Gauge theory and variational principles

David Bleccker

Jackozee retyping
2
Chapter 0

Preliminaries

0.1 Multilinear algebra and forms

0.1.0 Notation. Throughout this retyping we will make use of Penrose’s abstract index
notation interchangeably with index free notation, depending on the situation. The first
letters of the latin alphabet (a, b, c, . . . , etc) will be used as abstract indices. I.e, they will
denote the slots of an object. For instance, the action of a dual vector ω on a vector v will be
denoted equivalently both by index free notation and by abstract indices.

ω(v) ≡ ωa v a

At the same time, it will sometimes be necessary to express objects by their components
with respect to a basis. For this purpose we will use greek indices. That is, if the vectors
{vµ }µ∈In (where In represents the set {1, 2, . . . , n}) constitute a basis, we can represent any
vector w as
w = w a = µ w µ vµ a
P

Here we won’t use Einstein’s summation convention, in order to make explicit the fact that
greek indices represent components and hence take numerical values.
When we are working with various basis at the same time, it will be necessary to distin-
guish between the components taken with respect to each basis. We will do this by adding
carets ( ˆ) or primes ( 0 ) to the indices of these components. For instance, the same vector w in
the previous example can be decomposed with respect to a different basis {vµ̂ }µ̂∈Iˆn (where
Iˆn represents the set {1̂, 2̂, . . . , n̂}) by

w = wa = µ̂ v a
P
µ̂ w µ̂

Also, when we need to use numeric indices that are just labels (i.e, not components), we
will use the middle part of the latin alphabet (i, j, k, . . . , etc).

0.1.1 Notation. Let E, F be R vector spaces, with dim E = n

Let Ê be the dual space of E.
Let T 0,0 (E, F ) := F .
For p, q > 0, T p,q (E, F ) is the space of multilinear functions (F-valued tensors)

f : Ê ×p × E ×q → F

We denote T p,q (E, R) by T p,q (E).

3
4 CHAPTER 0. PRELIMINARIES

If {ui }i∈Ip ⊆ E and {v̂ i }i∈Iq ⊆ Ê, then the tensor product
p
O q
O
ui v̂ j = u1 ⊗ · · · ⊗ up ⊗ v̂ 1 ⊗ · · · ⊗ v̂ q ∈ T p,q (E)
i=1 j=1

is defined by  
Op q
O p
Y q
Y
j 1 p i
 ui v̂ (x̂ , . . . , x̂ , y1 , . . . , yq ) := ui (x̂ ) v̂ j (yj )
i=1 j=1 i=1 j=1

for any {x̂i }i∈Ip ⊆ Ê and {yi }i∈Ip ⊆ E.

Let {vµ }µ∈In ⊆ E be a basis for E, with {v̂ µ }µ∈In ⊆ Ê its dual basis. Then any f ∈ T p,q (E)
can be uniquely written as
a ...a P P µ ...µ
fb11...bqp = µi νi fν11...νqp vµ1 a1 ⊗ · · · ⊗ vµp ap ⊗ v̂ ν1 b1 ⊗ · · · ⊗ v̂ νq bq

Let Λk (E, F ) be the subspace of T 0,k (E, F ) consisting of the f A a1 ,...,ak ∈ T 0,k (E, F ) such
that
f A a1 ,...,ak = f A [a1 ,...,ak ]
where the uppercase A is an abstract index corresponding to the vector space F , and the
brackets around the indices are the antisymmetrization brackets.
Also, Λ0 (E, F ) = T 0,0 (E, F ) = F ,
and Λk (E) := Λk (E, R).
The elements of Λn (E) are called top forms.

0.1.2 Definition. For α ∈ Λp (E) and β ∈ Λq (E), we define the wedge α ∧ β ∈ Λp+q (E) by

(p + q)!
(α ∧ β)a1 ...ap+q = (α ⊗ β)[a1 ...ap+q ]
p!q!

0.1.3 Definition. The determinant of an element φ ∈ T 1,1 (E) is defined by

ω(φ(v 1 ), . . . , φ(v n ))
det(φ) :=
ω(v 1 , . . . , v n )

where ω ∈ Λn (E) is an arbitrary top form, and the vectors {v µ }µ∈In are an arbitrary basis of
E. One can verify this definition is independent of both choices.
A metric on E is a choice of an element g ∈ T 0,2 (E) such that g is symmetric and non-
degenerate. i.e
∀u,v∈E : g(u, v) = g(v, u)
∀u∈E : g(u, v) = 0 =⇒ v=0
An orthonormal (o.n.) basis of E relative to g is a basis {eµ }µ∈In such that
(
a b ±1 if µ = ν
gab eµ eν =
0 otherwise

Then, for this orthonormal basis, we define

ηαβ := gab eα a eβ b
0.1. MULTILINEAR ALGEBRA AND FORMS 5

A volume element of E relative to g is a top form µ ∈ Λn (E) given by

µ = ê1 ∧ · · · ∧ ên
for an orthonormal basis {eµ }µ∈In of E.
Note that êi ∈ Λ1 (E) and that µ(e1 , . . . , en ) = 1.
Actually µ depends on the choice of orthonormal basis, but only by a factor of ±1. A choice
of the two possible volume elements is called an orientation of E.
If µ is an orientation for E, then a basis {vµ }µ∈In of E is called positively oriented if
µ(v1 , . . . , vn ) > 0
0.1.4 Definition. If gab is a metric on E, then there is an induced metric g̃ ∈ T 0,2 (Λk (E))
defined as follows.
If the metric gab ∈ T 0,2 (E) is non degenerate, then it follows that there exists a unique tensor
g ab ∈ T 2,0 such that
g ac gcb = δba
where δ is the identity map on both the vector space E and its dual Ê. i.e.
∀v∈E : ∀ω∈ê : δba v b = v a and ωb δab = ωa
Using g ab we define g̃ by
1 a1 b1 a2 b2
g̃(α, β) := g g · · · g ak bk αa1 ...ak βb1 ...bk
k!
0.1.5 Definition. Suppose that we are given metrics gab on E and hAB on F . Then there is
a metric (gh) on Λk (E, F ) defined by
1
[gh](α, β) := hAB g a1 b1 g a2 b2 · · · g ak bk αA a1 ...ak β B b1 ...bk
k!
0.1.6 Theorem. Let g be a metric on E, suppose that µ is a volume element of E relative to
g. For n = dim(E), there is a unique linear isomorphism ? : Λk (E) → Λn−k (E), such that
∀α,β∈Λk (E) : α ∧ ?β = g̃(α, β)µ

Proof. For every γ ∈ Λn−k (E), define a function ϕγ : Λk (E) → R such that
∀α∈Λk (E) : ϕγ (α)µ = α ∧ γ
We can proof that
∀α∈Λk (E) : ϕγ (α) = 0 =⇒ γ=0
Thus, γ 7→ ϕγ defines a one-to-one linear map Λn−k (E) → Λk (E). Since
n!
dim(Λk (E)) = = dim(Λn−k (E))
k!(n − k)!
this map is an isomorphism. Thus,
∀β∈Λk (E) : ∃!γ∈Λn−k (E) : ∀α∈Λk (E) : ϕγ (α) = g̃(α, β)
We then take ?β to be this γ. Then
α ∧ ?β = α ∧ γ = ϕγ (α)µ = g̃(α, β)µ
6 CHAPTER 0. PRELIMINARIES

0.1.7 Notation. Let gab be a metric on E. For any ωa1 ...ak ∈ Λk (E), we define

ω a1 ...ak := g a1 b1 · · · g ak bk ωb1 ...bk

Let g ∈ T 0,2 (E) be a metric on E, µ ∈ Λn (E) be an orientation of E with respect to g, and

g̃ ∈ T 0,2 (Λk (E)) be the inner product on Λk (E) induced by g. We define the sign of the metric
by
sgn(g) := g̃(µ, µ) = ±1
note that this definition is orientation independent, since the possible difference in the sign
of the volume form is cancelled anyway, because µ appears quadratically.
Using the identity map δ we define the kth delta system as
[a a ] [a a ]
δba11...b
...ak
k
a1
:= k!δ[b11 · · · δbkk] = k!δb11 · · · δbkk = k!δ[b1
· · · δbakk]

This implies that contracting k lower or upper indices with the kth delta system has the
effect of antisymmetrizing those indices, plus a factor of k!.

δba11...b
...ak
k
ωa1 ...ak = k! ω[b1 ...bk ] and δba11...b
...ak b1 ...bk
k
x = k! x[a1 ...ak ]

Using the formula for the trace of the identity map (δaa = n) we can deduce the behaviour of
the delta systems when contracted with temselves.
a ...a c c a ...a a ...a
δb11...bppc = δc b11...bpp = (n − p) δb11...bpp
a ...a c ...c c ...c a ...a (n − p)! a ...a
δb11...bppc11...cqq = δc11...cqqb11...bpp = δ 1 p
(n − (p + q))! b1 ...bp

Using the nth delta system, we define the permutation symbols εµ1 ...µn and εµ1 ...µn with
respect to the basis {vµ }µ∈In by

εµ1 ...µn := δba11...b

...an 1
n
v̂ a1 · · · v̂ n an vµ1 b1 · · · vµn bn

εµ1 ...µn := δba11...b

...an
n
v1 b1 · · · vn bn v̂ µ1 a1 · · · v̂ µn an
these symbols satisfy

1
 if {µ1 , . . . , µn } is an even permutation of In
εµ1 ...µn = εµ1 ...µn = −1 if {µ1 , . . . , µn } is an odd permutation of In

0 otherwise


in any basis.
It is noteworthy that neither of the permutation symbols constitute the components of a ten-
sor. Nevertheless, as one can verify from their definition, both symbols together constitute
the components of the nth delta system.

εµ1 ...µn εν1 ...νn = δνµ11...ν

...µn
n

For a matrix aµ ν we define it’s ”numerical determinant” by

det(a· · ) := µ1 µn
P
µi εµ1 ...µn a 1 . . . a n
0.1. MULTILINEAR ALGEBRA AND FORMS 7

Since the right hand side of the equation is antisymmetric in the indices 1 to n, we can write
it as
det(a· · )εν1 ...νn = µi εµ1 ...µn aµ1 ν1 . . . aµn νn
P

multiplying both sides by εν1 ...νn and solving for det(φ· · ) we get

1 P P ν1 ...νn µ1
det(a· · ) = δ a ν1 . . . aµn νn
n! µi νi µ1 ...µn
repeating this process with a matrix whose indices are both up or both down, we get

1P P
det(a· · ) = εµ ...µ εν ...ν aµ1 ν1 . . . aµn νn
n! µi νi 1 n 1 n
1 P P µ1 ...µn ν1 ...νn
det(a· · ) = ε ε aµ1 ν1 . . . aµn νn
n! µi νi
We reckon that we can also calculate the sign of the metric by taking the determinant of
the matrix ηµν defined in 0.1.3 by ηµν := g(eµ , eν ) (for an orthonormal basis {eµ }µ∈In ).

det(η·· ) = sgn(g)
This method of calculation is also well defined, since it is well known (Silvester’s ’inertia of
signature’ theorem) that for all orthogonal bases, the number of positive self products (and
hence the determinant of ηµν ) is invariant.
We define the determinant of the metric with respect to the basis {vµ }µ∈In by
P 
|g·· | := | det(g·· )| =  µi ε
 µ1 ...µ n gµ1 1 . . . gµn n 


This quantity is not basis independent. Nevertheless, it will give us information about the
relation between the basis and the metric.
Using the permutation symbol, we make a reformulation of the expression for the volume
form µ. We begin by noting that

1P
µ = ê1 ∧ · · · ∧ ên = εµ ...µ êµ1 ∧ · · · ∧ êµn
n! µi 1 n
for an orthonormal basis {eµ }µ∈In of E.
Now we also introduce an arbitrary basis {vµ̂ }µ̂∈In of E. Since both are bases of E, we know
that there exists an endomorphism φ ∈ T 1,1 (E) that maps eµ 7→ vµ̂

vµ̂ a = φa b eµ b

Or, in terms of components

vµ̂ a = ν a
P
ν φ µ̂ eν

where in this exceptional case, both indices ν and µ̂ of φ represent its components taken
with respect to {eµ }µ∈In .
Then
gµ̂ν̂ = gab vµ̂ a vν̂ b
= γ δ gab φγ µ̂ eγ a φδ ν̂ eδ b
P P

= γ δ ηγδ φγ µ̂ φδ ν̂
P P
8 CHAPTER 0. PRELIMINARIES

hence
det(gˆ·ˆ· ) = sgn(g)(det(φ·ˆ· ))2
= ±(det(φ·ˆ· ))2
1
=⇒ | det(φ·ˆ· )| = |gˆ·ˆ· | 2
It follows that
1P
µ= εµ ...µ êµ1 ∧ · · · ∧ êµn
n! µi 1 n
1P P
= εµ ...µ φµ1 ν̂1 · · · φµn ν̂n v̂ ν̂1 ∧ · · · ∧ v̂ ν̂n
n! µi νi 1 n
1P
= det(φ·ˆ· )εν̂1 ...ν̂n v̂ ν̂1 ∧ · · · ∧ v̂ ν̂n
n! µi
And if {vµ̂ }µ̂∈In is positively oriented w.r.t µ (and hence det(φ·ˆ· ) > 0), then

1P 1
ν̂1 ν̂n
µ= µi |gˆ·ˆ· | εν̂1 ...ν̂n v̂ ∧ · · · ∧ v̂
2
n!
hence, we have obtained the components of µ relative to any positively oriented basis
{vµ }µ∈In (Dropping hats)

1P 1
µ1 ∧ · · · ∧ v̂ µn
µ= µi |gˆ·ˆ· | εµ1 ...µn v̂
2
n!
1
= µi |gˆ·ˆ· | 2 εµ1 ...µn v̂ µ1 ⊗ · · · ⊗ v̂ µn
P

1
=⇒ µµ1 ...µn = |g ·· | 2 εµ1 ...µn
If we use the metric to raise all the indices of µ, we obtain a totally antisymmetric tensor
in Λk (E) := Λk (Ê).
µa1 ...an = g a1 b1 . . . g an bn µb1 ...bn
Now we calculate its components in a basis.

µµ1 ...µn = µ 1 ν1 . . . g µn νn µν1 ...νn

P
νi g
1
µ1 ν1 . . . g µn νn εν1 ...νn
P
= |g ·· | 2 νi g
1
= |g ·· | 2 det(g ·· )εµ1 ...µn

now, since g µν is the inverse matrix of gµν , we have

1
|g ·· | 2 µ1 ...µn
= ε
det(g·· )

if the basis is positively oriented, det(g·· ) = sgn(g)|g·· |, and

sgn(g)
= 1 εµ1 ...µn
|g ·· | 2

Using this information, we obtain the identity

µa1 ...1n µb1 ...bn = sgn(g) δba11...b

...an
n
0.1. MULTILINEAR ALGEBRA AND FORMS 9

0.1.8 Theorem. Let g be a metric on E with orientation µ. For ω ∈ Λk (E), we have

1 a1 ...ak
(?ω)ak+1 ...an = ω µa1 ...ak ak+1 ...an
k!
˜ω denote the right hand side of the above equation. By the definition of ?, we
Proof. Let ?
only need to verify that α ∧ ?β = g̃(α, β)µ for all α, β ∈ Λk (E). Now
n!
(α ∧ ˜
?β)a1 ...an = (α ⊗ ˜?β)[a1 ...an ]
k!(n − k)!
n!
= α (˜?β)ak+1 ...an ]
k!(n − k)! [a1 ...ak
n! 1
= α β b1 ...bk µ|b1 ...bk |ak+1 ...an ]
k!(n − k)! k! [a1 ...ak
where the bars || mean the antisimmetrization is not taken over those indices
1 1
= αc ...c β b1 ...bk µb1 ...bk ck+1 ...cn δac11...c n
...an
k!(n − k)! k! 1 k
sgn(g) 1
= αc ...c β b1 ...bk µb1 ...bk ck+1 ...cn µc1 ...cn µa1 ...an
k!(n − k)! k! 1 k
1 1
= αc ...c β b1 ...bk δbc11...b
...cn
k ck+1 ...cn
µa1 ...an
k!(n − k)! k! 1 k
1 1
= αc ...c β b1 ...bk δbc11...b
...ck
µa1 ...an
k! k! 1 k k

1
= αb1 ...bk β b1 ...bk µa1 ...an
k!
= g̃(α, β)µa1 ...an

0.1.9 Theorem. For ω ∈ Λk (E), we have ? ? ω = sgn(g)(−1)k(n−k) ω.

Proof. We have
1
(? ? ω)b1 ...bk = (?ω)ak+1 ...an µak+1 ...an b1 ...bk
(n − k)!
1 1 a1 ...ak
= ω µa1 ...ak ak+1 ...an µak+1 ...an b1 ...bk
(n − k)! k!
1 1
= ωa ...a µa1 ...ak ak+1 ...an µak+1 ...an b1 ...bk
(n − k)! k! 1 k
sgn(g) 1 a ...a a ...a
= ωa ...a δ 1 k k+1 n
(n − k)! k! 1 k ak+1 ...an b1 ...bk

the interchange of indices in the delta system introduces a factor of (−1)k(n−k)

sgn(g) 1 a ...a a ...a
= (−1)k(n−k) ωa ...a δ 1 k k+1 n
(n − k)! k! 1 k b1 ...bk ak+1 ...an
1
= sgn(g)(−1)k(n−k) ωa1 ...ak δba11...b
...ak
k! k

k(n−k)
= sgn(g)(−1) ωb1 ...bk
10 CHAPTER 0. PRELIMINARIES

since ω is already antisymmetric.

0.1.10 Remark. There is a star operator ? : Λk (E, F ) → Λn−k (E, F ) defined componentwise
relative to a basis of F using ? on Λk (E).

0.2 Manifolds and tensor analysis