such that π ◦ s = identity

on B. This generalises functions on B.

Example. Vector fields are sections of the tangent bundle T B, xa ∈ B → V b (xa ) ∂x∂ b .

Lemma 2. A principal bundle admits a global section iff it is trivial.

Proof. Let x ∈ B. Suppose s : B → E is such a section. Then for all p ∈ π −1 (x) = Gx , we can
write p = s(x) · Rg (p), where Rg (p) is the right translation of G on Gx moving s(x) to p. Then
E → (x 6= π(p), g(p) ∈ B × G is a global trivialisation.

– 27 –
 5 Gauge Theory 5.4 Fibre bundles & connections

Definition. A connection on a principal bundle (E, π, B, G) is a g-valued 1-form whose vertical

component (i.e. the part cotangent to G) is the Maurer-Cartan form on G.

In local coordinates we can write a connection as ω = γ −1 Aγ + γ −1 dγ, where A is a g-valued

1-form on B, γ ∈ G are fibre coordinates, and γ −1 dγ is the Maurer-Cartan 1-form.
Let U, U 0 be two overlapping open sets on B, and g = gU U 0 be the transition function on B defining
E, so that γ 0 = gγ. We want A0 such that
( )
γ −1 Aγ + γ −1 dγ on U × G
ω= 0 −1 0 0 0 −1 are equal on U ∩ U 0 .
γ Aγ +γ dγ on U 0 × G
0

From this, we obtain

A0 = gAg −1 − dg · g −1 ,
so A and A0 are related by a gauge transformation on B.

Definition. The curvature is a g-valued 2-form on E defined as

Ω = dω + ω ∧ ω.

Using the form of ω above, we have

Ω = dγ −1 Aγ + γ −1 dγ + γ −1 A ∧ Aγ + γ −1 A dγ + γ −1 dγ γ −1 ∧Aγ
| {z }
=−d(γ −1 )

= γ −1 (dA + A ∧ A)γ
= γ −1 F γ.

Any local section γ : B → G can be used to pull back (ω, Ω) from E to B such that A = γ ∗ (ω) is
the gauge potential, and F = γ ∗ (Ω) is the gauge field. Gauge transformations on B are then just
equivalent to a change of sections.
If E → B is non-trivial, then a global section does not exist, and A can only be defined locally on
Lecture 16 B.
10/03/16

Horizontal decomposition
Let (E, B, π, G) be a principal bundle. Given a connection ω on that bundle, we can define a
splitting T E = H(E) ⊕ V (E) in the following way.

Definition. The horizontal subbundle is

H(P ) = {X ∈ T E s. t. Xy ω = 0}.

Horizontal vectors X annihilate ω = γ −1 Aγ + γ −1 dγ, so they also annihilate γωγ −1 = A + dγ · γ −1 .

We write dγ · γ −1 = λα ⊗ Tα , where Tα is a basis of g with structure constants given by [Tα , Tβ ] =
γ
fαβ Tγ .

– 28 –
5 Gauge Theory 5.5 Back to Yang-Mills instantons
∂
We claim that a basis of H(E) is Da = ∂xa − Aαa Rα , where Rα are right-invariant vector fields on
G. As a check, we have
∂
 
Da y(γωγ −1 ) = − Aαa Rα y(Ab dxb + λβ ⊗ Tβ ) = Aa − Aαa Tα = 0.
∂xa
Frobenius integrability of H(E) requires [H(E), H(E)] ⊂ H(E), so we see that curvature is an
obstruction to this, since
α
[Da , Db ] = −Fab Rα where Fab = ∂a Ab − ∂b Aa + [Aa , Ab ],

and this only belongs to H(E) if F = 0.

5.5 Back to Yang-Mills instantons

For any instanton on R4 , there exists a bundle E over S 4 which (under stereographic projection
S 4 → R4 ) maps a pullback of a connection to Yang-Mills potentials. Boundary conditions on R4
are equivalent to an extension of the instanton to S 4 .
We cover S 4 by two regions U± such that U+ ∩ U− = S 3 × [−, ] for some . Consider E → S 4 ,
G = SU (2), with a connection
−1 −1
(
γ+ A+ γ + + γ + dγ+ on U+ ,
ω= −1 −1
γ− A− γ+ + γ− dγ− on U− .

On U+ ∩ U− , γ− = gγ+ for some g : U+ ∩ U− → SU (2). The curvature is

−1
(
γ+ F+ γ+ on U+ ,
Ω = dω + ω ∧ ω = −1
γ− F− γ− on U− .

Introduce Chern-Simons 3-forms Y± on U± . Then we have

1
Z
k=− 2 Tr(Ω ∧ Ω)
8π S 4
"Z #
1
Z
=− 2 Tr(F+ ∧ F+ ) + Tr(F− ∧ F− ) in the limit  → 0
8π U+ U−
1 1 1
Z     
=− 2 Tr F+ ∧ A+ − A3+ − Tr F− ∧ A− − A3−
8π S 3 3 3
(now use the gauge equivalence of A+ and A− )
1 1
Z   h i
=− Tr (dg · g −1 )3 − d (A+ ∧ dg) · g −1
8π 2 S3 3 | {z }
vanishes in integral since ∂S 3 = ∅
1
Z  
=− Tr (dg · g −1 )3 ∈ Z.
24π 2 S3

This is the same formula as before, but we have a different interpretation. g is a transition function
for E → S 4 .
All information about instantons is in these topological numbers; they are connection invariant.

Fin
– 29 –