Applications of Differential Geometry To Physics: Cambridge Part III Maths
Applications of Differential Geometry To Physics: Cambridge Part III Maths
Lent 2016
Applications of Differential
Geometry to Physics
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
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
Gauge theory (both Maxwell and Yang-Mills), which will lead to discussions on an object known
as the connection on a principal bundle.
1.1 Manifolds
–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).
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 .
In fact:
–3–
1 Introduction 1.2 Vector Fields
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:
V |p γ
M p
–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:
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 :
∂ ∂
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) + ).
–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 :
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 ))
∂ ∂ ∂ ∂
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.
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.
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
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:
∂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:
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:
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
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
–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 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.
σ1 = dx , σ2 = dy , σ3 = dz − x dy .
– 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.
h = hαβ σ α ⊗ σ β
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
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
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
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
– 12 –
3 Hamiltonian Mechanics/Symplectic Geometry 3.1 Symplectic manifolds
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
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
Example. Now restrict the Poisson structure from the previous example to S 2 ⊂ R3 , and
reparametrise to spherical coordinates
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
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
– 14 –
3 Hamiltonian Mechanics/Symplectic Geometry 3.3 Canonical transformations
– 15 –
3 Hamiltonian Mechanics/Symplectic Geometry 3.5 Geodesics in non-Riemannian geometries
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.
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.
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
– 17 –
4 Topological charges in field theory 4 Topological charges in field theory
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.
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.
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
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
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
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=∅
∂y i
where J(x) = det ∂xj
, and y is regular.
– 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
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. 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
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
– 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.
– 23 –
5 Gauge Theory 5.2 Yang-Mills equations
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.
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
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
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 ).
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.
– 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
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 .
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.
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
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
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 .
Example. Vector fields are sections of the tangent bundle T B, xa ∈ B → V b (xa ) ∂x∂ b .
– 27 –
5 Gauge Theory 5.4 Fibre bundles & connections
Ω = dω + ω ∧ ω.
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.
H(P ) = {X ∈ T E s. t. Xy ω = 0}.
– 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 ],
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 –