Algebraic Quantum Field Theory

Wojciech Dybalski
Literature:
1. R. Haag: Local Quantum Physics, Springer 1992/1996
2. H. Araki: Mathematical Theory of Quantum Fields, Oxford University Press
2000.
3. D. Buchholz: Introduction to Algebraic QFT, lectures, University of Goettingen, winter semester 2007. (Main source for sections 1,2,5 below).
Programme of the lectures:
1. Algebraic structure of quantum theory
(a) quantum mechanics: Heisenberg, Weyl and resolvent algebra.
(b) infinite quantum systems.
2. Operator algebras and local (relativistic) quantum physics
(a) abstract algebras, representations
(b) locality, covariance
(c) vacuum
3. Construction of models
(a) free theories, conformal field theories
(b) wedge-local theories and Rieffel deformations
4. Scattering theory
(a) Scattering matrix
(b) Asymptotic completeness
(c) Infrared problems
5. Superselection structure and statistics
(a) DHR analysis (charges, statistics etc.)
(b) charged fields, gauge groups
(c) Infrared problems
1

Algebraic structure of quantum theory

1.1

Quantum systems with a finite number of degrees of

freedom

Observables describe properties of measuring devices (possible measured values, commensurability properties).
States describe properties of prepared ensembles (probability distributions
of measured values, correlations between observables)
Mathematical description based on Hilbert space formalism, Hilbert space H.
Observables: self-adjoint operators A on H.
States: density matrices on H (i.e. 0, Tr = 1).
Expectation values A, 7 TrA.
Remark 1.1 pure states (optimal information)= rays ei H, kk = 1 =
orthogonal projections 2 = . ( Question: Why equivalent? Express in a basis,
there can be just one eigenvalue with multiplicity one).
Usual framework : fixed by specifying H. E.g. for spin H = C2 , for particle L2 (R3 ). Question: What is the Hilbert space for a particle with spin?
L2 (R3 ; C2 ).
Question: Does every s.a. operator A correspond to some measurement?
Does every density matrix correspond to some ensamble which can be
prepared? In general no. Superselection rules. For example, you cannot
superpose two states with different charges.
New point of view: Observables are primary objects (we specify the family
of measuring devices). The rest of the theory follows.
1.1.1

Heisenberg algebra

Quantum Mechanics. Observables:

Qj , j = 1, . . . , n and Pk , k = 1, . . . , n.
(n = N d, N -number of particles, d-dimension of space).
We demand that observables form (generate) an algebra.
Definition 1.2 The free (polynomial) -algebra P is a complex vector space
whose basis vectors are monomials (words) in Qj , Pk (denoted Qj1 . . . Pk1 . . . Qjn . . . Pkn ).
1. Sums: Elements of P have the form
X
cj1 ...kn Qj1 . . . Pkn .
2

(1)

2. Products: The product operation is defined on monomials by

(Qj1 . . . Pk1 . . . Qjn . . . Pkn ) (Qj10 . . . Pk10 . . . Qjn0 . . . Pkn0 )
= Qj1 . . . Pk1 . . . Qjn . . . Pkn Qj10 . . . Pk10 . . . Qjn0 . . . Pkn0
3. Adjoints: Qj = Qj , Pk = Pk ,
X
 X
cj1 ...kn Qj1 . . . Pkn =
cj1 ...kn Pkn . . . Qj1 .

(2)

4. Unit: 1.
The operations (+, , ) are subject to standard rules (associativity, distributivity,
antilinearity etc.) but not commutativity.
Quantum Mechanics requires the following relations:


[Qj , Pk ] ij,k 1 = 0.

[Qj , Qk ] = [Pj , Pk ] = 0,

Consider a two-sided ideal J generated by all linear combinations of

A[Qj , Qk ]B, A[Pj , Pk ]B, A [Qj , Pk ] ij,k 1 B

(3)

(4)

for all A, B P.
Definition 1.3 Quotient P\J is again a -algebra, since J is a two-sided ideal
and J = J . We will call it Heisenberg algebra. This is the free algebra modulo
relations (3).
1.1.2

Weyl algebra

The elements of polynomial algebra are intrinsically unbounded (values of position

and momentum can be arbitrarily large). This causes technical problems. A way
n
out is to consider their
P bounded functions. For z = u + iv C we would like
to set W (z) exp(i k (uk Pk + vk Qk )). We cannot do it directly, because exp is
undefined for symbols Pk , Qk . But we can consider abstract symbols W (z) satisfying the expected relations keeping in mind the formal Baker-Campbell-Hausdorff
(BCH) relation. The BCH formula gives
1

eA eB = eA+B+ 2 [A,B]

(5)

We have z = u + iv, z 0 = u0 + iv 0 , W (z) = eA , W (z 0 ) = eB , A = i(uP + vQ),

B = i(u0 P + v 0 Q) and [Q, P ] = i. Thus we have
[A, B] = (1)[uP + vQ, u0 P + v 0 Q] = (1)(ivu0 iuv 0 ) = i(uv 0 vu0 ).

(6)

On the other hand:

Imhz, z 0 i = Imhu + iv, u0 + iv 0 i = Im(ivu0 + iuv 0 ) = uv 0 vu0 .

(7)

Hence
i

W (z)W (z 0 ) = e 2 Imhz,z i W (z + z 0 ).
3

(8)

Definition 1.4 The (pre-)Weyl algebra W is the free polynomial -algebra generated by the symbols W (z), z Cn modulo the relations
i

W (z)W (z 0 ) e 2 Imhz|z i W (z + z 0 ) = 0, W (z) W (z) = 0,

P
where hz|z 0 i = k zk zk0 is the canonical scalar product in Cn .

(9)

The Weyl algebra has the following properties:

1. We have W (0) = 1 (by the uniqueness of unity).
2. By the above W (z)W (z) = W (z) W (z) = 1 i.e. Weyl operators are unitary.
3. We have
X
z

X

X
i
0
az W (z)
bz0 W (z 0 ) =
az bz0 e 2 Imhz,z i W (z + z 0 ).
z0

(10)

z,z 0

Thus elements of W are linear combinations of Weyl operators W (z).

1.1.3

Representations of the Weyl algebra

Definition 1.5 A -representation : W 7 B(H) is a homomorphism i.e. a

map which preserves the algebraic structure. That is for W, W1 , W2 W:
1. linearity (c1 W1 + c2 W2 ) = c1 (W1 ) + c2 (W2 ),
2. multiplicativity (W1 W2 ) = (W1 )(W2 ),
3. symmetry (W ) = (W ) .
If in addition (1) = I, we say that the representation is unital. (In these lectures
we consider unital representations unless specified otherwise).
R
Example 1.6 Let H1 = L2 (Rn ) with scalar products hf, gi = dn x f (x)g(x). One
defines


i
1 (W (z))f (x) = e 2 uv eivx f (x + u), z = u + iv.
(11)
(Note that for u = 0 1 (W (z) is a multiplication operator and for v = 0 it is a
shift). This is Schrodinger representation in configuration space.
Remark 1.7 Heuristics: Recall that W (z) = e(i
Hausdorff
i

(ei(uP +vQ) f )(x) = e 2 uv (eivQ ei

= e

i
uv
2

ivx

(e

iuP

uP

k (uk Pk +vk Qk ))

f )(x)

(12)
i
uv
2

f )(x) = e

For the last step note (eiuP f )(x) = (eiu i x f )(x) = (

4

and Baker-Campbell-

ivx

(f )(x + u)

un n
n n! x f )(x)

(13)

= f (x + u).

R
Example 1.8 Let H2 = L2 (Rn ) with scalar products hf, gi = dn x f (x)g(x). One
defines


i
2 (W (z))f (x) = e 2 uv eiux f (x v), z = u + iv.
(14)
This is Schrodinger representation in momentum space.
Relation between (1 , H1 ), (2 , H2 ) is provided by the Fourier transform
Z
n/2
(Ff )(y) := (2)
dn x eixy f (x),
Z
1
n/2
(F f )(y) := (2)
dn x eixy f (x).

(15)
(16)

F is isometric, i.e. hFf, Ff i = hf, f i, (Plancherel theorem) and invertible (Fourier

theorem). Hence it is unitary. We have
2 (W ) = F1 (W )F 1 ,

W W.

(17)

Definition 1.9 Let (a , Ha ), (b , Hb ) be two representations. If there exists an

invertible isometry U : Ha Hb (a unitary) s.t.
b ( ) = U a ( )U 1

(18)

the two representations are said to be (unitarily) equivalent (denoted (a , Ha ) '

(b , Hb )). As we will see, equivalent representations describe the same set of states.
Is any representation of W unitarily equivalent to the Schrodinger representation
1 ? Certainly not, because we can form direct sums e.g. = 1 1 is not
unitarily equivalent to 1 . We have to restrict attention to representations which
cannot be decomposed into smaller ones.
Definition 1.10 Irreducibility of representations: We say that a closed subspace
K H is invariant (under the action of (W)) if (W)K K. We say that a
representation of (, H) of W is irreducible, if the only closed invariant subspaces
are H and {0}.
Remark 1.11 The Schroedinger representation 1 is irreducible (Homework).
Lemma 1.12 Irreducibility of (, H) is equivalent to any of the two conditions
below:
1. For any non-zero H
{(W ) | W W } = H
(i.e. if every non-zero vector is cyclic).
5

(19)

2. Given A B(H),
[A, (W )] = 0

for all

W W

(20)

implies that A CI (Schur lemma)

(i.e. the commutant of (W) is trivial).
Remark 1.13 Recall that the commutant of (W) is defined as
(W)0 = { A B(H) |[A, (W )] = 0 for all W W}.

(21)

Proof. For complete proof see e.g. Proposition 2.3.8 in [1]. We will show here
only that 1. 2.: By contradiction, we assume that there is A
/ CI in (W)0 .

If A (W)0 then also A (W)0 hence also s.a. operators A+A

and AA
are
2
2i
in (W)0 . Thus, we can in fact assume that there is a s.a. operator B (W)0 ,
B
/ C1. Then also bounded Borel functions of B are in (W)0 . In particular
characteristic functions (B), R (spectral projections of B) are in (W)0 .
Since B
/ C1, we can find 0 6= (B) 6= I. Let Ran (B) i.e. = (B).
Then for any W W
(W ) = (W ) (B) = (B)(W ),

(22)

hence cannot be cyclic because (B) projects on a subspace which is strictly

smaller than H. 
Question: Are any two irreducible representations of the Weyl algebra unitarily
equivalent?
Answer: In general, no. After excluding pathologies yes.
Example 1.14 Let H3 be a non-separable Hilbert space with a basis ep , p Rn .
Elements of H3 :
X
X
f=
cp ep , with
|cp |2 <
(23)
p

(i.e. all cp = 0 apart from some countable set). hf |f 0 i =

i

3 (W (z))ep = e 2 uv eiup ep+v .

0
p cp cp .

We define
(24)

This representation is irreducible but not unitarily equivalent to (1 , H1 ) ' (2 , H2 )

because H1,2 and H3 have different dimension.
Criterion: Representation (, H) of W is of physical interest if for any f H
the expectation values
z 7 hf, (W (z))f i
6

(25)

depend continuously on z.
Physical meaning of the Criterion: Set v = 0. Then u 7 (W (u)) is an
n-parameter unitary representation of translations on H. Hence, by the Criterion
and Stones theorem
(W (u)) = ei(u1 P,1 ++un P,n ) ,

(26)

where P,i is a family of commuting s.a operators on (a domain in) H. They can
be interpreted as momentum operators in this representation. Analogously, we
obtain the position operators Q,i . By taking derivatives of the Weyl relations
w.r.t, ul , vk one obtains [Q,j , P,k ] = ij,k 1 on a certain domain (on which the
derivatives exist).
Theorem 1.15 (Stone-von Neumann uniqueness theorem) Any irreducible representation of W, satisfying the Criterion, is unitarily equivalent to the Schrodinger
representation.
For a proof see Theorem 4.34 and Theorem 8.15 in [2].
Remark 1.16 This theorem does not generalize to systems with infinitely many
degrees of freedom (n = ). In particular, it does not hold in QFT. This is one
reason why charges, internal (gauge) symmetries, and groups play much more
prominent role in QFT than in QM. As we will see in Section 5, they will be
needed to keep track of all these inequivalent representations.
1.1.4

States

Definition 1.17 A state of a physical system is described by

1. specifying a representation (, H) of W,
2. specifying a density matrix on H.
Then (W ) = Tr(W ).
Lemma 1.18 A state is a map : W 7 C which satisfies
1. linearity (c1 W1 + c2 W2 ) = c1 (W1 ) + c2 (W2 ).
2. normalization (1) = 1.
3. positivity (W W ) 0 for all W W.
Proof. The only non-trivial fact is positivity: Write =
ki k = 1. Then, if the sum is finite, we can write
X
(W W ) =
pi Tr(|i ihi |(W W ))

pi |i ihi |, pi 0,

pi hi |(W W )i i =

X
i

pi kW i k2 ,

(27)

by completing i to orthonormal bases.

In the general case we can use cyclicity of the trace
Tr (W W ) = Tr (W ) (W ) = Tr (W )(W )
XX
=
pj |hei , (W )j i|2 .
i

(28)
(29)

The result is finite (because (W W ) is trace-class) and manifestly positive. 

Definition 1.19 We say that a representation (, H) is cyclic, if H contains a
cyclic vector . (Cf. Lemma 1.12). Such representations will be denoted (, H, ).
For example, any irreducible representation is cyclic.
Theorem 1.20 Any linear functional : W C, which is positive and normalized, is a state in the sense of Definition 1.17 above. More precisely, it induces a
unique (up to unitary equivalence) cyclic representation (, H, ) s.t.
(W ) = h, (W )i,

W W.

(30)

Proof. GNS construction (we will come to that). 

Lemma 1.21 If (1 , H1 ) ' (2 , H2 ) then the corresponding sets of states coincide.
Proof. Let 1 be a density matrix in representation 1 and W W. Then
Tr 1 1 (W ) = Tr 1 U 2 (W )U 1 = Tr U 1 1 U 2 (W ).

(31)

Hence it does not matter if we measure W in representation 1 on 1 or in 2 on

2 = U 1 1 U .
1.1.5

Weyl C -algebra

Definition 1.22 We define a seminorm on W:

kW k := sup k(W )k,

W W,

(32)

where the supremum extends over all cyclic representations. The completion of

W/ ker k k is the Weyl C -algebra which we denote W.

A few remarks about this definition:
1. The supremum is finite because for any representation we have
k(W (z))k2 = k(W (z)) (W (z))k = k(1)k = 1
and thus k(W )k for any W W is finite.
8

(33)

2. We cannot take supremum over all representations because this is not a set.
In fact, take the direct sum of all the representations which do not have
themselves as a direct summand and call this representation . Then we get
the Russels paradox:
M
:=
{ |
/ } then
/ ,
(34)
where 1 2 means here that 1 is contained in 2 as a direct summand.
3. Using the GNS theorem one can show that
kW k = sup (W W )1/2 .

(35)

Here the supremum extends over the set of states. Indeed:

sup (W W )1/2 = sup h, (W W )i sup k(W )k.

(36)

(,)

On the other hand

sup k(W )k = sup sup k(W )k = sup sup h, (W W )i1/2

kk=1

kk=1

sup (W W )1/2 .

(37)

4. In the case of the Weyl algebra ker k k = 0 so the seminorm (32) is actually
a norm. [5]
Apart from standard properties of the norm, it satisfies
kW1 W2 k kW1 k kW2 k Banach algebra property
kW W k = kW k2 C -property

(38)
(39)

This is adventageous from the point of view of functional calculus: For W W

we have f (W ) W for polynomials f , but for more complicated functions there
we have f (W ) W
for any continuous function f .
is no guarantee. For W W
Nevertheless, in the next few subsections we will still work with the pre-Weyl
algebra W.
1.1.6

Symmetries

Postulate: Symmetry transformations are described by automorphisms (invertible homomorphisms) of W.

Definition 1.23 We say that a map : W W is an automorphism if it is a
bijection and satisfies
1. (c1 W1 + c2 W2 ) = c1 (W1 ) + c2 (W2 )
9

2. (W1 W2 ) = (W1 )(W2 )

3. (W ) = (W )
4. (1) = 1.
Automorphisms of W form a group which we denote Aut W.
Example 1.24 If U W is a unitary, then U (W ) = U W U 1 is called an
inner automorphism. Inner automorphisms form a group In W. For example, for
U = W (u0 ) we have
u0 (W (z)) = W (u0 )W (z)W (u0 )1 = eihu0 ,vi W (z)

(40)

This is translation of coordinates, as one can see in the Schroedinger representation

1 :
1 (u0 (W (z))) = eihu0 ,vi ei(uP +vQ) = ei(uP +v(Q+u0 )) .

(41)

Similarly, for v0 Rn
iv0 (W (z)) = W (iv0 )W (z)W (iv0 )1 = eihv0 ,ui W (z)

(42)

is a translation in momentum space.

Example 1.25 Let R SO(n). Then
R (W (z)) = W (Rz)

(43)

is an automorphism which is not inner. (Set n = 3 and let R be a rotation around

the z axis by angle . Then, in the Schrodinger representation
1 (R (W (z))) = U 1 (W (z))U 1

(44)

U = eiLz , where Lz = Qx Py Qy Px . Clearly, U is not an element of W).

Automorphisms which are not inner are called outer automorphisms. They form
a set OutW which is not a group.
As we have seen above, even if an automorphism is not inner, it can be implemented
by a unitary in some given representation.
Definition 1.26 Let (, H) be a representation of W. Then AutW is said
to be unitarily implementable on H if there exists some unitary U B(H) s.t.
((W )) = U (W )U 1 ,

10

W W.

(45)

Example 1.27 A large class of automorphisms is obtained as follows

(W (z)) = c(z)W (Sz)

(46)

where c(z) C\{0} and S : Cn Cn a continuous bijection. Weyl relations

impose restrictions on c, S:
c(z + z 0 ) = c(z)c(z 0 ), c(z) = c(z), |c(z)| = 1,
S(z + z 0 ) = S(z) + S(z 0 ), S(z) = S(z), ImhSz, Sz 0 i = Imhz|z 0 i.

(47)
(48)

The latter property means that S is a real-linear symplectic transformation.

For continuous c and S such automorphisms are unitarily implementable in
all irreducible representations satisfying the Criterion (consequence of the v.N.
uniqueness theorem). See Homeworks.
Remark 1.28 (z1 , z2 ) := Imhz|z 0 i is an example of a symplectic form. In general, we say that a bilinear form is symplectic if it is:
1. Antisymmetric: (z1 , z2 ) = (z2 , z1 )
2. Non-degenerate: If (z1 , z2 ) = 0 for all z2 , then z1 = 0.
1.1.7

Dynamics

Definition 1.29 A dynamics on W is a one-parameter group of automorphisms

on W i.e. R 3 t 7 t s.t. 0 = id, t+s = t s .
Proposition 1.30 Suppose that the dynamics is unitarily implemented in an irreducible representation i.e. there exists a family of unitaries s.t.
(t (W )) = U (t)(W )U (t)1 ,

W W.

(49)

Suppose in addition that t 7 U (t) continuous (in the sense of matrix elements)
and differentiable (i.e. for some 0 6= H, t U (t) exists in norm).
Then there exists a continuous group of unitaries t 7 V (t) (i.e. V (0) = 1,
V (s + t) = V (s)V (t)) s.t.
(t (W )) = V (t)(W )V (t)1 .

(50)

Remark 1.31 By the Stones theorem we have V (t) = eitH for some self-adjoint
operator H on (a domain in) H (the Hamiltonian). Whereas t is intrinsic, the
Hamiltonian is not. Its properties (spectrum etc.) depend in general on representation.
Proof. We have s t = s+t . Hence
U (s)U (t)(W )U (t)1 U (s)1 = U (s + t)(W )U (s + t)1 ,
U (s + t)1 U (s)U (t)(W ) = (W )U (s + t)1 U (s)U (t).
11

(51)
(52)

By irreducibility of
U (s + t) = (s, t)U (s)U (t), where |(s, t)| = 1.

(53)

By multiplying U by a constant phase ei0 we can assume that U (0) = I, hence

(0, t) = (s, 0) = 1.

(54)

Now consider a new family of unitaries V (s) = (s)U (s), |(s)| = 1. We have
V (s + t) = 0 (s, t)V (s)V (t) = (s + t)U (s + t)
= (s + t)(s, t)U (s)U (t) = (s + t)(s, t)(s)1 (t)1 V (s)V (t).(55)
Hence
0 (s, t) =

(s + t)
(s, t).
(s)(t)

(56)

The task is to obtain 0 (s, t) = 1 for all s, t for a suitable choice of (depending on
). The key observation is that associativity of addition in R imposes a constraint
on : In fact, we can write
U (r + s + t) = (r, s + t)U (r)U (s + t) = (r, s + t)(s, t)U (r)U (s)U (t),(57)
U (r + s + t) = (r + s, t)U (r + s)U (t) = (r + s, t)(r, s)U (r)U (s)U (t).(58)
Hence we get the cocycle relation (cohomology theory)
(r, s + t)(s, t) = (r + s, t)(r, s).

(59)

Using this relation one can show that given one can find such that 0 = 1.
cocycle is a coboundary (Howework). Important intermediate step is to show,
using the cocycle relation that
(s, t) = (t, s).

(60)

To express as a function of we will have to differentiate . By assumption,

there is H, kk=1 s.t. t U (t) exists. By (53), we have
(s, t) = U (t) U (s) U (s + t) = h, U (t) U (s) U (s + t)i
= hU (t), U (s) U (s + t)i.

(61)

Hence t (s, t) exists and by (60) also s (s, t). 

Example 1.32 Isotropic harmonic oscillator: In the framework of the polynomial

algebra P we have (heuristically)
t (Qi ) = cos(0 t)Qi sin(0 t)Pi ,
t (Pi ) = cos(0 t)Pi + sin(0 t)Qi .
12

(62)
(63)

In the Weyl setting t (W (z)) = W (eit0 z). This defines a group of automorphisms
from Example 1.27 with St z = eit0 z, c(z) = 1. (St is complex-linear). This
dynamics is unitarily implemented in the Schrodinger representation:
1 (t (W )) = U (t)1 (W )U (t)1 ,
q
P Pi2 kQ2i

k
itH
U (t) = e , H = i 2m + 2 , 0 = m
.

W W,

(64)

Example 1.33 Free motion in the framework of P:

t (Qj ) = Qj +

t
Pj ,
m

(65)

t (Pk ) = Pk .

(66)

t (W (z)) = W (Rez + (t/m + i)Im z)

(67)

In the framework of W:

We have that St (z) = Rez + (t/m + i)Im z is a symplectic transformation, but only
real linear. This dynamics is unitarily implemented in the Schrodinger representation:
1 (t (W )) = U (t)1 (W )U (t)1 ,
U (t) = eitH , H =

W W,

(68)

Pi2
i 2m .

By generalizing the above discussion, one can show that dynamics governed by
Hamiltonians which are quadratic in Pi , Qj correspond to groups of automorphisms
of W. But there are many other interesting Hamiltonians, for example:
H=

P2
+ V (Q)
2m

(69)

where n = 1, V C0 (R)R (smooth, compactly supported, real).

Theorem 1.34 (No-go theorem) Let H =
U (t) = eitH . Then
f
U (t)1 (W )U (t)1 1 (W),

P2
2m

+ V (Q), V L1 (R) L (R) and

f t R.
W W,

(70)

implies that V = 0.
Proof. See [3]. 
does not contain dynamics corresponding to Hamiltonians (69). A
Thus AutW
recently proposed solution to this problem is to pass from exponentials W (z) =
ei(uP +vQ) to resolvents R(, z) = (i uP vQ)1 and work with an algebra
generated by these resolvents [4].
13

1.1.8

Resolvent algebra

Definition 1.35 The pre-resolvent algebra R is the free polynomial -algebra generated by symbols R(, z), R\{0}, z Cn modulo the relations
R(, z) R(, z)
R(, z)
[R(, z), R(, z 0 )]
R(, z)
R(, z)R(, z 0 )

i( )R(, z)R(, z),

R(, z),
iImhz, z 0 iR(, z)R(, z 0 )2 R(, z),
R(, z),
R( + , z + z 0 )(R(, z) + R(, z 0 )
+ iImhz, z 0 iR(, z)2 R(, z 0 )),
1
R(, 0) =
,
i
=
=
=
=
=

(71)
(72)
(73)
(74)
(75)
(76)

where , , R\{0} and in (75) we require + 6= 0.

Remark 1.36 Heuristically R(, z) = (i uP vQ)1 . Realtions (71), (72)
encode the algebraic properties of the resolvent of some self-adjoint operator. (73)
encodes the canonical commutation relations. (74), (75), (76) encode linearity of
the map (u, v) 7 uP + vQ.
Definition 1.37 The Schrodinger representation of R is defined as follows: Let
(1 , H1 ) be the Schrodinger representation of W. Since it satisfies the Criterion
(i.e. it is regular) we have Pi , Qj as self-adjoint operators on L2 (Rn ). Thus we
can define
1 (R(, z)) = (i uP vQ)1 .

(77)

One can check that this prescription defines a representation of R which is irreducible.
Definition 1.38 We define a seminorm on R
kRk = sup k(R)k,

R R,

(78)

where the supremum is over all cyclic representations of R. (A cyclic representation is a one containing a cyclic vector. In particular, irreducible representations
e is defined as the completion of R/ ker k k.
are cyclic). The resolvent C -algebra R
Remark 1.39 The supremum is finite because for any representation we have
k(R(, z))k

1
,

(Homework).

(79)

and thus k(R)k for any R R is finite. It is not known if ker k k is trivial.
To show that it would suffice to exhibit one representation of R which is faithful
14

(i.e. injective: (R) = 0 implies R = 0). A natural candidate is the Schrodinger

representation. In this case one would have to check that if
X


ci1 ,...in 1 R(i1 , zi1 ) R(in , zin ) = 0
(80)
f inite

Then all ci1 ,...in = 0.

is regular if there exist self-adjoint
Definition 1.40 A representation (, H) of R
operators Pi , Qj on H s.t. for R\{0}
(R(, z)) = (i uP vQ)1 .

(81)

e is regular.
For example, the Schrodinger representation 1 (of R)
Fact: Any regular irreducible representation of R is faithful [4]. Hence, the
e is faithful. This does not imply however that the
Schrodinger representation of R
Schrodinger representation of R is faithful since we divided by ker k k!
Proposition 1.41 There is a one-to-one correspondence between regular repree and representations of W
f satisfying the Criterion. (The latter are
sentations of R
also called regular). Hence, by the Stone-von Neumann uniqueness theorem, any
e is unitarily equivalent to the Schrodinger
irreducible regular representation of R
representation.
Proof. (Idea). Use the Laplace transformation
Z
(R(, z)) = i
et (W (tz))dt,

= sgn

(82)

out of a regular representation of W.


to construct a regular representation of R
Remark 1.42 The Laplace transform can also be useful in checking if ker k k is
trivial.
Up to now, we found no essential difference between the Weyl algebra and the
f is
resolvent algebra. An important difference is that the Weyl C -algebra W

simple, i.e. it has no non-trivial two sided ideals. The resolvent C -algebra has
many ideals. They help to accommodate interesting dynamics.
e s.t. in any irreducible
Theorem 1.43 There is a closed two-sided ideal J R
regular representation (, H) one has (J ) = K(H) where K(H) is the algebra of
compact operators on H.
Remark 1.44 We recall:
A is a compact operator if it maps bounded operators into pre-compact operators. (On a separable Hilbert space if it is a norm limit of a sequence of
finite rank operators).
15

 A is Hilbert-Schmidt (A K2 (H)) if kAk2 := Tr(A A)1/2 < . HilbertSchmidt operators are compact.
A convenient way to show that an operator on L2 (Rn ) is Hilbert-Schmidt is
to study its integral kernel K, defined by the relation:
Z
(Af )(p) = dp0 K(p, p0 )f (p0 ).
(83)
If K is in L2 (Rn Rn ) then A K2 (L2 (Rn )) and kAk2 = kKk2 .
For example, consider A = f (Q)g(P ). Its integral kernel in momentum space
is determined as follows:
Z
1
0
(f (Q)g(P ))(p) =
dp0 eiQp (Ff )(p0 )(g(P ))(p)
2 Z
1
=
dp0 (Ff )(p0 )(g(P ))(p p0 )
2 Z
1
=
dp0 (Ff )(p0 )g(p p0 )(p p0 )
2 Z
1
dp0 (Ff )(p p0 )g(p0 )(p0 ).
(84)
=
2
Hence the integral kernel of f (Q)g(P ) if K(p, p0 ) = (Ff )(p p0 )g(p0 ). If f, g
are square-integrable, so is K.
Proof. (Idea). By the von Neumann uniqueness theorem we can assume that
e con is the Schrodinger representation 1 . Then it is easy to show that (R)
tains some compact operators: For example, set ui = (0, . . . , 1, . . . , 0) and vi =
| {z }
i

(0, . . . , 1, . . . , 0). Then the operator

| {z }
i

A := 1 (R(1 , iv1 )R(1 , u1 ) . . . R(n , ivn )R(n , un ))

n
n
Y
Y
1
=
(ij Qj )
(ij Pj )1
j=1

(85)

k=1

is Hilbert-Schmidt for all i , i R\{0}. (This can be shown by checking that it

has a square-integrabe kernel). In particular it is compact. Now it is a general
fact in the theory of C -algebras that if the image of an irreducibe representation
contains one non-zero compact operator then it contains all of them (Howework
or Corollary 4.1.10 of [6]). Thus, since 1 is faithful, we can set J = 11 (K(H)).
e since K(H) is a closed two-sided ideal in
This is a closed two-sided ideal in R
B(H). 

16

Theorem 1.45 Let n = 1, H = P 2 + V (Q), where V C0 (R)R real, continuous

vanishing at infinity and U (t) = eitH . Then
e
U (t)1 (R)U (t)1 1 (R),

e t R.
for all R R,

(86)

Remark 1.46 Since 1 is faithful, we can define the group of automorphisms of

R


(87)
t (R) := 11 U (t)1 (R)U (t)1 ,
which is the dynamics governed by the Hamiltonian H.
R
Remark 1.47 For simplicity, we assume that V S(R)R and dx V (x) = 0.
General case follows from the fact that such functions are dense in C0 (R)R in
supremum norm.
Proof. Let U0 (t) = eitH0 , where H0 = P 2 . Since this is a quadratic Hamiltonian,
we have
e 0 (t)1 1 (R).
e
U0 (t)1 (R)U

(88)

Now we consider V (t) := U (t)U0 (t)1 . It suffices to show that V (t) 1 are
e by Theorem 1.43 and hence
compact for all V C0 (R)R since then V (t) 1 (R)
e (t)1 = V (t)U0 (t)1 (R)U
e 0 (t)1 V (t)1 1 (R),
e
U (t)1 (R)U

(89)

e
using V (t)1 = V (t) 1 (R).
We use the Dyson perturbation series of V (t):
V (t) =

X
n=0

Z
dtn

tn

t2

dtn1 . . .
0

dt1 Vt1 Vt2 . . . Vtn ,

(90)

where Vt := U0 (t)V (Q)U0 (t)1 and the integrals are defined in the strong-operator
topology, that is exist on any fixed
R t vector. (Cf. Proposition 1.50 below).
The key observation is that 0 ds Vs are Hilbert-Schmidt. To this end compute
the integral kernel Ks of Vs :
1
2
2
(Ks )(p1 , p2 ) = eip1 s (FV )(p1 p2 )eip2 s .
2

(91)

This
R t is clearly not Hilbert-Schmidt. Now let us compute the integral kernel Ks of
ds Vs :
0
s )(p1 , p2 ) =
(K

Z
0

1 ei(p1 p2 )t 1
ds (Ks )(p1 , p2 ) =
(FV )(p1 p2 ).
2 i(p21 p22 )

17

(92)

This is Hilbert-Schmidt. In fact:

Z
Z
Z
sin2 (tq1 q2 )
2
2

dp1 dp2 |(Ks )(p1 , p2 )| = c dq1 |(FV )(q1 )|

dq2
(q1 q2 )2
Z
Z
sin2 (r)
|t|
= c dq1 |(FV )(q1 )|2
dr
|q1 |
r2
Z
|(FV )(q1 )|2
= c0 |t| dq1
|q1 |

(93)

Since (FV )(0) = 0 we have (FV )(q1 ) c|q1 | near zero so the integral exists.
Consequently, the strong-operator continuous functions
Z t2
n1
dt1 Vt1 Vt2 . . . Vtn
(94)
R
3 (t2 , . . . , tn ) 7
0

have values in the Hilbert-Schmidt class and their Hilbert-Schmidt (HS) norms
are bounded by

1/2
Z
|(FV )(q1 )|2
0
c |t2 | dq1
kV kn1
|q1 |

(95)

(since kABk2 kAk2 kBk). The integral of any strong-operator continuous HSvalued function with uniformly bounded (on compact sets) HS norm is again HS.
(See Lemma 1.49 below). So each term in the Dyson expansion (apart from n = 0)
e and the expansion converges uniformly in norm. So V (t) 1 is a
is in 1 (R)
compact operator. 
Remark 1.48 The resolvent algebra admits dynamics corresponding to H = P 2 +
V (Q). But there are other interesting Hamiltonians which are not covered e.g.
H = P 2 + M 2 . So there remain open questions...
In the above proof we used two facts, which we will now verify:
Lemma 1.49 Let Rn 3 t 7 F (t) K2 (H) be continuous in the strong operator
topology and suppose that for some compact set K Rn we have
sup kF (t)k2 < ,

(96)

tK

where kF (t)k2 = Tr(F (t) F (t))1/2 . Then

Z
F :=
dt F (t)
K

is again Hilbert-Schmidt.

18

(97)

Proof. We have
kF k22 = Tr F F = |

XZ
i

XZ

dt1 dt2 |hei , F (t1 ) F (t2 )ei i|

KK

XZ

dt1 dt2 hei , F (t1 ) F (t2 )ei i|

KK

dt1 dt2 kF (t1 )ei k kF (t2 )ei k.

(98)

KK

Since the summands/integrals are positive, I can exchange the order of integration/summation. By Cauchy-Schwarz inequality:
Z
X

1/2 X

1/2
2
kF k2
dt1 dt2
kF (t1 )ei k2
kF (t2 )ei k2
KK

Z
dt1 dt2 kF (t1 )k2 kF (t2 )k2

=
KK
2

|K| sup kF (t)k22 < .

(99)

tK

Where in the last step we use the assumption (96). 

Lemma 1.50 (Special case of Theorem 3.1.33 of [1]) Let R 3 t 7 U0 (t) be a
strongly continuous group of unitaries on H with generator H0 (i.e. U0 (t) = eitH0 ,
above we had H0 = P 2 ) and let V be a bounded s.a. operator on H. Then H0 + V
generates a strongly continuous group of unitaries U s.t.
U (t) = U0 (t)
X Z
+
in
n1

dt1 . . . dtn U0 (t1 )V U0 (t2 t1 )V U0 (tn tn1 )V U0 (t tn )

0t1 tn t

(100)
For any H. (To get the expression for V (t) it suffices to set = U0 (t)1 0 ).
Proof. Strategy: we will treat (100) as a definition of a t 0 dependent family
of operators t 7 U (t). We will use this definition to show that it can be naturally
extended to a group of unitaries parametrized by t R. Then, by differentiation,
we will check that its generator is H0 + V . Hence, by Stones theorem we will have
U (t) = eit(H0 +V ) .
Let U (n) (t) be the n-th term of the series of U . We have, by a change of
variables,
Z t
(0)
(n)
U (t) = U0 (t), U (t) =
dt1 U0 (t1 )iV U (n1) (t t1 ).
(101)
0

19

Iteratively, one can show that all U (n) (t) are well defined and strongly continuous.
It is easy to check that this is a series of bounded operators which converges in
norm: In fact
X
tn
kU (n) (t)k < .
(102)
kU (n) (t)k kV kn kk, hence
n!
n
By taking the sum of both sides of the recursion relation (101), we get
Z t
ds U0 (s)iV U (t s).
U (t) = U0 (t) +

(103)

Now we want to show the (semi-)group property:

Z t1
ds U0 (s)iV U (t1 s)U (t2 )
U (t1 )U (t2 ) = U0 (t1 )U (t2 ) +
0
Z t2
= U0 (t1 + t2 ) +
ds U0 (t1 + s)iV U (t2 s)
0
Z t1
+
ds U0 (s)iV U (t1 s)U (t2 )
0
Z t2
= U (t1 + t2 ) +
ds U0 (t1 + s)iV U (t2 s)
0
Z t1
+
ds U0 (s)iV U (t1 s)U (t2 )
0
Z t1 +t2

ds U0 (s)iV U (t1 + t2 s)

(104)

R t +t
Rt
Now t11 2 part of the last integral cancels the 0 2 integral (change of variables).
We are left with
Z t1
U (t1 )U (t2 ) U (t1 + t2 ) =
ds U0 (s)iV U (t1 s)U (t2 ) U (t1 + t2 s)).(105)
0

Now let U (t) be defined by replacing V with V in (100), R. It is clear from

(100) that the function
Ft1 () = U (t1 )U (t2 ) U (t1 + t2 )

(106)

is real-analytic. By (105) we get

Z
Ft1 () =

t1

ds U0 (s)iV Ft1 s ().

(107)

Clearly, Ft1 (0) = 0. Using this, and differentiating the above equation w.r.t. at
0, we get Ft1 (0) = 0. By iterating we get that all the Taylor series coefficients
of Ft1 at zero are zero and thus Ft1 () = 0 by analyticity. We conclude that the
semigroup property holds i.e.
U (t1 + t2 ) = U (t1 )U (t2 ).
20

(108)

Now we want to show that U (t) are unitaries. A candidate for an inverse of
U (t) is U 0 (t) defined by replacing H0 with H00 := H0 and V by V 0 = V . (JUMP
DOWN). We also set U00 (t) = ei(H0 )t . Let t2 t1 . Then
Z t1
0
0
ds U0 (s)iV U (t1 s)U 0 (t2 )
U (t1 )U (t2 ) = U0 (t1 )U (t2 ) +
Z t02
ds U00 (t1 + s)iV 0 U 0 (t2 s)
= U0 (t1 t2 ) +
Z 0t1
ds U0 (s)iV U (t1 s)U 0 (t2 )
+
Z0 t2
ds U00 (t1 + s)iV 0 U 0 (t2 s)
= U 0 (t2 t1 ) +
Z 0t1
ds U0 (s)iV U (t1 s)U 0 (t2 )
+
Z0 t2 t1

ds U00 (s)iV 0 U 0 (t2 t1 s)

(109)
0

R t1

R t +t
In the last integral the part 0 combines with the second line and t11 2
cancels the first line. Thus we get
Z t1


0
0
U (t1 )U (t2 ) U (t2 t1 ) =
ds U0 (s)iV U (t1 s)U 0 (t2 ) U 0 (t2 (t1 s))
0

(110)
(JUMP TO HERE). By an analogous argument as above we obtain
U (t1 )U 0 (t2 ) = U 0 (t2 t1 ),
0

(111)
0

In particular, U (t)U (t) = 1 and we can consistently set U (t) := U (t) for t 0.
Moreover, it is easily seeen from (100), by a change of variables, that U 0 (t) = U (t) .
Thus we have a group of unitaries. By Stones theorem it has a generator which
can be obtained by differentiation: Clearly we have for in the domain of H0 :
t |t=0 U0 (t) = iH0

(112)

Now we write
It :=

tn

dtn

dtn1 . . .

n1

t2

dt1 Vt1 Vt2 . . . Vtn U0 (t)

(113)

We have
t It = i

n1

n1

0
n

dt1 Vt1 Vt2 . . . Vtn1 Vt U0 (t)

0

Z
dtn

t2

dtn1 . . .

n1

tn

Z
dtn1 . . .

t2

dt1 Vt1 Vt2 . . . Vtn U0 (t)iH0 .

(114)

Taking the limit t 0 the second term tends to zero and the first term tends to
zero apart from n = 1 (since then there are no integrals). The n = 1 term gives
iV , thus, together with (112) we get that the generator of U is H0 + V . 
21

1.2

Algebra of bounded operators on a Hilbert space

Motivation: Most algebras of interest in physics (e.g. C -algebras, W -algebras)

can be realized as certain subalgebras of the algebra B(H) of all bounded operators
on some suitable Hilbert space. Important advantage: on a Hilbert space it is
easy to introduce various concepts of convergence (strong-operator, weak-operator
topology).
Definition 1.51 B(H) is the space of linear maps A : H H s.t.
kAk := sup kAk =

sup

kk=1

kk=1,kk=1

|h, Ai| < .

(115)

Lemma 1.52 (Basic properties):

1. B(H) is a normed complex vector space which is complete. (Banach space).
2. B(H) is equipped with operator product B(H) B(H) B(H). We have
kABk kAk kBk.

(116)

i.e. B(H) is a Banach algebra (B-algebra).

3. B(H) is equipped with -operation B(H) B(H). We have
kA k = kAk

(117)

i.e. B(H) is a Banach algebra (B -algebra).

4. C -property:
kA Ak = kAk2 .

(118)

i.e. B(H) is a C -algebra.

Proof. (Of the C -property). On the one hand
kA Ak = sup |h, A Ai| sup kAk kAk = kAk2 .
,H1

(119)

,H1

On the other hand

kA Ak sup |h, A Ai| = kAk2 .
H1


Basic terminology in the theory of bounded operators:
self-adjoint: A = A .
positive: (A 0) if h, Ai 0, H. (Positive eigenvalues).
22

(120)

 projection: A = A = A2 .
isometry: kAk = kk, H. (Equivalently, A A = 1).
partial isometry A A = E, E-projection. (Then also AA = F , F projection).
unitary: A A = AA = 1.
finite rank: dim(AH) = n < .
compact operators K(H): A maps bounded sets into pre-compact. Equivalently, on a separable Hilbert space, kA An k < for operators An of finite
rank n and sufficiently large n (dep. on ).

1/2
Hilbert-Schmidt K2 (H): kAk2 := Tr A A
< .
Trace-class K1 (H): A = B C, B, C are Hilbert-Schmidt. If A positive,
Tr A < .
Useful facts:
A 0 iff there is a (non-unique) B
s.t. A = B B. If we require that B 0
then it is unique and we write B = A.

polar decomposition: A = U |A|, where |A| := A A and U partial isometry. U U projection onto A H, U U projection onto AH. Decomposition is
unique.
Let us look at B(H) as an abstract algebra (defined by its relations) and consider
its representations:
The defining representation of B(H) is denoted (, H), i.e.
(A) := A,

A B(H),

H.

(121)

Note that K2 (H), equipped with the scalar product

hH1 |H2 i = Tr H1 H2 ,

(122)

is a Hilbert space. Also K2 (H) is a left and right -ideal in B(H), that is
B(H) K2 (H) = K2 (H) B(H) = K2 (H),

K2 (H) = K2 (H).

(123)

Thus one can define a -representation of B(H) in K2 (H) as follows:

HS (A)|Hi := |AHi.

(124)

hH|HS (A)Hi = hH|AHi = Tr H AH = Tr HH A

(125)

Note that
Note that HH is positive and TrHH < . If it is normalized, i.e.
Tr HH = 1, then := HH is a density matrix. Hence all mixed states in
QM can be described in the Hilbert space formalism using this representation.
23

Remark 1.53 An abstract state (positive, normalized, linear functional)

on a C -algebra is called pure if the equation
= p 0 + (1 p) 00 ,

where 0 < p < 1,

0 , 00 states,

(126)

has only one solution: = 0 = 00 . General fact: is pure iff its

GNS representation is irreducible. In an irreducible representation the
physicists
Pdefinition of pure states as pure = |ih| and mixed states as
mixed = i pi |i ihi | works.
Remark 1.54 In terms of Theorem 1.20 (GNS construction) the situation
is the following: Consider a state (A) = Tr A, A B(H), where is a
density matrix (mixed in the physicists sense). This state induces a cyclic
representation ( , H , ) s.t.
(A) = h , (A) i.

(127)

This representation is unitarily equivalent to a subrepresentation of HS . The

isometry V : H K2 (H) given by

(128)
V (A) = |A i
satisfies V (A) = HS (A)V . Hence HS is reducible.
Pathological representations/states: By the Hahn-Banach theorem there exist positive, linear and normalized functionals on B(H) s.t. (C) = 0 for
any C K(H) but (1) = 1. The GNS construction gives a representation
which maps all compact operators to zero.
Also ( ) is less continuous than ( ) = Tr ( ). Any state on a C -algebra
is continuous w.r.t. the norm topology, but not necessarily in terms of the
weak topology (i.e. convergence of matrix elements).
N
X

1 = (1) = ( lim

|en ihen |) = lim (|en ihen |) = 0.

N

n=0

(129)

On the other hand

1 = (1) = Tr (lim
N

= lim
N

N
X

|en ihen |) =

n=0

N
XX

he0` , en ihen , e0` i

n=0

To exchange limN with

the bound

he0` ,

lim
N

N
X

he0` , en ihen , e0` i

n=0

N
X
= lim Tr (
|en ihen |)
N

(130)

n=0

we used the dominated convergence theorem and

N
X

en ihen , e0` i he0` , e0` i.

n=0

24

(131)

Let us consider more systematically various notions of convergence in B(H).

A sequence {An B(H)}nN is said to be convergent to A B(H) in:
(a) weak operator topology (weakly) if h, (A An )i 0 for any , H.
(b) strong operator topology (strongly) if k(A An )k 0 for any H.
(c) norm if kA An k 0.
P
For example, limN N
n=1 |en ihen | = 1 exists in weak and strong operator topology, but not in norm (if dim H = ). In general (c) (b) (a) but the converse
implications do not hold.
Definition 1.55 A positive linear and normalized functional : B(H) C
(state) is called normal if for every sequence of projections Qn , n N, which
converges strongly to some projection Q one has
(lim Qn ) = (Q) = lim (Qn ).
n

(132)

Note: is not normal in this sense. (H is assumed to be separable here. For nonseparable H one has to use generalized sequences (nets) {Qi }iI . Here I is an
index set together with a partial ordering (reflexive, transitive amd antisymmetric)
which satisfies: For any i, i0 I there is j s.t. j > i, j > i0 ).
Proposition 1.56 [1] Let be a normal state on B(H). Then there exists a
density matrix s.t.
(A) = Tr A,

A B(H).

(133)

It turns out that topological and algebraic concepts are closely tied for -subalgebras
of B(H):
Theorem 1.57 [1] (von Neumann bicommutant theorem) Let A be a unital algebra of operators on a Hilbert space. Then A is dense in A00 in the weak and
strong topology.
Remark 1.58 We note/recall the following:
1. The commutant of A in B(H) is defined as follows:
A0 = { B B(H) | [B, A] = 0 for all A A }.

(134)

2. A unital -algebra of operators on a Hilbert space s.t. A00 = A is called a

von Neumann algebra. In particular, it is a C -algebra.
3. For separable H it suffices to add limits of strongly convergent sequences to
obtain the strong closure of a -algebra. (Nets not needed).
25

1.3

Weyl algebra for systems with infinitely many degrees

of freedom

Algebraic approach is adventageous in order to perform the transition from finite

to infinite systems.
Finite systems: Cn , h , i, (z, z 0 ) = Imhz, z 0 i. Pre-Weyl algebra W is the
free -algebra generated by W (z), z Cn , subject to relations
i

W (z)W (z 0 ) = e 2 (z,z ) W (z + z 0 ),

W (z) = W (z),

z Cn .

(135)

Remark 1.59 This form of Weyl relations corresponds to W (z) = ei(uP +vQ) ,
z = u + iv via BCH. If we wanted Wnew (z) = ei(vP +uQ) , z = u + iv, that
would lead to a minus sign in front of :
i

Wnew (z)Wnew (z 0 ) = e 2 (z,z ) Wnew (z + z 0 )

(136)

This convention will be more convenient in the case of systems with infinitely
many degrees of freedom.
Infinite systems: infinite dimensional complex-linear space D with scalar
product h , i (pre-Hilbert space). Define the symplectic form (f, g) =
Im hf, gi, f, g D. Pre-Weyl algebra W is the free -algebra generated by
W (f ), f D, subject to relations
i

W (f )W (g) = e 2 (f,g) W (f + g),

W (f ) = W (f ),

f, g D.

(137)

Example 1.60 : D = S(Rd ),

Z

dd x f (x)g(x).

hf, gi =

(138)

i (Re f )+(Im f )

Heuristics: W (f ) = e
, where
Z
Z
d
(g) :=
d x g(x)(x), (h) :=
dd x h(x)(x)

(139)

are spatial means of the quantum field operator (x) and its canonical conjugate
momentum (x). The fields , satisfy formally
[(x), (y)] = i(x y)1,
[(x), (y)] = [(x), (y)] = 0.

(140)
(141)

(x), (y) are not expected to be operators, but only operator valued distributions.
But (g), (h) are expected to be operators and we have
Z
[(g), (h)] = i dd x g(x)h(x)1 = ihg, hi1.
(142)
26

Example 1.61 : D = S(Rd ),

Z
hf, gi =

dd p f (p)g(p).

Here Rd is interpreted as momentum space.

i (a (f )+a(f ))
Heuristic interpretation: W (f ) = e 2
where
Z
Z

a (f ) = d p f (p)a (p), a(f ) = dd p f (p)a(p).

(143)

(144)

are creation and annihilation operators of particles with momentum in the support
of f . The commutation relations are
[a(p), a (q)] = (p q)1,
[a(p), a(q)] = [a(p), a (q)] = 0.

(145)
(146)

Similarly as before a priori these are only operator valued distributions. For
smeared versions we have:
Z

(147)
[a(g), a (h)] = dd p g(p)h(p)1 = hg, hi1.
1.3.1

Fock space

We recall the definition and basic properties of a Fock space over h := L2 (Rd , dd x).
We have for n N
n h = h h = L2 (Rnd , dnd x),
ns h = Sn (h h) = L2s (Rnd , dnd x),
0s h := C, where is called the vacuum vector.

(148)
(149)
(150)

Here Sn is the symmetrization operator defined by

Sn =

1 X
, where (f1 fn ) = f(1) f(n) ,
n! P

(151)

Pn is the set of all permutations and L2s (Rnd , dnd x) is the subspace of symmetric
(w.r.t. permutations of variables) square integrable functions. The (symmetric)
Fock space is given by
(h) := n0 ns h = n0 L2s (Rnd , dnd x).

(152)

We can write (h) in terms of its Fock space components = {(n) }n0 . We
define a dense subspace fin (h) (h) consisting of such that (n) = 0 except
for finitely many n. Next, we define a domain
D := { fin (h) | (n) S(Rnd ) for all n }.
27

(153)

Now, for each p Rd we define an operator a(p) : D (h) by

(a(p))(n) (k1 , . . . , kn ) = n + 1(n+1) (p, k1 , . . . , kn ),

In particular a(p) = 0.

(154)

Note that the adjoint of a(p) is not densely defined, since formally
n

(n)

(a (p))

1 X
(k1 , . . . , kn ) =
(p k` )(n1) (k1 , . . . , k`1 , k`+1 , . . . , kn ) (155)
n `=1

However, a (p) is well defined as a quadratic form on D D. Expressions

Z
Z
d

a(g) = d p a(p)g(p), a (g) = dd p a (p)g(p), g S(Rd ),

(156)
give well-defined operators on D which can be extended to fin (h). On this domain
they act as follows
Z

(n)
n + 1 dd p g(p)(n+1) (p, k1 , . . . , kn ),
(157)
(a(g)) (k1 , . . . , kn ) =
n

(n)

(a (g))

1 X
(k1 , . . . , kn ) =
g(k` )(n1) (k1 , . . . , k`1 , k`+1 , . . . , kn ).(158)
n `=1

These expressions can be used to define a(g), a (g) for g L2 (Rd ). Since these
operators leave fin (h) invariant, one can compute on this domain:
[a(f ), a (g)] = hf, gi1

(159)

for f, g L2 (Rd ). (Formally, this follows from [a(p), a (q)] = (p q)).

Now we are ready to define canonical fields and momenta: Let : Rd 7 R+ be
d
1/2
1/2
positive, measurable function of momentum
f
p s.t. if f S(R ) then f,
L2 (Rd ). (Examples: (p) = 1, m (p) = p2 + m2 , m 0). We set for f, g
S(Rd )


1
(f ) := a (1/2 f) + a(1/2 f) ,
2


1
(g) := a (i1/2 g) + a(i1/2 g) ,
2

(160)
(161)

where f(p) := (Ff )(p). For := m this is the canonical field and momentum
of the free scalar relativistic quantum field theory of mass m 0. From (159) we
have

i
1
gi) = ihf, gi, (162)
i
[ (f ), (g)] =
hig, fi + hf,
g i = (hg, fi + hf,
2
2
where in the last step we made use of Plancherel theorem and
Z
h
g , f i = dd x g(x)f (x) = hf, gi.
28

(163)

Remark 1.62 Note that (160), (161) arise by smearing the operator-valued distributions:
Z


1
dd k
ikx
ikx
p
e
a
(k)
+
e
a(k)
,
(164)
m (x) =
(2)d/2
2m (k)
r
Z


i
m (k) ikx
m (x) =
e
a (k) eikx a(k) .
(165)
dd k
d/2
(2)
2
Consider a unitary operator u on h. Then, its second quantization is the
following operator on the Fock space:
(u)|(n) (h) = u u,
(u) = .

(166)
(167)

where (n) (h) is the n-particle subspace. We have the useful relations:
(u)a (h)(u) = a (uh),

(u)a(h)(u) = a(uh).

(168)

(Note that a (h) = a(h)).

Consider a self-adjoint operator b on h. Then, its second quantization is the
following operator on the Fock space:
d(b)|(n) (h) =

n
X

1 b 1,

(169)

i=1

d(b) = 0.

(170)

Suppose that b = b(k) is a multiplication operator in momentum space on h =

L2 (Rd ). Then as an equality of quadratic forms on D D we have
Z
d(b) = dd k b(k)a (k)a(k).
(171)
Moreover, suppose that U (t) = eitb . Then
(U (t)) = eitd(b) .
1.3.2

(172)

Representations of the Weyl algebra

Now we are ready to define several representatons of W on (h). We set D = S(Rd )

and (f, g) := Im hf, gi with standard scalar product in L2 (Rd ):
Definition 1.63 Let be as above. The corresponding Fock space representation
of W is given by
(W (f )) = ei( (Re f )+ (Im f )) .

(173)

In terms of creation and annihilation operators, we have

(W (f )) = e

i (a (f )+a(f ))
2

(174)

1
d + i 21 Imf
d )(p). Note that for = 1 we have f (p) = f(p)
where f (p) := ( 2 Ref
and thus we reproduce Examples 1.60,1.61.

29

Theorem 1.64 Representations m are faithful, irreducible and m1 is not unitarily equivalent to m2 for m1 6= m2 . (So Stone-von Neumann uniqueness theorem does not hold for systems with infinitely many degrees of freedom).
Proof. See Theorem X.46 of [7].
1.3.3

Symmetries

Symmetries are represented by their automorphic action on the algebra.

Definition 1.65 Let (D, ) be a symplectic space. A symplectic transformation S
is a linear bijection S : D D s.t.
(Sf, Sg) = (f, g),

f, g D.

(175)

Note that S 1 is also a symplectic transformation.

Fact: Every symplectic transformation induces an automorphism of W according
to the relation:
S (W (f )) = W (Sf ),

f D.

(176)

[) k = kf k.
Proposition 1.66 Let S be a symplectic transformation s.t. also k(Sf

(For = 1 this is just unitarity of S w.r.t. the scalar product in L2 (Rd )). Then
there exists a unitary operator U,S on (h) s.t.

U,S (W )U,S
= (S (W )),

W W,

(177)

and U,S = . (Converse also true).

Proof. We skip the index . Since we know that W acts irreducibly on (h), we
have that
D := { (W ) | W W }

(178)

is dense in (h). On this domain we set

US (W (f )) = (W (Sf )),

(179)

and extend by linearity to W. By invertibility of S this has a dense range. We

check that it is an isometry on this domain. For this it suffices to verify
hUS (W (f )), US (W (g))i = h(W (f )), (W (g))i.

(180)

We have
l.h.s. = h(W (Sf )), (W (Sg))i = h, (W (Sf )W (Sg))i
i

= e 2 Imhf,gi h, (W (S(g f )))i,

30

(181)

where we made use of the fact that S is symplectic. Let us set h := S(g f ). We
have
h, (W (h))i = h, e
1

)+a(h
))
i (a (h
2

= e 2 kh k h, e
k2
12 kh

)
i a (h
2

)
i a(h
2

\ )) k2
12 k(S(gf

= e
=e
= r.h.s. of (180),

i
1

= e 2 k(gf ) k
[

(182)

where we used Baker-Campbell-Hausdorff (which can be justified by expanding

exponentials into convergent series) and the additional assumption on S.
Now the converse: suppose that S is unitarily implemented in by a unitary
U,S s.t. U,S = . Then, in particular,
h, (W (Sf ))i = h, (S (W (f )))i

= h, U,S (W (f ))U,S
i = h, (W (f ))i. (183)
Hence,
1

e 2 k(Sf ) k = e 2 kf k
c

(184)

which concludes the proof. 

1.3.4

Symmetries in the case = 1 (non-local quantum field)

R
We set D = S(Rd ), hf, gi = dd xf (x)g(x), (f, g) = Im hf, gi, m > 0.
Note that any unitary u on h = L2 (Rd ), which preserves D, gives rise to a
symplectic transformation S = u|D .
By Proposition 1.66, the automorphism induced by S is unitarily implemented on (h).
A natural candidate for the implementing unitary is (u).
iak
\
1. Space translations: (Sa f )(x) = f (xa) (or (S
f (p)). Obviously
a f )(p) = e
Z
(185)
h(Sa f ), (Sa g)i = dd x f (x a)g(x a) = hf, gi.

(This implies that Sa is symplectic). The implementing unitary is U (a) =

(eipa ) = eiad(p) , where p meansR the corresponding multiplication operator on L2 (Rd , dd p). P := d(p) = d3 k ka (k)a(k) can be called the total
momentum operator. Indeed by (168):


i
a (eiap f)+a(eiap f)
2
a (W (f )) = W (Sa f ) = e


i

ipa 2 a (f)+a(f)
= (e
)e
(eipa ) .
(186)
31

2. Rotations: (SR f )(x) = f (R1 x), R SO(d).

Z
h(SR f ), (SR g)i = dd x f (R1 x)g(R1 x) = hf, gi

(187)

The implementing unitary is U (R) = (uR ), where (uR g)(x) = g(R1 x) is a

unitary representation of rotations on L2 (Rd ).
it(p)
[
f (p) where (p) is a reasonable disper3. Time translations: (S
t f )(p) = e
sion relation
pof a particle. Since we want to build a relativistic theory, we
set (p) = p2 + m2 , m > 0. Clearly:

h(St f ), (St g)i = hf, gi.

(188)

The implementing unitary is U (t) = (eit(p) ) = eitd((p)) , where

Z
H := d((p)) = d3 k (k)a (k)a(k),

(189)

can be called the total energy operator or the Hamiltonian.

Remark 1.67 Note that ft := St f satisfies the Schrodinger equation:
it ft (x) = (i)ft .

(190)

4. Lorentz transformations
Minkowski spacetime: (Rd+1 , g), g = (1, 1, 1, 1).
Lorentz group: L = O(1, d) = { GL(1 + d) | gT = g }
Proper Lorentz group: L+ = SO(1, d) = { O(1, d) | det = 1 }
(preserves orientation).
Ortochronous Lorentz group: L = { O(1, d) | eT e 0 }, where
e = (1, 0, 0, 0). (Preserves the direction of time)
Proper ortochronous Lorentz group: L+ = L L+ is a symmetry group
of the SM of particle physics.
The full Lorentz group consists of four disjoint components:
L = L+ L+ L L

(191)

For d = 3 they can be defined using time reversal T (t, x) = (t, x) and
parity P (t, x) = (t, x) transformations:
L+ = T P L+ ,

L = P L+ ,

32

L = T L+ .

(192)

Now we set
s
(S f )(p) =

(1 p)
f (1 p),
(p)

f D,

(193)

where 1 p is defined by 1 ((p), p) = ((1 p), 1 p). We have

h(S f ), (S f )i = hf, gi.

(194)

d p
is a Lorentz invariant measure
This can be shown (Homework) using that (p)
(unique for a fixed m and normalization, see Theorem IX.37 of [7]). Formally
Z
Z
dd p
d+1
2
2
0
d p (
p m )(
p )F (
p) =
F ((p), p),
(195)
2(p)

where p = (p0 , p), p2 = (p0 )2 p2 .

S arises by restriction to D of a unitary representation u of L+ acting on
h = L2 (Rd ) by formula (193). The implementing unitary is U () := (u ).
5. Poincare transformations: The (proper ortochronous) Poincare group P+ =
Rd+1 o L+ is a set of pairs (
x, ) with the multiplication:
(
x1 , 1 )(
x2 , 2 ) = (
x1 + 1 x2 , 1 2 ).

(196)

It acts naturally on Rd+1 by (

x, )
y =
y + x. (Here we set x = (t, x)).
Note that (
x, ) = (
x, I)(0, ). Accordingly, we define
S(x,) := Sx S = St Sx S

(197)

as a symplectic transformation on D corresponding to (

x, ). We still have
to check if (
x, ) 7 S(x,) is a representation of a group, that is whether
S(x1 ,1 ) S(x2 ,2 ) = S(x1 ,1 )(x2 ,2 ) .

(198)

We use the fact that all these automorphisms can be implemented in the
(faithful) representation =1 . We have




1 ((x,) (W (f ))) = 1 W (S(x,) f ) = 1 W (St Sx S f )




= U (t)U (x)U ()1 W (f ) U (t)U (x)U () (199)
To verify (198) it suffices to check that
U (
x, ) := U (t)U (x)U () = (ei(p)t )(eipx )(u )
= (ei(p)t eipx u )

(200)

is a unitary representation of P+ on (h). For this it suffices that

u(x,) = ei(p)t eipx u
is a unitary representation of P+ on h = L2 (Rd ). (Homework).
33

(201)

Summing up, for any m > 0 we have a representation P+ 3 (

x, ) 7 (x,) of
the Poincare group in Aut W. In the representation =1 automorphisms (m,=1)
are unitarily implemented by a representation P+ 3 (
x, ) 7 U (
x, ).
(m,=1)

Nevertheless, (W, (m,=1) , =1 ) does not give rise to a decent (local) relativistic QFT. Problem with causality:
W (f ), suppf O should be an observable localized in an open bounded
region O Rd at t = 0.
t (W (f )) should be localized in {O + |t|~n , |~n| = 1} in a causal theory.
i(p)t
[
f (p) thus St f is not com However, t (W (f )) = W (St f ), (S
t f )(p) = e
pactly supported. (Infinite propagation speed of the Schrodinger equation).
In fact, since ei(p)t is not entire analytic (cut at p = im), its inverse Fourier
transform cannot be a compactly supported distribution (see Theorem IX.12
of [7]).

p
Symmetries in the case (p) = p2 + m2 (local quantum field)
R
We set D = S(Rd ), hf, gi = dd xf (x)g(x), (f, g) = Im hf, gi.
1.3.5

Recall that we need symplectic transformations S s.t. k(Sf ) k = kf k,

1
d + i 12 Imf
d )(p).
where f (p) := ( 2 Ref
Note that k(Sf ) k = kf k does not imply in this case that S is symplectic.
Strategy: Take the unitary u on h corresponding to a given symmetry (which
we know from = 1 case) and find S s.t. uf = (Sf ) . Then check that S
is symplectic.
iap d
\
\
1. Space translations: We have Re(S
Ref (p) and
a f )(p) = Sa Ref (p) = e
iap
\
\
analogously for Im. Thus (Sa f ) (p) = e
f (p) and therefore k(S
a f ) k =
kf k so the symmetry is unitarily implemented. The implementing unitary
is the same as in the = 1 case.

\
d
\
2. Rotations: Again Re(S
R f )(p) = SR Ref (p) = uR Ref (p) and analogously for
Im. Since is rotation invariant (depends only on p2 ), we have uR (p)uR =
\
b

(p) and therefore Sd

R f (p) = (uR f )(p). Thus k(SR f ) k = kf k so the
symmetry is unitarily implemented. The implementing unitary is the same
as in the = 1 case.
it(p)
0
[
3. Time translations: First note that (S
f (p) does NOT satisfy
t f )(p) = e
the additional condition. For example, for f real we have

1
21
0
[
(p) cos ((p)t) + i 2 (p) sin ((p)t) f(p).
(S
t f ) (p) =
34

(202)

The L2 norm of this (St0 f ) does depend on t. (Thus St0 is not implemented
in this representation by unitaries preserving the vacuum).
Instead, we consider the following group of transformations:
(St f )(x) = (cos(t) + i1 sin(t))Re f (x)
+i(cos(t) + i sin(t))Im f (x).

(203)

Think of as a function of p2 = 2x . Thus we can compute real and

imaginary parts as for functions:
Re (St f ) = cos(t)Re f sin(t)Im f,
Im (St f ) = 1 sin(t))Re f + cos(t)Im f

(204)
(205)

This is a symplectic transformation

(St f, St g) = hRe (St f ), Im(St g)i (f g)

(206)

We note that terms involving Re f Re g and Im f Im g cancel because are

invariant under (f g). The remaining two terms give
(St f, St g) = hcos2 (t)Re f, Im gi hsin2 (t)Im f, Re gi (f g)
= hRe f, Im gi hIm f, Re gi = (f, g).
(207)
Next, we check k(St f ) k = kf k:
1

(St f ) = 2 Re (St f ) + i 2 Im (St f )

1
1
= cos(t) + i sin(t) 2 Re f + i 2 Im f
= eit f .

(208)

Hence clearly k(St f ) k = kf k and this group of automorphisms is unitarily

implemented on Fock space by unitaries preserving the vacuum. They are
by U (t) = (eit ) = eid((p)) . Thus the Hamiltonian is d((p)) =
Rgiven
d
d k (k)a (k)a(k).
Remark 1.68 Note that ft := (St f ) in (203) is the unique solution of the
Klein-Gordon equation:
(t2 2x + m2 )ft (x) = 0

(209)

with the initial conditions ft=0 (x) = f (x) and (t f )t=0 (x) = (2x m2 )Im f (x)+
iRe f (x). In contrast to the Schrodinger equation, KG equation has finite propagation speed: If suppft=0 , supp t ft=0 O then supp ft {O +
|t|~n , |~n| = 1}. This theory has good chances to be local.
4. Lorentz transformations: There exist symplectic transformations S which
satisfy k(S f ) k = k(S ) k and preserve localization (for f C0 (Rd ) we
have (S f ) C0 (Rd )) (Homework).
35

5. Poincare transformations: For (

x, ) P+ we define
S(x,) := Sx S = St Sx S

(210)

as a symplectic transformation on D corresponding to (

x, ). Obviously,
k(S(x,) f ) k = kf k, since the individual factors satisfy this. (We note that
Sx is as in the = 1 case but St , S are different). The proof that (
x, ) 7
S(x,) is a representation of a group goes as in = 1 case, exploiting that
these automorphisms are implemented on Fock space by the same group of
unitaries as in the = 1 case.
Summing up, for any m 0 we have a representation P+ 3 (
x, ) 7 (x,)
of the Poincare group in Aut W. In the representation m automorphisms (m)
are unitarily implemented by the representation P+ 3 (
x, ) 7 U (
x, ), the same
as in the = 1 case. Time evolution is governed by the KG equation which has
finite propagation speed and Lorentz transformations act locally: we expect that
(W, (m) , m ) gives rise to a local (causal) relativistic QFT.
(m)

1.3.6

Spectrum condition (positivity of energy)

In this subsection we study the spectrum

of the group of unitaries on (h) imp
2
plementing translations in , = p + m2 . (The discussion below is equally
valid for =1 since m (p) = (p), hence unitaries implementing translations are
the same in both representations).
U (t, x) = eiHtiP x = eid((p))tid(p)x

(211)

H, P 1 , . . . , P d is a family of commuting s.a. operators on (h). Such a family has

a joint spectral measure E: Let Rd+1 be a Borel set and its characteristic
function. Then E() := (H, P 1 , . . . , P d ). The joint spectrum of H, P 1 , . . . , P d ,
denoted Sp (H, P ) is defined as the support of E. Physically, these are the measurable values of total energy and momentum of our system.
Theorem 1.69 Sp (H, P ) V + , where V + = { (p0 , p) Rd+1 | p0 |p| } is the
closed future lightcone.
Proof. We have to show that for V + = , bounded Borel set, we have
E() = 0. Let C0 (Rd+1 ) approximate pointwise as 0. (This
regularization is needed because the Fourier transform of a sharp characteristic
function may not be L1 ). Note that (H, P ) leaves (n) (h) invariant, thus it
suffices to show that its matrix elements vanish on these subspaces. We have for

36

, (n) (h):
h, (H, P )i
= limh, (H, P )i
0
Z
(d+1)
2
= lim(2)
dt dxh, U (t, x)i (t, x)
0
Z
Z
(d+1)
2
dt dx dnd p ( )(p1 , . . . , pn )ei(p1 ++pn )x (t, x)
= lim(2)
0
Z
= dnd p ( )(p1 , . . . , pn ) (
p1 + + pn ),
(212)
where we made use of Fubini and dominated convergence. Note that p = ((p), p)
V + for all p Rd . Since V + is a cone, also p1 + + pn V + . Thus the last
expression is zero if V + = . 
Remark 1.70 In the proof above we used the following conventions for the Fourier
transform on Rd+1 :
Z
(d+1)
0
0
2

dd xdt eip tipx f (t, x),

(213)
f (p , p) := (2)
Z
(d+1)
0
2

f (t, x) := (2)
dd pdp0 eip t+ipx f (p0 , p).
(214)
A more detailed analysis of the spectrum exhibits that
for m > 0
Sp (H, P ) = {0} {Hm } G2m , where
p
Hm := { (p0 , p) Rd+1 | p0 = p2 + m2 },
p
G2m := { (p0 , p) Rd+1 | p0 p2 + (2m)2 }.

(215)
(216)
(217)

{0} is a simple eigenvalue corresponding to the vacuum vector . Hm is called

the mass hyperboloid. The corresponding spectral subspace E(Hm )(h) satisfies
E(Hm )(h) = (1) (h) = h.

(218)

Thus it is invariant under (

x, ) 7 U (
x, ). In fact it carries the familiar
irreducible representation of u(x,) given by (201). According to Wigners
definition of a particle, E(Hm )(h) describes single-particle states of a particle of mass m and spin 0. G2m can be called the multiparticle spectrum.
(PICTURE).
For m = 0 we have
Sp (H, P ) = V+ .

(219)

Again, there is a simple eigenvalue at {0} (embedded in the multiparticle

spectrum) which corresponds to the vacuum vector . Hm=0 is the boundary
of V+ . The subspace E(Hm=0 )(h) = h carries states of a single massless
particle of mass zero.
37

1.3.7

Locality and covariance

(m)

We fix m 0 and use automorphisms (x,) := (x,) constructed in m .

We set D = C0 (Rd ), (f, g) = Imhf, gi. (We can restrict the symplectic
space from S(Rd ) to C0 (Rd ) because it is preserved by all the symplectic transformations we considered in = m case, in particular by timetranslations and Lorentz transformations).
We call Or = { (t, x) Rd+1 | |t| + |x| < r} the standard double cone of
radius r. Its base is the ball Br = {x Rd | |x| < r}.
W(Or ) := Alg {W (f ) | suppf Br } is the algebra of observables1 localized (physically measurable) in Or .
W(Or + x) := x (W(Or )) = Alg{x (W (f )) | suppf Br } is the algebra
of observables localized in Or + x where x = (t, x).
Theorem 1.71 Suppose that Or1 and Or2 + x are spacelike separated. Then
[W1 , W2 ] = 0,

for all

W1 W(Or1 ),

W2 W(Or2 + x).

(220)

Proof. By Weyl relations

W (f1 )W (Sx f2 ) = eiIm hSx f2 ,f1 i W (Sx f2 )W (f1 ),

(221)

so we have to show that Im hSx f2 , f1 i = 0 for suppf1 Br1 and supp f2 Br2 .
First suppose that x = (0, x), then we simply obtain that Or1 and Or2 + x are
disjoint and so are Br1 and Br2 +x. Hence hSx f2 , f1 i = 0 simply due to disjointness
of supports of the two functions.
Now the general case: Write Sx f2 = St (Sx f2 ). As before Sx f2 is supported
in Br2 + x. Thus, by propagation properties of solutions of the KG equation,
St (Sx f2 ) is supported in (Br2 + x) + |t|B1 . But spacelike separation of Or1 and
Or2 + x implies that (Br2 +x)+|t|B1 is disjoint from Br1 . So, again, hSx f2 , f1 i = 0.
(PICTURE). 
W(Or + x) := (x,) W(Or ).
W(O) := Alg{ W(Or + x) | r, , x s.t. Or + x O} is the algebra of
observables localized in an arbitrary open bounded region O Rd+1 .
f
f
f
W(O)
are the norm closures of W(O)
in the C -algebra W.
1

-subalgebra of W generated by all such W (f )

38

1.3.8

Haag-Kastler axioms

f
Theorem 1.72 The net of C -algebras O 7 W(O),
labelled by open bounded
d+1
subsets O R , satisfies:
f 1 ) W(O
f 2 ),
1. (isotony) O1 O2 W(O
f 1 ), W(O
f 2 )] = 0,
2. (locality) O1 , O2 spacelike separated [W(O
f
f
3. (covariance) x, (W(O))
= W(O
+ x), for all (
x, ) P+ ,
f
f=S
4. (generating property) W
ORd+1 W(O).
Remark 1.73 The above properties are called the Haag-Kastler axioms. Any
triple
Rd+1 O 7 A(O),

A,

P+ 3 (
x, ) 7 (x,) ,

(222)

(not necessarily coming from a Weyl algebra), satisfying the above properties is
called a Haag-Kastler net of C -algebras.
Proof. 1. is obvious from definition of W(O).
2. For O1 , O2 translated double-cones (no Lorentz transformations involved) this
follows from Theorem 1.71. General case: Homework.
3. Recall definition of W(O):
W(O) := Alg{ (x0 ,0 ) (W(Or )) |(
x0 , 0 )Or O}.

(223)

Then we have
x, (W(O)) = Alg{ (x,)(x0 ,0 ) (W(Or )) |(
x0 , 0 )Or O}
= Alg{ (x,)(x0 ,0 ) (W(Or )) |(
x, )(
x0 , 0 )Or (
x, )O}
= Alg{ (x00 ,00 ) (W(Or )) |(
x00 , 00 )Or (
x, )O}.
(224)
f
4. Since D = C0 (Rd ), every Weyl operator
S W (f ) W belongs to W(O) for
sufficiently large O. Thus we have W ORd+1 W(O) and inclusions survive
taking closures. The opposite inclusion is obvious. 
f satisfies:
Theorem 1.74 The (irreducible) representation m of W
x, ) 7 (x,) are unitarily implemented by a
1. The automorphisms P+ 3 (
(strongly continuous) group of unitaries P+ 3 (
x, ) 7 U (
x, ).
2. (positivity of energy) The joint spectrum of the generators (H, P ) of U (
x) =
U (
x, I) is contained in V + .
3. (uniqueness of the vacuum vector) There is one (up to phase) unit vector
s.t. U (
x, ) = .
39

Remark 1.75 A representation of an abstract Haag-Kastler net satisfying the

above properties is called a vacuum representation.
Remark 1.76 Note that =1 is not a vacuum representation because the automorphisms := (m ) appearing in 1., which are compatible with locality, are not
unitarily implemented in this representation. Automorphisms =1 , constructed in
Subsection 1.3.4, are unitarily implemented in =1 but are not compatible with
locality.

1.4

Haag-Kastler net of von Neumann algebras

f
f as above, in the vacuum represenGiven the Haag-Kastler net O 7 W(O),
W,
tation m , one can proceed to a net of von Neumann algebras:
00
f
O 7 A(O) := m (W(O))
,

A :=

A(O),

(x,) := U (
x, ) U (
x, )

ORd+1

(225)
Note that A is defined only as a C -algebra by taking the norm closure (and not
the strong closure) of the local von Neumann algebras. Strong closure would be
too large: Since m is irreducible, we have A00 = B(H) (where H = (h) in this
case).
Theorem 1.77 All local algebras A(O), for different open bounded O, are isomorphic to a unique von Neumann algebra called type III1 hyperfinite factor.
We are not going to prove this theorem, but let us (partially) explain the vocabulary:
1. A center Z of a von Neumann algebra R is defined as Z = R R0 .
2. A von Neumann algebra is called a factor if its center is trivial i.e. Z = CI.
3. A von Neumann algebra is called hyperfinite if it is a weak closure of an increasing sequence of finite dimensional algebras (W -inductive limit of finitedimensional algebras).
4. Classification of factors (Murray-von Neumann):
Def: Two projections P1 , P2 R are called equivalent (denoted P1
P2 ) if there exists a partial isometry V R from H1 = P1 H to H2 =
P2 H s.t.
P1 = V V,

P2 = V V .

(226)

Def: We say that P2 < P1 if the two projections are not equivalent, but
there exists a subspace H1,1 H1 whose projection P1,1 is in R and is
equivalent to P2 .
40

 Thm: Let R be a factor and P1 , P2 R projections. Then precisely

one of the following holds:
P1 > P2 ,

P1 P2 ,

P1 < P2 .

(227)

Thm: There exists a unique, up to normalization, dimension function

Dim( ) from projections in R to non-negative real numbers, s.t.
(a)
(b)
(c)
(d)
(e)

DimP1 = DimP2 if P1 P2 ,
DimP1 > DimP2 if P1 > P2 ,
DimP1 < DimP2 if P1 < P2 ,
If P1 P2 = 0 then Dim(P1 + P2 ) = DimP1 + DimP2 ,
Dim0 = 0.

Def: R is type I if (suitably normalized) Dim ranges through 0, 1, 2, . . . , n,

possibly n = . In this case one can construct a decomposition
H = H1 H2 s.t. R = B(H1 ) 1H2 .
Def: R is type II1 if (suitably normalized) Dim ranges through [0, 1].
Def: R is type II if Dim ranges through [0, ].
Def: R is type III if Dim takes only values 0 and . Then all proper
projections in R have infinite dimension and (for separable H) are all
equivalent.
Sub-index 1 in III1 comes from a finer classification of Connes which will not be
explained here. (See e.g. Chapter V of [8]).
There is a variant of Theorem 1.77 for general Haag-Kastler nets of von Neumann algebras, not necessarily coming from the Weyl algebra and free fields. Here
we give an imprecise formulation:
Theorem 1.78 Let O 7 A(O) be a Haag-Kastler net of von Neumann algebras
coming from a quantum field theory which has an ultraviolet fixed point and good
thermal properties. Then, for any open bounded O
A(O) ' R Z,

(228)

where R is the unique hyperfinite type III1 factor and Z is the center of A(O).
See [9] for a precise statement.
1.4.1

Interacting Weyl systems with infinitely many degrees of freedom

(Outline)

1. Consider the representation m of W. In Subsection 1.3.5 we constructed

a group of automorphisms t0 , implementing time-translations. (We add an
index zero in this subsection to distinguish it from interacting dynamics to
be defined below). That is
m (t0 (W )) = U0 (t)m (W )U0 (t)1 ,
41

f
W W.

(229)

f r ) then t (W )
These time translations acted locally, that is if W W(O
f r+|t| ).
W(O
2. We also had U0 (t) = eitH0 , where
Z
H0 = d(m (p)) =
dd k m (k)a (k)a(k)
Z


1
=
dd x : 2 (x) : + : 2 (x) : +m2 : 2 (x) : ,
2
(230)
and : ( ) : means Wick ordering (shifting creation operators to the left and
annihilation operators to the right, ignoring the commutators). For example
: (a (k1 )a (k2 ) + a (k1 )a(k2 ) + a(k1 )a (k2 ) + a(k1 )a(k2 )) :
= a (k1 )a (k2 ) + a (k1 )a(k2 ) + a (k2 )a(k1 ) + a(k1 )a(k2 ).
, will denote m , m in this subsection, that is:
Z


1
dd k
p
(x) =
eikx a (k) + eikx a(k) ,
d/2
(2)
2m (k)
r
Z


i
m (k) ikx
d
ikx
(x) =
d
k
e
a
(k)

e
a(k)
.
(2)d/2
2

(231)
(232)

(233)
(234)

3. We would like to construct a group of automorphism governed by the (formal)

Hamiltonian:
Z
H = H0 + HI , HI :=
dd x : (x)4 :
(235)
Rd

HI is a well defined quadratic form on D D, where

D = { fin (h) | (n) S(Rnd ) for all n }.

(236)

However, it does not come from a densely defined operator containing in

its domain. Two problems when computing HI :
Integration over whole space generates expressions involving (k1 + +
k4 ), which are thus not in L2 . (Infrared (IR) problem).
Decay of (m (k1 ) . . . m (k4 ))1 is too slow to get a vector in L2 for d > 1
(Ultraviolet (UV) problem).
4. Solution of the U V problem: set d = 1.
Solution of the IR problem: Consider a family of Hamiltonians
Z
H(g) = H0 + HI (g), HI (g) := dx g(x) : (x)4 :, g C0 (Rd )R (237)
R

Thm: H(g) are well-defined symmetric operators on D. Domains of essential self-adjointness are known. Each H(g) has a unique (up to phase,
normalized) ground state g
/ C. g tends weakly to zero when g 1.
42

5. We construct the dynamics as follows: For A A(Or ) we set

t (A) := eitH(g) AeitH(g)

(238)

Thm: t (A) is independent of g provided that g = 1 on Br+|t| . Then also

t (A) A(Or+|t| ). Thus t can be extended to a group of automorphisms
on A which respects the local structure. (PICTURE).
6. Ug (t) := eitH(g) does not implement t in the defining representation of A
(i.e. on Fock space) because g must be modified depending on t. It turns
out that t is not unitarily implemented in this representation (i.e. this is
not a vacuum representation of the theory).
7. We want to find a representation of A s.t. is unitarily implemented. Define
the family of states on A as follows
g (A) := hg , Ag i,

A A.

(239)

Although g itself tends weakly to zero as g 1, the states g have weak*

limit points (by Banach-Alaoglu theorem) which are non-zero simply because
g (1) = 1.
8. Thm: There is a limit point of g as g 1 s.t. the dynamics t is unitarily
implemented in the corresponding GNS representation ( , H , ) by a
group of unitaries U (t) = eitH . The physical Hamiltonian H is positive
and H = 0.
9. By additional work one can extend t 7 t to P+ 3 (
x, ) 7 (x,) s.t. the
resulting net (A, ) satisfies the Haag-Kastler axioms and is its vacuum
representation. Sp (H , P ) looks like in the free scalar QFT, (at least for
small ). However, it can be shown that the resulting theory (called (4 )2 ,
where 2 stands for the dimension of spacetime d + 1) is different from the
free scalar QFT: its scattering matrix (to be defined later) is non-trivial.
In particular there is non-trivial 2-body scattering. In this sense (4 )2 is
interacting.
10. The fact that the interacting time-translations t cannot be unitarily implemented in the defining representation of A (i.e. on the Fock space) can be
expected on general grounds: In fact suppose the opposite and let t 7 V (t)
be the group of unitaries on (h) implementing the interacting dynamics.
Then (t,
x) := V (t)(x)V (t) ,
(t, x) := V (t)(x)V (t) would define a
local relativistic quantum field (on Fock space). This gives a unitary equivalence between the free field , and the interacting field ,

at any fixed
time
(t,
x) := V (t)U0 (t) (t, x)U0 (t)V (t) ,

(t, x) := V (t)U0 (t) (t, x)U0 (t)V (t) .

43

(240)
(241)

By a general result (Haags theorem [13]) we have

h, (
x1 ) . . . (
xn )i = h, (
x1 ) . . . (
xn )i,

n = 1, 2, 3, 4.

(242)

But 4-point functions govern 2-particle scattering (via analytic continuation

to Green functions and the LSZ reduction formulae). So there would be no
2-particle scattering in the theory of the interacting field ,

. (By slightly
generalizing this discussion, one concludes that the representation above
cannot be unitarily equivalent to the defining representation of A).
(Somewhat imprecisely, Haags theorem means that there is no interaction
picture in local relativistic QFT).
11. Let us compare systems satisfying canonical commutation relations with
finitely and infinitely many degrees of freedom:
Finitely many degrees of freedom:
(a) By the von Neumann theorem only one (up to unitary equivalence) irreducible representation available (the Schrodinger representation) in which Q and P operators available.
(b) Only Hamiltonians quadratic in Q, P give rise to a dynamics (group
of automorphisms) on the Weyl algebra.
(c) To study non-quadratic Hamiltonians we have to change the algebra
(from Weyl to resolvent algebra).
(d) The interacting and the free dynamics on the resolvent algebra
unitarily implemented in the same (Schrodinger) representation.
That is, the interaction picture exists.
Infinitely many degrees of freedom:
(a) Due to the breakdown of the von Neumann uniqueness theorem,
there are many non-equivalent irreducible representations in which
and exist.
(b) Some Hamiltonians non-quadratic in , (e.g. (4 )2 ) give rise to a
dynamics on the Weyl algebra. (No urgent need to use the resolvent
algebra).
(c) To treat such non-quadratic Hamiltonians we have to consider unitarily non-equivalent representation of the Weyl algebra (Haags
theorem forces us to do it, breakdown of the von Neumann theorem makes it possible).
(d) The interacting and the free dynamics of the Weyl algebra unitarily
implemented in different (unitarily non-equivalent) representations.
That is the interaction picture does not exists.
(e) However, non-quadratic Hamiltonians giving rise to the dynamics
on the Weyl algebra which is compatible with the Haag-Kastler
axioms, only available for d = 1, 2. In the physically relevant case
d 3 there is again a no-go theorem [11].
44

12. Thus Weyl algebras are too narrow to describe interacting Haag-Kastler theories in physical spacetime. Many other approaches have been tried and are
still tried (see [12] for an overview), so far without success. (4 )4 is expected
to be trivial due to severe UV problems (Landau pole), similar problems
with QED. Promising candidates are non-abelian Yang-Mills theories due
to their mild UV properties (asymptotic freedom). But there are difficulties
in the IR regime (confinement of gluons). This question is a part of the
Yang-Mills and Mass-Gap Millenium Problem.
13. Recent progress in d 1 [14]: Take the Haag-Kastler net of v.N. algebras in
the vacuum representation describing the free field theory: O 7 A(O), H0 ,
P0 . Let E be the joint spectral measure of H0 , P0 . Let Q be an antisymmetric
matrix in Rd+1 (i.e. p Q
q =
q Q
p, where is Minkowski scalar product).
For A A(O), where O is contained in the right wedge W (PICTURE)
define
Z
(0)
AQ := dE(
p)Qp (A).
(243)
Not well defined as it stands, but one can make sense out of it as a bounded
operator, and as an observable localized in W . Similarly, for A0 A(O0 ), O0
contained in the left wedge, one defines
Z
(0)
0
AQ = dE(
p)Qp (A)
(244)
This can be interpreted as an observable localized in the left wedge W 0 . In
fact, we have
[AQ , A0Q ] = 0.

(245)

Let AQ (W ) be the v.N. algebra generalted by all AQ as above. We define

AQ (W 0 ) := AQ (W )0 , which is non-empty as it contains A0Q . We keep H0 , P0
as in the free theory. This gives a wedge-local, relativistic theory which
turns out to be interacting. (With two opposite wedges one can separate
two particles and define 2-body scattering matrix. It is non-trivial).
The expressions AQ , A0Q are called warped convolutions. They are closely
related to the concept of Rieffel deformations from non-commutative geometry.

Haag-Kastler theories

In this section we will consider a Haag-Kastler net of von Neumann algebras in a

vacuum representation. It is given by the following objects:
1. A net of von Neumann algebras O 7 A(O) B(H), labelled by open
bounded subsets O Rd+1 .
45

2. The global C -algebra of this net A =

ORd+1

A(O).

3. A unitary representation P+ 3 (
x, ) 7 U (
x, ) on H. (We will write
U (t, x) = U (
x) := U (
x, I)).
4. The group of automorphisms (x,) ( ) := U (
x, ) U (
x, ) of B(H).
These objects satisfy the following properties:
1. (isotony) O1 O2 A(O1 ) A(O2 ).
2. (locality) O1 X O2 [A(O1 ), A(O2 )] = 0, where X denotes spacelike separation.
3. (covariance) (x,) (A(O)) = A(O + x).
4. (irreducibility) A00 = B(H).
5. (spectrum condition) Sp(H, P ) V + , where U (t, x) = eiHtiP x , x = (t, x).
6. (uniqueness of the vacuum vector) There is a unique (up to phase) unit vector
H s.t. U (
x, ) = for all (
x, ) P+ .
Remark 2.1 We stress that in this section is not necessarily a Fock space vacuum, H is not necessarily a Fock space and A(O) may not come from the Weyl
algebra. Here we consider abstract Haag-Kastler nets and we will use only the
above properties, unless stated otherwise.
Remark 2.2 Note that the covariance property implies that leaves A invariant
and thus is a group of automorphisms of this subalgebra of B(H).

2.1

Haag-Ruelle scattering theory

In this section we assume:

Sp(H, P ) = {0} Hm G2m

(246)

as in the case of the massive free field or (4 )2 at small . We want to construct

vectors in H which describe an asymptotic configuration of several particles living
on Hm .
For example, for two particles with energy-momenta near p1 , p2 Hm we pick
1 , 2 in E(i )H, where i are small neighbourhoods of pi . We want to construct
vectors of the form
1 out 2 H,
1 in 2 H,

(247)
(248)

which describes two (Bosonic) particles which are independent at asymptotic times
t . Hence out should have properties of a symmetric tensor product, but
should take values in H not in H H.
46

How to do it, we know from our experience with the Fock space: we need to

A (in
, B2,t
construct in our general framework certain creation operators B1,t
this case time-dependent) s.t.

lim B1,t
= 1 ,

(249)

lim B2,t
= 2 .

(250)

Then

1 out 2 = lim B1,t

B2,t
,
t

1 in 2 =

lim B1,t
B2,t
.

(251)
(252)

The two-body scattering matrix is a map defined by

S2 (1 out 2 ) = 1 in 2

(253)

S2 = I means that there is no two body scattering (free field). S2 6= I means

that there is scattering. This is the situation in (4 )2 for example. Theories with
S2 6= I (or more generally with the full S-matrix S 6= I) are called interacting.
2.1.1

Energy-momentum transfer (Arveson spectrum)

To construct the creation operators mentioned in the previous section, we need to

control the energy-momentum (EM) transfer of the operators (to get vectors with
EM localized in small neighbourhoods of points on the mass hyperboloid).
Definition 2.3 The energy-momentum transfer (or Arveson spectrum) of B A,
denoted SpB , is defined as the support of the operator-valued distribution
Z
0
0
d+1

2
B(p , p) = (2)
dd xdt ei(p tpx) B(t, x),
(254)
where B(t, x) = (t,x) (B). Thus SpB is simply the support of the inverse Fourier
transform of (t, x) 7 (t,x) (B). More precisely, we can write
SpB =

)i
supp h, B(,

(255)

,H

The following theorem justifies the name EM transfer:

Theorem 2.4 Let Sp(H, P ) be a Borel subset. Then
BE() = E( + SpB )BE().

(256)

This will be called the EM transfer relation.

Before we list properties of the Arveson spectrum, we need the following fact:
47

Lemma 2.5 Let B A, f S(Rd+1 ). Then

Z
B(f ) = dtdx f (t, x)B(t, x),

(257)

defined as a weak integral (i.e. in the sense of matrix elements) is an element of

A. (Homework. It is important here that local algebras are von Neumann and we
can use the bicommutant theorem).
Lemma 2.6 Basic properties of the EM-transfer relation: (Here B A)
1. SpB = SpB .
2. SpB(t0 ,x0 ) = SpB for (t0 , x0 ) Rd+1 .
3. SpB(f ) supp fb, f S(Rd+1 ).
Proof. Part 1 follows from the relation
Z
0
0
d+1

(B )(p , p) = (2) 2
dd xdt ei(p tpx) B (t, x)

Z
0 tpx)
d
i(p
d+1
d xdt e
B(t, x)
= (2) 2


0

= (B)(p
, p) .

(258)

Part 2 follows from the change of variables

Z
d+1
0
0

B(t0 , x0 )(p , p) = (2) 2

dd xdt ei(p tpx) B(t + t0 , x + x0 )
Z
0
0
0
i(p0 t0 px0 )
d+1
= e
(2) 2
dd xdt ei(p (t+t )p(x+x )) B(t + t0 , x + x0 )
0 t0 px0 )

= ei(p

0 , p).
B(p

(259)

0 , p) because ei(p0 t0 px0 ) 6= 0.

This distribution has clearly the same support as B(p
Part 3 is a consequence of (259) and the following computation. Recall that
Z
B(f ) = dd x0 dt0 B(t0 , x0 )f (t0 , x0 )
(260)
Hence
)(p0 , p) =
B(f
=

dd x0 dt0 B(t0 , x0 )(p0 , p)f (t0 , x0 )

dd x0 dt0 ei(p

= (2)

d+1
2

0 t0 px0 )

0 , p)f (t0 , x0 )
B(p

0 , p).
fb(p0 , p)B(p


48

(261)

To construct creation operators of particles, one should thus pick some A A

and f S(Rd+1 ) s.t. supp fb is in a small neigbourhood p of some point p Hm
and set B = A(f ). Then, by the EM transfer relation, we have
B E(p)H.

(262)

However, there is another constraint on creation operators. We want particles to

be localized excitations. This means B should have good localization properties
(not just an element of A). A seemingly natural choice is to pick B strictly local,
that is
[
B
A(O) =: Aloc .
(263)
ORd+1

But it turns out that strictly local B cannot have compact SpB . We need a
larger class of operators, which is still not the whole A.
2.1.2

Almost local observables

Definition 2.7 We say that A A is almost local if there exists a sequence

Ar A(Or ) s.t.
kA Ar k = O(r ).

(264)

That is for any n N there is a constant cn s.t.

kA Ar k

cn
,
rn

r > 0.

(265)

We denote the -algebra of almost local observables by Aaloc .

For strictly local operators the commutator is equal to zero if we shift one of them
sufficiently far in space. For almost local operators we have:
Lemma 2.8 Let A1 , A2 Aaloc . Then
k[A1 , A(y)]k = O(|y| ),

y Rd .

(266)

Proof. By almost locality we find Ai Ai,r = O(r ), Ai,r A(Or ). Thus we

get
[A1 , A2 (y)] = [A1,r , A2,r (y)] + O(r ).

(267)

Setting r = |y| we obtain that for sufficiently small > 0 and |y| sufficiently large
Or and Or + y are spacelike separated and thus the first term o the r.h.s. above
is zero. 
The following theorem gives invariance properties of Aaloc :
Theorem 2.9 Let A Aaloc . Then
49

1. A(t, x) Aaloc for (t, x) Rd+1 ,

2. A(f ) Aaloc for f S(Rd+1 ).
Proof. As for part 1, we have
kA Ar k = O(r )

(268)

kA(t, x) Ar (t, x)k = O(r ).

(269)

hence

But Ar (t, x) A(Or + (t, x)) A(Or+|t|+|x| ). We set A(t, x)r0 = Ar0 |t||x| (t, x)
A(Or0 ) for r0 > |t| + |x|. For r0 |t| + |x| we can define A(t, x)r0 arbitrarily, e.g.
A(t, x)r0 = I.

References
[1] O. Bratteli, D.W. Robinson, Operator algebras and quantum statistical mechanics I, Springer 1987
[2] J. Derezi
nski, C. Gerard, Mathematics of quantization and quantum fields,
Cambridge University Press 2013.
[3] M. Fannes, A. Verbeure, On the time evolution automorphisms of the CCRalgebra for quantum mechanics. Commun. Math. Phys. 35, 257264 (1974).
[4] D. Buchholz, H. Grundling, The resolvent algebra: A new approach to canonical quantum systems. Journal of Functional Analysis 254, 27252779 (2008).
[5] D. Petz, An invitation to the algebra of canonical commutation relations. Leuven University Press (available online).
[6] J. Dixmier, C -algebras. North Holland Publishing Company, 1977.
[7] M. Reed, B. Simon, Methods of modern mathematical physics II: Fourier
analysis, self-adjointness. Academic Press, 1975.
[8] R. Haag, Local quantum physics. Springer 1996.
[9] D. Buchholz, C. DAntoni, K. Fredenhagen, The universal structure of local
algebras. Commun. Math. Phys. 111, 123135 (1987).
[10] L. Rosen, A 2n theory without cut-offs. Commun. Math. Phys. 16, 157183
(1970).
[11] K. Baumann, On relativistic irreducible quantum fields fulfilling CCR. J.
Math. Phys. 28, 697 (1987).
50

[12] S. Summers, A perspective on constructive quantum field theory.

arXiv:1203.3991
[13] R. Streater, A.S. Wightman, PCT, spin and statistics and all that. Princeton
University Press 2000.
[14] D. Buchholz, G. Lechner, S.J. Summers, Warped Convolutions, Rieffel Deformations and the Construction of Quantum Field Theories. Commun. Math.
Phys. 304, 95-123, (2011).
[15] M. Reed, B. Simon, Methods of modern mathematical physics III: Scattering
theory. Academic Press, 1975.

51