Mathematical Structures in Physics

Winter term 2009/10

Christoph Schweigert
Hamburg University
Department of Mathematics
Section Algebra and Number Theory
and
Center for Mathematical Physics
(as of 02.02.2010)

Contents
1 Newtonian mechanics
1.1 Galilei space, equations of motion . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2 Lagrangian mechanics
2.1 Variational calculus . . . . . . . . .
2.2 Systems with constraints . . . . . .
2.3 Lagrangian systems . . . . . . . . .
2.4 Symmetries and Noether identities
2.5 Natural geometry . . . . . . . . . .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

1
1
8
10
10
17
22
35
39

3 Classical field theories

42
3.1 Maxwells equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.2 Special relativity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.4 Electrodynamics as a gauge theory . . . . . . . . . . . . . . . . . . . . . . . . . 53
4 Hamiltonian mechanics
55
4.1 (Pre-) symplectic manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.2 Hamiltonian dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5 Quantum mechanics
5.1 Deformations . . . . . . . . . . .
5.2 C -algebras and states . . . . . .
5.3 Strong quantizations . . . . . . .
5.4 Schrodinger picture and examples

.
.
.
.

.
.
.
.

.
.
.
.
i

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

68
68
71
79
86

6 A glimpse to quantum field theory

91

A Differentiable Manifolds
A.1 Definition of differentiable manifolds . . . .
A.2 Tangent vectors and differentiation . . . . .
A.3 Fibre bundles . . . . . . . . . . . . . . . . .
A.4 Vector fields and Lie algebras . . . . . . . .
A.5 Differential forms and the de Rham complex
A.6 Riemannian manifolds and the Hodge dual .

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

94
94
97
100
103
106
113

The current version of these notes can be found under

http://www.math.uni-hamburg.de/home/schweigert/ws09/pskript.pdf
as a pdf file.
Please send comments and corrections to schweigert@math.uni-hamburg.de!
These notes are based on lectures delivered at the University of Hamburg in the fall term
2007/2008 and in the fall term 2009/2010 as part of the master programs in physics and
mathematical physics. I am grateful to Linda Sass for notes taken in 2007/08.

References:
References per Chapter:
1. Classical Mechanics
Arnold, Vladimir I.: Mathematical methods of classical mechanics, Springer Graduate
Text in Mathematics 60, Springer, New York, 1978.
Goldstein, Herbert; Poole, Charles and Safko, John: Classical mechanics, Pearson/Addison Wesley, Upper Saddle River, 2002
etc.
2. Variational bicomplex and Lagrangian mechanics
Anderson, Ian M.: Introduction to the variational bicomplex, in M. Gotay, J. Marsden
and V. Moncrief (eds), Mathematical Aspects of Classical Field Theory, Contemporary
Mathematics 132, Amer. Math. Soc., Providence, 1992, pp. 5173. http://www.math.usu.
edu/~fg_mp/Publications/ComtempMath/IntroVariationalBicomplex.ps
Olver, Peter: Applications of Lie Groups to Differential Equations. Springer Graduate
Texts in Mathematics 107, 1986
Reyes, Enrique G.: On covariant phase space and the variational bicomplex. Internat. J.
Theoret. Phys. 43 (2004), no. 5, 12671286. http://www.springerlink.com/content/
q151731w71l71q23/
3. Hamiltonian mechanics and symplectic geometry
Berndt, Rolf: An introduction to symplectic geometry, American Mathematical Society,
Providence, 2001
ii

 Souriau, Jean-Marie: Structure of dynamical systems. Birkhauser, Boston, 1997. http:

//www.jmsouriau.com/structure_des_systemes_dynamiques.htm
4. Quantum mechanics
Takhtajan, Leon A.: Quantum Mechanics for Mathematicians. Graduate Studies in Mathematics, American Mathematical Society, 2008.
Messiah, Albert: Quantum mechanics. Dover Publications, 1999.
Strocchi, F.: An introduction to the mathematical structure of quantum mechanics. Advanced Series in mathematical physics 27, World Scientific, Singapur, 2005.
Giulini, Domenico: That Strange Procedure Called Quantisation. Lecture Notes in Physics
631, Springer, Berlin, 2003. http://de.arxiv.org/abs/quant-ph/0304202
5. Electrodynamics
Romer, Hartmann, M. Forger Elementare Feldtheorie. VCH, Weinheim 1993.
http://www.freidok.uni-freiburg.de/volltexte/405/pdf/efeld1.pdf
Appendix A: Manifolds
Bredon, Glen E.: Topology and geometry. Springer Graduate Texts in Mathematics 139,
1993.
Madsen, Ib and Tornehave, Jrgen: From Calculus to cohomology: de Rham cohomology
and characteristic classes. Cambridge University Press 1997.
Bott, Raoul and Tu, Loring W.: Differential forms in algebraic topology. Springer Graduate Texts in Mathematics 82, New York, 1982.

iii

Newtonian mechanics

1.1

Galilei space, equations of motion

A basic postulate of classical physics requires (empty) space and time to be homogenous. In
classical physics, the idea of empty space should be accepted as central. This is afforded by the
notion of an affine space over a k-vector space V . Here k is a field, in our application the field
R of real numbers.
Definition 1.1.1
(i) An affine space is a pair (A, V ), consisting of a set A and a k-vector space V together
with an action of the abelian group (V, +) underlying V on the set A that is transitive
and free.
In more detail, an action is a map
:V AA
such that
(v + w, a) = (v, (w, a)) for all v, w V, a A .
An action is called transitive, if for all p, q A exists v V such that (v, p) = q; an
action is called free, if this v V is unique.
We call dimk V the dimension of A and write dim A = dimk V .
We also say that the affine space (A, V ) is modelled over the vector space V .
(ii) We introduce the notation (v, p) = p + v and, if v is the unique vector in V such
that q = p + v, we write q p = v. We also say that A is a (V, +)-Torsor or a
principal homogenous space for the group (V, +). The group (V, +) is also called the
difference space of the affine space.
(iii) As morphisms between affine spaces (A1 , V1 ) (A2 , V2 ) modelled over vector spaces V1 , V2
over the same field, we consider affine maps, i.e. maps
: A1 A2
for which there exists a linear map A : V1 V2 such that
(p) (q) = A (p q)

for all p, q A .

Remarks 1.1.2.
1. Note that the (V, +)-equivariant morphisms are those morphisms for which A = idV .
2. While a vector space has the zero vector as a distinguished element, there is no distinguished element in an affine space.
3. Any two affine spaces over the same vector space are isomorphic, but not canonically
isomorphic.
1

4. Recall the definition of a semi-direct product of two groups H, N . Given a group homomorphism : H Aut(N ) the set N H can be endowed with the structure of a group
N o H by
(n, h) (n0 , h0 ) := (nh (n), hh0 ) .
The automorphism group of an affine space is isomorphic to the semi-direct product
Aut(A) = V o GL(V ) ,
where V acts on A by translations.
5. The choice of any point p A induces a bijection (, p) : V A of sets. As a finitedimensional R-vector space, V has a unique topology as a normed vector space. By considering preimages of open sets in V as open in A, we get a topology on A that does not
depend on the choice of base point. We endow A with this topology.
6. We will see in the appendix that affine space An is an n-dimensional manifold. The choice
of a point p A provides a natural global coordinate chart.
Suppose that we wish to model space; since space is three-dimensional, we have n = 3.
To be able to talk about lengths, we have to model measurements with rods. Our idealization
implies that there are infinitely long rods and rulers. In a first step, the edge of the ruler has a
distinguished direction and a marked point on it.
Definition 1.1.3
1. A ray L in a real vector space V is a subset of the form L = R0 v with v V \ {0}.
2. A halfplane in a real vector space V is a subset H V such that there are linearly
independent vectors v, w V with H = Rv + R0 w. The boundary of a halfplane is the
only line through the origin contained in it. (For the halfplane just given, this is the line
Rv.)
3. A rotation group for a real three-dimensional vector space V is a subgroup D GL(V )
which acts transitively and freely on the set of pairs consisting of a halfplane and a ray
on its boundary.
Proposition 1.1.4.
For any three-dimensional vector space, the map
{Scalar products on V }/R>0 Rotation groups
which maps the scalar product b to its special orthogonal group SO(V, b) is a bijection.
Proof:
We have to construct an inverse of the map. If one assumes that the rotation group is compact,
an invariant scalar product can be obtained by integration. For an elementary (but lengthy)
proof without this assumption, we refer to W. Soergel, Herleitung von Skalarprodukten aus
Symmetrieprinzipien. Mathematische Semesterberichte 68 (2008) 197-202.


Next we wish to talk about length scales. Such a length scale is e.g. the prototype meter.
It consists of a ruler with two marked points. Fixing one point, the other point moves on the
orbit of the rotation group.
Definition 1.1.5
1. Given a rotation group D, we call an orbit l V \ {0} a unit length.
2. Given a unit length, we define a norm on V as follows. We first remark that any ray
intersects a given unit length l in precisely one point. If the ray through w V intersects
l in v l and w = v with 0, we define the norm on W by |w| := .
We introduce the measurements of lengths with the help of a ruler with two marked points;
we find that the three-dimensional space of our intuition should have the following mathematical
structure:
Definition 1.1.6
1. An n-dimensional Euclidean space En is an n-dimensional affine space An together with
the structure of a Euclidean vector space on the difference space.
2. As morphisms of Euclidean spaces, we only admit those affine maps for which the linear
map A is an isometry, i.e. an orthogonal map.
3. The group of automorphisms of a Euclidean space is called a Euclidean group.

Proposition 1.1.7.
The Euclidean group of a Euclidean vector space En is a semi-direct product of the subgroup V
of translations given by the action of V and the rotation group SO(V, b):
Aut(E) = V o SO(V, b)
It is a non-compact Lie group of dimension n +

n(n1)
.
2

Including also time, we obtain the model of empty space in classical physics:
Definition 1.1.8
(i) A Galilei space (A, V, t, h, i) consists of
An affine space A over a real four-dimensional vector space V . The elements of A
are called events or space time points.
A non-zero linear functional
t:V R
called absolute time difference function.
The structure of a Euclidean vector space with positive definite scalar product h, i
on ker t.
(ii) t(a b) R is called the time difference between the events a, b A.
3

(iii) Two events a, b A with t(a b) = 0 are called simultaneous. This gives an equivalence
relation on Galilei space. The equivalence class
Cont(a) = {b A | t(b a) = 0}
is the subset of events simultaneous to a A.

Remarks 1.1.9.
(a) Cont(a) for any a A is a three-dimensional Euclidean space.
(b) While it makes sense to talk about simultaneous events a, b at different places, the phrase:
The two events a, b are at different time, but at the same place in three-dimensional space.
does not make sense.
Definition 1.1.10
As the morphisms of two Galilei spaces (A1 , V1 , t1 , h, i1 ) and (A2 , V2 , t2 , h, i2 ) we consider those
affine maps
: A1 A2
which respect time differences
t1 (b a) = t2 ((b) (a)) for all a, b A1 ,
and the Euclidean structure on space in the sense that the restriction A : ker t1 ker t2 is an
orthogonal linear map.
Remarks 1.1.11.
(a) An example for a Galilei space is given by the Galileian coordinate space G = (R1
R3 , R4 , pr1 , h, i). This is R R3 seen as an affine space over the vector space R4
= R R3
with the projection on the first component
t = pr1 : R4
= R1 R3 R
as the time difference functional and the standard Euclidean scalar product h, i on R3 .
(b) Two Galilei spaces are isomorphic, but not canonically isomorphic. The automorphism
group Aut(G) of the Galileian coordinate space G is called the Galilei group.
(c) We consider three classes of automorphisms in Aut(G):
Uniform motions with velocity v R3 :
g1 (t, x) = (t, x + vt)
Spacial translations by x0 R3 combined with time translations by t0 R:
g2 (t, x) = (t + t0 , x + x0 )

 Rotations and reflections in space with R O(3):

g3 (t, x) = (t, Rx)
One can show that Aut(G) is generated by these elements. It is a ten-dimensional noncompact Lie group.
Definition 1.1.12
1. Let X be any set and
:XG
be a bijection of sets. (On the right hand side, it would be formally more correct to write
the set underlying G.) We say that provides a global Galilean coordinate system on X.
2. We say that two Galilean coordinate systems 1 , 2
relative uniform motion, if
1 1
2 Aut(G) .

G are in

Any global Galilean coordinate system : X G endows the set X with the structure of
a Galilean space over the vector space R4 :
we define on X the structure of an affine space over R4 by requiring to be an affine map:
x + v := 1 ((x) + v)

for all

x X, v R4 .

Then we use the standard time difference functional pr1 : R4

= R R3 R and the standard
3
4
Euclidean structure on R to get a Galilei space (X, R ) that depends on . For this Galilei
space, we have an isomorphism

(X, R4 ) G
of Galilei spaces such that the linear map A associated to is the identity.
Lemma 1.1.13.
1. Being in relative uniform motion is an equivalence relation on the set of global Galilean
coordinate systems on X.
2. Two Galilean coordinate systems 1 , 2 in relative uniform motion endow the set X with
the structures of Galilei spaces for which the identity is an isomorphism of Galilei spaces.
This leads us to the following
Definition 1.1.14
1. A Galilean structure on a set X is an equivalence class of Galilean coordinate systems.
2. Given a Galilean structure on a set X, any coordinate system of the defining equivalence
class is called an inertial system or inertial frame for this Galilean structure.

Remarks 1.1.15.
5

1. By definition, two different inertial systems for the same Galilean structure are in uniform
relative motion.
2. There are no distinguished inertial systems.

Principle 1.1.16. (Galilean principle of relativity)

All laws of nature are of the same form in all inertial systems.
To make this principle more precise for mechanical system, we have to describe the motions
of mass points.
Definition 1.1.17
(i) A trajectory of a mass point in R3 is an (at least twice) differentiable map
x : I R3
with I R an interval.
(ii) A trajectory of N mass points in R3 is an N -tuple of (at least twice) differentiable maps
x(i) : I R3
with I R an interval. Equivalently, we can consider an (at least twice) differentiable
map
~x : I (R3 )N .
(iii) The velocity in t0 I is defined as the derivative:
x(t
0) =

dx
(t0 ) .
dt

(iv) The acceleration in t0 I is defined as the second derivative:

x(t0 ) =

d2 x
(t0 ) .
dt2

(v) The graph

{(t, x(t) | t I} R R3
of a trajectory is called the world line of the mass point. We consider a world line as a
subset of Galilean coordinate space G.
(vi) If we parameterize a world line by a different function I G with I an interval, we call
the parameter I the eigentime of the mass point.
The following principle is the basic axiom of Newtonian mechanics. It cannot be derived
mathematically but should rather seen as description of observations in nature.
Definition 1.1.18 [Newtonian determinism]
6

(a) A Newtonian trajectory of a point particle

: I R3 ,

with

I = (t0 , t1 ) R

is completely determined by the initial position (0 ) and the inital velocity


d

.

dt
t=t0
(b) In particular, the acceleration at t0 is determined by the initial position and the initial
velocity. As a consequence, there exists a function
F : R3N R3N R R3N ,
where the first factor are spacial coordinates, the second are velocities and the third is time,
such that for all Newtonian trajectories the Newtonian equation
d2 i
x = xi = F i (x, x,
t), i = 1, ..., N.
dt2
holds. This is a second order ordinary differential equation that completely determines the
system.

Remarks 1.1.19.
(a) The function F has to be measured experimentally and describes the acting forces. It determines the physical system. For the particular case F = 0, no forces act and x = 0. As
a consequence, the trajectory is a uniform motion. (Aristotle had a different idea about the
situation!)
(b) Standard mathematical propositions assure that a sufficiently smooth function F determines, together with the initial conditions on position and velocity, the trajectory. Implicitly, we will frequently assume the existence of global solutions or of solutions for sufficiently
large times.
(c) We will see that not all graphs (t, x(t)) G describe physical motions.
So far we have described trajectories using maps : I R3 giving rise to a graph (t, x(t)
G. Suppose we are given a Galilei space A with internal frames : A G. We then get a map
Definition 1.1.20
Let A be a Galilei space, I an interval of eigentime and : I A a smooth function. Then
is called a physical motion if for all inertial frames : A G the function : I G is
the graph of a Newtonian trajectory.
This is can be seen as a more precise version of the Newtonian principle of relativity.
Remarks 1.1.21.
1. We immediately have the following consequences of the Newtonian principle of relativity.
Invariance under time translations: the force F does not depend on time t.
7

 Invariance under spacial translations: F depends only on the relative coordinates

xi x1 .
Invariance under relative uniform motion: F only depends on the relative velocities.
Invariance under rotations: the force F is a vector, i.e. transforms in the same
representation of the rotation group as x and x.

2. Altogether, this implies that the motion of n points is described by a function

x(k) (t) = f (x(j) x(l) , x (s) x (r) )
with x(i) : I R3 .
3. We deduce Newtons first law: a system consisting of a single point is described in an
inertial system by a uniform motion (t, x0 + tv0 ). In particular, the acceleration vanishes,
x = 0.

1.2

Examples

It should be emphasized that the discussion of Newtonian relativity was for a system in empty
space. In general, it is interesting to consider systems where F is a function violating these
principles. In this way, we can conveniently summarize the effect of other mass points in the
system. For example, one might switch the perspective from considering the two-body system
consisting of the sun and the earth by considering just the motion of earth in the background
of the force exerced by the mass of the sun.
We investigate some examples in a fixed inertial system with the help of a function x : R
3
R . We call R3 the configuration space of the system.
Examples 1.2.1.
(a) Freely falling particle:
x = g
x3 with g 9, 81 ms2 .
One should notice that a direction is distinguished by the unit vector x3 in the direction
of the x3 -axis. Hence the isotropy of space is violated. We introduce a function, called
potential energy
U (x1 , x2 , x3 ) : R3 R such that grad U = F ;
in our case U (x1 , x2 , x3 ) = gx3 . The function U is not unique, but can be added by any
function whose derivative vanishes. The equations of motion then read x = grad U .
(b) To write down Newtons law of gravity, we consider a potential energy depending only on
the distance r from the center of gravity:
q
k
U (x1 , x2 , x3 ) =
with r := x21 + x22 + x23
r
x = grad U =

k
x
r3

(c) One-dimensional oscillator. To investigate the equation of motion x = 2 x, we introduce

2 2
the potential U = 2x . This can be realized e.g. using a mass connected to a spring.
Experimentally, one finds that for equal springs with equal initial conditions, but different
masses, the ratio
x1
= const1,2
x2
only depends on the balls, but not on the initial conditions. We put
m2
x1
=
x2
m1
with a quantity mi called inertial mass which is a property of the i-the ball. Again, we
introduce a physical unit by comparing to a standard, e.g. the prototype kilogram in Paris.
Observation 1.2.2.
Consider a mechanical system given by a force F (x, x,
t) which we suppose right away to be
given by a potential of the form
U : R3 R
and thus independent of t and x.
We have to study the coupled system of ordinary differential
equations
x = grad U (x)
()
of second order. For any solution x : I R3 of () we consider the real-valued function
: IR
1
2
(t) = kx(t)k

+ U (x(t)) .
2
For its derivative, we find
d
(t) = hx,
xi + hgrad U , xi
= 0.
dt
This suggests the following point of view:
Lemma 1.2.3.
The system () is equivalent to the following system of ordinary differential equation of first
order for the function y : I R6
y 1 = y2 , y 2 = grady1 U (y1 , y3 , y5 )
y 3 = y4 , y 4 = grady3 U (y1 , y3 , y5 )
y 5 = y6 , y 6 = grady5 U (y1 , y3 , y5 )
The solutions of this system are called phase curves.
We thus enlarged the space by considering the first derivatives as independent geometric
quantities.

Observation 1.2.4.
This suggests to introduce R6 with coordinates x, y, z, ux , uy , uz . We call this the phase space P
of the system. We equip the phase space P with the energy function
1
E(x1 , x2 , x3 , ux1 , ux2 , ux3 ) := (ux1 2 , ux2 2 , ux3 2 ) + U (x1 , x2 , x3 ) .
2
It should be appreciated that the first term in E is a quadratic form.
On phase space, we consider the ordinary differential equations
dxi
duxi
= u xi ,
= gradi U (x1 , x2 , x3 ) () .
dt
dt
The solutions of () are called phase curves. They are contained in subspaces of P of constant
value of E.
Observation 1.2.5.
1. Assume that for any point M P a global solution xM (t) of () with initial conditions
M exists. This allows us to define a mapping
gt : P P

by gt (M ) = xM (t).

2. Standard theorems about ordinary differential equations imply that gt is a diffeomorphism,

i.e. a differential map with differentiable inverse mapping.
3. We find gt1 gt2 = gt1 +t2 and thus a smooth action
g: RPP
(t, M ) 7 gt (M )
called the phase flow.

2
2.1

Lagrangian mechanics
Variational calculus

We start by considering the following mathematical problem:

Observation 2.1.1.
1. We wish to associate to trajectories : I RN , I = [t1 , t2 ] scalar values. Examples of
such functionals are the length of a trajectory:
Z
L1 () = kkdt,

or its energy
Z
L2 () =

kk
2 dt .

One can ask for which trajectories one obtains extremal values for a given functional.
This leads to an ordinary differential equation for the trajectory. In fact, this is a convenient way to describe (certain) ordinary differential equations in terms of scalar-valued
functionals which are sometimes much easier to handle, in particular as far as covariance
properties and coordinate transformations are concerned.
10

2. The functional we are interested in is allowed to depend on the trajectory and its derivatives. To this end, we need apart from position space
RN with coordinates (x1 , ..., xN )
also recipients for higher derivatives of . We therefore introduce
J1 (RN ) = RN RN
with coordinates (x1 , ..., xN , x1t , ..., xN
t ). This comes with a natural injection
RN , J1 (RN )
(x1 , ..., xN ) 7 (x1 , ..., xN , 0, ..., 0).
Since the equations of motion are second order differential equations, we continue by
adding a recipient for the second derivatives:
J2 (RN ) = RN RN RN with coordinates (xi , xit , xitt ) .
We also use the abbreviation xi2 := xitt . If we continue this way, we get a system of vector
spaces for derivatives up to order
J (RN )
= (RN )
with embeddings
J , J+1 (RN ).
3. Suppose we are given a smooth trajectory
: I RN .
It comes with derivative functions of all orders which we can use to extend to a function
to the larger spaces we just constructed:
j : I J R N
where the components are given by
(j )in =

dn i
.
dtn

4. The functionals of length and energy can now be expressed in terms of integrals over the
following functions on J1 RN
l1 , l2 : J1 (RN ) R
N
1
X
i
i
i 2 2
l1 (x , xt ) =
(xt )
l2 (xi , xit ) =

i=1
N
X

(xit )2

i=1

as

Z
Li () =
I

11

dt li (j1 ).

5. The variational problem consists of finding trajectories

: I RN
which for a given function
l : J (RN ) R
are stationary points for the functional on trajectories
Z
L() = dt l(j ).
I

To make the problem well-defined, we fix boundary values ai , bi RN , i = 0, 1, ..., 1

and restrict ourselves to trajectories with I = [t1 , t2 ] in the subset
n
o
di
di
()
T(ai )(bi ) = : I RN : i (t1 ) = ai , i (t2 ) = bi .
dt
dt
()

6. We wish to find extremal trajectories in T(ai )(bi ) . To this end, we consider differentiable
one-parameter families of trajectories
: I (0 , 0 ) RN .
We introduce for  (0 , 0 ) the notation
 : I RN
t 7 (t, ),
()

where we require that for all values of  that  T(ai )(bi ) . An application of the chain
rule yields
Z
Z
d 
d 

l j ( ) = 
dt l( ,  ,  , . . .)

d =0 I
d =0 I
Z
l 
d
l d i
=
+ ...
 i d i i + i
i
xt d 
I x x = dt d
We assume for simplicity that l restricts to a function on J1 RN . We notice that
2 (t, )
2 (t, )
d i
 =
=
d
t
t
By partial integration with respect to the variable t, we find
Z 
Z
d
d  l di 
l 
d l 
i

+
=


i
i i d i
i

dt xit xi = ddt
d 
I dt x d
I x x = dt

Z 
Z
 l di 

X 2l
l
d i
j

1
=

j
()

+
dt
.

j
i
i d
i +1
x
d
x
x
x
I
I
t

j,
|
{z
}
important expression

12

Definition 2.1.2
To any function
l : J1 (RN ) R
the Euler-Lagrange operator associates an RN -valued function
E(l) : J2 (RN ) RN
X 2l
l
E(l)i =

xj+1
j
i
i
x
x xt
j,
We rewrite it using a total derivative operator
D :=

xi+1

on functions of x, xt , xtt , . . . by
E(l)i =

xi

l
l
D i .
i
x
xt

()

Given a trajectory in T(ai )(bi ) , we obtain an RN -valued function on I R:

7 E(l)i j () =: E(l) []
In terms of this function, the derivative with respect to  becomes
Z

d
d 
L() =
E(l) [=0 ] ,
.
d
d
I
The boundary conditions at t0 and t1 enforce
d 
 = 0.

d t=t0 ,t1
We are now ready to apply the next lemma which uses standard facts from real analysis:
Lemma 2.1.3.
A continuous real-valued function
f : I = [t1 , t2 ] R
which has the property that for any smooth function
h: IR
vanishing at the end points of the interval, h(t1 ) = h(t2 ) = 0, the integral over the product
vanishes,
Z t2
f h dt = 0 ,
t1

is identically zero, f (t) = 0 for all t [t1 , t2 ].

13

Proof:
Suppose that f does not vanish identically. Possibly after replacing f by f , we find t0 <
t < t1 such that f (t ) > 0. Since f is continuous, we find a neighbourhood U of t such that
f (t) > c > 0 for all t U .
We can now find a smooth function h with support in U such that h|U = 1 for some
neighborhood of t contained in U . We thus find the inequalities
Z t1
Z
Z
f h=
f h>
f > |U |c > 0
t0

contradicting our assumptions.

From this we deduce the following

Proposition 2.1.4.
A trajectory : I RN is a stationary point for a functional on trajectories given by a function
l : J (RN ) R,
if and only if the N ordinary differential equations
E(l) [] = 0
hold. This set of ordinary differential equations is called Euler-Lagrange equations for the function l on the trajectory .
Examples 2.1.5.
(a) A so-called natural system of classical mechanics is given by a function of the form
N

l(x, xt ) =

1X i 2
(x ) V (xi )
2 i=1 t

We compute
l
= gradi V
xi
and find for the Euler-Lagrange operator

l
= xit
xit

E(l)i = gradi V xitt .

This gives us a system of N equations
xitt = gradi V
which gives the following Euler-Lagrange equations on a trajectory


i = gradi V ( ) ,
i.e. Newtons equations of motion in the presence of a force given by the gradient of the
potential V . If there is a potential for the forces of a system, all information about the
equations of motion is thus contained in the function
l : J(RN ) R
the Lagrangian.
14

(b) The length of the curve



= (t, x) : x = (t) mit t0 6 t 6 t1 R2
is given by
Z

t1

L() =

p
1 + 2 dt .

t0

We thus consider the function

p
1 + x2t .

l(x, xt ) =

To find the Euler-Lagrange equations, we compute

l
=0
x

l
xt
=p
xt
1 + x2t

and

and find
l
xtt
xtt =
3 = 0
xt xt
(1 + x2t ) 2
which reduces to xtt = 0. Hence we get the differential equation = 0 on the trajectories
which have as solutions (t) = ct + d, i.e. the stationary points are straight lines.
(c) To illustrate that a Lagrangian can be more easily re-expressed in other coordinates than
an equation of motion, we consider the two-dimensional case without potential given in
Cartesian coordinates by
l(x1 , x2 , x1t , x2t ) =

1
(x1t )2 + (x2t )2 .
2

In polar coordinates, we find by applying the chain rule

x1,t = (r cos )t = rt cos r sin t
x2,t = (r sin )t = rt sin + r cos t
(1)
and thus for the function
l(r, , rt , t ) =

1
1
1
rt cos r sin t )2 + rt2 sin + r cos t )2 = (rt2 + r2 2t ).
2
2
2

It has as solutions straight lines. In radial coordinates, we compute

l
=0

and

l
= r 2 t
t

which implies the equation for the trajectory

d 2
(r )
=0
dt
that expresses the conservation of angular momentum with respect to the origin. For the
radial coordinate we find
l
= r 2t
r

and

l
= rt
rt
15

and thus

r = r 2 .

(d) To investigate the motion of earth around the sun, we consider Keplers problem that is
defined by the following function
lk : J1 (R3 ) R
k
lk (x, xt ) = 21 kxt k2 + kxk
.
Here k is a constant that we keep for convenience. Instead of solving the corresponding
equation of motion exactly, we show how to use a scaling argument to learn more about
solutions.
To this end, we consider for any > 0 the map
s : J1 (R3 ) J1 (R3 )
s (x, xt ) := (2 x, 1 xt ).
It is chosen such that for the Lagrangian we have
l s = 2 l
and therefore for the Euler-Lagrange operators
E [l s ] = 1 E [l] .
Suppose we have a trajectory : R RN that is a global solution of the Euler-Lagrange
equations
E [l] j1 = 0 .
Consider for any R>0 the trajectory
: R R3
(t) := 2 (3 t).
We then find
j1 (t) = (2 (3 t), 2 3 (
3 t)) = s j1 (3 t)).
We find
E [l] j1 (t) = E [l] s j1 (3 t) = 1 E [l] j1 (3 t) = 0.
and conclude that the trajectory is a solution of the equations of motion as well.
An explicit calculation reveals that there solutions for which the trajectory is an ellipsis
with focal point x = 0. The scaling affects the semimajor axis as
a = 2 a.
If T is the time for one period,
(t + T ) = (t),
then
(t + 3 T ) = 2 (3 t + T ) = 2 (3 t) = (t),
and hence we find for the periods
T = 3 T.
16

We have found from scaling considerations Keplers third law:

6 a3
a3
a3
=
=
T2
6 T 2
T2
is independent of .
We introduce some terminology:
Definition 2.1.6
R
1. Given a Lagrangian l(xi , xit , t), we call L() = I dt l j the corresponding action for the
trajectory .
2. A coordinate function xi is called a generalized coordinate, xit a generalized velocity.
i

called the generalized momentum canonically conjugate to x .

force.

l
xi

l
xit

is

is called the generalized

3. A coordinate is called cyclic if the Lagrangian does not depend on it, i.e.
l
= 0.
xi

Proposition 2.1.7.
The momentum canonically conjugate to a cyclic coordinate is constant for any solution of the
Euler-Lagrange equations.
Proof:
For any solution
: I RN
of the Euler-Lagrange equations, we have for a cyclic coordinate xi
l
d l
j1 =
j1 = 0.
i
d t xt
xi


2.2

Systems with constraints

Frequently, one observes that N particles can only move on a submanifold of R3N . Examples
include:
A roller coaster is (hopefully) more or less obliged to move on a one-dimensional submanifold.
The points of a rigid body are required to move in such a way that their distances remain
constant.

17

In each situation, this is achieved by forces that are not known explicitly, but whose effect is
known.
Hence we consider an embedded submanifold M RN which we can at least locally describe
as the zero locus of r functions
g : RN R,

= 1, ..., r ,

i.e. as


M = x RN |g (x) = 0 for all .
For example, a rigid body consisting of N distinct particles is described by
one for each pair of distinct points,

N (N 1)
2

functions,

kxi xj k2 cij = gij (x1 , ..., xN )

with constants cij R>0 . As a regularity assumption we require the rank of the Jacobian to be
maximal,
g 
rang
= r = maximal.

xi xM
As a warm-up problem, we want restrict a smooth function f : RN R to M and describe
the stationary points of this restriction f |M .
Proposition 2.2.1.
Consider the auxiliary function
f : RN Rr R
P
(x, ) 7 f (x) + r=1 g (x) .
The additional parameters Rr are called Lagrangian multipliers. Then the restriction f |M
has a stationary point in x0 M , if and only if the function f has a stationary point in (x0 , 0 )
for some 0 Rr .
Proof:
The function f has a stationary point, if and only if the two equations hold
r
X
f
f

0=
= g (x0 ) and
=
i g

x
x
=1

The first equation is equivalent to x0 M while the second equation requires the gradient of f to be normal to M in x0 , ensuring that x0 is a stationary point of the restriction f |M . 
Observation 2.2.2.
1. We now wish to consider the natural system on RN given by a potential U : RN R,
N

1X i 2
l(x, xt ) =
(x ) U (x) ,
2 i=1 t
18

that is constrained by unknown forces to a submanifold M RN of dimension r. We

assume that M is given by N r smooth functions g = g (xi ).
We thus consider trajectories
: I RN
with im M . For any family of trajectories  we are looking for stationary points of
Z
l j1 
I

subject to the constraints g ( ) = 0 for all = 1, ..., r and all . We introduce Lagrangian
multipliers and minimize the functional
Z


X
f (, ) = dt l j1  +
g ( ) .
I

The derivative with respect to  yields the following additional terms

X

g di 


xi d =0

so that the equation of motion becomes

E(l) j1 () = xtt + gradU =

grad g .

(2)

The right hand side describes additional forces whose concrete form is unknown that
constraint the motion to the submanifold M .
2. We need to describe the geometry of the situation: using the theorem of implicit functions,
we find an open subset U RN r with coordinates {q }=1,...,r and a chart with local
coordinates (q ) on U M that locally parametrizes M . We identify Tp M with a subspace
of Tp RN and express the basis vectors as

xi
=
,
q
q xi

= 1, ..., N r.

In subsequent calculations, the notation is simplified by introducing the vector x RN

with coordinates xi and writing

x
=
.
(3)
q
q
3. Next, we have to relate the jet spaces J2 M and J2 RN . Since they are designed as recipients
of trajectories and their derivatives, we consider a trajectory with values in U M
: IU M
which we continue to a trajectory with values in RN : The chain rule yields
di
xi d
=
.
d
q
d
19

This is, of course, just the usual transformation of tangent vectors under a smooth map,
in the case the embedding M , RN . From this, we deduce the following transformation
rules for the coordinates xt :
xit = xit (q , qt , t) =

xi
q
q t

(4)

between the coordinates q , qt on J1 M and xit on J1 RN . We find as an obvious consequence

xit
xi
.
=
qt
q
Similarly, to find the transformation rules for the coordinates xitt describing the second
derivative, we compute the second derivative of a trajectory:
d2
xi d2
2 xi
d d
=

+
.
dt2
q q dt dt
q dt2
This yields the following relation between the coordinates qtt on J2 M and xitt on J2 RN :
xitt = xitt (q , qt , qtt , t) =

2 xi
xi
q
q q .
+
q tt q q t t

(5)

4. Since we do not know the right hand side of the equation of motion (2), we take the scalar
product in RN with the tangent vectors (3). This yields N r equations




x
x
xtt , = grad U,
()
q
q
Our goal is to rewrite this as the Euler-Lagrange equation for a function on J1 M . The
right hand side is easily rewritten using the chain rule:


x
U xi
U
=
grad U, =
i

q
x q
q
This suggests to take the restriction of U to M as the potential for the Lagrangian system
on M .
5. To rewrite the left hand side, we use the total differential operator on smooth functions
of xi , xit , xitt , . . . introduced in definition 2.1.2:
X

Dx =
xi+1 i .
q
i,
Note that the operator is linear and obeys the Leibniz rule. Recall that the Euler-Lagrange
equations read in terms of this operator
Dx

l
l
=
.
xit
xi

Similarly, we introduce the differential operator

X

.
D=
q+1

,
on smooth functions of q , qt , qtt , . . .
20

6. We need the following result:


 
 

x
x
xt
D
xt ,
= xtt , + xt ,
.
q
q
q
To see this, we first apply the Leibniz rule for D to find
 
 


x
x
x
D xt , = xt , D + Dxt ,
q
q
q
and compute for the second term
D xt =

xt xt
x
2x
q
+
qt qt + qtt = xtt
q
=
t
tt

q
qt
q q
q

according to (3) and (5). For the first term, we compute

 i
x
2 xi
D
=
q ,
q
q q t
which turns out to be the same as

xit
=
q
q

xi
q
q t


=

2 xi
q .
q q t

7. We now introduce the following function on J1 M depending on the coordinates q , qt :

N

T =

1X i 2 1X i 2
(x ) =
x (q , qt )
2 i=1 t
2 i=1 t

To compute Euler-Lagrange equations for T in terms of q and qt , we first note that the
chain rule implies


T
xt
= xt , .
(6)
q
q
Next, we calculate, again using the chain rule,

 

T
xt
x
= xt , = xt ,
qt
qt
q
where we have used the relation derived in the third step. The left hand side of the EulerLagrange equation for T then reads



 
 


x
xt
x
D
T = D xt , = xtt , + xt , ,
qt
q
q
q
where we used in the last equalities identities derived in step 6. We rewrite the first
summand using the equation of motion () and the replace the second summand by identity
(6):



U
T
D
T = + .

qt
q
q
Taking into account that U is independent of xit and thus of qt , we realize that these are
the Euler-Lagrange equations for the Lagrangian
L(q , q ) = T (xt (q, qt )) U (x(q)) , .
21

We have thus shown:

Proposition 2.2.3.
Suppose a natural system on RN with potential U (x) is constrained by unknown forces to a
submanifold M RN . Then the equations of motion on M are the Euler-Lagrange equations
for a Lagrangian l(q, qt ) on J1 M obtained from the function on J1 RN by restricting U to M
and transforming the coordinates qt as in (4).
This result clearly shows that, for the description of derivatives, we should associate bundles
to a Lagrangian system that can be pulled back consistently.

2.3

Lagrangian systems

We will now look at Lagrangian systems in a more abstract way. We start with the following
Observation 2.3.1.
1. A first datum in the description of a mechanical system is a surjective submersion :
E I, where I R is an interval of eigentime parametrizing the system and E is the
extended configuration space.
2. Global trajectories are global sections of . Local sections give local trajectories. They form
a sheaf on I.
3. A system with constraints is described by a submanifold : E 0 , E such that restricts
to a surjective submersion on E 0 .
Remarks 2.3.2.
1. Upon choosing a space-time point p0 A in a Galilei space A, the motion of a point
particle in A is described by taking as the surjective submersion the absolute time difference
functional
A R
p 7 t(p p0 )
2. A mass point moving in a manifold M with motion parametrized by an interval I R is
described by the surjective submersion p1 : I M I given by the projection on the first
factor.
3. We do not require E to be a bundle over I. Any smooth bundle over I is globally trivial,
but E might carry additional structure that depends on I. Moreover, we will generalize
the situation.
Observation 2.3.3.
1. It turns out that we do not gain much by requiring the target of the surjection to be an
interval. We therefore consider the case of surjective submersions : E M with M any
smooth manifold. One application is classical field theory: then M has the interpretation
of space-time that parametrizes a field like the electromagnetic field E(x) that depends
on the point x M in space time. In this situation, the (local) sections of : E M
are called (local) field configurations. They generalize trajectories which arise in the case
when dim M = 1.
22

2. The following crucial difference between field theory and mechanics should be noted: in
mechanics, there is one parameter which can be chosen to be identical to time. Local spacial
coordinates are coordinate functions on the manifold E of the extended configuration space.
In a field theory, on the other hand, space and time are locally defined parameters and
are to be considered as local coordinate functions on M .
3. In certain situations, it might be necessary to endow either of the manifolds E or M with
more structure. Examples in field theory include metrics endowing E with the structure
of a Riemannian manifold or, to be able to include fermions, a spin structure on M .
4. To be able to formulate a dynamical principle in terms of a partial differential equation,
we need to work with derivatives of any order of local sections. This is achieved by the
following construction.
Definition 2.3.4
1. Let E, M be smooth manifolds, dim M = m and let : E M be a surjective submersion. We say that defines a fibre bundle in the category of smooth manifolds.
2. For any open subset U M , we denote by (U ) the set of local smooth sections of ,
i.e. the set of smooth functions sU : U E such that sU = idU . These sets form a
sheaf.
3. Consider a point p M and two open neighbourhoods U, U 0 of p. We identify two local
sections (U ) and 0 (U 0 ) if they agree on a common open refinement U
U U 0 . The equivalence classes are called germs of local sections in the point p. The set
of equivalence classes is denoted by (p).
4. For any multi-index I = (I(1), ..., I(m)), we introduce the derivative operators
|I| :=
|I|
:=
xI

m
X

I(i)

i=1
m 
Y
i=1

xi

I(i)

I = is admitted as a value, || = 0 and the corresponding derivative operator is the

identity.
5. Let p M and U, U 0 be open neighborhoods of M . Consider two local sections (U )
and 0 (U 0 ) whose value in p agrees. Possibly after a further restriction, we can
assume that the two sections are defined on the same coordinate neighborhood of p that
we call for simplicity U . We fix local coordinates xi on M around p and y on E around
(p) = 0 (p).
We say that and 0 have the same r-jet in p, if in local coordinates all partial derivatives
up to order r coincide,
|I| 
|I| (0 ) 
 =
 , 0 6 |I| 6 r .
xI p
xI
p
23

Since differentiation is a local operation, local sections with the same germ in p M give
rise to the same r-jet. We have thus defined an equivalence relation on germs of sections
that is moreover independent of the choice of local coordinates.
6. An r-jet with representative will be denoted by jpr ; the natural number r is called the
order of the jet. The point p M is called the source and (p) E the target of the jet.
7. We introduce the set

 
Jr := jrp p M, (p)
(The notation Jr E is also in use, but we wish to stress that the set is determined by a
morphism and not only by the manifold E.)
We endow these sets with more structure
Observation 2.3.5.
1. We have the following obvious projections:
r :
r,0 :
and for r > k > 1
r,k :

Jr
jrp ()
Jr
jrp ()

M
(source projection)
p
E
(target projection)
(p)

Jr Jk (jet projections)
jrp 7 jkp

Such that the following diagram commutes:

r,r1

Jr Jr1

r
r1
y
y
M

1,0

...

J1

1
y

...

2. In a next step, we introduce local coordinates to endow Jr with the structure of a smooth
manifold and the surjection r : Jr M with the structure of a bundle map, the
jet bundle of with fibre Jrp . We include the case J0 := E.
To this end, we choose a bundle chart (U, u) for E, where U M is an open subset of M
and u is a pair u = (xi , u ) consisting of a local coordinate chart x : U M Rm of M
and u : 1 (U ) Rn of E. A common terminology is to call the local coordinates x of M
the independent coordinates and the local coordinates u of E the dependent coordinates.
This language is motivated by the situation where we describe local sections s : M
U E leading to an expression of the local coordinates u of E in terms of the local
coordinates xi of E.
3. The induced bundle chart of Jr is defined on
 

U r := jrp  (U )
24

and consists of maps ur = (xi , uI ) with |I| r with |i| r defined by the coordinates xi
of the base point p M


xi jrp = xi (p)
and the partial derivatives of the jet in local coordinates (u , xi ),
uI

jrp

|I| 
=
.
xI p

The jet bundle Jr is in particular finite-dimensional.

4. Let be a local section of defined on an open subset U M . Its r-th jet prolongation
is the local section
j r : U Jr
given by
(jr ) (p) = jrp .
In local coordinates, it is given by



|I|
, I .
x

Lemma 2.3.6.
Let E be a smooth manifold and p1 : R E R a trivial bundle. This situation is relevant
Then there is a canonical
for classical mechanics with time-independent configuration space E.
isomorphism
J1 R T E .
Given any section
: R R E
t 7 (t, (t))

the isomorphism maps the one-jet of to the pair with

where is a smooth function R E,
second entry the tangent vector
 !

d


j1t0 7 t0 ,
.
dt t=t0
This allows us to relate 1-jets to the tangent bundle. Certain authors, e.g. Arnold, therefore use
the tangent bundle as a framework for discussing Lagrangian mechanics. This has, however,
the disadvantage that one cannot describe geometrically equations of motions. Moreover, the
language does not generalize to field theory, i.e. higher dimensional parameter spaces.
Proof:
The isomorphy is immediate in coordinates:



1
d 
i
i
i
i
t0 , x , xt jt = t0 , x (t),  x (t) .
dt t=t0

25

Definition 2.3.7
The infinite jet bundle : J M is defined as the projective limit (in the category of
topological spaces) of the jet bundles
M
E J1 J2 ...

1,0

2,1

It comes with a family of natural projections

,r J Jr .

Proposition 2.3.8.
The infinite jet bundle has the following universal property: for any bundle : F M with
bundle morphisms sn : F Jn such that all diagrams of the form
F 1 RRRR

11 RRR sr
RRR
11
RRR
RRR
11
( r
1
J
ss 11
11
11
s,0
1

/E
Js
s,0

commute, there exists a unique bundle morphism : F J such that the following diagram
commutes:
F 2 DRRRR
22 D RRRR sr
22 D RRRRR

RRR
22 D"
R)
22
/ r
ss 2 J ,r J
22
22 ,s
s,0
2 

/E
Js
s,0

Locally, the infinite jet bundle J

is described by coordinates (x , uI ) and all multi-indices I,
where we admit I = . In particular, the jet bundle is infinite-dimensional.

Proposition 2.3.9.
Let E M be a fibre bundle and U M be an open subset. Any local section : U E can
be extended to a section j (s) = j s of J . In local coordinates, this extension is given by
collecting all partial derivatives:


|I| i
i i
x , s (x ), ..., I s (x )...
x
In other words, the extension to j (s) amounts to keeping all partial derivatives, or put
differently, to keeping the Taylor series as a formal series.
Definition 2.3.10
26

1. A real-valued function f : J R is called smooth, if it factorizes through an finite jet

bundle Jr , i.e. there exists r N and a smooth function fr : Jr R such that
f = fr r .
Note that this implies that the algebra of smooth functions on J is defined as the
inductive limit of the injections

k+1,k
: C Jk C Jk+1 .
It has the structure of a filtered commutative algebra.
2. A (geometric) partial differential equation of order r is a closed submanifold Jr . A
local solution on an open subset U M is a section : U E with jrp for all
p W . Sometimes, this submanifold is also called the shell. This terminology comes from
field theory jargon where one speaks of the mass shell and of on shell conditions for
relations holding only on the submanifold.

Example 2.3.11.
Consider the bundle p1 : Rn R Rn whose sections are real-valued functions on Rn . We have
global bundle coordinates (x1 , ..., xn , u). Consider as a submanifold the zero locus of the smooth
function
F : J2 R
i

F (x , u, ui , uij ) =

n
X

uii

i=1

This submanifold describes Laplaces equation.

Definition 2.3.12
(i) A vector field on the jet bundle J is defined as a derivation on the ring C (J ). In
local coordinates (xi , uI ) on J , this is a formal series of the form
X = Ai

+
BI
i
x
uI
,I

with smooth functions Ai = Ai (xj , uI ) and BI = BI (xj , uI ).

(ii) The prolongation of a vector field X (U ) on U M is the unique vector field pr X
(J ) acting in the point j (s)(p) J given by the prolongation of the section
s (U ) like as the following derivation on a smooth function f C (J )
pr Xj s(p) f := Xp (f j s) .
Notice that f j s is a real-valued function on U M .

27

Proposition 2.3.13.
Let (x, u) be local bundle coordinates for a bundle : E M . Then



Dj := pr
=
+
u
.
I+(j)
xj
xj
uI
In particular, when M = I R is an interval, we obtain the differential operator D appearing
in the Euler-Lagrange operator in proposition 2.1.2
Proof:
We compute using f C (J )



f
f |I|+1 s

pr

f
=
(f

j
s)
=
+
xj j s(p)
xj
xj uI x|I|+(j)

Our next goal is to endow the jet bundle J with a connection:
Definition 2.3.14
Let : E M be a fibre bundle. Then V = V () := ker (d : T E T M ) T E is vector
subbundle of rank dim E dim M over E, the vertical bundle of . It comes with a natural
surjective submersion to M .
(i) An Ehresmann connection on E is a smooth vector subbundle H T E over E such that
TE = H V
where the direct sum is to be understood fibrewise. H is then called the horizontal bundle
of the Ehresmann connection. The rank as a vector bundle over E is dim M .
(ii) Equivalently, an Ehresmann connection can be defined as an T E-valued 1-form on E,
i.e. an endomorphism of the vector bundle T E such that
2 = (idempotent)
im = V
Then H := ker is the horizontal subbundle.
Given an Ehresmann connection, we can decompose any vector field X vect(E) on E in
its horizontal part XH and its vertical part XV :
X = XH + XV .
We need a few properties of Ehresmann connections:
Remarks 2.3.15.
Let : E M be a smooth fibre bundle with an Ehresmann connection.
28

(a) Let : [0, 1] M be a smooth curve in M . A lift of is a smooth curve : [0, 1] E

such that = . As a diagram:
E
y<
yy
y
y

yy
yy

/ M
[0, 1]

A lift is called horizontal, if we have for the derivative vector

0 (t) H(t)

for all t [0, 1].

By standard theorems about ordinary differential equations, horizontal lifts locally exist for
a given initial condition (0) 1 (0). If global lifts exist for all curves, the connection is
called complete.
(b) Given an Ehresmann connection, we define a 2-form R with on E with values in T E which
is called the curvature of the Ehresmann connection. The 2-form is defined by its value on
two vector fields X, Y vect(E):
R(X, Y ) = [XH , YH ]V
(c) An Ehresmann connection is called flat, if the 2-form R vanishes. Flat connections are
exactly Frobenius integrable connections.
We next wish to convince ourselves that the notion of Ehresmann connection comprises
other notions of connection if we combine it with suitable compatibility conditions.
Observation 2.3.16.
Consider a real vector bundle over M , i.e. a bundle : E M where each fibre 1 (x) carries
the structure of an R-vector space. (The case of complex vector bundles is completely analogous.)
1. For any x M , the fibre Ex = 1 (x) has the structure of an R-vector space, i.e. we have
for each e Ex a translation operator Te : Ex Ex . Similarly, we have for any scalar
s R a scalar multiplication Ss : Ex Ex . For a linear connection, the subspaces Ee for
different e Ex in the same fibre are required to be linearly related:
He+e0 = (dTe0 )e He

and

Hse = dSs (He ) .

Moreover, we require the vertical lift

vlE : E M E V
d 
vlE (ux , vx ) =  (ux + tvx )
dt t=0

with

ux , vx Ex

to be an isomorphism of vector bundles over M .

2. This allows us to define the connector of a connection as the following homomorphism of
vector bundles over M :

K : TE
V 1
E M E E

vlE

29

pr2

3. Let now N be any smooth manifold. We can now define the covariant derivative for a
smooth map f : N E and a vector field X vectN :
OX f : N
T N T E
E.
X

Tf

In the special case N = M and the tangent bundle, E = T M , the smooth map f : M
T M is just a vector field on M and we have the usual notion of the covariant derivative
of a vector field.
In the next step, we will see that any jet bundle carries a natural Ehresmann connection.
Proposition 2.3.17.
Let : E M be a fibre bundle and J be the corresponding jet bundle. For any x J
consider the subspace of Tx J spanned by all prolongations pr X of tangent vectors X
T x M . The the following holds:
1. The subspaces endow
Cartan connection.

with

an

Ehresmann

connection,

the

so-called

2. One has
pr [X1 , X2 ] = [pr X1 , pr X2 ] ;
hence the Cartan connection is flat.
In local coordinates, one can be more explicit:
Definition 2.3.18
1. Given a multi-index I = (i1 , i2 , . . . , ir ), we define a differential operator DI := Di1 Di2
. . . Dir . Given finitely many local functions Z I , we call a differential operator of the form
Z I DI acting on local functions a total differential operator.
2. Local differential forms on the jet bundle J are defined as the inductive limit of the
system

k+1,k
: Jk J k+1
i.e.


(J ) = lim ind Jk .
A local p-form on J is thus a finite linear combination of expressions of the form

A dxi1 ... dxip duj11...jpq ... dul1q...lpq

with p + q = k and with A a smooth function on J .
3. A differential form (J ) is called a contact form, if
j (s) = 0 (M )
for all local sections s (E).

Remarks 2.3.19.
30

(a) The subspace of contact forms is a differential ideal of (J ), the contact ideal I.
(b) One can check that a vector field on the jet bundle X Vect (J ) is horizontal, if and
only if its inner product with all contact forms vanishes,
X = 0 for all I
(c) Locally, the contact ideal I is generated by the one-forms
I := duI uI+(j) dxj .
We convince ourselves that these differential forms are indeed contact forms: given a local
section s : U E, we consider its prolongation
f := j s : U J
and pull back using the usual formula dy for one-forms:
f =
This yields
f I =

f j
dx .
xj


(j s)I j


j
dx

u
I+(j)  dx = 0.
xj
j (s)

We are now in a position to introduce a natural splitting on the complex of local differential
forms on a jet space.
Definition 2.3.20
(i) A p-form p (J ) is said to be of type (r, s), if at all points = j s J one has
(X1 , ..., Xp ) = 0,
as soon as either more than s vectors are vertical or more than r vectors are horizontal.
In local coordinates, is then a sum of terms
f dxi1 ... dxir I11 ... Iss
with a smooth function f . The subspace of local differential forms of type (r, s) is denoted
by r,s (J ).
(ii) We decompose
p (J ) =

r,s (J ) .

r+s=p

Correspondingly, the exterior derivative

d : p (J ) p+1 (J )
decomposes into horizontal and vertical derivatives
d = dH + dV
31

with
dH : r,s (J ) r+1,s (J )
dV : r,s (J ) r,s+1 (J )
The relation d2 = 0 implies the three identities
d2H = 0, d2V = 0, dH dV = dV dH .

Remarks 2.3.21.
(a) On smooth functions, we have the following formula in terms of local coordinates:
dV f =

uI I


f
f

+ uI+(i) dxi
dH f = (Di f ) dx =
xi
uI
i

Moreover, one checks

dH (dxi ) = 0

and dV (dxi ) = 0

as well as
dH I = dxj I+(j)

and dV I = 0.

(b) We wish to comment on the variational problem given by a smooth function on a jet bundle
J . Again we consider a one-parameter family s of local sections of E
M and want

to determine
d 
 f (j s ) .
d =0
Then the infinitesimal version of the variation
  
ds 

X :=

d =0 u

gives us a vertical vector field on E. We denote by pr

E X the unique vector field on J
which acts on functions on E as X and preserves the contact ideal. Then the variational
formula expresses the derivative as the contraction of a vector field and a one-form on J :
d 
 f (j s ) = h pr
E X, dV f i.
d =0

(c) Given a total differential operator Z, we define a new total differential operator Z + called
the adjoint of Z by
Z
Z

(j ) (F Z(G))dvolM =
(j ) (Z + (F )G)dvolM
M

for all sections : M E and all local functions F, G LocE . It follows that
F Z(G)dvolM = Z + (F )GdvolM + dH
for some n1,0 (E).
If Z = Z J DJ in local coordinates, then integration by parts yields Z + (F ) = (D)J (Z J F ).
32

Definition 2.3.22
(i) The variational bicomplex of a fibre bundle : E M is the double complex

0 0,2

1,2

dV
0 0,1

dV
1,1

dV
0 R 0,0

dV
1,0

dH

dH

dH

...

n,2

...

dV
n,1

dH

2,0

dH

dH

...

n1,0

dV
n,0
dH

(ii) A Lagrangian system consists of a smooth fibre bundle : E M together with a form
l n,0 (J ), called the Lagrange function or Lagrangian density of the system. The
manifold E is called the (extended) configuration space of the system.
(iii) A Lagrangian system is called mechanical, if M is an interval I R.
(iv) A global section : M E is a called a motion of the Lagrangian system, if it is a
stationary point for the functional, called the action
Z
F (s) =
j (s) l .
M

(v) Given a Lagrangian density

l = l xi , uI dx1 ... dxn n,0 (J ) ,
the Euler-Lagrange form E(l) n,1 (J ) is defined as
E(l) = E (l) dx1 ... dxn
with
E (l) =

l
l
l
Di + Dij + . . . .

u
ui
uij

Example 2.3.23.
g). The metric gives us a smooth function
1. Consider a Riemannian manifold (E,
T : T E R
v 7 12 gp (v, v) for v Tp E
We wish to construct a classical Lagrangian system on p1 : E := I E I with
we pull
I R an open interval. Using the canonical identification J1
= R T E,
1

back T along the projection R E E to a function on J . This function is called

the kinetic energy and the mechanical Lagrangian system described by it is called the
g). The corresponding Lagrangian density is
free system on the Riemannian manifold (E,
T(xt )dt g n,0 (E) with n := dim E = dim E + 1.
The trajectories for this system can be seen to be geodesics for the Riemannian manifold.
33

2. A smooth function
U : E R
is called a potential energy. (In interesting problems, one might be obliged to allow U
to have singularities and to impose more conditions on U .) Again, we pull back U to a
function U on J1 and obtain a n-form on E by combining it with the volume form.
3. A Lagrangian system with Lagrangian l := T U is called a natural system.
We next formulate the algebraic Poincare lemma:
Proposition 2.3.24 (algebraic Poincare lemma).
On good open subsets of J , we have dH = 0 for s,0 if and only if there are local
functions j on J such that
= j d1 x . . . dxs .
Let M be a smooth manifold of dimension n and : E M a fibre bundle. Forms on J
of type (n, s) with s 1 are dH -closed, but not even locally exact.
To improve the situation, we introduce for s 1 coaugmentations, called the
inner Euler operators
I : n,s (J ) n,s (J )
by taking the inner product
X
1
I() :=
(1)|I| .
u
s
I
I
Lemma 2.3.25.
We have the following identities for the inner Euler operators:
(i) For n1,s (J ), we have
I (dH ) = 0 .
(ii) For any n,s (J ) there exists n1 (J ) such that
= I() + dH .
(iii) I is an idempotent, I 2 = I.
(iv) For any Lagrangian density Lagrange-Dichte n,0 (J ), we have
E() = I (dV )
This suggests to extend the bicomplex. We introduce the following subspace of n,s ((J )):
F s (J ) := im {I : n,s n,s } = { n,s (J ) : I = }
The elements of F s (J ) are called source forms.
Moreover, we introduce maps
V : F (J ) F s+1 (J )
V () = I(dV )
One can check that the following identity holds: V2 = 0.
Definition 2.3.26
34

(i) The augmented variational bicomplex for a fibre bundle : E M is the double complex
0 0,3

0,2
0
1,2 ...

n,2

0,1
0
1,1 ...
dV
0 R

0,0

n1,1

...

n1,0

dH

1,0

n,1

F2

V
F1

n,0

(ii) The Euler-Lagrange complex is the edge complex

d

V
H
H
F 2 ...
n,0
F 1
1,0 ...
0 R 0,0

where V is called the Helmholtz operator.

We recognize n,0 as the recipient for the Lagrangian densities and F 1 as the recipient
for the equations of motion. The size of the kernel of V describes the obstruction to finding
Lagrangian densities for equations of motion.

2.4

Symmetries and Noether identities

Our goal in this subsection is to describe symmetries of a Lagrangian system ( : E M, l).

Observation 2.4.1.
Suppose a Lie group G acts by automorphisms of a vector bundle E M (over the identity
of M ) in such a way that it leaves the action S of a Lagrangian l : J R invariant.
Then the group action induces a vertical vector field on E, for each element of the Lie
algebra of G, such that the prolongation prE (
) of to J has the property that dl(prE (
))volM

is dH exact. Here volM denotes both a volume on M and its pullback to J via the projection
J M .
We only need infinitesimal aspects of the group action and hence formulate symmetries in
terms of vector fields. We introduce the following notions:
Definition 2.4.2
1. An evolutionary vector field is a mapping from J into the vertical vector fields on E.
In local coordinates Q = Q u where the functions Qa are local functions on J .
It can be thought of as an infinitesimal variation depending on the fields and their
derivatives.
For every evolutionary vector field Q on E, there exists its prolongation, prE (Q), the
unique vector field on J preserving the contact ideal such that (dE )(prE (Q)) = Q.
2. An evolutionary vector field QE on E is called a variational symmetry of a Lagrangian l,
if it has the property that dl(pr(QE ))volM = pr(QE )(l)volM is dH exact.
35

Remarks 2.4.3.
1. In local coordinates, QE is a variational symmetry iff
prE (QE )(l) = DK (QaE )

l
uaK

is a divergence on jet space, i.e., iff it is equal to D j for some set {j } of local functions
defined on J .
2. Integrating by parts shows that this condition is equivalent to requiring that
QaE (D)K ( ula ) be a divergence.
K

3. But the Euler Lagrange operator Ea acting on the Lagrangian l is defined by the equation
Ea (l) = (D)K ( ula ). Thus an evolutionary vector field QE is a variational symmetry of
K
a Lagrangian L iff QaE Ea (l) is a divergence.
We rewrite these results and obtain Noethers first theorem:
Theorem 2.4.4.
Let QE be a variational symmetry, i.e. there are local functions j K on jet space such that
prE (QE )(l) = DK (QaE )

l
= DK j K .
uaK

Then we have on jet space


DK

QaE

l
jK
uaK


=0

so that the function on jet space

QaE

l
jK
uaK

is conserved for all solutions s : M E of the Lagrangian system:

K s
a l

a l
(Q
jK ) j s = DK (QE a jK ) K = 0
xK E uaK
uK
x
Example 2.4.5.
Consider a natural mechanical system on E = RN , i.e. l(x, xt ) = 12 |xt |2 U (x), and suppose that
the potential U is invariant under the family of translations given by a fixed vector T RN \{0}:
TT : R RN RN
x 7 x + tT ,

i.e. T i x
i = 0. Then ( t , T ) is a variational vector field on E = R E and

Ti

l
= T i (xt )i
i
xt

is a conserved quantity. It is called momentum in the direction T .

Our next example is the rigid body:
36

Observation 2.4.6.
A rigid body is a system of N 3 mass points moving in R3 with the standard Euclidean
structure, subject to the constraints
kxi xj k = rij = const.
To avoid degenerate cases, we require that not all points are on a line.
To determine the configuration space, we choose three mass points x[1] , x[2] , x[3] not on
aline whose position determine the position of the body. As the first three coordinates, we
take x(1)
Next consider the differences v1 = x2 x1 , v2 = x3 x1 ; there is exactly one rotation
in SO(3) which maps these vectors to a fixed reference pair (w1 , w2 ) of vectors having the
same lengths and including the same angle:
kv1 k = kw1 k, kv2 k = kw2 k

and

hv1 , v2 i = hw1 , w2 i

i.e. g(v1 , v2 )wi = vi . As coordinates, we choose (x1 , g(v1 , v2 )) and identify the configuration
space with R3 SO(3).
We consider the system in the absence of external forces, i.e. we assume that it is described
by the Lagrangian:
N
X
1
l=
m hx,t , x,t i with m > 0
2
=1
Noethers theorem gives three conserved quantities for the translations in R3 : the vector
field on E = R3N induced by a translation by T i is
i
Qi,
E = T .

with i = 1, ..., 3 and for all = 1, ..., N . For the derivative of the action, we find
l
= m xi
t
i
xt
Hence the conserved quantity is
X l  dhs i, X
i
=
m xi,
I=
t T .
i
x
ds
t
i,
i,
Since this applies to all T R3 , we have three conserved quantities
Ii =

N
X

m xi,
t

i = 1, ..., 3

=1

Denote by M :=

PN

=1

m > 0 the total mass of the body and introduce the center of mass
xcm :=

N
1 X
m x
M =1

37

of the rigid body, then the conservation law implies

d2
xcm = 0.
dt2
In other words, the center of mass is in uniform motion.
This allows us to continue our discussion in a reference frame where the center of mass
is at rest.
The invariance under rotations gives, again by Noethers theorem three further conserved
quantities; one more conserved quantity is energy, as follows from the time independence
of the action. These four functions determine for any given solution of the equations of
motion a submanifold C of the six-dimensional phase space SO(3). It is
Two-dimensional, dimC = 2.
Bounds on the tangent vectors from energy conservation imply that it is a compact
submanifold of the compact manifold SO(3).
Orientable, since T SO(3)
= SO(3) R3 as a smooth manifold. Thus T SO(3)is parallelizable and in particular orientable and any submanifold defined by global regular
equations is orientable as well.
The derivative of the motion gives a vector field on C that for non-vanishing energy
vanishes nowhere.
A standard theorem of topology then implies that C is a torus. One can choose coordinates
1 , 2 R/2Z on C such that the equations of motion have the form
1 = 1

2 = 2 .

Obviously, the two angular frequencies i depend on the initial conditions.

Noethers second theorem deals with more structure than just symmetries.
Definition 2.4.7
A gauge symmetry of a Lagrangian system ( : E M, l) consists of a family of local functions
RaI : J R
such that for any local function
 : J R
the vector field RaI (DI ) u a on E is a variational symmetry of l.
Remarks 2.4.8.
1. Loosely speaking, a gauge symmetry is a linear mapping from local functions on J into
the variational vector fields on E.
2. Notice that the coefficients of the vector field depend linearly on  and on its derivatives.
3. By the results just obtained, it follows that being a gauge symmetry is equivalent to requiring (RaI DI )Ea (l) to be a divergence for each local function  on J .
38

4. This in turn is equivalent to saying that (RaI DI )+ (Ea (l)) is a divergence for each local
function , where (RaI DI )+ is the formal adjoint of the differential operator RaI DI .
Since the adjoint of a total differential operator is again a total differential operator, there
exist local functions R+aI : J E R such that (RaI DI )+ = R+aI DI . These functions
can be found by working out the iterated total derivatives (D)I (RaI F ). In many cases it
is easier to use an integration by parts procedure to obtain the coefficient functions on
J , {R+aI }.
5. It follows that  7 RaI (DI ) u a defines a gauge symmetry iff R+aI DI (Ea (l)) is a divergence for each . This condition is equivalent to saying that R+aI DI (Ea (l)) is identically
zero on the jet bundle.
Such an identity is called a Noether identities One thus has a one-one correspondence
between gauge symmetries of a Lagrangian and Noether identities.
We have thus found:
Theorem 2.4.9 (Noether).
For a given Lagrangian system ( : E M, l) and for local real-valued functions {RaI } defined
on J , the following statements are equivalent:
1. The functions {RaI } define a gauge symmetry of l, i.e., RaI (DI ) u a is a variational
symmetry of l for any local function  : J R.
2. RaI (DI )Ea (l) is a divergence for any local function ,
3. The functions {RaI } define Noether identities of l, i.e. R+aI DI (Ea (l)) is identically zero
on the jet bundle.
One should be aware that gauge symmetries induce differential identities between the equations of motions, i.e. dependencies between the equations of motions and their derivatives.

2.5

Natural geometry

A natural differential operator is a recipe that constructs from a geometric object another one,
in a natural fashion, and which is locally a function of coordinates and their derivatives. They
are thus intimately related to jet bundles and have to be compatible with smooth maps between
manifolds.
Examples 2.5.1.
1. Let M be a n-dimensional smooth manifold. The classical Lie bracket X, Y 7 [X, Y ] is a
natural operation that constructs from two vector fields on M a third one. Given a local
coordinate
(x1 , . . . , xnP
) on M , the vector fields X and Y are locally expressions
P system
i
i
X = 1in X /x , Y = 1in Y i /xi , where X i , Y i are smooth functions on M .
If we define Xji := X i /xj and Yji := Y i /xj , 1 i, j n, then the Lie bracket is


P
locally given by the formula [X, Y ] = 1i,jn X j Yji Y j Xji /xi .
2. The covariant derivative (, X, Y ) 7 X Y is a natural operator that constructs from a
linear connection and vector fields X and Y , a vector field X Y . In local coordinates,


X Y = ijk X j Y k + X j Yji
,
(7)
xi
where ijk are Christoffel symbols.
39

3. Natural operations can be composed into more complicated ones. Examples of composed operations are the torsion T (X, Y ) := X Y Y X [X, Y ] and the curvature
R(X, Y )Z := [X,Y ] Z [X , Y ]Z of the linear connection .
4. Let X be a vector field and a 1-form on M . Denote by (X) C (M ) the evaluation
of the form on X. Then (X, ) 7 exp((X)) defines a natural differential operator
with values in smooth functions. Clearly, the exponential can be replaced by an arbitrary
smooth function : R R, giving rise to a natural operator O (X, ) := ((X)).
5. Randomly generated local formulas need not lead to natural operators. As we will see
later, neither O1 (X, Y ) = X31 Y 4 /x2 nor O2 (X, Y ) = X j Yji /xi behaves properly under
coordinate changes, so they do not give rise to vector-field valued natural operators.
In the examples, the natural differential operators are recipes given as a smooth function in
coordinates and derivatives that are covariant under changes of local coordinates.
Definition 2.5.2
1. Denote by Mann the category of n-dimensional manifolds and open embeddings. Let Fibn
be the category of smooth fiber bundles over n-dimensional manifolds with morphisms
differentiable maps covering morphisms of their bases in Mann .
2. A natural bundle is a functor B : Mann Fibn such that for each M Mann , B(M ) is a
bundle over M . Moreover, B(M 0 ) is the restriction of B(M ) for each open submanifold
M 0 M , the map B(M 0 ) B(M ) induced by M 0 , M being the inclusion B(M 0 ) ,
B(M ).
For each s 1 we denote by GLn(s) the group of s-jets of local diffeomorphisms Rn Rn at 0,
n
n
so that GL(1)
n is the ordinary general linear group GLn of linear invertible maps A : R R .
s
Let Fr (M ) be the bundle of s-jets of frames on M whose fiber over z M consist of s-jets of
local diffeomorphisms of neighborhoods of 0 Rn with neighborhoods of z M . It is clear that
Frs (M ) is a principal GLn(s) -bundle and Fr1 (M ) the ordinary GLn -bundle of frames Fr(M ).
Theorem 2.5.3 (Krupka, Palais, Terng).
For each natural bundle B, there exists l 1 and a manifold B with a smooth GL(l)
n -action
such that there is a functorial isomorphism
B(M )
B := (Frl (M ) B)/GL(l)
= Frl (M ) GL(l)
n .
n

(8)

Conversely, each smooth GLn(l) -manifold B induces, a natural bundle B. We will call B the
fiber of the natural bundle B. If the action of GLn(l) on B does not reduce to an action of the
quotient GLn(l1) we say that B has order l.
Examples 2.5.4.
1. Vector fields are sections of the tangent bundle T (M ). The fiber of this bundle is Rn , with
the standard action of GLn . The description T (M )
= Fr(M ) GLn Rn is classical.
2. De Rham m-forms are sections of the bundle m (M ) whose fiber is the space of antisymmetric m-linear maps Lin(m (Rn ), R), with the obvious induced GLn -action. The presentation
m (M )
= Fr(M ) GLn Lin(m (Rn ), R)
40

is also classical. A particular case is 0 (M )

= Fr(M ) GLn R
= M R, the bundle whose
sections are smooth functions. We will denote this natural bundle by R, believing there
will be no confusion with the symbol for the reals.
3. Linear connections are sections of the bundle of connections Con(M ) which we recall
below. Let us first describe the group GL(2)
n . Its elements are expressions of the form
n
n
A = A1 + A2 , where A1 : R R is a linear invertible map and A2 is a linear map from
the symmetric product Rn  Rn to Rn . The multiplication in GL(2)
n is given by
(A1 + A2 )(B1 + B2 ) := A1 (B1 ) + A1 (B2 ) + A2 (B1 , B1 ).
The unit of GL(2)
n is id Rn + 0 and the inverse is given by the formula
1
1
1
(A1 + A2 )1 = A1
1 A1 (A2 (A1 , A1 )).

Let C be the space of linear maps Lin(Rn Rn , Rn ), with the left action of GL(2)
n given as
1
1
1
(Af )(u v) := A1 f (A1
1 (u), A1 (v)) A2 (A1 (u), A1 (v)),

(9)

n
for f Lin(Rn Rn , Rn ), A = A1 + A2 GL(2)
n and u, v R . The bundle of connections
is then the order 2 natural bundle represented as

Con(M ) := Fr2 (M ) GL(2)

C.
n
Observe that, while the action of GL(2)
n on the vector space C is not linear, the restricted
action of GLn GL(2)
on
C
is
the
standard action of the general linear group on the
n
space of bilinear maps.
For k 0 we denote by B(k) the bundle of k-jets of local sections of the natural bundle B
so that B(0) = B. If g is represented in this way, then
B(k) (M )
B(k) ,
= Fr(k+l) (M ) GL(k+l)
n
where B (k) is the space of k-jets of local diffeomorphisms Rn B defined in a neighborhood
of 0 Rn .
Definition 2.5.5
Let F and G be natural bundles. A (finite order) natural differential operator O : F G is
a natural transformation (denoted by the same symbol) O : F(k) G, for some k 1. We
denote the space of all natural differential operators F G by Nat(F, G).
If F and G are natural bundles of order l, with fibers F and G, respectively, then each
natural operator in Definition is induced by an GL(k+l)
-equivariant map O : F (k) G, for
n
some k 0. Conversely, such an equivariant map induces an operator D : F G. This means
that the study of natural operators is reduced to the study of equivariant maps. The procedure
described above is therefore called the IT reduction (from invariant-theoretic).
Examples 2.5.6.
1. Given natural bundles B0 and B00 with fibers B 0 resp. B 00 , there is an obviously defined
natural bundle B0 B00 with fiber B 0 B 00 . With this notation, the Lie bracket is a natural
operator [, ] : T T T and the covariant derivative an operator : ConT T T ,
where T is the tangent space functor and Con the bundle of connections. The corresponding
equivariant maps of fibers can be easily read off from local formulas given in Examples.
41

2. The operator D : T 1 C from the Example above is induced by the GLn equivariant map O : Rn (Rn ) R given by o (v, ) := ((v)).

3
3.1

Classical field theories

Maxwells equations

We start considering Maxwells equation on a Galilei space A and will see only later that this
is not an appropriate conceptual setting, because the laws of electrodynamics are incompatible
with the Galilean principle of relativity.
Observation 3.1.1.
(a) Electrodynamics leads to a new conserved quantity: electric charge. We introduce a charge
density. It assigns to a space-time point (x, t) A the density of electric charges. Given any
measurable subset U of a hypersurface Ht A of simultaneous events, the electric charge
present in the space region U should be given by an integral over U . It is therefore natural
to consider the charge density t as a three-form on Ht :
t := dx1 dx2 dx3 3 (Ht )
We will endow the hypersurfaces Ht with an orientation. By the basic axioms of Galilei
space, these hypersurfaces are endowed with a Euclidian metric. We can thus use threedimensional Hodge duality to identify it with a function on Ht .
(b) The charge density is not constant in time. If we take a volume V Ht with smooth
boundary V , then electric charge can pass through its boundary V . The amount of charge
passing per time should be described by the integral over a two form j 2 (Ht ). We then
have the natural conservation law for any closed volume V Ht
Z
Z
d
=
j
dt V
V
which by Stokes theorem takes the form
Z
Z
Z
d
=
j=
dj .
dt V
V
V
Since this holds for all volumes V Ht , we have the infinitesimal form of the conservation
law
d
+ dj = 0
dt
which is an equality of three-forms on Ht for all t. We can write the conservation law more
compactly in terms of the three-form form
j := dt j 3 (A)
defined on the four-dimensional space A as dj = 0.
Using the Hodge star on Ht , we can relate the three-form j on Ht to the 1-form j which
in turn can be identified with a vector ~j. This allows us to recover the conservation law in
the classical form

+ div~j = 0 .
t
42

(c) Moreover, observations tell us that distributions of electric charge experience a force. For
simplicity, we consider a point particle carrying an electric charge q and moving with velocity
d~x(t)
~v (t) =
Tx(t) Ht
dt
It turns out that the force is proportional to q and depends on two R3 -valued functions on A,
~ x, t) and the magnetic field B(~
~ x, t). The Lorentz force force experienced
the electric field E(~
is
~
~
F~Lor (x(t)) = q E(x(t)
+ q~v (t) B(x(t)
Using the metric on A.
The electric and the magnetic field are to be thought of as sources of this force. Again using
the Euclidean structure on the hypersurface Ht of simultaneous events in a Galilei space A,
we identify the vector of force with a one-form on Ht and thus introduce a one-form for the
electric field
E = Ei dxi 1 (Ht ) .
To understand the second term, we introduce a 2-form to describe the magnetic field
B :=

1
klm Bk dxl dxm 2 (Ht )
2

and compute its contraction with the vector ~v of velocity:

~v B = ~v (Bx dy dz + By dz dx + Bz dx dy
= Bx vy dz Bx vz dy + By vz dx By vx dz + Bz vx dy Bz vy dx
= (By vz Bz vy )dx + (Bz vx Bx vz )dy + (Bx vy By vx )dz

This gives a one-form ~v(t) B Tx(t)

Ht on the world line of the charged point particle.

Again, we would like to rewrite the equation in a more uniform way on the four-dimensional
Galilei space A. To this end we introduce a vector on A by
j = q

+ q~v vect(A)
t

that combines the electric charge and the electric current associated to the moving pointlinke
source. We have added a minus-sign for later convenience. We also combine electric
and magnetic field to get a two-form on the four-dimensional Galilei space A, called the
electromagnetic field strength
F := dt E + B =

3
X

Ei dt dxi + B1 dx2 dx3 + B2 dx3 dx1 + B3 dx1 dx2 ) 2 (A) .

i=1

We then find in terms of 1-forms the Lorentz force as a contraction:

(FLor ) = j F .

43

(d) Maxwells equations conversely describe the dynamics of the field for given charges and
sources.
To state them concisely, we endow Galilei space A with the structure of a pseudo-Euclidean
manifold by extending the metric on the spacial hypersurfaces Ht in such a way that the

is normal to all surfaces Ht and normalized by

global vector field t
g(

, ) = 1
t t

The minus sign is here just an ad hoc convention.

This allows us to introduce a one-form on A
j := dt + j1 dx + j2 dy + j3 dz 1 (A)
with the property that j = j where the Hodge star is now the four-dimensional Hodge
operator. We compute explicitly the Hodge dual
F = Bi dt dxi E1 dx2 dx3 E2 dx3 dx1 E3 dx1 dx2
This two-form is sometimes called the dual field strength.
Maxwells equations can be stated concisely as
dF = 0 F = d F = j
Here the first equality is in 3 (A) and the second equality in 1 (A).
(e) To make the comparison to the classical formulation, we compute
dF

(divB)dx dy dz + (t B1 + 2 E3 3 E2 )dt x2 dx3

+(t B2 + 3 E1 3 E1 )dt x3 dx1 + (t B3 + 1 E2 2 E1 )dt x1 dx2

we find
divB = 0

and

~ =0
t B + rotE

The first equation expresses the absence of magnetic charges. The second equation is Faradays law of induction. They are the homogeneous Maxwell equations.
(f ) Similarly, we compute
~ + (t E1 + 2 B3 3 B2 )dx
d F = divEdt
+(t E2 + 3 B1 1 B3 )dy + (t E3 + 1 B2 2 B1 )dz
We thus find
divE =

and

rotB t E = ~j

These equations are called inhomogeneous Maxwell equations. The first equation is called
Gau law and the second equation Amp`eres law with Maxwells term.
We spend a moment to use Stokes theorem discuss the interpretation of these equations
which are the basis of all electrical engineering.
Interpretation 3.1.2.
44

(i) Gau law reads in integral form

Z

Q=

d x=
V

~ d~j.
E

The electric flux through a surface is thus proportional to the electric charge included by
the surface. It should be appreciated that this holds even for time dependent electric fields.
We use this law to determine the electric field of a point charge in the origin:
(~x, t) = q0 (~x).
Symmetry considerations lead to the ansatz:
~ x, t) = f (r)~
E(~
er .
Integrating over the two-sphere with center 0 and radius r, we find
Z
~ df~ = 0 4 r2 f (r)
q=
E
Sr2

and thus Coulombs law:

~ x, t) = q 1 e~r .
E(~
4 r2
In the case of static field, Coulombs law and the principle of superposition of charges
implies the first Maxwell equation.
(ii) Faradays law of induction yields a relation between the induced voltage along a loop F
bounding a surface F
Z
~
Uind
dx E
F

and the magnetic flux

mag

Z
=

~
df~B

through the surface which reads

Z

Z

Z
d
d
d
mag
~
~
~
~
d~j B = f
df B
Uind
dx E =
dt
dt
dt
F
F
F
This law is the basis of the electric motor and the electrical generator.
Remarks 3.1.3.
1. We consider the Maxwell equation on a star-shaped region U A on which we can apply
Poincares lemma. From dF = 0, we conclude that there exists a one-form A 1 (U )
such that dA = F |U . The one-form A is called a gauge potential for the electromagnetic
field strength F .
A is not unique: taking any function 0 (U ), we find that
A0 := A + d1 (U )
also obeys the equation dA0 = F |U . This arbitrariness in choosing A is called the
gauge freedom and choosing one A is called a choice of gauge.
One can impose additional gauge conditions on A to restrict the choice. For example, for
a Lorentz gauge, one requires A = 0.
45

2. We also write these equations in the three-dimensional language of vector calculus

~ = rot A
B
~
where A is the 3-vector dual to the spacial part of the one-form A 1 (U ). The field A
is called a vector potential. For the electric field, we find
~ = d A
E
t

with a scalar function called scalar potential. It is given by the time component of the
one-form A 1 (U ). Indeed, the homogeneous Maxwell equation yields
~+
d(E

A
) = 0,
t

and by Poincares lemma, we write the expression in parenthesis locally as the gradient
of a smooth function.
3. In this language, the gauge freedom is expressed by the fact that vector potential and scalar
potential are not unique, but can be change for any smooth function (~x, t) to
~0 = A
~ + grad
A

and

0 =

d
dt

which are again potentials for the same field strength.

4. Typical gauge conditions then read:
~=0
the Coulomb gauge: divA
for which the remaining gauge transformations are those with 4 = 0.
~ + 12
the Lorentz gauge: divA
c t
x
for which the remaining gauge transformations are those with  = c12 t
2 4 = 0.
5. We pause a moment to consider the special case of so called static situations in which
fields, currents and charge densities are time independent, the Maxwell equations decouple:
~ = div B
~ =0
div E
0
~ = 0 rot B
~ = 0~j with div~j = 0
rot E
On a star-shaped region, we find E = grad with 4 =  . Mathematically, electrostatics is thus the theory of the (inhomogeneous) Poisson equation. Similarly, we find for
~ = rot A
~ the equation
the magnetic field in terms of the vector potential B
4A grad divA = 0~j,
which reduces in the Coulomb gauge to 4A = 0~j.
A characteristic feature of electrodynamics is the existence of a quantity c having the dimension of a velocity.
Observation 3.1.4.
46

1. The force between two point charges q1 and q2 at a distance r is

 m
F~  =

1 q1 q2
.
40 r2

The magnetostatic force between two parallel conductors with currents I1 and I2 at a
distance r can be seen to be
 m  2 0 2lI1 I2
F~  =
.
4
r
The ratio of these two forces is dimensionless and independent of the system of units.
Hence the ratio
1
2 0 0 =: 2
c
is a quantity with the dimension of a velocity that is characteristic for eletrodynamics.
2. Another consideration yielding this velocity concerns Maxwells equations with vanishing
external sources = 0, ~v = 0.
~ = 0 rot E
~ = B~
div E
t
~ = 0 rot B
~ = E~
div B
t
Taking the rotation of the second equation yields
4E = 4E grad(divE) = rot rotE =
2

~
~ = E,
=
rot B
= rot B
t
t
t2
A dimensional analysis shows that we should have a factor with the dimension of a velocity
c2 . Hence we get


2
1

~ :=
~ =0
E
4 E
c2 t2
with  the so-called dAlembert operator. Similarly, one finds
~ :=
B


1 2
~ = 0.
4 B
c2 t2

To find solutions, we make the ansatz

~ ~x) := E~0 ei(t~k~x)
E(t,
~ ~x) := B~0 ei(t~k~x)
B(t,
~

of plane waves with direction of propagation |~kk| . These waves decribe the propagation of
light or other electromagnetic radiation (radio waves, gamma rays, . . . ) in empty space.
Hence c can be identified with the velocity of light.
We also get a dispersion relation
2 = c2 k 2
between wave length =

2
k

and frequency =
47

.
2

Maxwells equations imply moreover that

~k E~0 = 0
~k B~0 = 0
~k E~0 = B~0 ~k B~0 = 2 E~0 .
c
~ B)
~ form an oriented basis of R3 : electromagnetic waves are transverIn other words, (~k, E,
sal. Depending on the initial conditions, they can be polarized, either linearly or elliptically,
in particular in a circular way.

3.2

Special relativity

The existence of a distinguished velocity that is the same in all reference frames in relative
uniform motion has been experimentally verified to very high precision. But it is incompatible
with the transformation of velocities under the Galilei group. It thus forces us to revise our
ideas about space and time and gives rise to the theory of special relativity.
Our postulates are
(i) Space and time are homogeneous and spacially isotropic, i.e. our model for space time is
still still be the one of an affine space A over R4 .
(ii) The laws of physics should be of the same form in all reference frames in relative uniform
motion.
(iii) In particular, the velocity of light is fixed by Maxwells equations and thus should be the
same for all reference frames, independently of the motion of the source of light. It should
be a universal limit velocity that cannot be reached by an observer at rest.
Homogeneity in the first postulate is taken into account by modelling space-time again by
an affine space over the vector space R4 . It has as a symmetry group R4 o GL(4, R). Spacial
isotropy means that we distinguish a subgroup isomorphic to the rotation group SO(3). It is
important to note that the notion of spacial depends a priori on an observer We are lead to a
symmetry group R4 o G where G should contain a subgroup isomorphic top SO(3) which might
depend on the observer.
We have to specify the set of reference frames. They correspond to observers moving a
set of distinguished lines that is invariant under the action of R4 o G. It will be our postulates
that these are certain affine lines.
Together with postulate (iii), one can show that this leads to the unique solution:
Definition 3.2.1
(i) An affine space M over R4 together with a metric of signature (1, +1, +1, +1) on its
difference space is called a Minkowski space.
(ii) Standard Minkowski space is R4 with the diagonal metric. It plays the role of standard
Galilei space. Using the velocity of light, we endow it with coordinates (ct, x1 , x2 , x3 ) whose
dimension is length.
(iii) A Lorentz system of M is an affine map
: M R4
which is an isometry on the difference space.
48

(iv) The light cone in Tp M is the subset


LCp := {x Tp M p (x, x) = 0}.
(v) A Poincare transformation is a diffeomorphism of M whose differentials respect the light
cones in all tangent spaces, the so-called causal structure. A Lorentz transformation is
the linear map induced on the difference space.
Lemma 3.2.2.
Let
:MM
be a diffeomorphism that preserves light cones. Then is an affine mapping and we have on
tangent spaces
p (p x, p y) = ap ()p (x, y) for all x, y Tp M.
Neglecting translations, we have thus the direct product
R>0 SO(3, 1),
with R>0 the subgroup of dilatations.
Remarks 3.2.3.
(a) The Lorentz group L = O(3, 1) is a Lie group with four connected components:



L = L  det = , 00 = |00 |
with , = {1}.
(b) The connected component of the neutral element, L+1
+1 , are called proper orthochronous
Lorentz transformations.
Using the parity transformation
P = diag(+1, 1, 1, 1)
and time reversal
T = diag(1, +1, +1, +1)
we can write every element of L uniquely in the form
= P n T m 0

with n, m {0, 1}, 0 L+1

+1 .

(c) The Lorentz group L+1

+1 is not compact. A maximal compact subgroup is the group of rotations
of the Euclidean subspace spanned by span(e1 , e2 , e3 ) which is isomorphic to the group SO(3)
of rotations. Since the Lorentz group is semi-simple, any other maximal compact subgroup
is conjugate to this subgroup.
Consider the element

cosh sinh
sinh cosh
1 () =
0
0
0
0
49

0
0
1
0

0
0
+1
+1 .
0
1

It is called a boost in e1 -direction with rapidity . For y = ()x, we have y 2 = x2 , y 3 = x3

and
y 0 = x0 cosh x1 sinh
y 1 = x1 cosh x0 sinh
In particular, y 1 = 0 is equivalent to x1 = ct tanh = vt so that the origin of the y
coordinate system is in uniform relative motion, with velocity v = c tanh . In particular,
we find the velocity c of light as a limit that cannot be reached: |v| < |c|. Similarly, one
defines boost n () in any direction n and finds
n (1 )n (2 ) = n (1 + 2 ) = n ()
which reads in terms of coordinates
v=

v1 + v2
.
1 + v1c2v2

(d) Every Lorentz transformation can be written uniquely as a product of a boost B() and
a rotation (), but the boosts are not a subgroup.
(e) The Poincare group turns out to be the semi-direct product of the Lorentz group and the
translation group R4 . It is a ten-dimensional non-compact Lie group.
We reconsider the geometry of Minkowski space:
Definition 3.2.4
(i) A non-zero vector x in Minkowski spaces is called
time like, if x2 = (x, x) < 0
light like, if x2 = (x, x) = 0
space like, if x2 = (s, s) > 0

(ii) A time like or light like vector is called future directed, if x0 > 0 and past directed if
x0 < 0. The light cone decomposes into a forward and backward light cone.
(iii) The future I + (x) and the past I (x) of a point x M are the sets
I (x) := {y M|(y x)2 0, y x future / past directed }.
The future of a subset S M is defined by
I (S) := xS I (x)
We next introduce the notion of an observer. A basic postulate requires massive particles
and thus observers to move, in the absence of external forces (and gravitational forces) on affine
lines in Minkowski space with time like velocity vectors that are future directed. Light rays are
required to have lightlike velocity vectors.
Definition 3.2.5
50

1. A velocity unit vector w is a future directed timelike vector normalized to (w,

w)
= 1.
2. For any a M4 , the affine line a + Rw is the worldline of an observer which we abbreviate
with Oa,w or sometimes even with Ow , when the basepoint a does not matter.
3. The time interval elapsed between two events x, y R4 as observed by observer Ow is
given by the observer-dependent time function
tw (x, y) := (w,
x y) .
Two events are simultaneous, if tw (x, y) = 0. We say that x happens before y, if
tw (x, y) < 0 and x happens after y if tw (x, y) > 0.

Remarks 3.2.6.
(a) The time interval between two events x, y M depends on the observer.
(b) The set of events simultaneous to x for the observer w is the affine hyperplane perpendicular to w containing x. This gives a foliation of Minkowski space into space-like affine
hyperplanes that depends on the velocity unit vector w.

(c) More generally, a massive of massless point particle is described by its trajectory
x: I M
7 x ( )
where we require for physical motions the velocity vector to be future directed and timelike
or lightlike, respectively,

2
dx0
dx
> 0
0.
d
d
The Lorentz invariant quantity
12

1
=
c

s 
2
dx
d
d

is called eigen time of the particle elapsed between the two points x(1 ) and x(2 ) on its
trajectory.
We can now formulate the postulate of causality:
an event x M can influence an event y M (slang: a signal can be sent from x to y) if and
only if y lies in the future of x.
If x 6 I + (y) and y 6 I + (x), the the events are called causally disconnected.
The following lemma guarantees that causality is reasonably related to the time differences
observed by all observers:
Lemma 3.2.7.
For any observer Ow and for any events x, y M simultaneous for Ow , the relative position
y x is space like.

51

The following lemma ensures that all observer dependent time functions are consistent with
causality:
Lemma 3.2.8.
1. For any two observers Ow and Ov with different velocity unit vectors w and v, there
exist events x, y M such that x happens before y for w and y happens before x for v.
(Relativity of simultaneity). However, such events are always causally disconnected and
cannot influence each other so that causal paradoxes are excluded.
2. The future I + (x) of an event x M equals
I + (x) = {y M|(v, y x) < 0 for all observersOv}
i.e. the set of events that happens after x for all initial observers.
Observation 3.2.9.
We discuss some famous relativistic phenomena:
a) Time dilatation:
Consider two observers w and v and two events x, y M. Assume that they occur at rest
for the observer w,
i.e. x y = tw with some t R. One could think about a clock travelling
with observer w on which this observer looks twice; he measures a time interval
p
tw (x, y) = |x y| = (x y)2 .
In contrast, observer Ov measures a time difference
tv(x, y) = |(
v , x y)|
Introduce
a := (x y) + (x y, v) v
which is a space-like vector orthogonal to v. We have thus
x y = tv(x, y) v + a
which implies
tw (x, y)2 = |(
v , x y)|2 + a2 = tv(x, y)2 + a2
Hence
|tv(x, y)| > |tw (x, y)|
Thus the moving observer Ov measures a longer time interval than the observer Ow at rest.
This effect has been measure with very high precision in observations of the life time of
instable particles in motion.
b) Similarly, lengths in motion see to be contracted: one has the phenomenon of
(length contraction).

52

c) We finally discuss the twin paradoxon as the geometry of triangles given by future-directed
timelike vectors in M. Consider three future directed timelike vectors in M that obey
z = x+y
with the interpretation that z is the twin at rest, x describes the way forth and y the way
back. One has
(z, z) = (x, x) + 2(x, y) + (y, y)
and thus in the rest system of x where x takes the form x = (t, 0, 0, 0):
p
p
2(x, y) = 2x0 y 0 = 2 x2 y 2 + ~y 2 > 2 x2 y 2 .
One should appreciate that this has essentially the structure of the Cauchy-Schwarz inequality, but with the opposite inequality. It implies

p 2
x2 + y 2
z2 >
p

z 2 > x2 + y 2
so that the twin at rest is older.

3.4

Electrodynamics as a gauge theory

We have seen that the homogeneous Maxwell equation dF = 0 suggests to apply the Poincare
lemma on a star-shaped subset U R4 (or more generally on a Lorentzian manifold modelling
space time). and to write the two-form F as the exterior derivative dA of a one-form A 1 (U ).
On a general manifold, this is possible only locally. Hence, we fix a good open cover
U = (U ) of our (Lorentz-)manifold. This means that arbitrary intersections
U1 ...n := U1 . . . Un
of coordinate neighbourhoods are either empty or contractible.
2
1
Applying the
 Poincare lemma to F |U (U ) we find A (U )
such that F U = d A .

On U = U U we find a one-form


d (A A ) = F U U F U

= 0.

and thus, again by the Poincare lemma a function C (U , R) with

(A) := A A = d =

1
1 i i
e
de
= d log ei
i
i

0 (U ) := C (U , U (1)) .
We therefore put g := ei
On triple overlaps U , we find the so-called cocycle relation more
(g) := g g1 g = 1.
53

 Define
V n,m :=

n U1 ,...,m+1

1 ,...,m+1

and form the double complex

...

V 0,1 V 1,1
d

0,0
0 V
V 1,0
d

with d2 = 0, 2 = 0, d + d = 0. Then D = d + (1)n

total complex
M
V n :=
V k,l

is the differential on the

k+l=n
1
Hence a class in HD
is represented by an element in

0 (U ),
V 1,0 V 0,1 = 1 (U ) ,
and hence by a pair (A , g ) as above, for which the identity holds
0 = D (A , g ) = (( A) dlog g , ( g) )
1,1 (U ) 0,2 (U ).
The class is determined up to the imppage of
D : V 00 0 (U ) V 1,0 V 0,1


D( ) = d , 1
.

For a geometric interpretation, we consider on the disjoint union

G
[
=
L
(U C ) =
(U {} C)

the equivalence relation

(x, , g z) (x, , z) .
Then the quotient


L := L

comes with a natural projection to M which provides a complex line bundle : L M .

This bundle comes with a linear connections
5 = d +

1
A
i

which furnishes a curvatures which is just the electromagnetic field F by


F U := d A .
54

Because of
d A d A = d ( A) = d dlog g = 0
this is globally defined. One should appreciate that this imposes on the cohomology class
of the two-form F the condition to be integral,
[F ] H2 (M, Z) H2dR (M, R) .
One should also note that if the fundamental group 1 (M ) 6= 0 is non-trivial, there are
non-isomorphic line bundles with the same field strength F . They can be dinstinguished
by quantum mechanical measurements in the Aharonov-Bohm effect.

Hamiltonian mechanics

4.1

(Pre-) symplectic manifolds

We will now get to learn a different formulation of mechanical systems that turns out to be
a good starting point for quantum mechanics as well. To present it, we need a few geometric
notions.
Definition 4.1.1 Let M be a smooth manifold.
(i) A 2-form 2 (M ) is called a presymplectic form, if it is closed and of constant rank.
(ii) A non-degenerate pre-symplectic form is called a symplectic form.
(iii) A smooth manifold M together with a (pre-)symplectic form is called a
(pre-)symplectic manifold.
(iv) A morphism f : (M, M ) (N, N ) of (pre-)symplectic manifolds is a differentiable map
f : M N that preserves the (pre-)symplectic form, f N = M . Such morphisms are
also called canonical transformations.
Remarks 4.1.2.
1. Since for any point p of a symplectic manifold M the tangent space Tp M is a symplectic
vector space, the dimension of a symplectic manifold is necessarily even. A presymplectic
manifold, in contrast, can have odd or even dimension.
2. One checks readily that if M is a symplectic manifold of dimension dim M = 2n, the
n-the power n of the symplectic form is a volume form on M .
Example 4.1.3.
For any smooth manifold M of dimension n, the cotangent bundle T M has a natural structure
of a symplectic manifold of dimension 2n.
Indeed, local coordinates (xi )i=1...n for M give local coordinates (xi , pi ) for the cotangent
bundle T M , by describing local one-forms as
(x) =

n
X

pi (x)dxi .

i=1

55

Under changes of coordinates, we have

n
X
xi j
d
x =
dx .
j
x
j=1
i

Consider the locally defined two-form on the total space T M of the cotangent bundle
0 :=

n
X

dpi dxi .

i=1

Locally, it can be written as d with

:=

n
X

pi dxi .

i=1

Indeed,
d =

n
X

dpi dqi = 0 .

i=1

Hence the form 0 is closed; one checks that it is non-degenerate. Moreover, it is independent
of local coordinates: if xi are different coordinates, we find
n
X
i=1

n
X
xi xj j
dp dxk = 0 .
d
p d
xi =
k x
i
x

i,j,k=1
i

As a consequence, the locally defined two-forms patch together to a globally defined symplectic form on the cotangent bundle T M . The symplectic manifold (T M, 0 ) is called the
canonical phase space associated to the manifold M .
The following result is central to symplectic geometry:
Theorem 4.1.4 (Darboux theorem).
Let (M, ) be a (2n + k)-dimensional presymplectic manifold with rank rank = 2n.
Then we can find for any point m M a neighborhood U and a local coordinate chart
: U R2n+k
(u) = q 1 , ..., q n , p1 , ..., pn , 1 , ..., k


P
such that U = ni=1 dpi dq i . Such coordinates are called Darboux coordinates. One can even
choose a covering such that the induced coordinate changes are canonical transformations.
Darboux theorem implies that symplectic geometry is locally trivial - in contrast to Riemannian geometry where curvature provides a local invariant.
Observation 4.1.5.
Let (M, ) by a symplectic manifold and f C (M, R) be a smooth function. Since is not
degerate, we can find a unique vector field Xf on M such that
df (Y ) = (Xf , Y )
56

holds for all local vector fields Y on M holds. The vector field Xf is called the symplectic gradient
of f . This way, we obtain an embedding
.
C (M, R) R , VectM .
We compute the symplectic gradient in local Darboux coordinates defined on U M . We
make the ansatz
n
X

Xf =
i i + i
vect(U )
q
pi
i=1
with local coordinate functions i and i . We evaluate both sides on a local vector field
Y =

n
X
i=1

vect(U ) .
+ i
i
q
pi

Since we work in Darboux coordinates, we find

(Xf , Y ) =

n 
X

i i i i

i=1

which has to equal

n
X
f i f
df (Y ) =
+
i
q i
pi
i=1

The comparison of the coefficients yields for the symplectic gradient the local expression
Xf =

n
X

i=1

f
f
+
vect(U ) .
pi q i q i pi

The (commutative) algebra of smooth functions on a symplectic manifolds carries additional

algebraic structure.
Definition 4.1.6
Let k be a field of characteristic different from zero.
(i) A Poisson algebra is a k vector space P together with two bilinear products and {, }
with the following properties
(P, ) is an associative algebra.
(P, {, }) is a Lie algebra.
The bracket {} provides derivations for the associative product:
{x, y z} = {x, y}z + y{x, z}.
(ii) A Poisson manifold is a smooth manifold M with a Lie bracket
{, } : C (M ) C (M ) C (M )
such that (C (M ), {, }) together with the pointwise product of functions is a (commutative) Poisson algebra.
57

Examples 4.1.7.
(i) Every associative algebra (A, ) together with the commutator
[x, y] := x y y x
has the structure of a Poisson algebra. One should notice that this commutator is trivial,
if the algebra is commutative.
(ii) Consider the associative commutative algebra A = C (M ) of smooth functions on a
smooth manifold M . If M has a Poisson structure, we can using the symplectic gradient
to define the following Poisson bracket
{F, G} := hdG, Xf i
for F, G A. From the last equation in observation 4.1.5, we find
{f, g} = dg(Xf ) =

g f
g f
i
.
i
pi q
q pi

It is an instructive exercise to verify antisymmetry and the Jacobi identity from this
local formula. Also, the reader should check the following equation which asserts that the
symplectic gradient provides a morphism of Lie algebras from the Lie algebra structure on
smooth functions given by the Poisson structure to the Lie algebra of vector fields:
X{f,g} = [Xf , Xg ] .
(iii) For any smooth function h C (M ) on a Poisson manifold M , the map
{h, } : C (M ) C (M )
f 7 {h, f }
is a derivation and thus provides a global tangent vector field. We consider its integral
curves s : I M which are solutions of the following first order ordinary differential
equations
d
f s(t) = {h, f }(s(t))
dt
for which we introduce the short hand notation:
f = {f, h} .
In local Darboux coordinates, we find for the special case of a local coordinate function
f = q i the equation
h
qi =
pi
and for f = pi
pi =

h
.
q i

These equations are called Hamiltons equations for the Hamilton function h.

58

(iv) For a simple time independent mechanical system with configuration space M , the equations of motion are second order on M and thus first order on T M . They correspond to
a vector field on T M .
The cotangent bundle, on the other hand, is a symplectic manifold. This raises the question
on whether there is a function h on T M such that the vector field {h, } on T M contains
information equivalent to the mechanical system.
Observation 4.1.8.
Before we explain this point, we wish to have another look at presymplectic manifolds. For every
point p M , consider the subspace
kerp = {v Tp M | p (v, ) = 0} = {v Tp M | v p = 0} Tp M
which gives a distribution on M
[

ker =

kerp T M

pM

of constant rank. We use Cartans formulae to check that the commutator of two vector fields
v, w with values in ker is contained in ker .To this end, we have to show that v = w = 0
implies [v,w] = 0. We find
[v,w] = Lv w w Lv = 0 iw (dv + v d) = 0 .
This allows us to apply a theorem of Frobenius that asserts that there is a foliation of M ,

i.e. M can be written as a disjoint union of submanifolds L A , called the leaves of the
foliation, whose tangent spaces are just ker , i.e. for p L
Tp L = kerp
or, more globally,
[ [

Tp L = ker .

A pL


In general, the set of leaves UM = M ker is not a smooth manifold. If this happens to be
the case, then it is of dimension dim M dim ker . In this case, UM carries a symplectic
structure
such that the projection

: M M ker = UM
becomes a morphism of presymplectic manifolds,
= .
One sufficient condition is the existence of local slices: for every point p M , one can find
a submanifold p that intersects every leaf at most once and whose tangent space complements
ker in every point q ,
Tq M = kerq Tq .

59

4.2

Hamiltonian dynamics

Observation 4.2.1.
The kinematical framework for Lagrangian mechanics is a fibre bundle = pr1 : E =
I E I with R and interval and a Lagrangian function
l : J R.
The dynamics is given by a differential equation R J given by the variational problem
for l.
We now invoke the Newtonian principle that R is second order. This implies that the
Lagrangian function l factorizes to a function on J1
= T E I, cf. Lemma 2.3.6
Our idea is to describe the motion of the mechanical system, i.e. solutions to the variational problem given by l in terms of the integral curves of a vector field of the form {h, }
Since the generalized momenta li are cotangent vectors to E,
this amounts to a
on T E.
st
to generalized momenta.
transition from generalized velocities, i.e. tangent vectors to E,
Hence we consider the symplectic manifold
T E
= T E T I
with local coordinates (q, p, t, h) and and the local canonical 1-form
= p dq h dt
which induces the global symplectic form = d.
The submanifold relevant for the equations of motions is defined as the zero locus of a
function
R
: T M
We will from now on assume that is of the form
(p, q, t, h) = pr1 H h0
with a constant h0 and a function H : T M R called the Hamilton function.
We could interpret as a moment map with values in the Lie algebra R. By a so-called
symplectic reduction, we eliminate h and obtain the manifold
T E I,
called the extended phase space or evolution space. The local one-forms
= pdq pr1 H dt = pr1 E pr1 H dt
provide us the global closed 2-form
= d = pr1 T E pr1 dH dt.
on T E I.
60

Lemma 4.2.2.
(T M I, ) is a presymplectic manifold with


ker(m,t) = span ((XH )m , 1)
Hence the integral curves of the symplectic gradient XH on T E are the images of the leaves of
T E I under projection on the first factor.
Proof:
In local coordinates (pi , q i , t) on T E I we find
d = dp dq

H
H i
dp dt i dq i dt .
pi
q

The corresponding antisymmetric matrix reads

0
Enn

0
A = Enn
H
H
pi
q
i

H
pi
H
q i

It has obviously rank at least 2n. On the other hand, the vector


H H
i,
, 1 = (XH , 1)
q pi
is in the kernel of A, hence the rank equals 2n.

One can even show that the space of leaves


T E I ker
is a symplectic manifold and isomorphic, as a symplectic manifold, to the cotangent space
). The cotangent bundle can thus be interpreted as the space of trajectories or classical
(T E,
solutions. Following Souriau, we call it the space of motions.
More generally, we define:
Definition 4.2.3
1. A Hamiltonian system consists of a presymplectic manifold (M, ), called evolution space
with one-dimensional leaves and a smooth space of leaves, called the phase space of the
system, together with a smooth function
H:M R,
called the Hamiltonian function. A leaf of the foliation is also called a trajectory. The
are called (Hamiltonian)
images of trajectories in evolution space M in phase space M
trajectories.
,
2. A time-independent Hamiltonian system consists of a symplectic manifold (M
), together with a smooth function
:M
R,
H
also called the Hamiltonian function.
61

Lemma 4.2.4.
,
be a time-independent Hamiltonian system. Then for any interval I R,
1. Let (M
, H)
, where the presymplectic form
one has a Hamiltonian system with evolution space I M

:= pr2
is the pullback under the
;
pr2 : M M

and the Hamiltonian is H := pr2 H.

,
2. Then the symplectic manifold (M
) is the phase space of the Hamiltonian system
(M, , H). Hamiltonian motions are then integral curves of the symplectic gradient XH .
Let us discuss an important example of Hamiltonian systems:
Example 4.2.5.
A (time-independent) natural Hamiltonian system consists of the cotangent space M = T E
of a smooth manifold E with its canonical symplectic structure, together with a Hamiltonian
function H : T E R that is the sum of a positive definite quadratic form h, i on the
fibres of T E and the pull back of a smooth function
V : E R
In local Darboux coordinates:
under the projection : T E E.
1
h(p, q) = hp, pi + V (q) .
2
We still have to understand how to relate for natural systems a Lagrangian function
l : T E R
to a Hamiltonian function
H : T E R .
Definition 4.2.6
Let I R be an interval and f : I R be a piecewise smooth continuous convex function, i.e.
f (tx + (1 t)y) 6 t f (x) + (1 t) f (y)
for all x, y I and all t [0, 1]. This implies that f 00 (x) > 0, wherever the second derivative is
defined.
Consider the function
F :RI R
(p, x) 7 px f (x) .
for a fixed p R. Suppose maxxI F (p, x) exists. The fact that f is convex implies that it is
assumed for a unique x = x(p) I. In case f is differentiable in x, we find f 0 (x) = p. This
defines implicitly x as a function of p.
This allows us to introduce a function of P by
g(p) := F (p, x(p)) = maxxI F (p, x).
The function g is called the Legendre transform of f .

62

Remarks 4.2.7.

(i) Consider as an example the function f (x) = m x on R with m > 0 and > 1. Then the
function
x
F (p, x) = px m

1
has extrema for p = mx . Hence we find for the Legendre transform
1

g(p) = m 1

with

1
1
+ = 1.

As special cases, we find for = 2

f (x) =

m 2
x
2

with Legendre transform

g(p) =

p2
2m

and for m = 1
f (x) =

with Legendre transform

g(p) =

p 1
1
,
+ =1

(ii) The definition via a maximum implies the inequality

F (x, p) = xp f (x) 6 g(p)
for all values of x, p where the expression makes sense. We rewrite this as the so-called
Youngs inequality
px 6 f (x) + g(p).
Applying this to the second example just discussed, we find the classical inequality
px 6

1 1
1
1
x + x with
+ = 1.

(iii) If f is a smooth function with f 00 > 0, then its Legendre transform g is again a convex
function. It turns out that the Legendre transform is an involution.
(iv) We mention the following interpretation: the enveloping curve of the family of lines {y =
px g(p)}, with g(p) a convex function, has the equation y = f (x), where f and g are
related by a Legendre transform.
(v) We finally mention the generalization to finite-dimension R-vector spaces. Let f : V R
2f
be a convex (smooth) function, i.e. the Hessian xi x
j is positive definite. Then defined a
function on the dual space
g : V R
by
F : VV R
F (p, q) = hp, qi f (q)
and put g(p) = maxxV F (p, x). All results, including Youngs inequality, generalize.
63

(vi) It is instructive to specialize to a positive definite quadratic form

f : V R
with f (q) = 21 Aij q i q j . Minimizing
F (p, q) = hp, qi f (q)
for a fixed p yields pi = Aij q j and thus q i = (A1 )ij pi and thus
1
1
g(p) = pi (A1 )ij pj (A1 )ij pi pj = (A1 )ij pi pj .
2
2
First explain 4.2.3 4.2.5.
Next, we relate the Lagrangian and the Hamiltonian formulation for natural systems.
Observation 4.2.8.
1. Let E be a smooth manifold and I R an interval. Consider a natural mechanical
Lagrangian system with Lagrangian function
l : T E I R
Then, locally,
leading to second order equations of motions. Fix local coordinates q i on E.
we have
l = l(q i , qti , t)
2. On the other hand, consider the cotangent bundle T E of E with its canonical symplectic
structure as the phase space of a natural Hamiltonian system and the manifold T E I
with its canonical presymplectic structure as the evolution space. Local coordinates on E
then give local Darboux coordinates (q i , pi ) on the symplectic manifold T E and on the
presymplectic manifold I T E A (time dependent) Hamiltonian function
H : T E I R
is then given in local Darboux coordinates as
H = H(p, q, t) .
Theorem 4.2.9.
Suppose that the Lagrangian function l is for fixed q M convex in its second argument
qt Tq(t) M . Suppose that the Lagrangian function l and the Hamiltonian function H are related
fibrewise by a Legendre transform. In local coordinates:
H(p, q, t) = pqt l(q, qt , t)
with qt = qt (p, q, t) as defined by the Legendre transform and
l(q, qt , t) = pqt H(pi , q i , t)
with p = p(q, qt , t) respectively.
64

1. If
s : I E I
is a motion for the Lagrangian system given by l, i.e. a solution of the Euler-Lagrange
equations, then
s : I T E I
t 7 (ds(t), t) T E I
is a motion for the Hamiltonian system given by H.
2. Conversely, if
s : I T E I
is a motion for the Hamiltonian system given by H, then using the projection : T E E
of the cotangent bundle, we obtain a solution
s := (( idI ) s, id) : I E I
of the Lagrangian system given by l.
These statements are sometimes summarized that, if a Lagrangian function l and a Hamiltonian function H are related by a Legendre transform, then the Euler-Lagrange equations
D

l
l
=
qti
q i

are equivalent to the Hamiltonian equations

pi =

H
H
, qi = i ,
i
q
p

i.e. 2n ordinary differential equations of first order, where is the Legendre transform of l.
Proof:
The total derivative of the time-dependent Hamiltonian H is the one-form
dH =

H
H
H
dp +
dq +
dt 1 (T E I)
p
q
t

which we compute explicitly for

H = pqt l(q, qt , t)
where qt = qt (p, q, t) is defined implicitly by the Legendre transform as a function on evolution
l
space T E I. Using the identity pi = q
i of local functions on evolution space implied by the
t
Legendre transform, we find
d(pqt l(q, qt , t)) =
 q
 q
l qt 
l
l qt 
l
t
t
dp dq + p
dq dt
= qt dp + p

p
q p
q
q
q q
t
|
{z t }
|
{z t }
=0

=0

l
l
= qt dp dq dt
q
t

(10)
65

By comparison of coefficients, we find the following identities of locally defined functions on

T E:
H
H
l
H
l
,
=

,
=

qti =
pi
q i
q i
t
t
Suppose, s : I E I is a Lagrangian motion. Then the first equation becomes

qi =

H
.
pi

The Euler-Lagrange equations read

l
l
d
s=D
s=
q
qt
dt

l
s
qt


.

Substituting the second equation yields

pi =

H
q i

so that we have obtained the Hamiltonian equations.

l
l
From the Euler Lagrange equations q
= D q
= Dpt we now derive Hamiltons equations.
The second part of the assertion is shown analogously.

Observation 4.2.10.
,
The following observation is formulated for a Hamiltonian system with phase space (M
)
I with local Darboux
and local Darboux coordinates (p, q) and evolution space M := M
, we have the symplectic form which is locally
coordinates (p, q, t). On phase space M
!
X
X
i
i
;

=
dpi dq = d
pi dq = d
i

, we have the presymplectic form which is locally

on the evolution space M = I M
= p1
p1 dH p2 dt
!
X
=d
pi dq i Hdt
i

= d .
are solutions of the Hamiltonian equations
Motions in M
x = {x, H} with x(t1 ) = (p1 , q1 ).
They lead to a phase flow which we assume to be globally defined. It gives us diffeomor
phisms of M
M

gtt2 : M
1

for any pair t1 , t2 I. On evolution space M , we obtain parametrizations of the leaves:

t 7 gtt1 (x), t .
66

Proposition 4.2.11.
Let 1 and 2 be two curves in evolution space M that encircle the same leaves. Then we have
I
I

=
.
1

is called Poincare-Cartan integral invariant.

The one-form
Proof:
Let be the the surface formed by those parts of the leaves intersecting 1 (and thus 2 ) that
is bounded by the curves, = 2 1 . Such a surface is sometimes called a flux tube. Stokes
theorem implies
I
I
Z
Z
Z

=
=
d =

= 0,
2

since consists of leaves whose tangent space is by definition the kernel ker
of the presymplectic form.

Observation 4.2.12.
We now specialize to curves in evolution space M at constant time, i p1
2 (ti ), to find
I
I
p dq =
p dq
1

parametrized
We can obtain such a pair of curves from a closed curve in phase space M
by s [0, 1] and first lifting to a curve


1 (s) = (s), t1
parametrized by s [0, 1] at fixed time t1 I and and then applying a phase flow to obtain a
closed curve


2 (s) = gtt12 (s), t2
in evolution space at another fixed time t2 .
Using the surface
n
o


= gtt1 (s), t , s [0, 1] , t [t1 , t2 ]
formed by the leaves, we obtain by proposition 4.2.11 the equality
I
I
p dq =
p dq
t
1

gt 2

.
of integrals along curves in phase space M
in phase space for the integral over the
As a consequence, we find for any surface M
symplectic form
Z
Z
=
.
t
1

gt 2

The symplectic form on phase space is thus invariant under the phase flow. We conclude that
the maps gtt12 : M M describing phase flow are canonical transformations.
The phase flow has as further invariants the power k of the symplectic form. This includes

,
as a special case the canonical volume form
dimM /2 of the symplectic manifold (M
).
67

We need the following simple

Lemma 4.2.13.
Let M be a manifold with a volume form and g : M M be a volume preserving map that is
mapping a measurable subset D M of finite volume to itself. Given a point p D, we can
find in any neighborhood U of p a point x U which comes back to the neighborhood U , i.e.
there is n N such that g n x U .
Proof:
The images U, gU, . . . , g n U, . . . have all the same finite volume and are contained in the subset
D of finite volume. Hence they cannot be disjoint. Hence there are integers k and l with k > l
such that
g k U g l U 6= .
We deduce g kl U U 6= .

We deduce the following famous statement:

Proposition 4.2.14 (Poincares recurrence theorem).
Consider a (time-independent) natural Hamiltonian system consisting of a phase space

(T E,
0 ) and a Hamiltonian function
R
h:M
(p, q) 7 hp,pi
+ V (q)
2m
Then the associated phase flow
M

g :I M
we can find a point x U
has the property that in any neighborhood U of any point p M
which returns to U for sufficiently large time t, g(t, x) U .
Proof:
Consider the subset
D = {(p, q) | H(p, q) 6 const}
of finite volume. Since phase flow preserves the symplectic volume, we can apply the lemma. 

5
5.1

Quantum mechanics
Deformations

We start by considering the mathematical notion of a formal deformation of an associative

algebra.
Definition 5.1.1
Let k be a field of characteristic 0 and A an associative k-algebra with unit.

68

(i) An algebra homomorphism

: Ak
is called an augmentation, if ( 1A ) = holds for all k. An augmentation endows
the ground field k with the structure of an A-bimodule by
Ak
(a, )
kA
(, a)

k
(a )
k
( ) .

For example, the ring k [[t]] of formal power series with coefficients in k admits the augmentation
 : k [[t]] k
!
X

ai ti := a0 .
iN0

It can be shown that this ring is a local Noetherian ring.

(ii) A formal deformation of an associative algebra A is an associative k [[t]]-algebra B together
with an isomorphism of k-algebras
:= k k[[t]] B
: B
A .
0 A) and (B 00 , 00 : B
00 A) is a
A morphism of formal deformations (B 0 , 0 : B
morphism
: B 0 B 00
of k [[t]]-algebras such that
(00 )1 0 = idk k[[t]] .
Isomorphic deformations are also called equivalent.
Theorem 5.1.2.
A formal deformation of a k-algebra A is equivalent to a family of k-linear maps
{i : A A A}iN
with 0 (a, b) = a b and
X

i (j (a, b), c) =

i+j=k

i+j=k

i,j>0

i,j>0

i (a, j (b, c))

for a, b A.
Indeed, we can use these maps to endow the free k [[t]]-module B := k [[t]] A with the
product
a b := a b + t1 (a, b) + t2 2 (a, b) + ... for all a, b A
which we extend in a k [[t]]-linear way.
The associative product on k [[t]] A is also called a star product. The following lemma
relates Poisson algebra and formal deformations:
69

Lemma 5.1.3.
Let A be a commutative associative k-algebra. Then for any formal deformation of A, the kbilinear map
{a, b} := 1 (a, b) 1 (b, a) for all a, b A,
endows A together with its commutative product with the structure of a Poisson algebra.
Proof:
As we have seen in example 4.1.7 for any associative algebra, the commutator
[x, y] = x y y x
endows the associative algebra k [[t]] A with the structure of a Poisson algebra. A simple
calculation then shows due to commutativity of the product in A, the first non-vanishing term
is the bracket {, }:
[x, y] = {x, y}t + O(t2 ) .
The Jacobi identity follows from the identity
[x, [y, z]] = {x, {y, z}}t2 + O(t3 ) .
Similarly, one shows that {, } is a derivation for the commutative product of A.

This motivates the following

Definition 5.1.4
(i) Given a formal deformation (k [[t]] A, ) of an associative commutative algebra A, the
Poisson algebra (A, , { }) = P is called the classical limit of (k [[t]] A, ).
(ii) Conversely, given a Poisson algebra P = (A, , { }), a k [[t]]-algebra B is called a
deformation quantization of P if the classical limit of B is isomorphic to the Poisson
algebra P .
(iii) A deformation quantization of a Poisson manifold M is a deformation quantization of the
Poisson algebra (C (M ), , {, }) in which all maps i are differential operators.
Theorem 5.1.5. (Kontsevich 1997)
Any Poisson manifold admits a deformation quantization that is essentially unique.
Observation 5.1.6.
(a) The phase space of a Hamiltonian system is a Poisson manifold. The Poisson structure can
be interpreted as a hint to the existence of a family of associative algebras (which in general
are not commutative any longer). This idea is at the basis of quantum mechanics.
(b) The Poisson bracket on a phase space is in local Darboux coordinates
{f, g} =

f g
f g

.
q i pi pi q i

Hence it has dimension [length momentum]1 = action1 . As a consequence, the deformation parameter t has to be a dimensionfull quantity, t = ~ with the dimension of an
action, the Planck constant. It has the value of 1.0541034 Js and is one of the fundamental
constants of nature.
70

(c) A dimensionfull quantity like ~ is neither big not small per se. Only its ratio with a quantity
of the same dimension, i.e. a characteristic action of a system can be small or big.
(d) While ~ has a definite value, there are no evaluation homomorphisms for a formal deformations for other values of the deformation parameter than zero. Hence we have to look for
a different framework that allows to formulate convergence properties: normed algebras.

5.2

C -algebras and states

Definition 5.2.1
A (unital) C -algebra A is a complex (unital) associative algebra
which is a Banach algebra, i.e. it is endowed with a norm kk : A R>0 such that (A, k k)
is a complete topological vector space and such that the norm is submultiplicative
ka bk kak kbk .
with an involution generalizing complex conjugation and hermitian conjugation
: AA
a 7 a ,
which is an antilinear antihomomorphism of algebras:
+
(a + b) = a
b f
ur alle , C, a, b A

(ab) = b a
f
ur alle a, b A
The following identity holds
ka ak = kak2

for all a A.

Remarks 5.2.2.
(a) The identity ka ak = kak2 implies
k1k = 1, ka k = kak, ka1 a2 k 6 ka1 k ka2 k.
(b) Elements such that a = a are called self-adjoint. They form a real subspace, but in general
not an associative subalgebra, but only a Lie subalgebra.
(c) Let H be a separable Hilbert space, i.e. a complete unitary vector space with a topological
basis that is at most countable. The scalar product h, i on a Hilbert space H will be in
our conventions always linear in the first argument and antilinear A linear map
A: HH
is called bounded, if
kAk := supkxk=1 kAxk <
The space of bounded operators on a Hilbert space is a Banach algebra B(H).
71

(d) From Riesz representation theorem we deduce that for given A B(H) and v H the
continuous linear function
hA, vi : H C
can be represented by a unique vector A v H:
hA, vi = h, A vi

for all v H.

It is not hard to show that A B(H) and that the Banach algebra B(H) is endowed by
the map A 7 A with the structure of a C -algebra.
(e) A theorem of Gelfand and Naimark asserts that every C -algebra is isomorphic to a closed
-subalgebra of B(H).
The idea is that a C -algebra serves as an algebra of observables, as a quantization of the
Poisson algebra of functions on (extended) phase space. Quantum mechanics is an inherently
probabilistic theory. Hence we want to associate to any observable at least an expectation value.
This motivates the following definition:
Definition 5.2.3
Let A be a C -algebra. A state on A is a normed positive linear functional
: A C,
i.e. we have A and positivity and a normalization condition,
(a a) > 0

and

(1) = 1 .

Remarks 5.2.4.
(a) The set of all states is convex: if 1 and 2 are states, then for all t [0, 1] the convex
linear combination
t = t1 + (1 t)2
is a state.
(b) A point x of a convex subset X of a real vector space V is called extremal, if for every
segment yz X containing x one has either y = x or z = x.
(c) Extremal states on a C -algebra are called pure states.
(d) Such states exist: according to the Krein-Milman theorem every compact convex subset of
a locally convex vector space equals the closed convex hull of its extremal points.
Examples 5.2.5.
1. The complex valued continuous functions on R
C 0 (R) ={f : R C continuous | for all  > 0 exists N = N (f ),
such that |f ()| <  for all || > N }
72

form a commutative algebra without unit. Taking complex conjugation as an involution

and the supremum as a norm, we endow it with the structure of a C -algebra.
Since the algebra C 0 (R) is not unital, we need to modify the normalization condition. A
state is called normalizable, if there is k R such that
|(f )| 6 kkf k
for all f . We define kk as the infimum of such k. A state is called normalized, if kk = 1.
Thus states are normalized Radon measures, i.e. normalized probability measures.
The pure states of C0 (R) are Dirac measures, i.e. evaluations at some x R,
x (f ) = f (x) .
They are in bijection to points in R.
2. A second example is the C -algebra A = B(H) of bounded operators on a Hilbert space
H: an endomorphism of H is called a density matrix, if of all a A the endomorphism
a B(H) is trace class. In this case,
(a) =

Tr(a)
Tr()

is a state on B(H). Taking for the orthogonal projection on a unit vector H, we

obtain a pure state
(a) = (a, ) .
We next wish to endow the continuous dual of a Banach space with a topology. Two topologies are of particular importance:
Remark 5.2.6.
1. Let A be a complex Banach space and A its continuous dual, i.e. the continuous linear
functionals. It can be endowed with the structure of a Banach space by the operator norm
kk := sup

aA

|(a)| .

kak=1

2. There is a second important topology on the continuous dual A , the weak topology or
topology of pointwise convergence. It is generated by the following collection of subsets of
A : for 0 A , a A and > 0 we set



U (0 ; a) := A  |(a) 0 (a)| < .

Definition 5.2.7
Consider a C algebra A.
(i) An element A is called a character, if 6= 0 and
(ab) = (a)(b)
73

for all a, b A .

(ii) The subset M (A) A of all characters of A together with the topology induced by the
weak -topology on A is called the Gelfand spectrum of the C -algebra A.
Example 5.2.8.
Let X be a locally compact Hausdorff space and A = C 0 (X) be the algebra of continuous
functions on X. Then for every point p X, the evaluation map f 7 f (p) is a character.
Moreover, the map
X M (C 0 (X))
p 7 p (f ) = f (p)
that associates to every point p the character corresponding to evaluation in p is injective and
continuous.
Proof:
To see injectivity, consider two different points p 6= q. Since f is a locally compact Hausdorff space, we can find a continuous function f C 0 (X) with f (p) 6= f (q). The inequality
p (f ) = f (p) 6= f (q) = q (f )
implies that the characters are different, p 6= q .
Consider a convergent sequence pi p in X. Then for every f C0 (X), continuity of f
implies
pi (f ) = f (pi ) f (p) = p (f ) ,
which is just pointwise convergence pi p in A , i.e. convergence in the weak topology.


Definition 5.2.9
Let A be a C -algebra. For any element a A consider the following function on the Gelfand
spectrum of A:
a
: M (A) C
7 (a)
The map that associates to an element of A a function on the spectrum
G : A C 0 (M (A))
a 7 a

is called the Gelfand transform.

We quote the following theorem:
Theorem 5.2.10 (Gelfand-Naimark).
Let A be a commutative C -algebra. Then the Gelfand transform G is an isometric -algebra
isomorphism.
74

Remarks 5.2.11.
(a) A commutative C -algebra is thus in a natural way isomorphic to the algebra of coninuous
functions on a topological space that is encoded in the algebra, the Gelfand spectrum. The
point of view to see C -algebras as generalizations algebras of functions is the starting point
for non-commutative geometry.
(b) The pure states on a commutative C -algebra are its characters. Hence any commutative
C -algebra is in a natural way isomorphic to the C -algebra of continuous functions on its
pure states.
(c) The Gelfand spectrum M (A) is compact if and only of the C algebra is unital.
Definition 5.2.12
(i) A -representation of a C -algebra A on a Hilbert space H is a -preserving ring homomorphism
: A B(H) .
If the algebra A is unital, the map is required to preserve the unit element. Otherwise,
one requires the subspace



(x)  for all x A, H H
to be dense in H.
(ii) A vector H in a -representation : A B(H) is called cyclic, if the subspace
{ (x) | x A }
is dense in H. If a cyclic vector exists, the representation is called cyclic.
Remarks 5.2.13.
(a) Every non-zero vector in an irreducible representation is cyclic.
(b) In contrast, an arbitrary non-zero vector in a cyclic representation is not necessarily cyclic.
Definition 5.2.14
Consider a C -algebra A and a -representation : A B(H).
(i) For every non-zero vector H \ {0} the function
(a) :=

h(a), i
h, i

aA

is a state on A. Non-zero vectors in the same one-dimensional subspace of H give the

same state. Such a state is called a ray state.
(ii) For every density matrix on H the function
(a) = Tr (a)

aA

is a state on A. Such states are called normal states.

75

We will see that all states on a C -algebra can be obtained this way:
Theorem 5.2.15 (Gelfand-Naimark-Segal).
(i) Let be a state on a unital C -algebra A. Then there is a -representation of A with
a cyclic vector H such that
(a) = h (a), i
holds for all a A. This representation is unique up to unitary equivalence.
(ii) The representation is irreducible, if and only if is a pure state.
Proof:
We only sketch the first part and restrict to the case of unital algebras. We first endow A with
a degenerate hermitian product
hx, yi := (x y )

for all x, y A.

Then
I := {a A | (a a) = 0}
turns out to be a left ideal in A. The quotient A/I has a natural non-degenerate hermitian
product which endows it with the structure of a pre-Hilbert space. We consider its completion
H := A/I .
This gives a -representation of A with cyclic vector 1 H .

We need a final piece of mathematical theory to discuss quantum mechanics:

Definition 5.2.16
Let A be a unital C -algebra. Denote by B(R) the sigma-algebra of Borel subsets of R. A
(normalized) projector-valued measure on R with values in A is a mapping
P : B(R) A
satisfying the following axioms:
1. The measure is projector valued: For every Borel subset E R, P (E) is an orthogonal
projection, i.e.
P (E) = P (E)2 and P (E) = P (E) .
2. The measure is normalized: One has
P () = 0 and P (R) = 1 .
3. Additivity: for every countable disjoint union E = t
n=1 En of Borel subsets with pairwise
empty intersection, one has

X
P (E) = lim
P (Ei ) .
n

76

i=1

Remarks 5.2.17.
1. To every projector-valued measure one can associate a projector valued resolution of the
identity. This is the projector-valued function P : R A:
P () := P ((, ))
with the properties
P ()P () = P (min{, })
and
lim P () = 1
lim P () = P ()

lim P () = 0

2. Suppose, we are given a state on A. Then (P () defines a probability measure.

In the special case when A is the algebra of bounded operators on a separable Hilbert space
H, A = B(H), for any H \ {0} the distribution function
P () = hP (), i
defines a bounded measure on R. It is a probability measure, if || = 1. More generally,
for any pair of vectors , H \ {0}, the function hP (), i defines a complex measure
on R.
In quantum mechanics, unbounded linear operators on a separable Hilbert space H are
omnipresent. Their range of definition is only a dense subspace of H. There is a notion of selfadjointness for such operators. If A is bounded, it is an element of the C -algebra B(H), and
self-adjointness means A = A. One should be aware of the fact that the C algebra B(H) does
not contain unbounded operators.
We introduce the notion of the spectrum of an operator:
Definition 5.2.18
1. Let A be a linear operator on a separable Hilbert space H. The spectrum of A is defined
as



Spec(A) = C  A 1 has no bounded inverse .
2. Let a be a self-adjoint element of an abstract C algebra A. The spectrum of a is defined
as



Spec(a) = C  a 1 not invertible in A .

Remarks 5.2.19.
1. Both definitions agree for elements of the C -algebra B(H) of bounded operators on a
separable Hilbert space H.
2. Any eigenvalue of an operator A on the Hilbert space H is an element of the spectrum
specA. The collection of eigenvalues is also called the point spectrum. If H is infinitedimensional, the point spectrum is, in general, a proper subset of the spectrum.
We are now ready to formulate von Neumanns spectral theorem for self-adjoint operators
on a separable Hilbert space.
77

Theorem 5.2.20.
For every self-adjoint operator A on a Hilbert space H, there exists a unique projector-valued
resolution P () of the identity satisfying the following properties:
(Spectral decomposition)
The range of definition of A can be characterized as
Z
D(A) = { H :
2 dhP (), i < }

and for every D(A)

Z

A =

dP ()

defined as a limit of Riemann-Stieltjes sums. The support of the spectral measure dP ()

equals the spectrum Spec(A) of A.
(Functional calculus)
For every continuous function f on R, f (A) is a linear operator on H with dense domain
Z
D(f (A)) = { H :
|f ()|2 dhP (), i < }

and for every D(f (A))

Z

f (A) =

f ()dP ()

One has f (A) = f (A) with the complex conjugate function. The operator f (A) is bounded,
iff the continuous function f is bounded on the spectrum Spec(A). On bounded functions,
the map f 7 f (A) is an isomorphism of commutative C -algebras. In a sense, an unbounded self-adjoint operator A on a Hilbert space encodes such an isomorphism.
There is also a version of the theorem for a self-adjoint element a of an abstract C -algebra
A. Again, there is a projector-valued measure P () on Spec(a) such that for any continuous
function f on Spec(a), one has
Z
f (a) =
f ()dP () .
Spec(a)

In particular,
Z
a=

dP () .
Spec(a)

For any state , one obtains

Z
(a) =

(dP ())
Spec(a)

with a probability measure (dP ()) on Spec(a).

78

5.3

Strong quantizations

Definition 5.3.1
Consider a unital C -algebra A B(H). We endow A with a 1-parameter family of Lie brackets
1
[a, b] =: {a, b}~ R
i~
parametrized by ~ R . These Lie brackets preserve the real subspace {x | x = x } A of
self-adjoint elements and endow it with the structure of a real Lie algebra.
Let (A, {, }) be a commutative unital real Poisson algebra. A full quantization of A is a

C -algebra A B(H), together with a homomorphism of real Lie algebras

Q : (A, {, }) (A, {, }~ )
for some real value of ~ such that
(i) Q preserves the unit, Q(1) = 1.
(ii) The Lie-algebra representation of A on H induced by Q(A) B(H) is irreducible.

Remarks 5.3.2.
(a) We only require compatibility of Q with the Lie algebra structure on A, not with the commutative product.
(b) It can be helpful to admit unbounded operators with dense domain on the Hilbert space H in
the image of Q. In this case, the image of A is required to be contained in the real subvector
space in the symmetric operators on H.
In the case of a full quantization of the Poisson algebra of smooth functions on a symplectic
manifold M , the images of smooth function f C (M ) whose symplectic gradient Xf is
not completely integrable as a vector field on M take values in unbounded operators.
In this case, one adds the following further technical requirement: suppose the vector fields
for two smooth functions f, g C (M ) are completely integrable and one has {f, g} = 0.
One says that the two functions f and g are in involution. Then the two operators Q(f )
and Q(g) are required to commute in the strong sense that all spectral projectors commute.
We are now ready to formulate some axioms of a quantum mechanical system:
Definition 5.3.3
1. The first datum for a quantum system consists of a C -algebra A. The self-adjoint elements
of A are called observables. A state of a quantum system is a state of A.
2. Given a quantum mechanical system A in a state , the result of a measurement of an
observable a A cannot be predicted. The possible results of the measurements at a
given time t are given by the spectrum Spec(a) R and quantum mechanics predicts the
probability measure
a, = (dPa ())
79

for the outcome. In particular, the expectation value equals

Z
hai =
(dPa ()) = (a)
Spec(a)

and the variance

(a) = h(a hai)2 i = ha2 i (hai)2 = (a2 ) (a)2 .
3. The measurement process influences the state: if the value a0 spec(a) for an observable
a has been measured, the system is in a state where the osbervable a has variance zero.
~ := {A1 , A2 , . . . , An } can be measured simulta4. One says that a finite set of observables A
neously, if they commute pairwise. In this case, there is a a projector valued measure on
Rn given by
PA~ (E1 E2 . . . En ) = PA1 (E1 ) PA2 (E2 ) . . . PAn (En )
Then there can exist vectors in a -representation and thus states which are simultaneous
eigenstates so that the predictions with variance zero are possible for all observables
simultaneously.
Non-commuting observables a1 , a2 can be measured simultaneously; otherwise Heisenbergs uncertainty relation would be an unobservable statement. Every single measurement itself still produces a sharp results, i.e. a pair of values in (spec(a1 ), spec(a2 )).
5. The dynamics of the quantum mechanical system is given by a strongly continuous oneparameter group U (t) of unitary operators acting on observables as
a(t) = U (t)1 aU (t) .
The idea is thus to implement time evolution by inner automorphisms of A. States are
time-independent. This induces a time-dependent family of probability measures for each
pair (a, ) with a an observable and a state on A.
If we realize the C -algebra A as a subalgebra of B(H), there is an infinitesimal description in terms of an unbounded self-adjoint operator H on H, called
the Hamiltonian or Hamilton-operator such that any observable a A obeys the
Heisenberg equation of motion
d
i
a(t) = [H, a] .
dt
~
Heuristically, we see the observable H as the image of the Hamilton function h under a
strong quantization, H = Q(h).

Proposition 5.3.4 (Heisenbergs uncertainty relations).

Given a quantum mechanical system A and observables a, b A and a state of A, the following
relation for the variance holds:

1
(a) (b) > ([a, b]) .
2
80

In the special case when a = q, b = p with [a, b] = i~1, i.e. when a and b obey canonical
commutation relations, we find
~
(q)(p) > .
2
Hence there is no state in which position and momentum of a free system have variance zero.
Proof:
We decompose the product ab into the sum of an anti-commutator {a, b} := ab + ba and a
commutator:
1
1
1
1
ab = (ab + ba) + (ab ba) = {a, b} + [a, b] .
2
2
2
2
If a and b are self-adjoint, the anti-commutator is self-adjoint, the commutator is anti selfadjoint. We conclude for the absolute value
{a, b} = ab + ba = a b + b a = (ba) + (ab) = {a, b}
[a, b] = [b, a] .
Hence a state yields real values on the anticommutator and purely imaginary values on the
commutator:




(ab)2 = 1 ({a, b})2 + 1 ([a, b])2 .
4
4
We find an upper bound for this expression using a GNS represention:






(ab)2 = hab, i2 = hb, ai2 6
6 kbk2 kak2 = hb2 , iha2 , i = (b2 )(a2 ),
where we used the Cauchy-Schwarz inequality. Altogether we find
2
1
(a2 )(b2 ) > ([a, b]) .
4
By considering a := A (A)1 and b := B (B)1 and taking into account that
[a, b] = [A, B] ,
we deduce

1
(A)2 (A)2 > |([A, B])|2 .
4


Even when one admits unbounded operators, strong quantizations need not exist. We will
see that they do not exist for the Poisson algebra of smooth functions on the cotangent bundle
M = T Rn with its canonical symplectic structure. It is also known that the cotangent bundle
T S 2 of the 2-sphere does not admit a strong quantization while the cotangent T T 2 of the
2-torus admits a strong quantization.
Observation 5.3.5.
81

1. Consider a symplectic manifold (M, ) with local Darboux coordinates (pi , q i ) with i =
1, . . . , n. A strong quantization
Q : (A, {, }) (A, {, }~ )
provides (essentially) self-adjoint operators on H
Qi := Q(q i ),

Pi := Q(pi ) .

They obey Heisenbergs commutation relations or canonical commutation relations :



 i j
Q , Q = 0, [Pi , Pj ] = 0
Pj , Qk = i~ jk
Taking the trace of the last relation yields


0 = tr Pj , Qk = i~ jk tridH = i~ jk dim H
This implies dim H = 0; thus there are no strong quantizations on a finite-dimensional
Hilbert space H.
2. It is convenient to consider the commutative n-parameter groups generated by Qi and Pj
respectively:
j
j
U () = ei Pj
and
V () = eij Q
with Rn . The Weyl algebra is the closed subalgebra of B(H) generated by
W (, ) := U ()V () .
For such operators Weyls rules read
0

W (, )W (0 , 0 ) = ei((,),( , )) W ( + 0 , + 0 )
with the standard symplectic form.
In particular, we obtain two unitary representations
7 U ()
7 V ()
of Rn on H.
Definition 5.3.6
1. The Schrodinger representation of the Weyl algebra is the representation defined on the
Hilbert space H = L2 (Rn ) by
[U ()f ] (x) := f (x )
[V ()f ] (x) := eix f (x) .
2. More generally, given a finite or infinite-dimensional Hilbert space H, we obtain an
analogous representation of the Weyl algebra on the infinite-dimensional Hilbert space
HH = L2 (Rn , H).
82

These are indeed all representations of the Weyl algebra.

Theorem 5.3.7 (Stone - von Neumann).
(i) For any continuous unitary representation
U : Rn B(H)
V : Rn B(H)
of the Weyl operators on a Hilbert space H, there is a Hilbert space H and a unitary
operator
U : H HH
providing an isomorphism of -representations to the Schrodinger representation on HH
(ii) The representation of the Weyl algebra on H is irreducible, iff dimC H = 1.
Let us now consider the case of global Darboux coordinates, i.e. M = T Rn with the
canonical symplectic structure. For a full quantization of the Poisson algebra for M = T Rn ,
we can assume that the Hilbert space is HH with the Weyl operators acting on L2 (Rn , H)
as
Q(q i )(q) = q i (q)
and
Q(pi )(q) = i~ qi (q) .
Lemma 5.3.8.
For any full quantization Q of T R, we have the quadratic identities
Q(q 2 ) = (Q(q))2
Q(p2 ) = (Q(p))2
1
Q(qp) = (Q(q)Q(p) + Q(p)Q(q)) .
2
Proof:
We introduce the shorthand f := Q(f ). We first note that for any f with {f, q} = 0, f
has to be of the form
f(q) = A(q)(q) .
with A(q) a hermitian operator on H.
In particular, {q 2 , q} = 0 implies qb2 = A(q). From {p, q 2 } = 2q, we deduce
2
q which evaluation in the Schrodinger representation to


1 ~
q, A(q) (q) = A0 (q)(q)
i~ i

1
i~

i
p, qb2 =

This implies A0 (q) = 2q and thus

qb2 = q2 2e
with e a hermitian endomorphism of H. Analogously, we obtain pb2 = p2 + 2e+ with a
hermitian endomorphism e+ of H.
83

 From the Poisson bracket 4pq = {q 2 , p2 }, we deduce

1 h b2 b2 i
1  2 2 1
qp
b =
q ,p =
q , p [e , e+ ] =
i~
i~
i~
1
1
= (
q p + pq) + h with h = [e+ , e ]
2
i~
Further calculations furnish the relations
1
[e+ , e ] = h,
i~

1
[h, e ] = 2e
i~

which are just the commutation relations for three generators of the non-compact real Lie
algebra sl(2, R). This Lie algebra is known not to have finite-dimensional representations
with these hermiteicity properties.

Theorem 5.3.9 (Groenewold, van Hove).
No strong quantization can be extended from the Lie algebra of polynomials that are first order
in p and q to a Lie subalgebra of C (T R) containing Polynomials in q and p which are in both
variables of degree strictly bigger than two.
Proof:
In general, one has for real polynomials f, g the identities
fd
(q) = f (
q ),

fd
(p) = f (
p)

and

1
[ = 1 (g(
\
q )
p + pg(
q ))
and P
(g)q = (g(
p)
q + qg(
p)) .
g(q)p
2
2
We only need these identities for f (x) = x3 and g(x) = x2 and restrict our proof to these
cases.
Applying this result to the identity
1
1 3 3
{q , p } = {q 2 p, p2 q}
9
3
in the Poisson algebra, we find that a strong quantization of the left hand side yields
2
1  3 3
q , p = q2 p2 2i~
q p ~2 1,
9i~
3
while for the right hand side we obtain
1
= q2 p2 2i~
q p ~2 1 .
3
The two results differ, hence we have obtained a contradiction.
84

 Since the two operators q3 and qb3 commute with q, we conclude that their difference
equals a hermitian multiplication operator A(q),
qb3 q3 = A(q) .
The identity
h

i


3 , p} = 3i~qb2 == 3i~
\
qb3 , p = i~{q
q 2 = q3 , p

implies that A(q) = A is constant,

qb3 = q3 + A .
To determine the constant, we compute


1 \
1 h b3 i
1
1
3
3
3
b
q = {q , qp} =
q , qp
b ==
q + A, (
q p + pq) = q3 ,
3
3i~
3i~
2
hence A = 0. The identity p3 = pb3 is derived analogously.
A straightforward calculation shows


1 h b3 b2 i
1 3 2
1  3 2 1 2
2 p = {q[
,p } =
q , p ==
q , p =
q p + pq2 .
qc
6
6i~
6i~
2
2 q is derived analogously.
The expression for pc


Remarks 5.3.10.
While the Lie subalgebra of all polynomials does not admit a strong quantization, its
subalgebra F(2) = span(1, p, q, p2 , q 2 , pq) admits a strong quantization.
On the other hand, the Lie subalgebra of all polynomials is larger than required for our
purposes. Already the Lie subalgebra
F(1) = span(1, p, q) ,
the so-called Heisenberg Lie algebra contains sufficiently many variables to separate all
points of the phase space T Rn .
Besides F(2) also the Lie subalgebra F(,1) = span(q i , pq i )i=0,1,... can be quantized. These
two Lie algebras are the only maximal Lie subalgebras of Fpol containing the Heisenberg
Lie algebra F(1) .
This leads to the following definition:
Definition 5.3.11
An imperfect quantization of a Poisson algebra (A, {, }) is an injective linear map Q : A A
into a unital C -algebra A B(H) of bounded operators on a separable Hilbert space H such
that
85

1. The unit of the Poisson algebra is mapped to the unit of the C -algebra.
2. The representation of the subalgebra generated by Q(A) on H is irreducible.
3. There exists a Lie subalgebra A0 of A such that the polynomials in A0 are dense in A and
that the restriction Q|A0 is an injection of Lie algebras.
Remarks 5.3.12.
1. Imperfect quantizations exist for most physical systems.
2. For polynomials in A0 of higher order, the quantization condition only holds in the weakened form
1
Q({f, g}) = [Q(f ), Q(g)] + O(~2 ) .
i~

5.4

Schr
odinger picture and examples

Observation 5.4.1.
1. So far, we have discussed the so called Heisenberg picture which we summarize as follows:
The observables form a C -algebra A. The dynamics is given by a one-parameter
group U (t) of unitary operators. With a GNS construction, we can identify A with a
subalgebra of the algebra B(H) of bounded operators on H. Then the time-evolution
can be described in terms of an (in general unbounded) linear operator H by Heisenbergs equations
1
a
d
a = [a, H] +
.
dt
i~
t
We will suppress explicit dependence on time in our discussion.
States, in contrast, do not depend on time. With the GNS construction, we find a
Hilbert space H and a non-zero vector H such that
(a) = ha(t)0 , 0 i .
2. The self-adjoint operator H gives a unitary 1-parameter group
i

U (t) = e ~ Ht .
Consider the unitarily transformed operator
i

a
(t) := e ~ Ht a(t) e ~ Ht .
It does not depend on time:


i
d
i
~i Ht
a
(t) = e
H, a(t) e ~ Ht
dt
~


i
~i Ht i
+e
H, a(t) e ~ Ht = 0
~
We consider the expectation value in the state at the time t and find
(a(t)) = ha(t)0 , 0 i = hU (t)a(t)U 1 (t)U (t)0 , U (t)0 i
= h
a(t), (t)i .
In the last step, we introduce the time-dependent vector (t) := U (t)0 H.
86

3. This way, we obtain the Schrodinger picture which we summarize as follows:

Observables are time-independent.
(Pure) states are represented as vectors in a separable Hilbert space H. They are
time-dependent. They obey the (time-dependent) Schrodinger equation:
i~

d
= H .
dt

The eigenvector equation for the Hamiltonian H

H = E
is also called time-independent Schrodinger equation. Indeed, every eigenvector E of H
H E = EE
provides a solution of the Schrodinger equation:
i

(t) = e ~ Et E ,
since

d
i
(t) = i~
E E = H E .
dt
~
The (generalized) eigenvectors of a self-adjoint operator being complete, one can describe a
general state as a superposition of eigenstates and discuss time evolution componentwise.
i~

We finally discuss two important examples: the harmonic oscillator and the hydrogen atom.
Example 5.4.2 (Harmonic oscillator).
We discuss the harmonic oscillator in the Schrodinger picture using the Schrodinger representation. The classical phase space is the cotangent space T R with Hamilton function
p2
1
hcl =
+ m 2 x2 .
2m 2
The quantization in the Schrodinger representation uses the Hilbert space H = L2 (R) and yields
operators
d2
~ d
Q(x) = x
Q(p) =
and Q(p2 ) = ~2 2 .
i dx
dx
We thus obtain the Hamilton operator


2
~2 d2
m 2 2 ~
x2
2 d
H=
x0 2 + 2
+ x =
2m dx2
2
2
dx
x0
where we have introduced the parameter
r
x0 :=
which has the dimension of length.
87

~
m

It is helpful to introduce the operator

1
a :=
2

d
x
x0
+
dx x0

together with its adjoint



1
x
d
a = x0
+
.
dx x0
2
Heisenbergs commutation relations imply the commutation relation

[a, a ] = 1 .
The Hamiltonian now reads

1
H = ~ a a +
2


;


, .
since the operator a a is positive definite, the spectrum of H is contained in Spec(H) ~
2
To investigate the spectrum, we first convince ourselves that the smallest possible value in
the spectrum is an eigenvalue: the eigenvalue equation
H 0 =

1
~ 0
2

is equivalent to the equation a0 = 0 which in the Schrodinger representation gives the following
ordinary differential equation
d0
x
x0
= 0 .
dx
x0
It has Gaussian function


x2
0 (x) = N0 exp 2
2x0
as a solution where N0 is a normalization factor.
To discuss the spectrum further, we introduce the number operator n := a a which obeys
the commutation relations
[n, a ] = a

and

[n, a] = a .

Suppose that is an eigenvector of n to the eigenvalue . Then the commutation relations

imply the relations
n(a ) = a n + a = ( + 1)a
n(a) = an a = ( 1)a,
so that the two vectors a and a, if they are non-zero, are eigenvectors of n as well and
hence eigenvectors of H. This explains the name raising operator or creation operator for a
and lowering operator or annihilation operator for a.
One now concludes that all eigenvectors of H are multiples of vectors of the form (a )n 0
with Z>0 . We compute them explicitly:

n
 


x
d
x
x2
n (x) = Nn
+
0 = Nn Hn
exp 2
x0 dx
x0
2x0
88

where Nn is a normalization factor and Hn are Hermite polynomials.

One can show that the spectrum of H is a pure point spectrum so that

 

1 
spec(H) = ~ n +
n Z>0 .
2
Example 5.4.3 (Hydrogen atom).
For a central potential V (~x) = V (r) with r := |~x| with classical Hamiltonian
hcl =

1 2
p + V (r)
2m

we introduce the generators

Li = ijk xj Pk
for angular momentum. From the classical Poisson bracket, we find the commutation relations
for the operators
[Li , Lj ] = i~ ijk Lk .
We obtain a representation of the real Lie algebra su(2) by anti-self-adjoint operators iLi .
These operators have commutation relations
[Li , pj ] = i~ ijk pk
[Li , xj ] = i~ ijk xk
with the operators xj arising as the quantization of space coordinates and the momenta pj . One
computes that L2 = L2x + L2y + L2z is the quadratic Casimir operator in U(su(2)).
Let us obtain unitary representations of the Lie algebra su(2) by considering the Hilbert space
H = L2 (S 2 , C) of square integrable function on the two-dimensional sphere S 2 . The identity


S2 = SU(2) U(1) = SO(3) SO(2)
shows that S 2 has a natural left action. It factorizes to an action of SO(3) and is explicitly
obtained by restricting rotations of R3 to the two-sphere S 2 R3 .
One can show that a Hilbert space basis is given by the countable set of functions
Yl,m

with

l {0, 1, 2, ...},

m {l, l + 1, ..., l 1, l}

which has eigenvalues

L2 Yl,m = ~2 l(l + 1)Yl,m
L3 Yl,m = ~mYl,m
The functions Yl,m are called spherical harmonics. The non-negative integer l is
called angular quantum number or orbital quantum number, the integer m is called
magnetic quantum number.
To investigate the kinetic part of the Hamiltonian, we rewrite
p2 =

L2
+ pr
r2
89

where we have introduced the differential operator



~
1
pr :=
r +
i
r
for radial momentum. To solve the Schrodinger equation
 2

p
+ V (r) (r, , ) = E (r, , )
2m
in radial coordinates (r, , ), we make the ansatz
(r, , ) = R(r)Yl,m (, ) ,
which yields for the function u(r) = rR(r) the ordinary differential equation


~2 d2
~2 l(l + 1)
+
+ V (r) u(r) = Eu(r) .

2m dr2
2mr2
It suggests to introduce the effective potential
Vef f :=

~2 l(l + 1)
+ V (r) .
2mr2
2

In the special case of the Coulomb potential V (r) = er , we find eigenvalues

E=

E1
me4 z 2
= 2
2
2
2~ n
n

n = 1, 2, 3, . . .

The integer n is called principal quantum number. The range of the orbital quantum number
l$is restricted for given value n of the principal quantum number to l {0, 1, ..., n 1}.
We thus find the following eigenvectors:
Principal
Angular
Magnetic
Energy # states
quantum nr. quantum nr.
quantum nr.
n=1
l=0
m=0
E1
1 (ground state)
n=2
l=0
m=0
E1 /4
l=1
m {1, 0} E1 /4
4
..
..
.
.
E1 /n2

n
l = 0, . . . n 1
We finally present three comments:

n2

1. The energy eigenvalue does not depend on the quantum numbers l and m, yielding a
degeneracy of the eigenstates. This is specific for the Coulomb potential and can be explained by the fact that for this potential the so-called Runge Lenz vector provides three
more quantities commuting with the Hamiltonian (but not with angular momentum). In
the real hydrogen atom, this degeneracy is lifted by relativistic effects.
2. The spectrum has also a continuous part for E > 0; the corresponding generalized eigenstates are called scattering states.
3. When one measures radiative transitions in atomic

spectra, one measures energy differences. These are of the form E = |E1 | n12 m12 , with integers n and m, in very good
agreement with experiment.
90

A glimpse to quantum field theory

We start with a very naive idea on what a quantum field is. In the simplest case, for a configuration space of the forms p1 : M R M , a classical field is a function on space time.
In quantum mechanics, numbers are replaced by operators. So instead of ordinary functions,
we study operator-valued functions on space time. It turns out that we also have to deal with
something like a -function, we thus deal with operator-valued distributions, and they should
be mathematical formulations of quantum fields. An ordinary Schwartz distribution assigns a
number to each test function, by definition. An operator-valued distribution should assign an
operator, which may be unbounded, to each test function. There is a precise mathematical
definition of this notion of an operator-valued distribution, and we can further axiomatize a
physical idea of what a quantum field should be. Such an axiomatization is known as a set of
Wightman axioms.
We work on a Minkowski space R4 . The scalar product of x = (x0 , x1 , x2 , x3 ) and y =
(y0 , y1 , y2 , y3 ) is x0 y0 x1 y1 x2 y2 x3 y3 . The linear maps R4 R4 preserving the scalar
product are the Lorentz transformations. The restricted Lorentz group is the subgroup such
that 00 > 0 and det = 1. The semi-direct product with translations is called the Poincare
group; the subgroup of the Poincare group of elements x 7 x + b such that is in the
restricted Lorentz group is called the restricted Poincare group.
The universal cover of the restricted Poincare group is naturally identified with
{(A, a)|A SL(2, C), a R4 } .
We say that two regions O1 , O2 of Minkowski space are spacelike separated, if for any x =
(x0 , x1 , x2 , x3 ) O1 and y = (y0 , y1 , y2 , y3 ) O2 , we have (x0 y0 )2 (x1 y1 )2 (x2 y2 )2
(x3 y3 )2 < 0.
We are now ready to state the Wightman axioms.
1. We have closed operators 1 (f ), 2 (f ), . . . n (f ) on a Hilbert space H for each smooth
function f on R4 with compact support.
2. There exists a dense subspace D H that is contained in the domains of all i (f ), (f )
for all f CC (R4 ). For all such f , we have (f )D D and i (f )D D. For all v, s D,
the function f 7 hv, (f )wi is a Schwartz distribution.
3. We require that H carries a unitary representation U of the universal cover of the restricted
Poincare group and of an n-dimensional representation of SL(2, C) such that
U (A, a)D
= D
P
1
1
U (A, a)i (f )U (A, a) =
j S(A )ij j (f (A, a)), with f (A, a) = f ((A) (x a)).
Here the last identity is to be understood as identities on U .
4. If the supports of two smooth functions f and g are compact and spacelike separated,
then we [i (f ), j (g)] = 0 and [i (f ), j (g)] = 0
5. There is a non-zero vector D, the vacuum vector such that
U (A, a) = for all elements of the restricted Lorentz group

91

 the spectrum of the four-parameter unitary group U (I, a), a R4 , is in the positive
cone
{(x0 , x1 , x2 , x3 )|x0 > 0, x20 x21 x22 x23 0}
The subspace generated by finitely many applications of i (f ) and j (g) on is
dense in H.
Moreover, we require the subspace of such vectors to be one-dimensional.
Distributions and unbounded operators cause various technical difficulties in a rigorous
treatment. Bounded linear operators are much more convenient for algebraic handling, and we
seek for a mathematical framework using only bounded linear operators. There has been such
a framework pursued by Araki, Haag and Kastler, and such an approach is called algebraic
quantum field theory today. The basic reference is Haags book and we now explain its basic
ideas.
Suppose we have a family of operator-valued distributions {} subject to the Wightman
axioms. For an operator valued distribution and a smooth test function f Cc (R4 ) with
support in a bounded region O the expression (f ) gives an unbounded operator operator which
we assume to be self-adjoint.
In quantum mechanics, observables are represented as (possibly unbounded) self-adjoint
operators. We regard (f ) as an observable in the spacetime region O. We have a family of
such quantum fields and many test functions with supports contained in O, so we have many
unbounded operators for each O. Applying spectral projections of these unbounded operators,
we can consider the von Neumann algebra A(O) generated by these projections. In this way, we
obtain a family of von Neumann algebras {A(O)} on the same Hilbert space H parameterized
by bounded space time regions O. We axiomatize properties of this family.
We list the axioms for space time being Minkowski space R4 . In this case, it is enough to
consider as bounded regions double cones of the form (x + V+ ) (y + V ), where x, y R4 and
we introduce the forward and backward cones
V = {z = (z0 , z1 , z2 , z3 ) R4 |z02 z12 z22 z32 >, z0 > 0} .
1. (Isotony) For a larger double cone O2 O1 , we have more test functions, and operators:
A(O1 ) A(O2 ).
2. (Locality) Suppose two double cones O1 and O2 are spacelike separated so that no interaction between them is possible, even at the speed of light. Then the observables in the
two regions should commute, leading to the requirement that the elements in A(O1 ) and
A(O2 ) commute.
3. (Poincare Covariance) The (universal cover of the) natural symmetry group of Minkowski
space, the restricted Poincare group, is required to be unitarily represented on H such
that A(gO) = Ug A(O)Ug .
4. (Vacuum) There is a distinguished unit vector H, the vacuum vector such that
Ug = for all elements g in the restricted Poincare group.
5. (Irreducibility) We require that O A(O) is dense in H.
6. (Spectrum Condition) If we restrict the representation U to the translation subgroup, its
spectrum is contained in the closure of V+ .
92

It is clear that the above set of axioms is very similar to the Wightman axioms. The
assignment of 0 7 A(O) is called a net (of von Neumann algebras). It is very hard to construct
an example satisfying the above axioms. In the 4-dimensional Minkowski space, we have only
one example known and this is known under the name of free fields.

93

Differentiable Manifolds

A.1

Definition of differentiable manifolds

It is important to describe spaces which locally look like an open subset of Rn . Examples
are abundant and most easily visualized in the case of two-dimensions: a sphere or a doughnut
locally look like R2 . Indeed, we describe the earth in an atlas as follows:
we have a collection of charts which are pieces of R2 on which the earth is represented.
Admittedly with simplifications, but this is inessential for our purposes. What matters is
that every point appears in at least one chart.
At the boundary of each chart, we have a prescription on how to glue together the charts.
In fact, the points close to the boundary of any chart have to appear in at least one more
chart.
We formalize this in the following definition:
Definition A.1.1
(i) A topological manifold M is a Hausdorff topological space with a countable basis of open
sets which is locally homeomorphic to Rn . The natural number n is called the dimension
of the manifold. We sometimes write M n or dim M = n.
In more detail, this means: for every point p M there exists an open neighborhood
p U M and a homeomorphism xU : U V to some open subset V Rn .
(ii) A chart of a topological manifold is a homeomorphism x : U V from an open subset of
U M to an open subset of V Rn .
(iii) An atlas is a family I of charts x : U Rn such that the map
tI x1
: tI U M
given by the collection x1
is a surjection.
(iv) An atlas is called differentiable, if for all pairs x, y of coordinate charts the map
y x1 : Rn Rn
is differentiable bijection everywhere where it is defined.
(v) A topological manifold, together with the choice of a differentiable atlas is called a
differentiable manifold or, synonymously a smooth manifold.
Remarks A.1.2.
1. A Hausdorff space is a topological space M in which for any two disjoint points p, q M
there exist disjoint open sets U, V in M with p U and q V . It is necessary to impose
the Hausdorff property: identify all points with y = y 0 < 0 on two copies of the real axis
R with coordinates y and y 0 respectively. Then each point is contained in a coordinate
neighbourhood homeomorphic to an open interval in R, but there are no disjoint open
neighbourhoods of the two distinct points y = 0 and y 0 = 0.
94

2. The requirement of M to have a countable basis implies that M is paracompact: for every
atlas (U ), there exists a locally finite atlas (V )I with every V contained in some U .
Locally finite means that for every point p M , there exists an open neighborhood which
intersects only a finite number of the sets V .
In this situation, one can find a partition of unity: this is a set of smooth real-valued
functions g such that
(a) 0 g 1 in M for each I
(b) the support of g is contained in V
P
(c)
I g (p) = 1 for each p
Using a partition of unity, a smooth function f C (M ) can be decomposed
X
f =f 1=
(f g )

into a locally finite sum of functions f g with support contained in the coordinate chart
V . This is essential for setting up an integration theory on manifolds.
Examples A.1.3.
1. An affine space modelled over Rn is an n-dimensional manifold.
2. Any discrete topological space is a zero-dimensional manifold.
3. The circle S 1 is a one-dimensional manifold. More generally, the n-sphere in an n + 1dimensional Euclidean vector space V
S n := {x V |hx, xi = 1}
is an n-dimensional (compact) manifold.
4. A torus is a two-dimensional manifold. The Klein bottle is a two-dimensional manifold.
5. Let L Rn be a lattice, i.e. L is the integral span of a basis of Rn . A lattice is also a
discrete subgroup of the additive group of Rn , and the quotient Rn /L is an n-dimensional
manifold, the n-dimensional torus T n .
Definition A.1.4
(i) Let M, N be differentiable manifolds of dimensions m, n respectively. Let G be open in
M . A map
f :GN
is called differentiable or smooth , if for all charts the expression
y f x1 : Rm Rn
is differentiable as soon as it is defined.

95

(ii) Differentiable maps are the morphisms in the category of differentiable or smooth manifolds. This category is also called the category of smooth manifolds. The isomorphisms
are called diffeomorphisms. (Note that at this point, we do not know how to differentiate
diffeomorphisms!)
(iii) For any open subset U M , we consider the real-valued differentiable functions
C (U, R) =: F(U ). They form an R-algebra. For any inclusion U V we have a resctriction
resVU : F (V, R) F (U, R)
which is a morphism of R-algebras. Obviously, resUU = id and for U V W we have
V
W
resW
U = resU resV .

This means that we have a presheaf of R-algebras.

(iv) We have two more properties:
if (U )I is an open covering of an open subset U M and if for f, g F(U )
we have resUU (f ) = resUU (g) for all I, then f = g. This just formalizes that a
function is uniquely determined once we know it everywhere locally.
if (U )I is an open covering of an open subset U M and if we have a collection
of functions f F(U ) such that the restrictions agree on all intersections:
U

resUU U (f ) = resU U (f )
for all , I, there is f F(U ) such that f = resUU (f ). This formalizes the idea
that we can define functions by patching together locally defined functions, if they
coincide on overlaps.
This means that we have a sheaf of R-algebras.
(v) In our case, the restriction morphisms are surjective. One says that the sheaf is flabby.
(vi) Let U M be open and let for p U be Fp0 (U ) the ideal of functions that vanish in a
neighborhood of p. The quotient algebra

Fp (U ) := F(U ) F 0 (U )
p
is called the algebra of germs of differentiable functions on U in P .
Lemma A.1.5.
Given an inclusion U V , the restriction
F(V ) F(U )
induces a canonical isomorphism
Fp (M )
Fp (G).
We call this algebra the germs of differentiable functions in p M and denote it by Fp .
One should be aware that this lemma does not hold, if one works with analytic rather than
differentiable functions. In this case, one has to consider a direct limit.
96

A.2

Tangent vectors and differentiation

Differentiation requires only a local knowledge of the function. We should thus define differentiation on germs of functions. We consider all operations on germs of functions that obey the
algebraic rules for differentiation: linearity and the Leibniz rule.
Definition A.2.1
(i) A tangent vector at M in the point p is a derivation
v : Fp M R
i.e. an R-linear map for which the Leibniz rule
v(f g) = v(f )g(p) + f (p)v(g)
holds.
(ii) The set of all derivation in a point p forms an R-vector space, the tangent space Tp M .

Remarks A.2.2.
(a) Examples for tangent vectors of Rn are provided by partial derivatives:

f
(p) .
 : f 7
i
x p
xi
(b) More generally, a coordinate chart x around p M defines for all i = 1, . . . , n


: Fp M R,
xi 
p


f 7 x i (f x1 )
x(p)

a tangent vector.
Lemma A.2.3.


1. For any coordinate system, the family x i |p i=1,...,n is a basis of Tp M .
2. This implies in particular dimR Tp M = dim M and the statement that any tangent vector
v Tp M can be written uniquely as a linear combination
v=

n
X
i=1


 .
xi p

We claim that i = v(xi ), i.e. the coefficients are obtained by evaluating the derivation
on the germ of the coordinate function xi .
Proof:

97

1. For an open ball U around 0 Rn and a function C (U, R) with (0) = 0 consider
the function
Z 1

i (u) =
(tu) dt C (U, R).
i
x
0
A brief calculation shows
i (0) =
and

Z
(u) =
0

ui 0
n

X
d
ui i (u).
(tu) =
dt
i=1

2. Given a coordinate x around p with x(p) = 0 we use the preceding remark to deduce that
for any function f the expression f x1 can be written as
f x1 (u) =

n
X

ui i (u)

i=1

and thus
f=

n
X


xi i xx1 (U ) .

i=1

We conclude
v(f ) = v(f |

x1 (U )

)=

n
X

v(x )i (0) =

i=1

n
X
i=1


v(x ) i  f
x p
i


Lemma A.2.4.
If x and y are coordinate charts around p M , we have
n

X j 

=



y i p j=1 i xj p
with a square matrix given by the Jacobian matrix for locally defined function describeing the
change of coordinates x y 1 ,
xk
ik =
.
y i
Proof:
The existence of the coefficients follows from the fact that we have two bases of tangent space
Tp M . Applying both sides to the function on U M given by the k-th coordinate
xk : U R
gives the concrete form of the coefficients.

98

Example A.2.5.
Let (A, V ) be an affine space modelled over an R-vector space V . For p A consider on Fp A
the derivation
d 
f 7  f (p + tv) for v V,
dt t=0
which is, of course, just differentiation in the direction of the affine line {p + tv, t R}. For an
affine space, this provides a natural isomorphism of the tangent space Tp A and the difference
space V :
V
Tp A .
Definition A.2.6
Let f : M N be a differentiable mapping. Then f induces for any point p M a map on
germs of functions
f : Ff (p) N Fp M
7 f
a linear map
fp : Tp M Tf (p) N
by
(fp v) () = v( f ) for v Tp M, Ff (p) N.
For this linear mapping, we also use the notation fp0 , Df |p , Tp f
Remarks A.2.7.
(a) One shows that
idp = idTp M
and that the chain rule holds:
(g f )p = gf (p) fp .
(b) If x isna chart
around
n p and
o y a chart around f (p), then fp is given with respect to the
o


bases xi p and yj f (p) by the Jacobian matrix of y f x1 in the point x(p).
Definition A.2.8
(i) A differentiable application f : M N is called an immersion, if the linear map
fp : Tp M Tf (p) N
is injective for all p M . It is called a submersion, if the maps are surjective for all p M .
(ii) An embedding is an immersion that is a homeomorphism of M to a subspace of N .
(Exercise: find an immersion that is not an embedding.)
(iii) Let N be a smooth manifold. If the inclusion M N of a subset is an embedding, then
M is called a submanifold of N .

99

(iv) Let f : M N be a differentiable map. Then p M is called a regular point of f if the

linear map fp is surjective; otherwise it is called a critical point of f .
(v) A point q N is called a regular value of f , if all p f 1 (q) are regular points of f .
Otherwise, it is called a critical value.
The proof of the following statement uses the theorem of implicit functions. We leave it as
an exercise to the reader:
Lemma A.2.9.
Let M be a differentiable manifold of dimension n and N be a differentiable manifold of dimension k. The preimage of a regular value under a differentiable map
f : Mn Nk
is a submanifold of dimension n k.

A.3

Fibre bundles

Definition A.3.1
(i) The Cartesian product M N of two differentiable manifolds M, N of dimension n and
k respectively, is endowed with the structure of a differentiable manifold by products of
coordinate charts. One has dim M N = n + k.
The projections pr1 , pr2
M N
.

&

are surjective submersions. To this end, note that, in local coordinates, the projections
are the projections Rn+k Rn and Rn+k Rk .)
(ii) A surjective submersion : P M is called a differentiable fibration or fibre bundle
with total space P , base space M and fibre N , if for each point p M there exists a
neighborhood U and a diffeomorphism
x : 1 (U )
U N
such that the diagram

1 (U ) U N

p1
y
y
U
U
commutes. The diffeomorphism x is called a local trivialization or bundle chart. The manifold 1 (q) is called the fibre over q M .

(iii) A bundle is called trivial, if there exists global bundle chart P M N .

100

(iv) Let U be an open subset of M . A local section on U of a bundle : P M is a

differentiable mapping
s:U P
such that s = idU . The sets F(U ) of sections on all open subsets U M forms a sheaf.
The elements of F(M ) are called global sections.
The following lemma shows why surjective submersions are particularly important:
Lemma A.3.2.
Consider a surjective map : M N of smooth manifold. For any open subset U N , a
local section of is a map s : U M such that s = idU , i.e. a local one-sided inverse of .
A surjection : M N of smooth manifolds has local smooth sections, iff is a surjective
submersion.
Proof:
If is a surjective submersion, the existence of local sections is a consequence of the theorem
of implicitly defined functions. If admits a smooth section, the surjectivity of follows from
differentiating s = idU to find d ds = id.

Remark A.3.3.
If we describe a mechanical system of N mass points moving in a Galilei space A in terms
of world lines parametrized by an eigentime in an interval I, we start with the fibre bundle
: I An I given by projection on the first factor.
Trajectories are then sections of . If we impose constraints e.g. to describe a solid body,
we impose that the spacial distances between the mass points are constant in time , we restrict
An which still provides a fibre bundle over I. The manifold M

ourselves to a submanifold M
is then called the (extended) configuration space of the system. To describe (extended) configuration spaces that depend on time, we have to consider general surjective submersions.
We finally need manifolds with additional structure.
Definition A.3.4
An n-dimensional smooth manifold G with a group structure such that the map

: GGG
(g, h) 7 gh1 ,
is smooth, is called a Lie group.
Remarks A.3.5.
1. Note that this implies that both the inverse g 7 g 1 and the multiplication (g1 , g2 ) 7 g1 g2
are smooth maps.
2. Examples:
The general linear group GL(n, R) is an n2 -dimensional Lie group. It is non-abelian
for n > 1.
101

 The orthogonal group SO(n), defined as the structure group of a Euclidian vector
space of dimension n is an n(n1)
-dimensional Lie group. It is non-abelian for n > 2.
2
The unitary group U(n), defined as the structure group of a unitary vector space of
dimension n is an n2 -dimensional real Lie group. It is non-abelian for n > 2.
The unitary group SU(n), defined as the structure group of a unitary vector space
of dimension n consisting of maps of determinant 1 is an n2 1-dimensional Lie
group. It is non-abelian for n > 2.
The Galilei group is a ten-dimensional non-compact Lie group.
Observation A.3.6.
Let M be a differentiable manifold of dimension n.
S
The set p Tp M of all tangent vectors can be endowed with the structure of a differentiable
manifold as follows. For a local coordinate chart defined on U M
x : U Rn
we consider the local bijection, called the associated bundle chart,
[
Tp M Rn Rn
x :
pU

defined by


x(v) = x(p), v(x1 ), ..., v(xn )

for v Tp M

One deduces from Lemma A.2.4 the following identity for a change of local coordinates
x, y:
!
n
i
X
y
bi , ... ,
x y1 (a, b) = x y 1 (a),
i
x
i=1
so that we have differentiable maps which endow
[
Tp M
T M :=
pM

with the structure of a manifold of dimension 2n.

In local coordinates x and x, the map
:

TM M

with

(v) = p for v Tp M

is the projection on the first n components and thus a submersion. This way, T M becomes a bundle over M with fibre an n-dimensional vector space. This bundle is called
the tangent bundle.
Any smooth map f : M N gives rise to a map
Tf : TM TN
(v, p) 7 fp (v) Tf (p) N
102

This map is differentiable and we have a commuting diagram

Tf

T M

TN

M N
f

One says that T f covers f . Thus T is a (covariant) functor in the category of (smooth)
manifolds.
An important aspect of bundles is the fact that they can be pulled back : given a smooth
map of manifolds f : M N and a smooth bundle : E N , one defines
f E := M N E
= {(m, e)|m M, e E such that f (m) = (e)}
The bundle projection f E M is induced from the projection on the first factor. In
other words, the fibre of f E in the point m M is the fibre of E in the point f (m) N .
We then have a map such that the following diagram commutes:
M N E


A.4

E


Vector fields and Lie algebras

Definition A.4.1
Let M be a differentiable manifold. A differentiable vector field on M is a global section X
of the tangent bundle. A local vector field on an open subset U M is a local section of the
tangent bundle.
In other words, for every point p M we have Xp Tp M such that for all smooth functions
f C (M ) the function
X(f ) : M R
p 7 Xp (f )
is smooth.
Remarks A.4.2.
Let M be an n-dimensional smooth manifold.
(a) The set vect(M ) of all vector fields is an infinite-dimensional real vector space. One can
even define a scalar multiplication of smooth real-valued functions and vector fields. Thus
vect(M ) is a module over the algebra C (M, R) of smooth functions.
(b) For a local coordinate chart x with domain U M , we define n local vector fields on U :


(p) =
 with i = 1, 2, . . . n
xi
xi p
They form a basis of the C (U )-module vect(U ). Local vector fields thus form a free C (U )module; one says that vector fields form a locally free C (M )-module.
103

(c) One has a Leibniz rule for the action of vector fields on smooth functions f, g C (M )
X(f g) = X(f )g + f X(g).
Vector fields are thus globally defined R-linear differential operators on C (M ).
Definition A.4.3
A Lie algebra over a ring R is an R-module L together with an antisymmetric R-bilinear
mapping, called the Lie bracket or commutator
[, ]:LLL
such that the Jacobi identity
[x, [y, z]] + [z, [x, y]] + [y, [z, x]] = 0
holds for all x, y, z L.
Vector fields over any smooth manifold M have the structure of an (infinite-dimensional)
real Lie algebra:
Observation A.4.4.
(a) Given two vector fields X, Y Vect(M ), we consider the following map on germs of functions:
Xp Y : Fp M R
with (Xp Y )(f ) := Xp (Y (f ))
Because of
Xp Y (f g) = Xp (Y (f )g + f Y (g))
= Xp Y (f )g + Yp (f )Xp (g) + Xp (f )Yp (g) + f Xp Y (g)
the map Xp Y is not a derivation on germs of smooth functions and thus does not give rise
to a vector field. However,
[X, Y ]p := Xp Y Yp X
is a derivation and thus provides a vector field. One checks that this way Vect(M ) becomes
a Lie algebra over R. One should be aware that one does not obtain a Lie algebra over the
ring C (M ), because we find for any germ of a function:
[X, f Y ] = X(f Y ) f Y (X)
= X(f )Y () + f XY () f Y X()
= f [X, Y ] + X(f )Y ()
hence [X, f Y ] = f [X, Y ] + X(f )Y
Similarly, we find [f X, Y ] = f [X, Y ] Y (f )X.
(b) The local basis fields

xi

commute, since the partial derivatives commute.

104

(c) In local coordinates, we write

X=
Y =

n
X
i=1
n
X
i=1

xi

xi

Note that i = X(xi ) with xi the i-th coordinate function. We compute for the Lie bracket,
using observation A.4.4 (a):
P  i j 
[X, Y ] =
i,j xi , xj
= i x i ( j ) x j + j [ i x i , x j ]
j

= i
+ j i [ x i , x j ] j x
j xi
xi xj

P  l k
k

l
=
k,l xl xl
xk

If the manifold carries the additional structure of a Lie group G, we can single out a subclass
of vector fields:
Definition A.4.5
Let G be a smooth Lie group.
1. For any element g G, we call the smooth map
Lg : G G
h 7 gh
the left translation by g. This defines a smooth left action of G on itself. Similarly for any
g G , the map h 7 h g defines a right translation.
2. A global vector field V (T G) is called left invariant, if (Lg ) V = V . Right invariant
vector fields are defined analogously.
3. One can verify that left invariant and right invariant vector fields form a Lie subalgebra
of the Lie algebra of all vector fields, called the Lie algebra Lie(G) of the Lie group G.
Left translation acts transitively and freely on the Lie group G. Hence, a left invariant
vector field is determined by its value in the neutral element e G in Te G As a vector
space, the Lie algebra Lie(G) can be identified with the tangent space Te G. In particular,
it is finite-dimensional of dimension dim G.
Remark A.4.6.
1. We extend the notion of a trajectory to an arbitrary smooth manifold M . Consider an
interval I R and a smooth map
:IM
Since I is a subset of an affine space, we can identify the derivative with a linear function
on the difference vector space R
D : R
= Tt I T(t) M
105

which we describe by
(t)
:= D(1) T(t) M ,
i.e. by a tangent vector.
Given a vector field V on M , we call a trajectory such that
d
= V ((t0 ))
dt t=t0
the integral curve of the vector field V . In local coordinates
x : U M Rn
the vector field and the trajectory read
V =

n
X
i=1

Vi

xi

and = x : R Rn

Then the condition that the trajectory is an integral curve amounts to the system of
ordinaty differential equations
d i
(t) = V i ((t))
dt
This justifies the point of view that integral curves of vector fields are a geometric expression for solutions of a system of ordinary differential equations of first order.
2. Standard theorems assert the existence of local solutions of ordinary differential equations.
As a consequence, for each point q M , there is an open neighborhood U and  > 0
such that the vector field X defines a family t : U M of diffeomorphisms for all
|t| < , obtained by taking each point p U a parameter distance t along the integral
curves of X. In fact, the t form a one-parameter local group of diffeomorphisms, since
t+s = t s = s t for |t| < , |s| <  and |s + t| < .

A.5

Differential forms and the de Rham complex

Definition A.5.1
Let k be a field of characteristic different from two and V be a k-vector space.
(i) A multilinear map f : V p = V . . . V k is called alternating, if f (v1 , ..., vp ) = 0, as
soon as vi = vj for a pair i 6= j.
(ii) The vector space of alternating p-forms on V will be denoted by p V . We have 1 V = V ,
the dual vector space, and we set 0 V := k.
Lemma A.5.2.
(i) Given p one-forms 1 , ..., p 1 V = V we consider the multilinear map
1 ... p :

V p k
(v1 , ..., vp ) 7 det( i (vj ))

This defines an alternating p-form, 1 . . . p p V.

106

(ii) If {b1 , . . . bn } is a basis of V and {b1 , . . . bn } the corresponding dual basis of V , then

n
o
i1
ip 
b ... b  1 6 i1 < i2 < ... < ip 6 n
is a basis of p V.
(iii) For a finite-dimensional

vector space V of dimension n, the dimension of the space of p
n
p
forms is dim V = p .
Definition A.5.3
On the graded vector space

V :=

n
M

p V

p=0

we define the product on the basis vectors:

(bi1 . . . bik ) (bj1 . . . bjl )
:= bi1 . . . bik bj1 . . . bjl
This product is called exterior product, wedge product or Grassmann product.
Remarks A.5.4.
(a) The wedge product is independent of the choice of basis {bi } of V . More invariantly, we
n
define the exterior algebra (V ) as the quotient of the tensor algebra T V =
p=0 (V )
of the dual space modulo a two-sided ideal
(V ) = T V /h 0 + 0 i
generated by all pairs , 0 V .
(b) The wedge product is associative.
(c) The wedge product is graded commutative: for i V and 0 j V , we have
0 = (1)ij 0
(d) Alternating forms can be pulled back along a linear map f : V W : we define a linear
map
p f : p W p V
by
(p f ()) (v1 , ..., vp ) = (f v1 , ..., f vp )
for vi V and p W . For p = 1, this is just the dual map f : W V . For
p = n, n V is one-dimensional and n f is a 1 1-matrix which we canonically identify
with a scalar. From the axiomatic definition of the determinant as a normalized alternating multilinear form on square matrices, one easily derives that the scalar n f equals the
determinant.
107

(e) Given another linear map g : W U , we find

p (g f ) = p (f ) p (g).
We also use the short hand notation f instead of p (f ). This way, we have for each p a
contravariant functor
p : vect(k) vect(k) .
We now apply this construction of linear algebra to the tangent bundle fibrewise:
Definition A.5.5
(i) Let M be a smooth n-dimensional manifold. The set
[
p T M :=
p Tx M
xM

can be endowed with the structure of a smooth manifold as in observation A.3.6. The
evident

projection to M turns this into a fibre bundle with fibre a vector space of dimension
dimM
.
p
(ii) In particular, for p = n, we get a line bundle, the determinant line bundle.
(iii) The local sections of the bundle p T M on an open subset U M are called (local)
differential forms p (U ). They form a C (U )-module. The collection p (U ) for all open
subsets U M forms again a sheaf. The elements of p (M ) are called global differential
forms.
(iv) The exterior product of differential forms is defined fiberwise and gives for every open
subset U M an infinite-dimensional graded commutative algebra
: k (U ) l (U ) k+l (U ) .
We consider the situation also in local coordinates on the differentiable manifold M :
Remarks A.5.6.
(a) Let x : M U Rnbe a coordinate chart of a smooth manifold M of dimension n. We


is a basis of the tangent space Tp M . We denote the dual basis
have seen that x i 
p i=1,...,n
n o
of the cotangent space Tp M by dxi p
.
i=1...n

xi

Recall that
is a local section in the tangent bundle T M , i.e. a local vector field. Similarly,
i
dx is a local section in the cotangent bundle T M , i.e. a local one-form.
(b) Correspondingly, local p-forms can be written as
X
(x) =
i1 ...ip (x)dxi1 . . . dxip
16i1 <...<ip 6n

with smooth functions i1 ...ik (x) on U M . This implies that a local p-form defined on the
domain of definition of the coordinate chart x is determined by its values on the local vector
fields x i . The C (U )-module p (U ) is thus locally free.
108

(c) Under change of local coordinates, we deduce from

y j
=
xi
xi y j
that
i

dx =

X xi
j

y j

dy j

As a consequence, we obtain
1

dx . . . dx = det

xi
y j

dy 1 . . . dy n

which justifies the name determinant line bundle for the highest non-vanishing exterior
power n T M of the cotangent bundle.
In several contexts, it is helpful to have antisymmetrized expressions in derivatives: for
example, the curl of a vector field on R3 is given by (rotv)i := ijk j vk (with  totally antisymmetric in the indices) and the electromagnetic field strength is given in terms of the vector
potential as F := A A . This leads us to the following
Definition A.5.7
For any open subset U of a smooth manifold M , we define a map
d : p (U ) p+1 (U )
by defining d for p (U ) on p + 1 (local) vector fields v1 , ..., vp+1 as
d (v1 , ..., vp+1 ) =
=

p+1
X

(1)i+1 vi (v1 , ..., vi , ..., vp+1 ) +

i=1

(1)i+j [vi , vj ] , v1 , ..., vi , vj , ..., vp+1

i<j

Here the hat denotes expressions to be left out.

One now checks the following claims:
Remarks A.5.8.
(a) The map d is R-linear.
(b) If p (M ), then we have for all differential forms
d( ) = d + (1)p d .
(c) For any smooth map f : M N , we have as a consequence of the chain rule:
df = f d.
109

(d) We find d2 = 0 for all M .

We are now ready to introduce the Lie derivative:
Definition A.5.9
Let M be a smooth manifold and X be a smooth vector field on M . Recall from Remark
A.4.6 the local one-parameter group t of diffeomorphisms given by X. Since these maps are
invertible, we can push for q := t (p) vector fields
t := t : Tq M Tp M
and pull back differential forms
t := t : k (M )q k (M )p
The Lie derivative with respect to the vector field X is defined for T being either a differential
form or a vector field as
1
LX T |p := lim (Tp (t T )|p ) .
t0 t

Remarks A.5.10.
From the properties of t , it follows:
1. The Lie derivative LX v of a vector field v is a vector field; the Lie derivative LX of a
k-form is a k-form.
2. On vector fields, this operation reduces to the Lie bracket of vector fields, LX Y = [X, Y ].
3. The Lie derivative LX is linear and preserves operations from tensor calculus like contractions.
4. There is a Leibniz rule for the tensor product:
LX (S T ) = LX S T + S LX T .
5. For a one-form = i dxi , one finds LX = (LX )i dxi with
(LX )i =

i j
X j
X
).
+

(
j
xj
xi

For a k-form, the last term is replaced by k terms of similar form with insertions of
derivatives of the vector field X at any position.
6. One has
d(LX ) = LX (d)
Observation A.5.11.
Let M be a smooth manifold and X be a smooth vector field on M .

110

1. We define a linear map, called the interior product

X : k+1 (M ) k (M )
by
(X )(X1 , X2 , . . . , Xk ) = (X, X1 , X2 , . . . , Xk )
for any k-tuple of local vector fields X1 , . . . , Xk .
2. X is R-linear and an antiderivation,
X ( ) = X + (1)k (X )
for k (M ).
3. Cartans magic formula relates the Lie derivative to the interior product and the exterior
differential d:
LX = X d + d(X ) .
4. For any smooth function f , one has the relation
Lf X = f LX + df X .
The equality d2 = 0 leads us to the following
Definition A.5.12
Let M be a smooth manifold of dimension n and U M open.
1. A differential form p (U ) such that d = 0 is called closed.
2. A differential form p (U ) is called exact , if there exists p1 (U ) such that
= d.
3. Because of d2 = 0 for all , exact differential forms are closed. Hence we consider the
quotient space. The real vector space for global differential forms





p
p
p+1
im d : p1 M p M
HdR (M ) := ker d : M M
is called the p-th de Rham cohomology group of M . For convenience, we have adopted
here the convention n+1 (M ) = 0, 1 (M ) = 0.
4. Given a smooth map f : M N , the pull back f : p (N ) p (M ) of differential forms
commutes with the exterior derivative and thus gives rise to a pull back
f : HpdR (N ) HpdR (M )
of de Rham cohomology groups.

Lemma A.5.13.
111

1. For any smooth manifold M , the number of connected components of M equals

dimR H0dR (M ).
2. We have HkdR (Rn ) = 0 for all k > 1.
Example A.5.14.
Consider on the punctured plane R2 \ {0} the one-form
:=

x2

x
y
dy 2
dx 1 (R2 \ {0}) .
2
+y
x + y2

A simple computation shows that is closed:





x

dx dy
dy dx = 0 .
d =
x x2 + y 2
y x2 + y 2
In radial coordinates, we find
1
(r cos (dr sin + r cos d) r sin (cos dr r sin d))
r2
= d.

If would be the exterior derivative of a function f C (R2 \ {0}), i.e. = df we would

necessarily have f = + const. on R2 \ {half axis}. Such an f cannot exist on R2 \ {0}; hence
H1dR (R2 \ {0}) 6= 0 .
We conclude in particular that the punctured plane R2 \ {0} is not diffeomorphic to the plane
R2 .
Definition A.5.15
1. Let M be a differentiable manifold of dimension n. A nowhere vanishing n-form, i.e.
(p) 6= 0 for all p M , is called a volume form on M .
2. A manifold that admits a volume form is called orientable.
3. An orientation of M is the an equivalence class of volume forms that differ by a positive
function.
Proposition A.5.16.
A smooth manifold M is orientable if and only if there exists an atlas (xi )iI such that the
determinant of the Jacobian of all diffeomorphisms xi x1
is positive.
j
Facts A.5.17.
(a) An n-form can be integrated over any oriented n-dimensional manifold M with boundary:
there is a linear map
Z
: n (M ) R
Z
7

112

If : M N is a diffeomorphism of smooth manifolds of dimension n and if n (N ),

then
Z
Z

=
.
M

(b) Let M be a smooth n-dimensional manifold with boundary and let i : M M be the
embedding of the boundary. For any (n 1)-form n1 (M ) with compact support,
Stokes theorem asserts the following equality:
Z
Z

i=
d
M

(c) A smooth manifold M is said to be contractible to a point p0 M , if there exists a smooth

map
H : M [0, 1] M ,
called a homotopy, such that
H(p, 0) = p and H(p, 1) = p0

for all

pM .

(d) Poincare lemma asserts that any closed p-form on a contractible manifold M is exact.
For proofs and details, we refer to the book by Boot and Tu and by Madsen and Tornehave.

A.6

Riemannian manifolds and the Hodge dual

We start again with linear algebra: for any finite dimensional k-vector space V , the dual vector
space V has the same dimension as V . The two vector spaces are thus isomorphic, but there
is no canonical isomorphism. However, a non-degenerate symmetric bilinear form
g :V V k
provides an isomorphism
gV : V V
v 7 (g(v, ) : w 7 g(v, w))
To get a geometric notion derived from this notion of linear algebra, we endow the tangent
spaces in a smooth way with non-degenerate symmetric bilinear forms:
Definition A.6.1
1. A Riemannian manifold (M, g) is a smooth manifold M such that for every point p M
the tangent space Tp M has the structure of a Euclidean vector space, i.e. there is a positive
symmetric definite bilinear form
gp : Tp M Tp M R ,
called the metric of M , such that for all local coordinate charts x of M the locally defined
functions
gij : U R


 
gij (p) := gp
,

xi p xj p
are smooth for all 1 i, j, dim M .
113

2. More generally, we also consider non-degenerate symmetric bilinear forms

gp : Tp M Tp M R
of any signature s and obtain the notion of a pseudo-Riemannian manifold. For the case
of signature (1, n 1), the manifold (M, g) is called a Lorentzian manifold.
3. For any open subset U M of a (pseudo-)Riemannian manifold, a canonical isomorphism of bundles between the tangent bundle T M and the cotangent bundle is obtained
fibrewise:
gTp M : Tp M Tp M .
This gives a bijection between local one-forms (which are just local sections of the cotangent bundle) and local vector fields (which are just local sections of the tangent bundle).

gU : vect(U ) 1 (U ) .
of one-forms and vector fields induced fibrewise by the isomorphism V above.
An important aspect of Riemannian manifolds is the fact that we can define lengths of
curves on them (and also angles between intersecting curves). We can also define volumes:
Lemma A.6.2.
Let (M, g) be an oriented (pseudo-)Riemannian manifold of dimension n. If (dx1 , dx2 , . . . dxn )
is an oriented local basis of the cotangent bundle, then the n-forms
q
g (U ) := | det gij |dx1 dx2 . . . dxn n (U )
patch together to a globally defined n-form g n (M ) which is nowhere vanishing. It is called
the normalized volume form for the metric g.
If V is a finite-dimensional vector space of dimension n, the two vector spaces p V and
np V have the same dimension and are isomorphic. Again, there is no canonical isomorphism.
Lemma A.6.3.
1. Let V be an orientable finite-dimensional R-vector space of dimension n. Assume further
that there is a non-degenerate symmetric bilinear form g on V . Fix an orthonormal basis
(b1 , b2 , . . . , bn ) on V . We extend the bilinear form g to a non-degenerate symmetric bilinear
form g on p V for all p by setting for , 0 p V
X
h, 0 i :=
(bi1 , . . . , bip ) 0 (bi1 , . . . , bip )g(bi1 , bi1 ) . . . g(bip , bip )
i1 ,i2 ,...ip

2. If the vector space is even oriented, there exists a unique isomorphism of vector spaces
g : p V np V
such that for all , p V the equality
= g(, )g
holds. The isomorphism is called the Hodge operator.
114

Remarks A.6.4.
1. For V an oriented Euclidean vector space with oriented orthonormal basis (e1 , . . . en ), the
Hodge operator acts as
(e1 e2 . . . ep ) = ep+1 ek+1 . . . en .
2. In particular, for R3 with the standard orientation and the standard Euclidean structure,
one finds
dx = dy dz dy = dz dx dz = dx dy
Indeed, the three equations on dx
dx (dx) = g(dx, dx)dx dy dz = dx dy dz
dy (dx) = g(dy, dx)dx dy dz = 0
dz (dx) = g(dz, dx)dx dy dz = 0
have the unique solution dy dz.
This allows us to formulate the cross product of vectors in R3 with the standard structure
as an oriented Euclidean vector space:
R3 R3 R3
(v, v 0 ) 7 1 ((v) (v 0 ))
3. Another important example is R4 with coordinates (t, x, y, z) and a metric of signature
(+, , , ). We find, e.g. for one-forms
dt = dx dy dz

dx = dt dy dz

dy = dt dz dx

dz = dt dx dy .

and for 2-forms

dt dx = dy dz dt dy = dz dx dt dz = dx dy
dx dy = dt dz
dy dz = dt dx
dz dx = dt dy
4. If g is a non-degenerate bilinear symmetric form on a vector space V , then we have for
the component functions
()i1 i2 ...inp =

1 j1 ...jp p

| det g|j1 ...jp i1 ...inp

p!

where the indices of are raised and lowered with g and its inverse and where  is totally
antisymmetric in the indices with normalization 1,2,...n = 1.
Proposition A.6.5.
The Hodge operator has the following properties:
(i) 1 = g , g = (1)s
V
V
(ii) For p V, np V , we have
g(, ) = (1)p(np) g(, )
115

(iii) On

Vp

V , we find for the square of the Hodge operator

()2 = (1)p(np)+s idVp V

Observation A.6.6.
Let (M, g) be an oriented (pseudo-)Riemannian manifold.
1. In this case, we use the canonical volume form g on the manifold and extend the Hodge
operator fibrewise to a smooth map
p (U ) np (U )
for any open subset U M .
2. In this way, we obtain an L2 -norm on a suitable subspace of p (M ):
Z
0
h, i :=
0 .
M

3. In particular, we can define the adjoint

: p (U ) p1 (U )
of the exterior derivative d, called the codifferential by the equation
h, 0 i = hd, 0 i .
On p-forms, it reads explicitly
= (1)p 1 d .
Indeed, for p1 (M ) and 0 p (M ) we deduce from
d( 0 ) = d 0 + (1)p1 d 0
that
R
R
p
(1)p h, 1 d 0 i = (1)
(1Rd 0 i = (1)p M d 0 i
M
R
= M d( 0 ) + M d 0
= hd, 0 i
it satisfies 2 = 0.
4. The Laplace operator on differential forms is given by
:= ( + d)2 = d + d .
It is symmetric
h, 0 i = h, 0 i
and positive definite
h, i 0 .
The elements in its kernel are called harmonic forms. The vector space of harmonic forms
can be shown to be naturally isomorphic to de Rham cohomology. The Hodge operator
induces an isomorphism of harmonic forms
p
np
: H
(M ) H
(M )

which implements Poincare duality of de Rham cohomology.

116

We finally comment on the relation of the operations we just introduced to the classical
operations of gradient, curl and divergence in three-dimensional vector calculus. To this end,
we consider the special case of M being R3 with the standard scalar product and the standard
orientation.
Observation A.6.7.
1. For a function f 0 (U ) = C (U, R), the derivative d is a one-form. I. The vector
field corresponding to df is called the gradient of the function f C (U ):
grad(f ) = 1 (df ) vect(U ) .
2. For a one-form = Adx + Bdy + Cdz 1 (R3 ), we obtain a 2-form
d = (

C B
A C
B A

)dy dz + (

)dz dx + (

)dx dy
y
z
z
x
x
y

whose coefficient functions we recognize as the components of the curl.

We therefore define the curl operator as
curl : vect(U ) vect(U )
by
vect(U )

1 (U )

2 (U )

1 (U )

vect(U )

3. Finally, for a 1-form, we compute for = Adx + Bdy + Cdz 1 (R3 )

d = d (Ady dz + Bdz dx + Cdx dy)


B
C
dx dy dz = A
= A
+
+
+
x
y
z
x

B
y

C
z

We therefore define the standard divergence operator on vector fields as

div(v) : vect(U ) C (U, R)
by
vect(U )

1 (U )

2 (U )

3 (U )

C (U, R)

4. The relation d2 = 0 now implies the two classical identities

curl gradf = 0

and

117

div curl v = 0 .

Index
-representation, 75
acceleration, 6
action of a group, 1
action of a Lagragian system, 33
affine map, 1
affine space, 1
Aharonov-Bohm effect, 55
alternating form, 106
angular quantum number, 89
annihilation operator, 88
atlas, 94
augmentation, 69
augmented variational bicomplex, 35
boost, 50
bundle chart, 100
canonical commutation relations, 82
canonical phase space, 56
canonical transformations, 55
Cartan connection, 30
Cartans magic formula, 111
causality, 51
character, 73
chart, 94
choice of gauge, 45
classical limit, 70
closed differential form, 111
codifferential, 116
commutator, 104
configuration space, 22, 33, 101
connector, 29
contact form, 30
contact ideal, 31
contractible manifold, 113
Coulomb potential, 90
Coulombs law, 45
covariant derivative, 30
covariant derivative, 39
creation operator, 88
critical point, 100
critical value of a function, 100
curl operator, 117
curvature of a connection, 29

cyclic coordinate, 17
cyclic representation, 75
cyclic vector, 75
dAlembert operator, 47
Darboux coordinates, 56
Darboux theorem, 56
de Rham cohomology, 111
deformation quantization, 70
density matrix, 73
dependent coordinate, 24
determinant line bundle, 108
diffeomorphisms, 96
difference space, 1
differentiable fibration, 100
differentiable manifold, 94
differentiable vector field, 103
differential forms, 108
dispersion relation, 47
divergence operator, 117
dual field strength, 44
Ehresmann connection, 28
eigen time, 51
eigentime, 6
electric field, 43
electromagnetic field strength, 43
embedding of manifolds, 99
Euclidean group, 3
Euclidean space, 3
Euler-Lagrange complex, 35
Euler-Lagrange form, 33
Euler-Lagrange operator, 13
evolution space, 60, 61
evolutionary vector field, 35
exact differential form, 111
extended phase space, 60
exterior product, 107
extremal point, 72
fibre, 100
fibre bundle, 100
field configurations, 22
flabby sheaf, 96
flat connection, 29
118

foliation, 59
formal deformation, 69
formal power series, 69
free action of a group, 1
full quantization, 79
future of a point, 50
Galilean coordinate system, 5
Galilean structure, 5
Galilei group, 4
Galilei space, 3
Gau law, 44
gauge conditions, 45
gauge freedom, 45
gauge potential, 45
gauge symmetry, 38
Gelfand spectrum, 74
Gelfand transform, 74
Gelfand-Naimark theorem, 74
generalized coordinate, 17
generalized momentum, 17
generalized velocity, 17
germ of a function, 96
germ of a local section, 23
global section, 101
gradient, 117
gtime like vector, 50
Hamilton function, 60
Hamiltons equations, 58
Hamiltonian, 80
Hamiltonian function, 61
Hamiltonian system, 61
harmonic forms, 116
harmonic oscillator, 87
Heisenberg equation of motion, 80
Heisenberg Lie algebra, 85
Heisenberg picture, 86
Heisenbergs commutation relations, 82
Heisenbergs uncertainty relations, 80
Helmholtz operator, 35
Hilbert space, 71
Hodge operator, 114
homogeneous Maxwell equations, 44
homotopy, 113
horizontal bundle, 28
horizontal lift, 29
Hydrogen atom, 89

immersion, 99
independent coordinate, 24
inertial frame, 5
inertial mass, 9
inertial system, 5
inhomogeneous Maxwell equations, 44
inner Euler operators, 34
integral curve, 106
interior product, 111
invariant vector field, 105
Jacobi identity, 104
jet, 23
jet bundle, 24
jet prolongation, 25
Keplers problem, 16
kinetic energy, 33
Lagrange function, 33
Lagrangian, 14
Lagrangian action, 17
Lagrangian density, 33
Lagrangian multiplier, 18
Lagrangian system, 33
Laplace operator, 116
left invariant vector field, 105
left translation, 105
Legendre transform , 62
length contraction, 52
Lie algebra, 104, 105
Lie bracket, 104
Lie derivative, 110
Lie group, 101
lift of a curve, 29
light cone, 49
light like vector, 50
linear connection, 29
local differential form, 30
local section, 101
local trivialization, 100
Lorentz force, 43
Lorentz gauge, 45
Lorentz system, 48
Lorentzian manifold, 114
magnetic field, 43
magnetic quantum number, 89
119

metric on a Riemannian manifold, 113

Minkowski space, 48
natural bundle, 40
natural Hamiltonian system, 62
natural system, 14, 34
Newtons law of gravity, 8
Newtonian equation, 7
Newtonian trajectory, 7
Noether identities, 39
observable, 79
observer, 51
operator norm, 73
orbital quantum number, 89
orientable manifold, 112
parity transformation, 49
partial differential equation, 27
partition of unity, 95
past, 50
past of a point, 50
phase curves, 9, 10
phase flow, 10
phase space, 10, 61
physical motion, 7
Planck constant, 70
Poincare lemma, 113
Poincare transformation, 49
Poincares recurrence theorem, 68
Poincare-Cartan integral invariant, 67
point spectrum, 77
Poisson algebra, 57
Poisson equation, 46
Poisson manifold, 57
potential energy, 8, 34
presheaf, 96
presymplectic form, 55
principal homogenous space, 1
principal quantum number, 90
projector-valued measure, 76
prolongation, 27
pseudo-Riemannian manifold, 114
pull back of bundles, 103
pure states, 72
rapidity, 50
ray, 2

ray state, 75
regular point of a function, 100
regular value of a function, 100
relative uniform motion, 5
Riemannian manifold, 113
rigid body, 37
scalar potential, 46
scattering states, 90
Schrodinger equation, 87
Schrodinger picture, 87
Schrodinger representation, 82
self-adjoint element, 71
semi-direct product of groups, 2
sheaf, 96
shell, 27
simultaneous events, 4
slice of a foliation, 59
smooth manifold, 94
source forms, 34
source of a jet, 24
space like vector, 50
space of motions, 61
spectrum, 77
spherical harmonics, 89
star product, 69
state, 72
Stokes theorem, 113
Stone - von Neumann theorem, 83
submanifold, 99
submersion, 99
symplectic form, 55
symplectic gradient, 57
symplectic manifold, 55
tangent bundle, 102
tangent space, 97
tangent vector, 97
target of a jet, 24
Time dilatation, 52
time reversal, 49
topological manifold, 94
trajectory, 6, 61
transitive action of a group, 1
trivial bundle, 100
twin paradoxon, 53
vacuum vector, 91
120

variational bicomplex, 33
variational symmetry, 35
vector potential, 46
velocity, 6, 50
vertical bundle, 28
vertical lift, 29
volume form, 112
weak topology, 73
wedge product, 107
Weyl algebra, 82
Wightman axioms, 91
world line, 6
Youngs inequality, 63

121