Defining physics at imaginary time: reflection

positivity for certain Riemannian manifolds

A thesis presented
by
Christian Coolidge Anderson
[email protected]
(978) 204-7656
to
the Department of Mathematics
in partial fulfillment of the requirements for an honors degree.
Advised by Professor Arthur Jaffe.

Harvard University
Cambridge, Massachusetts
March 2013

Contents
1 Introduction

2 Axiomatic quantum field theory

3 Definition of reflection positivity

4 Reflection positivity on a Riemannian manifold M

4.1 Function space E over M . . . . . . . . . . . . . . . . . . . . .
4.2 Reflection on M . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3 Reflection positive inner product on E+ E . . . . . . . . . .

7
7
10
11

5 The
5.1
5.2
5.3
5.4

.
.
.
.

12
13
14
16
17

6 Reflection positivity on the level of group representations

6.1 Weakened quantization condition . . . . . . . . . . . . . . . .
6.2 Symmetric local semigroups . . . . . . . . . . . . . . . . . . .
6.3 A unitary representation for Glor . . . . . . . . . . . . . . . .

17
18
19
20

7 Construction of reflection positive measures

7.1 Nuclear spaces . . . . . . . . . . . . . . . . . .
7.2 Construction of nuclear space over M . . . . .
7.3 Gaussian measures . . . . . . . . . . . . . . .
7.4 Construction of Gaussian measure . . . . . . .
7.5 OS axioms for the Gaussian measure . . . . .

22
23
24
27
28
30

Osterwalder-Schrader construction
Quantization of operators . . . . . . . .
Examples of quantizable operators . . . .
Quantization domains . . . . . . . . . .
The Hamiltonian . . . . . . . . . . . . .

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.
.

.
.
.
.

.
.
.
.
.

.
.
.
.

.
.
.
.
.

.
.
.
.

.
.
.
.
.

.
.
.
.

.
.
.
.
.

.
.
.
.

.
.
.
.
.

.
.
.
.

.
.
.
.
.

.
.
.
.

.
.
.
.
.

.
.
.
.
.

8 Reflection positivity for the Laplacian covariance

31

9 Reflection positivity for the Dirac covariance

9.1 Introduction to the Dirac operator . . . . . . . . . . . . . . .
9.2 Proof of reflection positivity . . . . . . . . . . . . . . . . . . .

34
35
38

10 Conclusion

40

11 Appendix A: Cited theorems

40

12 Acknowledgments

41

Introduction

Two concepts dominate contemporary physics: relativity and quantum mechanics. They unite to describe the physics of interacting particles, which
live in relativistic spacetime while exhibiting quantum behavior. A putative
theory of particles is referred to as a quantum field theory (QFT), and the
most famous example is Yang-Mills theory, which is the basis of the Standard
Model and accurately predicts the behavior of all observed particles. A QFT is
a calculational framework, and efforts have been made to deduce the essential
features of QFTs and express them as mathematical axioms. The Wightman
axioms, formulated in the 1950s, provide the most prominent example [38].
Although satisfactory axiom schemes have been known for sixty years, it remains an open problem to construct an axiomatic version of Yang-Mills or
any other physically plausible QFT [26]. Mathematical quantum field theories
have been constructed for two and three spacetime dimensions, but the case
of four spacetime dimensions (our physical universe) is still open. This is not
the first moment in history when physics outpaced the ability of mathematics
to describe it, and such moments herald growth for both fields. The successful
constructions in two and three dimensions, as well as the current efforts in four
dimensions, rely on constructing a theory of Euclidean fields and analytically
continuing it to imaginary time. This analytic continuation is the topic of the
present paper.
A mathematical QFT is based in the static Lorentzian manifold that is
spacetime. The time coordinate is distinguished from the local spatial coordinates by a minus sign in the metric, and the condition that the manifold is
static means that time-translation on the manifold is well-defined. The QFT
is then built around a unitary representation of the isometry group of this
manifold. In the prototypical case that the manifold is flat Minkowski space
the metric is
ds2 = dt2 + dx21 + dx22 + dx23 ,
and the isometry group is the Poincare group. For a general spacetime, if
we pass to imaginary time t 7 it, then the Lorentzian signature becomes
Euclidean, and we achieve a new manifold with a Euclidean symmetry group.
Arthur Wightman and others understood in the early 1950s that any QFT can
be analytically continued in this way to a theory of Euclidean fields, meaning a theory of generalized functions with Euclidean symmetries. Passing to
imaginary time is now a staple of the study of physicists QFTs and incites
traditional suspicion from first-year students of the topic.
This suggests the question: can we go in the other direction? Given a theory of Euclidean fields, what conditions are sufficient to analytically continue
it to a mathematical QFT? Edward Nelson gave one set of conditions in the
early 1970s that has not been checked for non-trivial examples [32]. Konrad
Osterwalder and Robert Schrader made the breakthrough in 1973 when they
discovered reflection positivity, which makes analytic continuation possible in
1

this context and in many others [34, 35]. Osterwalder and Schrader showed
that any Euclidean field theory equipped with a reflection positive bilinear
form can be analytically continued to a mathematical QFT. They proved that
the Wightman axioms follow from reflection positivity along with other conditions. This approach has been used to construct mathematical QFTs in two
[17] and three [8, 14] spacetime dimensions.
In the present paper, we define reflection positivity and use it to give the
following construction: Given a static Riemannian manifold M with isometry
group Isom(M ), there is a related Lorentzian manifold Mlor with isometry
group Isom(Mlor ). Under certain physically plausible conditions on M , we
show how to construct a reflection positive, unitary representation of Isom(M )
that can be analytically continued to a unitary representation of Isom(Mlor ).
In addition to being mathematically interesting, this construction is at the
heart of quantum field theory, and we show how it relates to the Wightman
axioms for a non-interacting bosonic particle. In support of this construction,
we construct various reflection positive bilinear forms on spaces over M .
We keep our treatment of reflection positivity abstract enough that its
broad utility can be appreciated. The introductory discussion in Sections 3 and
4 encompasses the usage of reflection positivity in contexts with no immediate
connection to mathematical QFT. These include the use of reflection positivity
by Arthur Jaffe et al. to construct representations of the Heisenberg algebra
on a Riemann surface [22], and the use of a reflection positive form by Vasily
Pestun to compute the partition function for supersymmetric Yang-Mills on
the four-sphere [36].1

Axiomatic quantum field theory

Physics is set in a static d-dimensional Lorentzian manifold Mlor called spacetime. The prototypical example is d-dimensional Minkowski space, which is
the setting for Einsteins special relativity. In general, the case d = 1 corresponds to non-relativistic quantum mechanics, and the case d = 4 is believed
to correspond to the physical universe. For purposes of quantum field theory,
it must be possible to decompose the spacetime manifold as R where R is
the time axis and the hypersurface is a spatial cross-section. This decomposition is necessary because time-evolution of quantum states is fundamental
to the development of a QFT.
Quantum field theories describe particles in this spacetime. The central
object of the mathematical theory is a Hilbert space H of physical states
equipped with a continuous unitary representation U of the isometry group of
spacetime Isom(Mlor ). In the case that Mlor is Minkowski space, this means
that H is equipped with a representation of the Poincaree group.
1
It is an original observation of the author that the bilinear form defined by equations
3.2 3.4 of the cited paper is reflection positive. The reflection, denoted as conjugation, is
induced by a reflection on S 4 .

In addition to the representation U , the Hilbert space is equipped with

a field operator . The field operator is an operator-valued distribution, so
it is a map from some space of test functions to End(H). Now and in the
rest of the paper, this space of test functions will be denoted H (Mlor ) with a
continuous dual H 0 (Mlor ) containing the distributions. The test functions are
some refinement of L2 (Mlor ). In the case that Mlor is Minkowski space, it is
common to take H = S, Schwartz space. A discussion of test function spaces
is presented in 4.1.
Quantum field theory is defined by requiring that the the Hilbert space H,
representation U , and field operator satisfy the axioms of Arthur Wightman
[16, 21, 38]. These axioms were originally formulated only in the case that
Mlor is Minkowski space, but we have broadened their phrasing to apply to
the more general setting of this paper:
Axioms 2.1 (Wightman).
1. The representation U is positive-energy, which
means that the generator p0 of the representation of time translation satisfies p0 0.
2. There exists an invariant vacuum vector = U H.
3. The field transforms covariantly under U . This means that for g
Isom(Mlor ), we have U (g)(f )U (g) = (g f ) where g f = f g 1 .
4. Vectors of the form (f1 ) . . . (fn ) for fj H and arbitrary n span
H.
5. The quantum field is local, meaning that if f and g have space-like
separated support then (f )(g) = (g)(f ).
6. The vacuum vector is the unique vector (up to scalar multiplication) in
H that is invariant under time translation.
The success of Osterwalder and Schrader was to formulate equivalent axioms for a theory of generalized functions with Euclidean symmetries. They
constructed a Hilbert space E equipped with a unitary representation of the
Euclidean group (the symmetries of Rd ) and with an operator-valued distribution . In the Euclidean picture, (f1 ) and (f2 ) are commuting or
anti-commuting operators for any choice of f1 , f2 . This is in contrast to the
quantum field , which has non-trivial commutation relations. In [34, 35], Osterwalder and Schrader give conditions such that their Euclidean formulation
is equivalent to the Wightman formulation. The crucial condition is reflection
positivity. In recent papers by A. Jaffe and G. Ritter, much of the original
construction is generalized to the setting where Rd is replaced with a static
Riemannian manifold M [23, 24, 25]. This is the point of view that we follow
in this exposition.
The commutation relations for the Euclidean field determine whether it
is a bosonic particle or fermionic particle. The developments of Osterwalder
3

and Schrader apply to both cases, but only the bosonic case is considered here.
In the Osterwalder and Schrader picture for bosonic particles, we begin with a
space of test functions H (M ) and construct a Borel probability measure d()
on its continuous dual H 0 (M ). Developing the theory of possible measures
has been one of the major contributions of constructive quantum field theory
to mathematics. In this paper, we will study the prototypical example of
Gaussian measures. The measure gives the Euclidean field and will be used to
produce the quantum field . A natural representation of Isom(M ) on H (M )
will be used to produce H and U . Osterwalder and Schrader gave axioms
for d such that this Euclidean theory is equivalent to a Wightman quantum
theory. In outline, these are:
Axioms 2.2 (Osterwalder-Schrader). Properties of d():
1. A regularity condition.
2. A clustering condition.
3. Euclidean covariance.
4. Reflection positivity.
These axioms will be discussed as needed in Section 4. Axioms 3 and 4,
which are critical to our exposition, are detailed as Axiom 4.4 and Axiom 4.12.
For the proof of equivalence with the Wightman axioms, which is not proven
in the present exposition, see [16, 34, 35].

Definition of reflection positivity

In this section, we define reflection positivity by giving two examples that

build up to Definition 3.4.
Reflection positivity is a condition for a space E equipped with a bilinear
form h, i. In the presence of reflection positivity, the structure of E can be
analytically continued to a new Hilbert space H with inner product a continuation of h, i. Our first example is the following proposition due to Fitzgerald
[9]:
Proposition 3.1. Assume f (z1 , z2 ) is analytic for |z1 | < 1 and |z2 | < 1, and
choose 0 < < 1. If for any sequence of real numbers s1 , . . . , sm in [0, ] the
matrix ajk = f (sj , sk ) is positive definite then for any sequence of real numbers
r1 , . . . , rn in [0, 1], the matrix bjk = f (i rj , i rk ) is positive definite.
Proof. Fitzgerald proves that the positivity of ajk leads to the positivity of
bjk = f (zj , zk ) for any choice of zi in the unit circle. Here the bar denotes
complex conjugation.

This proposition is essentially the converse of a reflection positivity argument: We begin with a function f that can be analytically continued. At the
cost of a minus sign, we are able to analytically continue a positivity condition for real parameters to a positivity condition for imaginary parameters.
In the setting of this paper, the positivity condition for imaginary parameters
will imply the existence of analytic continuation, and the minus sign will be
replaced with a more general involution referred to as reflection.
The following example due to A. Uhlmann [41] illustrates how to define
reflection-positivity in a setting with a more complicated reflection. Consider
the infinite-dimensional real vector space E of distributions given by
f (~r) =

N
X

qj (~r ~rj ).

(1)

j=1

A physicist can think of these as charge distributions with finite self-energy.

Define the following bilinear form on this space:
Z
f 0 (~r1 )f (~r2 )
0
.
hf , f i = d3~r1 d3~r2
k~r1 ~r2 k
This form gives the interaction energy of charge distributions. Define a unitary
involution on E as the pullback of a reflection on R3 :
(f )(x, y, z) = f (x, y, z).
We will refer to as a reflection on E.
Proposition 3.2. Let E+ = {f E : f (x, y, z) = 0 if x 0}. Then
b(f, f ) hf, f i 0

for all f E+ .

Proof. Suppose that f E+ . Then we have that

f (~r) =

N
X

qj (~r ~rj ),

j=1

where the first component of ~rj is greater than or equal to zero. We write the
components of ~r as x, y, z, so this is the assertion that xj 0 for all j. Now
we wish to prove:
X q` qm
0.
k~r` ~rm k
`,m
Fourier transforming 1/r to 1/k 2 and writing
a` = q` exp[i(ky y` + kz z` )]
5

lets us rewrite this inequality as

Z X
a` a
m k 2 exp[i(x` + xn )kx ]d3 k 0.
`,m

Let c(ky , kz ) = ky2 + kz2 . Then performing the integration with respect to kx
gives the left-hand side as
XZ
exp[c1 (x` + xm )]
dky dkz
am a
`
c
`,m
which is greater or equal to zero.
This shows that b(, ) is a semi-positive bilinear form on E+ . Let N denote
the space of vectors that have norm zero with respect to b. Then the following
lemma allows us to construct a Hilbert space as the completion of E+ /N :
Lemma 3.3. Let b be a semi-positive bilinear form on a vector space E+ , and
let N be the null space of b, then N is a linear space and b gives a positivedefinite bilinear form on E+ /N .
Proof. We will show that for u, v E+ and N N , b(u + N, v) = b(u, v). This
follows from the fact that the Cauchy-Schwarz inequality holds for b:
|b(u, v)| b(u, u)1/2 b(v, v)1/2 .

The above discussion illustrates the following characteristic properties of

reflection positivity:
Definition 3.4 (Reflection positivity). Reflection positivity constructions have
the following components:
1. E: A real or complex vector space E equipped with a Hermitian form
h, i. Although this form was positive definite in the previous example,
this need not be the case.
2. : An operator on E such that 2 = id and hf, gi = hf, gi.
3. E+ : A linear subspace E+ E with the property that for all f E+ ,
hf, f i 0.
These conditions mean that h, i is a reflection positive form on E. We can
then construct a related Hilbert space H by beginning with E+ and modding
out by the nullspace of b(, ) = h, i. The art of reflection positivity is to
construct additional structures on E that survive the construction of H. We
use this notation consistently throughout the paper.
6

Reflection positivity on a Riemannian manifold M

In the case of quantum field theory, we will equip E with a unitary representation of a Euclidean symmetry group and with a Euclidean field. These
structures will continue to exist on H, and the construction of H will cause
them to be analytically continued to quantum structures. To achieve this
structure on E, we will choose it to be a particular function space over M .
For the remainder of the present paper, unless stated otherwise, let M
denote a complete, connected, Riemannian manifold. For the purposes of
quantum field theory, we further demand that M be static in the sense of the
following definition:
Definition 4.1 (Static manifold). A manifold with metric (M, g ) is static if
it possesses a globally defined, hypersurface orthogonal Killing field . Physicists refer to this as the generator of time translation, and the global coordinate along this field is written t. Then the manifold M can be decomposed as
M = R . The metric can be written locally as
2

ds = F (x)dt +

dim
X

Gij (x)dxi dxj

i,j=1

where F and G depend only on the coordinates.

Any static Riemannian manifold M is embedded in a complex manifold with
Euclidean section (see [18, 7] for discussion) and has a Lorentzian continuation
Mlor with a metric that can be written locally as
2

ds = F (x)dt +

dim
X

Gij (x)dxi dxj .

i,j=1

It makes sense to discuss time evolution on such a manifold, so it is a

natural setting for physics. Examples of static manifolds include Minkowski
space, de Sitter space, and anti de Sitter space. The fact that a static manifold has an analytic continuation means that it is an ideal setting for defining
mathematical quantum field theory. We will use reflection positivity to analytically continue a representation of Isom(M ) to a representation of the identity
component of Isom(Mlor ).

4.1

Function space E over M

Let M be a static Riemannian manifold. The vector space C0 (M ) consists of

the smooth functions of compact support on M , and the vector space L2 (M )
consists of functions integrable under the Riemannian volume form. In the
present paper, we will take both spaces to be real. These spaces of functions
can be completed under various norms to give important function spaces on
7

M . For quantum field theory, we construct a space H (M ) of test functions

such that its continuous dual H 0 (M ) is equipped with a Borel probability
measure. This is easiest to accomplish in the case that H is a nuclear space.
Nuclear spaces will be defined and discussed in section 7.1, but an example is:
Example 4.2 (Schwartz space). Schwartz space for the manifold M = Rd
with the standard metric is denoted S(Rd ). It is the space of functions f C
such that for all , we have kf k, < where


kf k, = sup x D f (x) .
xRn

Schwartz space is a nuclear space. The continuous dual of Schwartz space,

called the tempered distributions, is denoted S 0 (Rd ).
Schwartz space can, in principle, be defined in any setting where there is a
notion of decay at infinity. An example of such generalized Schwartz functions
is Harish-Chandras Schwartz space over a semisimple Lie group [19, Section
9].
In 7.1, we give an original construction of a nuclear space H (M ) over
any manifold M where the spatial hypersurface is compact. The papers
[23, 24] assert that a convenient choice of nuclear space exists for any static
Riemannian manifold M , but there is an error in their construction.2 The
question of how to construct a suitable space of test functions for an arbitrary
spacetime M remains open.
The nuclear space of test functions H (M ) will contain L2 (M ) and C0 as
dense sets. It is a refinement of these spaces in order to better accommodate
distributions. We will describe the elements of H (M ) as functions and refer
to properties such as support that are only defined for true functions. These
ideas can be extended to H (M ) from C0 (M ). Throughout the current paper,
all spaces of test functions are taken to be real.
Given a function space H and its continuous dual H 0 , we write (f ) for
the pairing of f H and H 0 . Alternatively, we write this in terms of
an integral kernel:
Z
(f ) = hf, i =
(x)f (x)dx R.
M

Given the measure d() on H 0 , we can integrate functions in H as

Z
(f )d().
M

The Bochner-Minlos theorem (see Appendix A) asserts that d is equivalent

to its generating function:
Z
S{f } = ei(f ) d.
2

The papers assume that the embedding of Sobolev spaces is Hilbert-Schmidt for any
manifold M . The error was reported to the authors in March 2013.

We will follow J. Frohlich [10] in using the generating function to state the
important properties of d.
We define our Euclidean Hilbert space as the complex vector space
E L2 (H 0 , d).
We write P () = (f1 ) . . . (fn ) for the monomial in E given by the finite
sequence of functions fj H . Given certain regularity conditions on d, the
complex span of these monomials is a dense set in E. The space E then carries
an action of Isom(M ) in the following way:
Definition 4.3 (Induced operator ()). Let Isom(M ). This induces an
action on C0 (M ) by f = ( 1 ) f = f 1 . Let P () = (f1 ) (fn )
E be a monomial. Then we define
()P ( f1 ) ( fn ).
This operator () extends linearly to the dense domain of polynomials in E.
This defines a group representation for Isom(M ), and if 1 , 2 Isom(M )
commute then [(1 ), (2 )] = 0. For quantum field theory, we demand that
the measure d be such that this representation is unitary:
Axiom 4.4 (Euclidean invariance of measure). This is one of the OsterwalderSchrader axioms, see Axioms 2.2. We demand that d be such that (Isom(M ))
is a strongly continuous unitary representation. This is accomplished when the
generating function of the measure satisfies
S{f } = S{ f },
for all Isom(M ). Section 7 gives a recipe for constructing such measures.
In addition to this action of Isom(M ), E also carries a Euclidean field
operator that, via analytic continuation, will give the quantum field operator
of the Wightman axioms:
Definition 4.5 (Euclidean field operator). Consider the operator-valued distribution on E defined in the following way on monomials: Given f H , the
operator-valued distribution gives the operator
(f1 ) (fn ) 7 (f )(f1 ) (fn ).
This extends by linearity to a densely defined operator on E. In fact, polynomials in this field operator give a densely defined set in E by acting on the
constant function 1. Note that these operators commute for all f . This is
the heart of the Euclidean field theory for bosonic particles: it is a theory of
commuting fields.

In sections 5 8, we use reflection positivity to construct a quantum Hilbert

space H from E. Operators on E that satisfy suitable conditions can be quantized to give operators on H (see section 5.1). The Euclidean field operator
will quantize to give the quantum field. The action of the Euclidean group
Isom(M ) will quantize to give a strongly continuous unitary representation of
the identity component of Isom(Mlor ) on H.
For the constructions in the following sections, we use the following notion
of domains in E: For any open set O M , the corresponding domain in E is
denoted EO , and it is defined as the closure of


EO = span ei(f ) : f H (M ), supp(f ) O .
We have:
Remark 4.6. Let Isom(M ), and suppose |N : N O where N, O are
open sets in M . Let EN and EO be the corresponding domains. Then
()EN = EO .

4.2

Reflection on M

We will characterize the elements Isom(M ) that are a reflection:

Definition 4.7 (Reflection on M ). Given a complete, connected, Riemannian
manifold M , an isometry Isom(M ) is a reflection if there is some p in
the fixed-point set M such that dp is a hyperplane reflection in the tangent
space.
The theory of these reflections is developed in [1] including the following
result:
Proposition 4.8. Given a reflection on complete connected Riemannian
manifold M , we have that the fixed point set M is a disjoint union of totally
geodesic submanifolds, including at least one submanifold of codimension one.
Proof. Let p M be a point such that Tp is a reflection. Then Tp Tp =
idTp M . Thus is an involution. It follows that Tx is a Euclidean involution for
each x M , and thus it is diagonalizable with eigenvalue 1 on the eigenspace
Tx N and 1 on the eigenspace Tx N where N is the connected component of
M containing x.
The connected component of M that contains p has codimension one.
Any codimension one component of M is called a reflection hypersurface. The image on the top of the next page is from [1] and illustrates examples
of reflections generated by multiple hypersurfaces. Note that this only possible for manifolds that are not simply connected. The manifold on the left of
the image is S 1 , not to be confused with the disk. The image is followed by
examples of reflections in the sense of Definition 4.7.
10

Example 4.9 (Static manifold). Suppose that M is static in the sense of

definition 4.1. Fix a particular hypersurface to which the global Killing field
is orthogonal. Let t be the one-parameter group of isometries determined
by . Define t : M R by setting t = 0 on and otherwise defining t(p) = T
such that T (x) = p for some x . Define to map p M to the unique
point on the same -trajectory with t((p)) = t(p). Note that none of the
constructions in this paper will depend on the arbitrary choice of .
A specific example is the case that M = Rd with coordinates (t, ~x), and is
reflection in the plane = M = {(t, ~x) : t = 0}. The plane is a reflection
hypersurface.
The above is the physically significant class of examples. The following,
though, has the mathematical merit that it works on Riemann surfaces (which
do not admit Killing fields):
Example 4.10 (Schottky double). Let S denote a compact Riemannian manifold which arises as a Schottky double of a bordered Riemannian manifold T
with boundary T . The Schottky double is equipped with an antiholomorphic
involution from T to its mirror image T. This involution is a reflection in
the sense of Definition 4.7. This reflection is considered in detail in [22]. A
specific example is the case that M is the Riemann sphere, (z) = 1/
z , and
= M = {z : |z| = 1} is the unit circle.
Maintaining our focus on static manifolds, we adopt the following definition
for the remainder of the paper:
Definition 4.11 (Quantizable manifold). A complete, connected, Riemannian
manifold M is called quantizable if it is static and equipped with a reflection
in the sense of Example 4.9. Such a manifold is decomposed as
M = t t +
where is the t = 0 hypersurface. The manifold M is equipped with a reflection
that fixes and exchanges + and .

4.3

Reflection positive inner product on E+ E

Let H be a space of test functions over a quantizable manifold and let H 0

be its dual equipped with the measure d. We have defined E = L2 (H 0 , d).
11

A transformation Isom(M ) acts on E via (). Let be the reflection of

definition 4.11. By abuse of notation, we let = () denote the action of
the reflection on E. We will use this reflection to define reflection positivity.
Axiom 4.12 (Reflection positivity). This is one of the Osterwalder-Schrader
axioms, see Axioms 2.2. Define E E to be equal to E in the sense
previously defined. We say that h, iE is reflection positive when
hA, AiE 0

for all A E+ .

Two equivalent definitions are the following:

1. If + : E E+ is the canonical projection, then
+ + 0,
as an operator on E.
2. If S{f } is the generating function of the measure d, then
0

n
X

ci cj S{fi fj },

i,j=1

for every finite sequence cj C and fj supported in + .

The Osterwalder-Schrader construction

In this section and section 6, we assume the existence of the Euclidean space E
defined in section 4.1. We show how to construct E for a quantizable manifold
M in sections 7 and 8. In this section, we give the analytic continuation of
the Euclidean structure of E to Lorentzian structure. This construction is
originally due to Osterwalder and Schrader [34, 35], and our treatment most
closely follows that of A. Jaffe [16, 23].
We begin with a Hilbert space E equipped with the inner product h, i and
the action of the isometry group Isom(M ). Recall that there exists a subspace
E+ such that hu, ui 0 for u E+ . We define a bilinear form h, iH on E+
by
hu, viH = hu, vi

for

u, v E+ .

By the self-adjointness of on E, this is a sesquilinear form:

Z
Z
hA, BiH = ABd = ABd = hB, AiH .
Let N denote the kernel of h, iH .

12

(2)

(3)

Definition 5.1 (Quantum Hilbert space). Let H denote the completion of

E+ /N with respect to the inner product h, iH . Let : E+ H denote the
natural quotient map, called the quantization map. We have an exact sequence:

0 N , E+
H 0.
The space H is the Hilbert space of quantum states. To each vector u E+ ,
there corresponds a quantum state (u) u.

5.1

Quantization of operators

Assume that T is a densely defined, closable operator on E. We give a condition

on T such that it induces a well-defined operator T on H.
Proposition 5.2 (Condition for quantization). Let T + = T . Assume
that there exists a domain D+ Dom(T ) Dom(T + ) E+ such that
T : D+ E+

and

T + : D+ E+ .

Assume that the projection (D+ ) is dense in H. Then T has a quantization

T : H H defined by the commutative diagram:
0

/N


/N

/ E+


/ E+

/H

/0


/H

/0

The adjoint of T is given by

T = (T + ).

(4)

Proof. Suppose u N D+ . Let S E+ denote a set of vectors in the domain

of T such that the image of this set under is dense in H. Then
0 = h(T S), uiH = hT S, uiE = hS, T uiE = h(S), (T u)iH .
Thus T u N and hence T is well-defined on D+ /(D+ N ). This implies that
T is well-defined on (D+ ).
For u, v (D+ ), let u, v E+ denote representatives in the preimage of
the projection. we have that


h
u, TviH = hu, T viE = hT u, viE = (T )u, v H
as desired. The operators T and (T ) extend uniquely to H from the dense
subset (D+ ).

13

Proposition 5.3 (Contraction property). Let T be a bounded operator on E

such that T and T + = T preserve E+ . Then


(5)
T kT kE .
H

Proof. We will prove that for all u H, Tu kT kE k
ukH . We use the
H
following lemma:
Lemma 5.4. For all u H with preimage u E, k
ukH kukE .
Proof. By the Cauchy-Schwarz inequality for E:
k
uk2H = hu, uiE kukE kukE = kuk2E .

Using the Cauchy-Schwarz inequality for H:

1/2

1/2
1/2

u, T T uiH k
ukH T T u .
T u = hT u, T uiH = h
H

Iterating the application of the Cauchy-Schwarz inequality gives that

2n
n

T)2n1 u
uk12
(
T
T u k


H
H
H

2n
n
n1


+
2
[(T
T
)
= k
uk12
u]


H
H


n
2
n
n1
k
uk12
(T + T )2 u
H
E


2n
12n + 1/2
T T E kukE
k
ukH
n

= k
uk12
H

kT kE kuk2E

which gives the desired result.

5.2

Examples of quantizable operators

In this section, we describe two classes of operators that satisfy the quantization condition of Proposition 5.2. The first class are particular unitary
operators that quantize to self-adjoint operators:
Proposition 5.5 (Unitary to self-adjoint). Let U be a unitary operator on E
that preserves E+ . If U 1 = U , then U admits a quantization U and U is
self-adjoint.
Proof. The operator U = U preserves E+ , so Proposition implies that U
is well-defined. Self-adjointness follows from Equation 4: T = (T + ).

14

The second class is particular unitary operators that quantize to unitary

operators:
Proposition 5.6 (Unitary to unitary). Let U be a unitary operator on E such
that U and U 1 preserve E+ . If [U, ] = 0, then U admits a quantization U
and U is unitary.
Proof. By assumption, the operator U = U 1 preserves E+ . Thus U has
a quantization. Similarly, U 1 has a quantization. We have that (U 1 ) is the
inverse of U , and Equation 4 implies that U = U 1 .
The first class of operators, unitary operators U such that U 1 = U ,
arise from the reflected isometries on M .
Definition 5.7 (Reflected isometry). An isometry Isom(M ) is reflected
if 1 = . If preserves + , then () preserves E+ . Then, Proposition

5.5 implies that ()

exists and is self-adjoint. If is reflected then so is 1 ,
1

and ( ) is the inverse of ().

The set of all such isometries is denoted GR . This set is closed under
inverses and contains the identity. It is not closed under multiplication.
Unitary operators U such that [U, ] = 0 arise from reflection-invariant
isometries on M :
Definition 5.8 (Reflection-invariant isometry). A reflection-invariant isometry is an isometry Isom(M ) that commutes with Isom(M ). It follows
that [(), ] = 0. If and 1 preserve + , then () and ( 1 ) preserve

E+ . Then, Proposition 5.6 implies that ()

exists and is unitary.
The set of all such isometries is denoted GRI . It is the stabilizer of a Z2
action and thus a subgroup of Isom(M ). We have that
GR GRI = {id, } GR GRI 6= Isom(M ).
The reflection-invariant and reflected isometries describes the structure of
Isom(M ) in a powerful way:
Proposition 5.9. Let G0 denote the connected component of the identity in
Isom(M ). The group G0 is algebraically generated by GR GRI .
This is a proof on the level of Lie algebras that follows from a Cartan
decomposition. Let g = Lie(Isom(M )) be the Lie algebra of Killing fields.
The reflection on M acts on these Killing fields by push forward: X 7
X = (1 ) X . This is a Lie algebra homomorphism that squares to the
identity. Therefore g can be decomposed as a vector space into g+ g where
is the identity on g+ and gives multiplication by 1 on g . By the fact
that is a Lie algebra homomorphism, we have that
[g+ , g+ ] g+ ,

[g+ , g ] g ,
15

[g , g ] g+ .

The vector fields in g+ are exactly the generators of reflection-invariant isometries, and g+ is the Lie algebra of the Lie group GRI . Likewise, vector fields
in g are the generators of the reflected isometries, although g is not a Lie
algebra.
To prove that GR GRI algebraically generates G0 , recall that exp(g) algebraically generates G0 . The result follows from the vector space decomposition
g = g+ g .

5.3

Quantization domains

The previous section leaves open the following issue: Suppose that Isom(M )
is, for instance, reflected so that 1 = . It is possible that the intersection of E+ and the preimage of E+ under is a proper subset O E+ such
that (O) is not dense in H. This issue of domains is an obstacle to quantizing otherwise well-behaved operators, and it is the major issue that we will
confront in section 6.
As an aside, in this section, we introduce the topic of quantization domains. A quantization domain is a set + such that (E ) is dense in H.
If the set O of the previous paragraph could be proved to correspond to E for
some quantization domain , then the issue would be resolved. The problem
of classifying quantization domains is believed to be open, and quantization
domains are the subject of current research [20]. The following theorem due
to A. Jaffe and G. Ritter [23] uses reflected and reflection-invariant operators
to find quantization domains:
Theorem 5.10. Let Isom(M ) be either reflected or reflection-invariant.
Let = (+ ) and suppose that + . Then is a quantization domain.
Proof. By remark 4.6, we have that E = ()E+ .
We will prove that the orthogonal complement in H of (E ) is zero, which
proves that (E ) is dense in H. Let u ((E )) with preimage u E+ .
Let v E+ . Then
0 = h
u, (()v)iH = hu, ()viE .
We have that ()1 = ( 1 ) is unitary, so
0 = h( 1 )u, viE .
First, suppose that is reflection-invariant, i.e. [( 1 ), ] = 0. This

implies by Proposition 5.6 that ()

is unitary. Furthermore
D
E
1 )
0 = h( 1 )u, viE = h( 1 )u, viE = (
u, v .
H

1 ).
Because this statement holds for all v E+ , this implies that u ker (

Because ()
is unitary, this implies u = 0, as desired.
16

Second, suppose that is reflected, i.e. () = ( 1 ). We know from

Proposition 5.5 that ()

exists and is self-adjoint on H. Similarly to the

above, let u ker (). Now, let {s : s R} be a strongly continuous one s ) is a

parameter semigroup of isometries such that = t . Then R(s) = (
semigroup, self-adjoint, a contraction, and strongly continuous. The last two
properties follow from Proposition 5.3. By Stones theorem (see Appendix A),
there exists self-adjoint K such that s = esK . Evidently, s then has zero
kernel.
Example 5.11 (Positive-time half-space). In the case that M = Rd and T
R, then
= {(t, ~x) Rd : t > T }
is a quantization domain.

5.4

The Hamiltonian

The quantization condition of proposition 5.5 enables a beautiful quantization

for the time-translation isometry on a quantizable manifold M :
Theorem 5.12. Recall that M is a static spacetime Riemannian manifold.

Let t
be the Killing field on M that gives time-translation. Let t be the
corresponding one-parameter group of isometries. For t 0, T (t) = (t ) has
a quantization R(t). This is a strongly continuous one-parameter semigroup of
self-adjoint contraction operators on H. The semigroup R(t) leaves invariant
where 1 is the constant function on M . Thus there exists a
the vector 0 = 1
densely defined, positive, self-adjoint operator H such that
R(t) = exp(tH)

and

H0 = 0.

Proof. For t 0, T (t) is a reflected isometry such that T (t)E+ E+ . Thus

R(t) = T(t) is a self-adjoint semigroup on its domain of definition in H. The
contraction property of R(t) follows from Proposition 5.3. The group T (t)
is strongly continuous. From this and the contraction property, the strong
continuity of R(t) follows. It is evident that R(t)0 = 0 . The proof then
follows from Stones theorem (see Appendix A).
This is the Hamiltonian H and ground-state 0 of quantum field theory,
as required by the Wightman Axioms (Axioms 2.1). The Hamiltonian gives
time-evolution of the physical states in H and satisfies the physically crucial
positivity condition H 0.

Reflection positivity on the level of group

representations

We have seen in the previous section that the one-parameter semi-group corresponding to time translation has a representation on the Hilbert space H of
17

physical states. For quantum field theory, H must be equipped with a unitary representation of the full symmetry group Glor of Lorentzian spacetime.
Reflection positivity allows the representation of G = Isom(M ) on E to be analytically continued to a representation of Glor on H. We give the construction
in this section. Recall that any static Riemannian manifold M can be analytically continued to a static Lorentzian manifold Mlor . Let Glor = Isom(Mlor ).
We will prove that the Osterwalder-Schrader construction extends to give a
representation of the identity component G0lor of this Lie group on H.
Let {i : 1 i n} be a basis for g. We will quantize each of the
semigroups (i ()) given by exponentiation of the i . This produces n oneparameter families of operators. We will prove that each of these families
is a one-parameter unitary group Ui () on H, and that these groups give a
representation of the action of G0lor . The crux of this argument is performing
the quantization. Reflection-invariant and reflected isometries will still be at
the heart of the discussion, but it is not generally possible, for the (i ()) to
satisfy the quantization condition given by Proposition 5.2. In particular, it
is not generally true that the quantizations have dense domain of definition in
H. For this reason, we cannot apply Stones theorem, as we did in Theorem
5.12, to prove that the Ui () are unitary groups. Instead, we use the theory
of symmetric local semigroups developed in [11, 30].

6.1

Weakened quantization condition

We again consider the quantization of an unbounded linear operator T on E

with partner operator T + = T .
The quantization condition of Proposition 5.2 demands that there exists
D+ (E+ Dom(T ) Dom(T + )) such that (D+ ) is dense in H and D+ is
sent into E+ by T and T + . We can weaken this condition in the following way
(due to Jaffe and Ritter):
Definition 6.1 (Quantization Condition II). The operator T satisfies
1. The domains Dom(T ) and Dom(T ) are dense in E.
2. There is a set D+ E+ that is in the intersection of the domains of T ,
T + , T + T , and T T + .
3. Each of those four operators, T , T + , T + T , and T T + maps D+ into E+ .
Proposition 6.2 (Quantization II). Given T satisfying Quantization Condition II, we have
1. The operators T |D+ and T + |D+ have quantizations with domain (D+ ).
2. For u, v D+ , we have h
u, TviH = h(T + )u, vi.

18

Proof. Let f D+ N where N is the null space of h, iE . We will prove

that (T f ) = (T + f ) = 0. This will prove that T and T + have quantizations
with domain (D+ ). We have that
h(T f ), (T f )iH = hT f, T f iE = hf, T T f iE = h(f ), (T + T f )iH .
Where we have used the fact that T + T maps D+ to E+ . The Cauchy-Schwarz
inequality on H gives the result. The proof for T + is identical and uses the
fact that T T + maps D+ to E+ .
Now suppose f, g D+ Then
hf, TgiH = hf, T giE = hT f, giE = h(T + )f, giH ,
as desired.
In the case that (T + ) = T, then T is symmetric:
Definition 6.3 (Symmetric operator). The operator T on H is symmetric
when for all u, v in Dom(T), we have
h
u, TviH = hTu, viH .
In the case that T is symmetric, this weak quantization condition is sufficient to apply the theory of symmetric local semigroups.

6.2

Symmetric local semigroups

The theory of symmetric local semigroups was simultaneously developed by

A. Klein and L. Landau in [30] and by J. Frohlich in [11]. Shortly thereafter,
these same groups of authors combined their result with Osterwalder and
Schraders reflection positivity to analytically continue a representation of the
Euclidean group to a representation of the Poincare group [12, 31]. In the
current paper, we are presenting a generalization of their result. We recount
the theory of symmetric local semigroups using the notation of Klein and
Landau:
Definition 6.4. A symmetric local semigroup on a Hilbert space H consists of
(S(), D , T ). The parameter T is a real number T > 0. For each [0, T ],
S() is a symmetric linear operator with domain Dom(S ) = D H. These
objects must satisfy the following properties:
1. D D if and D = [0,T ] D is dense in H.
2. S is weakly continuous.
3. S0 = I and S (D ) D for 0 T .
4. S S = S+ on D+ for , , + [0, T ].
19

This is a relaxation of the situation studied by Nussbaum in which the

semigroup operators S are densely defined and there is a common dense
domain on which the semigroup property holds (see [33]).
Then we have the theorem:
Theorem 6.5. For a symmetric local semigroup (S , D , T ), there exists a
unique self-adjoint operator A such that
D Dom(eA )

and

S = eA |D

for all [0, T ].

Remark 6.6. The authors who prove this result also prove that
"
#
[
[

D
S (D ) , where 0 < S T,
0<

0<<

is essentially self-adjoint.
is a core for A, i.e. (A, D)

6.3

A unitary representation for Glor

Recall that the Lie algebra g = Lie(G) can be decomposed as a vector space
into g+ g where the Killing fields in g+ generate reflection-invariant isometries and the Killing fields in g generate reflected isometries. We use the
theorem of previous section to prove that:
Theorem 6.7. Let be a Killing filed in g+
parameter isometry group { }. There exists
operator A on H such that
(
A
if
) = e
(
eiA if

or g that generates the onea densely defined, self-adjoint

g ,
g+ .

Proof. For the first case, let g . Then each isometry is reflected.
Define 1
(+ ). We have that 0 is the identity map, so 0 = + . The
continuity of (x) with respect to implies that for in some neighborhood
of zero, is a nonempty open subset of + . As 0 from above,
increases to fill + .
By remark 4.6, we have that ( )E E+ . Then, by Proposition 6.2,
) with domain D (E ). By the fact
( ) quantizes to an operator (
) is symmetric.
that is reflected, (
Fix > 0. Then
[
[
= +
E = E+ .
0<<

0<<

20

It follows that D = 0<< D is dense in H. It is routine to verify that

), D , ) satisfies Definition 6.4 for a symmetric local semigroup. Theo((
rem 6.5 then gives the desired result.
For the second case, let g+ . Then each isometry is reflectioninvariant, and we have that ( )+ = ( ) on E. We claim that ( )E+
E+ . Suppose that this is true. Then, ( ) satisfies Quantization Condition I,
) is defined on (E+ ) which is dense in H
given by Proposition 5.2. Then (

be definition. We have that ( ) extends by continuity to a one-parameter

unitary group on H. By Stones theorem (see Appendix A), there exists a
self-adjoint operator A on H such that
) = exp(iA).
(
To complete the proof in this case, we simply need to prove that ( )E+ E+ .
We will prove that any reflection-invariant isometry preserves and either
preserves or exchanges + and . The claim then follows from the fact that
is in the identity component of G.
Suppose that p and, WLOG, (p) + . Then, by reflectioninvariance, + contains ()(p) = ()(p) , which is a contradiction.
Thus M restricts to an isometry on and an isometry on + t . This gives
the desired result.
Now let {i+ : 1 i n+ } be a basis of g+ and let {i : 1 i n } be
a basis of g . Let A
i denote the corresponding densely defined self-adjoint
operators on H constructed in the previous theorem. We define G0lor as the
group generated by the one-parameter unitary groups
Ui () = exp(iA
i ),
and we claim that it is isomorphic to the identity component of Glor = Isom(Mlor ).
Because this claim only concerns the identity component, it can be checked
on the level of Lie algebras.
If g generates , then let () denote the generator of ( ). We
have that ([X, Y ]) = [(X), (Y )], so we have that gives a Lie group

). By the above
homomorphism. Then let ()
denote the generator of (
+
+

analysis, we have that (

j ) = iAj and (j ) = Aj . Then define the two
Lie algebras:

g {(X)
: X g }.
We have that g+ ig is a Lie algebra represented by skew-symmetric operators
on H. Then our claim is:
Theorem 6.8. Let glor be the Lie algebra of the connected component of
Isom(Mlor ). There is an isomorphim of Lie algebras
glor
= g+ ig .
21

Proof. Given coordinates x on M , let MC denote the manifold obtained by

allowing the x0 coordinate to take values in C. This manifold is equipped with
the involution J : MC MC given by x0 7 ix0 . Denote the induced map
on Lie algebras by J . The Lie algebra glor is generated by
{j+ }1jn+ {k }1kn ,

where

j iJ (j ).

We compute the commutation relations of this algebra. Let fijk be the set of
real structure constants such that
[i , j ]

n+
X

fijk k+ .

k=1

Applying J to both sides gives

[i , j ] =

n+
X

fijk k+ .

k=1

These, together with the inherited relations for g+ are precisely the commutation relations of g+ ig , which proves the result.

Construction of reflection positive measures

The constructions of the previous sections are the achievement of reflection

positivity. The remainder of this exposition will focus on constructing the reflection positive forms that enable these arguments to succeed. In this section,
we will construct a family of reflection positive measures d that give reflection
positive forms on E L2 (H 0 , d).
The situation is that we begin with the Hilbert space L2 (M ) and introduce
test functions in order to give it the structure of a rigged Hilbert space:
H (M ) L2 (M ) H 0 (M ),

(6)

where H (M ) is a nuclear space and H 0 (M ) is its continuous dual. These

ideas will be defined in section 7.1. In section 7.2, we construct this rigging
for a wide class of quantizable manifolds M . This construction replaces the
incorrect construction in [23, 24].
The nuclear structure of H (M ) allows us to construct the measure d on
H 0 . This measure must satisfy the Euclidean invariance condition of axiom
4.4 and the reflection positivity condition of axiom 4.12. As a sequence of
propositions over the course of sections 7.4 and 7.5, we prove the following
theorem:
Theorem 7.1. Let C be a positive, continuous, nondegenerate bilinear form
on the nuclear function space H (M ). Then there exists a unique Gaussian
measure dC on H 0 (M ) with covariance C and mean zero. This Gaussian
22

measure is Euclidean invariant if and only if C is Euclidean invariant in the

sense that for Isom(M ) and f H ,
h f, C f iL2 (M ) = hf, Cf iL2 (M ) .
This Gaussian measure satisfies reflection positivity if and only if C is reflection positive in the sense that
hf, Cf iL2 (M ) 0,
for functions f H supported at positive times.
This pushes the issue of reflection positive to bilinear forms on L2 (M ). An
example of a Euclidean invariant, reflection positive form C is constructed in
section 8.

7.1

Nuclear spaces

We will say what it means for a topological vector space to be nuclear. Several
slightly different definitions abound in the literature. Our definition follows
[13, Chapter 1].
Definition 7.2. Let H be a topological vector space equipped with a countable
family of inner products denoted h, in for n = 1, 2, . . .. Suppose that the inner
products give the topology of H in the sense that a neighborhood basis of zero
is given by the sets Un, = {f H : kf kn < }. Let Hn denote the Hilbert
space given by the completion of H under the nth inner product. The space
H is complete relative to the aforementioned topology if and only if it can be
written
\
H =
Hn .
n

In this case, H is called a countably Hilbert space. Suppose furthermore

that
for all f H .

kf kn kf kn+1

If this condition does not hold, then the inner products can be redefined as
hf, gi0n

n
X

hf, gii .

i=1

Then the embedding Hn , Hn+1 is a continuous map from an everywhere

dense set to an everywhere dense set. Extend it to a continuous linear map
n
Tn+1
: Hn+1 Hn . We define Tmn to be the map achieved in this way for

23

m > n. It is well-defined. The space H is nuclear if the following condition

holds: for all m, there exists n > m such that Tmn can be written
Tmn f

k hf, uk ivk

for all f Hm ,

i=1

where {uk } and {vk }P

are orthonormal systems of vectors in Hm and Hn respectively, k > 0, and k k < . This is the condition that Tmn is a nuclear
operator. In the case that H is a Hilbert space, we instead say that Tmn is a
trace class operator.
Remark 7.3. For a Hilbert space H , some authors substitute the condition
that for all m, there exists n such that Tmn is a Hilbert-Schmidt operator.
This is essentially the same definition, because every trace class operator is
Hilbert-Schmidt, and the composition of two Hilbert-Schmidt operators is trace
class.
Proposition 7.4. Give a countably Hilbert space and, in particular, a nuclear
space
\
H =
Hn ,
n

then its continuous dual H 0 is

[

Hn0
=

Hn .

Proof. The isomorphism

from the fact that each Hi is a Hilbert space.
S comes
0
0
We prove that H = Hn . Let Hn , then by construction is continuous
with respect to the topology of H, so H 0 . Given H 0 , the fact that
is continuous on H implies that there exists  and n such that is bounded in
the sphere kf kn . Then is continuous with respect to the norm kkn .

7.2

Construction of nuclear space over M

Let M be a quantizable manifold in the sense of definition 4.11, and suppose

that it satisfies the additional constraint that the spatial hypersurface is
compact. Then, we can construct the following nuclear space of test functions
over M . Note that for transparency of the construction, the indexing differs
from the above definition, but it is equivalent.
Proposition 7.5. Let M be the Laplace-Beltrami operator on L2 (M ). Let
Qt denote the operator on L2 (M ) that gives multiplication by the global time
coordinate. Then let
H=

1
1
M + Q2t + .
2
2
24

This is a hybrid between the Laplacian and the Harmonic oscillator operator.
For integer n, define the following inner product on L2 (M ):
hf, gin = hf, (H + I)n gi.
Let Hn be the Hilbert space achieved by completion under h, in . Then

Hn

n=

is a nuclear space.
Proof. The Laplace-Beltrami operator on M can be written as
= + t2 ,
where is the Laplace-Beltrami operator on .
The compactness of gives us that the positive operator has a countable set of eigenfunctions gj that give an orthonormal basis for the Hilbert
space L2 (). Let j be the corresponding eigenvalues with multiplicity. Arrange the indices such that j j+1 for all j.
The single-dimensional Harmonic oscillator operator 21 (t2 + t2 ) 12 has
the normalized Hermite polynomials as eigenfunctions. Denoted h0 , h1 , . . .,
these Hermite functions are an orthonormal basis for the Hilbert space L2 (R)
with corresponding eigenvalues 0, 1, . . ..
2 (R) where
denotes the completed Tensor
Note that L2 (M ) = L2 ()L
product. Then the operator H can be written as
1
1
H = id + id (t2 + t2 + 1).
2
2
By the previous discussion, H has eigenvalues gj hk , and they give an orthonormal basis for the Hilbert space L2 (M ).
We claim that for any > 0, (H + I)[(d+1)/2+] is trace class. We use
Weyls asymptotic formula [5, Page 172], which implies that
k const k 2/(dim ) = const k 2/(d1) ,
and we use the inequality

X
(a + j)b const a(b1) ,
j=0

valid for 1 < a, 1 < b to successively bound the asymptotic behavior of the
two sums
X


 X
tr (H + I)[(d+1)/2+] =
(j + k + 1)[(d+1)/2+] .
j=0 k=0

25

(Note that d = 2 is a special case in which we only apply the inequality once.)
For n Z, define the inner product h, in = h, (H + I)n iL2 (M ) . Let
Hn be the Hilbert space that comes from completion of the nth inner product.
Because 0 H, kkn kkn+1 , so there is an injection inn+1 : Hn+1 , Hn .
Note that (H + I)1/2 inn+1 : Hn+1 , Hn is a unitary map. Then,
Lemma 7.6. For D > (d + 1), the canonical injection
n+D1
inn+D = inn+1 in+D
: Hn+D , Hn

is trace class.
Proof. Write


inn+D = (H + I)D/2 (H + I)D/2 inn+D .
The second term is unitary and the first term is trace class. This gives the
result.
The proposition then follows follows from our previous definition of a nuclear space.
In the case that M = Rd , then it is possible to instead use H = S(Rd ).
This space has a very similar nuclear structure to the structure defined above:
Remark 7.7. Define the operator H on L2 (Rd ) as
H=

d
X
1
k=1

 1
k2 + Q2k ,
2
2

where the operator Qk is given by multiplication by xk . The operator H is

diagonalized in the basis of Hermite functions. For f and g finite linear combinations of Hermite functions, let
hf, gin = hf, (H + I)n gi.
This inner product extends by completion to define a Hilbert space Hn . Because
0 H, we have kf kn kf kn+1 , so there is an inclusion inn+1 : Hn+1 , Hn .
By a similar analysis to the previous proposition, we have that for D > 2d,
the canonical injection
: Hn+D , Hn
inn+D = inn+1 in+D1
n+D
T
is trace class. Schwartz space is given by n Hn .

26

7.3

Gaussian measures

Now we define the measure d on the space H 0 . There is no -finite Lebesgue

measure on an infinite dimensional space, so we will use infinite-dimensional
Gaussian measures. This is a concept that we will define.
For finite-dimensional spaces, Gaussian measures are defined from the
Lebesgue measure in the following way:
Definition 7.8. Let V be a finite-dimensional inner product space. Let B0 (V )
be the completion of the Borel -algebra on V . Let be the usual Lebesgue
measure on V . Then for A B0 (V ), the standard Gaussian measure is
defined by


Z
1
1
2
exp kxk d(x).
(A) = n
2
2 A
Remark 7.9. The standard Gaussian measure on a finite dimensional inner
product space is equivalent to its Lebesgue measure.
To define a Gaussian measure in an infinite-dimensional setting, suppose
that H is a nuclear space and H 0 is its dual.
Definition 7.10. For A a Borel set on Rn and f1 , . . . , fn H , then
SA;f1 ,...,fn { H 0 : (f1 (), . . . , fn ()) A}
is called a Borel cylinder set in H 0 . The smallest -algebra in H 0 that
contains all of the cylinder sets is called the cylinder -algebra.
This definition can be interpreted in the following way: For a finitedimensional subspace of H containing f1 , . . . , fn , let 0 H 0 be the linear
space of such that
f1 (), . . . , fn () = 0.
Then the map 7 f1 (), . . . , fn () factors through H 0 /0 . Choose A
H 0 /0 such that the image of A is a Borel subset of Rn . The preimage of A
under the canonical projection is a Borel cylinder set in H 0 . We say that it
is based in . Given measures d on H 0 /0 parameterized by all choices
of , then these measures are said to be compatible if we have the following:
for 1 2 and X H 0 /01 a Borel set,
d1 (X) = d2 (Q1 (X)),
where Q denotes the natural map from H 0 /02 into H 0 /01 . In this case, the
measure on the sets H 0 /0 lifts to a measure d on Borel cylinder sets. If
each d is a Gaussian measure, then d is said to be a Gaussian measure
on cylinder sets for H 0 . If d is countably additive on the -algebra of
Borel cylinder sets, then it is known that it can be uniquely extended to a
countably additive measure d on the class of all Borel sets (see, for instance,
[13, IV.2]). This is called a Gaussian measure for H 0 .
27

7.4

Construction of Gaussian measure

Given the groundwork of the previous section we prove the following proposition, which is part of Theorem 7.1. Our proof follows [13] and [16].
Proposition 7.11. Let C be a covariance operator defined on H . Then there
is a unique Gaussian measure, which we denote dC , defined on H 0 , and
having C as its covariance operator.
Proof. Let CV be the restriction of C to an n-dimensional subspace V H .
Then CV is the covariance of a unique finite dimensional Gaussian measure on
V:


1/2
det CV
1
1
(7)
dxCV = n exp hx, CV xi0 dx
2
2
where dx is the Lebesgue measure on V . Note that we use the nuclear inner
product h, i0 on H0 , which is equal in our case to the inner product on L2 (M ).
Let V 0 be the dual space of V under the H0 inner product. The inner
product relates these spaces and allows us to think of dxCV as a measure on
V 0 . As above, define V 0 = { H 0 : h, V iH0 = 0}. Then the Gram-Schmidt
process gives rise to a unique H0 -orthogonal decomposition H 0 = V 0 V 0 ,
i.e. V 0 is isomorphic to H 0 /V 0 . Thus dxCV gives a Gaussian measure on
H 0 /V 0 . The measures achieved in this way for different subspaces V are
compatible in the sense defined above. This follows immediately from the
fact that a finite-dimensional Gaussian measure is uniquely defined by its
covariance.
Thus we have a measure dC on the cylinder sets of H 0 . It is finitely
additive and regular by construction. It remains to prove that it is countably
additive, in which case it extends to the Borel algebra as discussed above.
Recall the nuclear space structure
H =

Hn

and

n=

H =
0

Hn ,

n=

where H0 = L2 . Let S(r, j) = { : kkj r} be the sphere of radius r in Hj .

Given our measure d on cylinder sets in H 0 , we say that it has vanishing
measure at infinity in Hj if for all > 0 there exists r such that for any
Borel cylinder set X disjoint from S(r, j), we have that (X) .
Lemma 7.12. Let d be a finitely additive, regular measure defined on Borel
cylinder sets in H 0 . Suppose that there exists j such that d has a vanishing
measure at infinity in Hj . Then d defines a countably additive measure on
the Borel cylinder sets of H 0 .
F
Proof. Let Y =
k=1 Yk be a disjoint
P union of Borel cylinder sets. Let Y0 =
0
H \ Y . We wish to prove that k=0 (Yk ) = 1. By finite additivity, the sum
28

P
is less than or equal to 1. By regularity, it is sufficient to prove
k=0 (Zk ) 1
where each Zk is a weakly open cylinder set containing Yk .
Fix > 0. The ball S(r, j) is weakly compact, so there is a finite union
Z ofPZk such that Z contains S(r, j). By hypothesis, (H 0 \ Z)
1
k=0 (Zk ), which proves the result.
If the hypotheses of this lemma hold, then the we have completed the
proof. By the continuity of C on H , there exists j such that C : Hj Hj
boundedly or, equivalently,
|hf, Cgi0 | const kf kj kgkj .

(8)

Fix this j (It is almost our choice of j to satisfy the above lemma. We will
increase it slightly over the course of the proof.).
Let Z be a cylinder set based on the finite-dimensional subspace V H
and suppose that Z S(r, j) = . Let SV = S(r, j) + V 0 . Because Z =
Z + V0 , we have that Z SV = . This gives us the first of the following
inequalities, and we will prove the second:
Z
dC .
C (Z)
H 0 \SV

Recall that H 0 = V 0 V 0 . Let PV denote the canonical projection onto

V 0 . Then


Z
Z
1
1/2
1
dim V /2
dC = (2)
det CV
exp hx, CV xi dx.
2
H 0 \SV
V 0 \PV S(r,j)
We simplify the integration through the change of variable
1/2

Y = V 0 \ CV

PV S(r, j).

Then the integral becomes

Z
H 0 \SV

dC = (2) dim V /2

Z
exp
Y

kyk
2

!
dy.

By definition, we have that for y Y ,

1/2

CV y V 0 \ PV S(r, j) H 0 \ PV S(r, j) H 0 \ S(r, j).

Furthermore, we have by the nuclear space structure of H , that S(r, j) =
(H + I)j/2 S(r, 0). Thus we have that
1/2

(H + I)j/2 CV y H 0 \ S(r, 0).

Thus
1/2

1/2

r2 hy, CV (H + I)j CV yiH0

29

which implies that

C (Z) (2) dim V /2
1/2

kyk2
1/2
1/2
hy, CV (H + I)j CV yiH0 r2 exp
2

!
dy.

1/2

Let AV = CV (H + I)j CV . The above equation gives that C (Z)

r2 trV AV where this is the H0 trace taken over V . If we can prove that trV AV
is bounded independent of V , then by taking r large enough we achieve the
desired result. By equation 8, there is a constant , independent of V , such
that



|hf, CV gi0 | (H + I)j/2 f 0 (H + I)j/2 g 0
for f, g H . Thus we have that for f, g (H + I)1/2 H ,


hf, (H + I)j/2 CV (H + I)j/2 gi0  kf k kgk .
0
0
1/2

This inequality extends to f, g H0 and so CV (H + I)j/2 is a bounded

operator on H0 with norm independent of V . By the nuclear property of
1/2
the Hj norms, we increase j so that CV (H + I)j/2 is Hilbert-Schmidt, with
norm independent of V . The required bound on trV AV follows. This j satisfies
Lemma 7.12, so the proof that dC exists is complete.
The uniqueness of dC follows from the details of the construction. The
uniqueness of finite-dimensional Gaussian measures implies that dC is unique
on cylinder sets. The fact that these generate all Borel sets under repeated
monotone limits then gives the result.
R
Corollary 7.13. The generating function S{f } = eih,f i dC defined by the
Gaussian measure dC is given by


1
S{f } = exp hf, Cf iH0 .
(9)
2
Proof. Let f H 0 be the dual vector to f H . By the previous construction, the generating function is given by a finite Gaussian integral over the
one-dimensional space V spanned by f . Suppose that C|V = V . Then
equation 7 for the finite Gaussian integral becomes:




Z
1
1
1/2
1
(2)
exp hx, iH0 exp(ipx)dx = exp hp, pi
2
2
where the integral is evaluated term-by-term in the power series.

7.5

OS axioms for the Gaussian measure

To complete the proof of Theorem 7.1, we show that the Euclidean invariance
and reflection positivity of d are equivalent to the Euclidean invariance and
30

reflection positivity of C. These conclusions follow from Corollary 7.13, which

gives the generating function of the Gaussian measure d as


1
S{f } = exp hf, Cf iL2 .
2
Recall from axiom 4.4 that d is Euclidean invariant when for all
Isom(M ) and f H , we have S{ f } = S{f }. It follows that this is equivalent to the condition
h f, C f iL2 = hf, Cf iL2 .
Recall from axiom 4.12 that d is reflection positive if for any finite collection of functions fi with positive support in H , we have that
Mij = S{fi fj }

(10)

is positive. By the formula for S, this is S{fi }S{fj } exphfi , Cfj i, so if C

is reflection positive then d is reflection positive. The converse is valid for
non-Gaussian measures and is stated as a separate proposition due to [16]:
Proposition 7.14. Let d be a measure on H 0 with generating functional
S{f }. Assume that S{f } is an entire analytic function of the complex variable
f H . If d is reflection positive in the sense of axiom 4.12, then so is the
two-point function of d.
Proof. Define Mij as in equation 10. Take any real f H . Let f1 = f ,
f2 = 0, 1 = 1 , and 2 = 1 . Then
Z
2
X
0
i Mij j (f )(f )d().
i,j=1

Reflection positivity for the Laplacian covariance

Theorem 7.1 reduces the reflection positivity of L2 (H 0 , d). to the reflection

positivity of a bilinear form C on H . In this section, we will construct a
reflection positive C of great physical interest:
P
Definition 8.1 (Resolvent of the Laplace-Beltrami operator). Let = di=1 x2i
be the Laplace operator, which is an essentially self-adjoint operator on
C0 (Rd ). Then the resolvent C = ( + m2 )1 is a bounded operator on
L2 (Rd ) that gives the following bilinear form on L2 :
Z
hf, CgiL2 C(f, g) d~x d~y f(~x)C(~x, ~y )g(~y ),
C = ( + m2 )1 .
31

C is the Greens function of the equation = m2 . This equation is of

physical significance because it is the analytic continuation to imaginary time
of the equation of motion for a free particle (a particle that does not interact).
More generally, given a d-dimensional Riemannian manifold M with LeviCivita connection , then let M denote the (negative-definite) covariant Laplace-Beltrami operator. If M is complete then M is essentially
self-adjoint as an operator on functions, forms, and tensors [39, Theorem 2.4].
Then, by the spectral theorem, the resolvent CM = (M + m2 )1 is a welldefined bounded operator that gives the inner product
hf, CM giL2 (M )

(11)

In Theorem 8.3, we prove that the inner product of Equation 11 is reflection

positive when M is the Laplace-Beltrami operator on any complete Riemannian manifold that admits a reflection. Then, in Theorem 8.4, we prove that
on Rd reflection positivity still holds when Dirichlet and Neumann boundary
conditions are applied to the Laplacian.
Let M be a d-dimensional, complete, connected Riemannian manifold, let
G Isom(M ) be the isometry group of the manifold, and let G be a
reflection in the sense of Definition 4.7. Suppose that is dissecting, meaning
that it partitions the manifold as t t + where the reflection hyperplane
disconnects its complement. Note that in this section we do not assume that
M is static, although the reflections that we defined for a static manifold are
dissecting. Let U denote the unitary representation of G on L2 (M ), defined
by U f f f 1 for G. Let M denote the Laplace-Beltrami
operator. This operator is an isometry invariant, which can be checked by
direct calculation on the coordinate expression:
d
1 X

j (g jk gk u).
M u =
g j,k=1

Then the following lemma gives us that [U , C] = 0, i.e. Cf = (Cf ) :

Lemma 8.2. Let T be an operator on a Banach space X such that the resolvent
set of T is non-empty. In order that T commute with a bounded operator A
on X, it is necessary that R(z, T ) = (T z)1 commute with A.
Proof. See [29, Theorem 3.6.5].
We then have the following theorem due to [25]:
Theorem 8.3 (Reflection Positivity for Laplace-Beltrami). Let M be a complete, connected Riemannian manifold with a dissecting reflection in the
sense discussed above. Let + t t denote the partition of the manifold
by the reflection hyperplane. For all f C02 (+ ),
0 hf , Cf iL2 .
32

(12)

Proof. For convenience, let u = Cf . Then

Z

1
hu, f iL2 = hu, C u iL2 =
uC 1 u dV
Z
Z
1
=
uC u dV

C 1 uu dV.

The first line uses the fact that [U , C] = 0 and that f has support only on
. The second line uses the fact that C 1 u = f is zero on . Replacing
C 1 with (M + m2 ) and integrating by parts, we find
Z

hf , uiL2 = [u n u un u ]dS,

where n is the normal vector to . By construction (see Proposition 4.8), we

have that for p , dp = d1
p = diag(1, 1, . . . , 1) in a coordinate basis
where the first coordinate is in the direction of np . Using (n u )p = (n u)p
and u = u on , the above equation simplifies to

Z

un udS .
hf , Cf iL2 = 2 Re

Now we show by manipulation that the quantity in brackets is real and positive:
Z
Z
a
a (ua u)dV
una udS =

Z
=
(a ua u + u
u)dV

Z
(|u|2 + m2 |u|2 )dV 0.
=

The last equality comes from the fact that u = m2 in , which holds
because f is supported in + .
Now we restrict our attention to the setting M = Rd . We let denote a
reflection on Rd , as well as the action of that reflection on L2 (Rd ). We prove
the result, due to Glimm and Jaffe [15], that reflection positivity still holds
in the presence of boundary conditions on the Laplacian. In this discussion,
we use the following notion of inequality for bilinear forms: A B means
that Dom(B) Dom(A) (form domains) and hx, Axi hx, Bxi for all x
Dom(B).
Theorem 8.4 (Reflection Positivity for Laplacian with boundary conditions).
Suppose that B is the Laplacian on Rd with boundary data B on a finite
union of piecewise smooth hypersurfaces. Suppose that the boundary data
consists of a mixture of Dirichlet and/or Neumann conditions, and suppose
that the boundary conditions are symmetric under the reflection . Let C =
(B + m2 )1 . Then C is reflection positive with respect to .
Note that this implies theorem 8.3 in the case that M = Rd .
33

Proof. By the fact that the commutator [, ] = 0 and the fact that preserves the boundary conditions, we have that [B , ] = 0. Then Lemma 8.2
implies that [C, ] = 0.
Let Rd denote the plane of reflection that decomposes Rd into Rd t
t Rd+ . Let be operators on L2 (Rd ) that give orthogonal projection onto
L2 (Rd ). Then we wish to prove that
+ C+ 0,

(13)

as a bilinear form on L2 (Rd ).

We can write this positivity condition as
+ [C (I )C]+ 0.

(14)

First, we will prove that (I )C restricted to is equal to CD = (0B +I)1

where 0B is the Laplacian operator with Dirichlet data on and B data on
\ . Second, we will prove that CD C.
The first claim comes from the fact that (I )C(x, y) vanishes on ,
and (B + I)(I )C(x, y) = (x y) on Rd , i.e. (I )C satisfies the
characteristic differential equation of CD .
The second claim follows from the fact that B 0B as bilinear forms
because 0B is defined as the restriction of B to functions vanishing on .
Taking inverses gives the desired result.
Remark 8.5. The positivity condition in equation 13 could alternatively be
written as
+ [(I + )C C]+ 0,

(15)

and the proof could be based on this fact by proving:

1. (I + )C restricted to is equal to CN = (0B + I)1 where 0B is the
Laplacian operator with Neumann data on and B data on \ .
2. C CN as bilinear forms.
Let be the free Laplacian with covariance C. In the proof of the previous
theorem, we discussed the monotonicity of bilinear forms CD C CN where
CD (resp. CN ) corresponds to the free Laplacian with Dirichlet (resp. Neumann) conditions on . This monotonicity can be proven from the coordinate
representations of the covariance forms. An analogous result for general Riemannian manifolds is proven in [25] using Theorem 8.3

Reflection positivity for the Dirac covariance

The entire paper up until now presents a unified discussion: we construct a

Euclidean space E over M equipped with a reflection positive bilinear form,
34

and we use this reflection positivity to analytically continue the theory to a

Lorentzian setting. The final step of our argument, carried out in the previous
section, was to prove the reflection positivity for the resolvent C of the LaplaceBeltrami operator. We found earlier that the reflection positivity of C implies
the reflection positivity of E L2 (H 0 , dC ).
In this final section of the paper, we branch out from the preceding argument and prove the reflection positivity of a related bilinear form over M
that does not directly give a measure. This form can instead be used to define a Berezin integral [2] and to develop the the physical theory of fermionic
particles. This is more than we will undertake in this paper. Our purpose in
this section is to illustrate the rich array of reflection positive forms over the
Riemannian manifold M .
In this section, we prove the reflection positivity of C = (D m)1 where
D is a Dirac operator. Dirac operators on Riemannian manifolds were introduced by Atiyah and Singer for their 1963 proof of the Atiyah-Singer Index
Theorem. This was a rediscovery of a physically significant idea of Dirac.
Working in 4 dimensional Minkowski spacetime, Dirac sought an operator D
such that D2 equals the Laplacian. This operator is given by / , where
denotes the Gamma matrices. The physics of electron motion in Minkowski
space is famously described by the Dirac equation:
(i/ + m) = 0.
In the following section, we generalize the definition of the Dirac operator
to the setting of Riemannian manifolds. In section 9.2, we prove the desired
reflection positivity result, which is due to [25].

9.1

Introduction to the Dirac operator

To generalize the theory of this operator /, we ask the following: Let M be

a Riemannian manifold and E a vector bundle over M . Let (M, E) denote
the space of smooth sections. What are the first-order differential operators
D that square to the generalized Laplacian on (M, E)? Write
X
D=
ak (x)k + b(x),
k

where ak (x) and b(x) are sections of End(E). Squaring this and comparing it
to the generalized Laplacian, we find that D2 is the generalized Laplacian if
and only if for u, v Tx M , we have
ha(x), uiha(x), vi + ha(x), viha(x), ui = 2hu, vix
where h, ix is the inner product on Tx M that arises from the metric on M .
The above equation is the defining relation for the Clifford algebra of Tx M .
This leads to the following definition for a Dirac operator:
35

Let M be a Riemannian manifold. Let Cl(M ) denote the Clifford bundle

of M . This is the bundle over M whose fiber at x M is the Clifford algebra of
Tx M with metric as inner product. These fibers are denoted Clx (M ). Recall,
by generalities of Clifford algebras, that there is a natural inclusion Tx (M ) ,
Clx (M ).
Let E denote a Hermitian vector bundle over M such that each fiber Ex
is a self-adjoint Clx (M ) module in a smooth fashion. The inclusion Tx (M ) ,
Clx (M ) gives rises to a bundle map m : T (M ) E E called Clifford
multiplication. For convenience, we denote Clifford multiplication as v.
If is a connection for E M , then we have a sequence

(M, E)
(M, T (M ) E)
(M, E).
Definition 9.1 (Dirac operator). Given a Riemannian manifold M and a
Hermitian vector bundle E M satisfying the above conditions, the Dirac
operator / is the composition of the above maps, i.e. / = m : (M, E)
(M, E).
A Dirac operator can be described in local coordinates. Let O M be an
open subset with an orthonormal frame {ej } of tangent vector fields, and let
{vj } denote a dual frame of 1-forms. For (M, E; O):
X
/ =
vj ej .
(16)
j

A Clifford connection on E is a metric connection that is compatible

with Clifford multiplication in the sense that
X ( v) = (X ) v + X v
for a vector-field X, a 1-form , and a section v of E. Here X arises from the
Levi-Civita connection on M . A Clifford connection has the powerful property
that the associated Dirac operator is skew-symmetric with respect to the inner
product of L2 (E) given by
Z
hu, vi =
hu(x), v(x)ix dV.
M

Proposition 9.2. If is a Clifford connection on E then i/ is symmetric.

Proof. Let , be smooth sections of compact support in (M, E). We wish
to prove that
Z


hi/, i h, i/i dV = 0.
M

It is sufficient to handle the case that , have support in a compact set U with
a local orthonormal frame ej for smooth vector fields, and a dual orthonormal
frame vj for 1-forms. On this set, we have
X
/ =
vj ej .
36

Now define a vector field X on U by hX, vi = h, v i for v 1 U . The

desired result will follow from the divergence theorem if we can prove


div X = h/, i + h, /i .
By definition
div X =

hej X, vj i.

Using the fact that is a metric connection and the definition of X, we have
X
 X

div X =
ej hX, vj i hX, ej vj i =
ej h, vj i h, (ej vj ) i .
j

Expanding and using the fact that is a Clifford connection, this becomes
X

div X =
hej , vj i + h, vj ej i ,
j

as desired.
To give a reflection positivity result, we need to introduce a condition on
M that enhances the symmetry of i/ to essential self-adjointness on smooth
sections of compact support. Such a condition is proven in [6], and we sketch it
here. For any first-order differential operator L, let (, x) denote its symbol.
For each x M and Tx M , the symbol is a linear map (, x) : Ex Ex
given by
(, x)e = L(gf )(x) g(x)(Lf )(x),
where f (M, E) is a section such that f (x) = e, and g C (M ) is
a function with dgx = . We wish to globalize this solution. When L is
symmetric, then standard results give existence, uniqueness, and smoothness
for the solutions of the hyperbolic system u/t = Lu. Define the local
propagation velocity of the system in the following way:
c(x) = sup{k(, x)k : Tx M, || = 1},
where kk is the operator norm on Ex . Then for M , we define c() =
sup{c(x) : x }. Then, define
c(r) = c(Sr ),
where Sr is the ball of radius r about an arbitrary reference point x0 M .
Then [6, Theorem 2.2] gives that L is essentially self-adjoint if the following
two conditions hold:
1. M is complete
37

2.

R
0

dr/c(r) = .

Thus, if the manifold M satisfies these two conditions, the operator i/ is

essentially self-adjoint. This will be an assumption in the reflection-positivity
theorem of the next section.
Note that our definition of the Dirac operator and our discussion of its
properties includes the most prominent example as a special case:
Remark 9.3 (Spinor bundle). Let M be a complete n-dimensional oriented
Riemannian manifold. Associated with M is the bundle P M of orthonormal frames. It is a principal SO(n) bundle. A spin structure on M is a
lift P M to a principal Spin(n)-bundle such that P is a double-covering
characterized in the following way: the action of Spin(n) on the fibers of P is
compatible with the action of SO(n) on the fibers of P via the covering homomorphism Spin(n) SO(n). When such a structure exists, there is a natural
associated Clifford bundle called the spinor bundle, which has a Clifford
connection. Details of this construction can be found in [3, 40]. The Dirac
operator i/ is essentially self-adjoint for this connection, and the theory that
we are developing applies [6, Section 3]

9.2

Proof of reflection positivity

Let M be a complete static Riemannian manifold. Let xi give local coordinates such that x 0 is the hypersurface-orthogonal Killing field that gives
time-translation. We let t = x0 . Locally the metric takes the form
ds2 = F (x)dt2 + Gjk (x)dxj dxk .
Let be the reflection map around the time-zero surface . Decompose
M = t t + . Let E M be a holomorphic Clifford bundle with
Clifford connection . Let denote the pullback of that acts on sections
of E. Let dxi denote the local frame of one-forms, and let i denote the Clifford
multiplication by dxi , i.e. i (v) = dxi v. Then the anti-commutator obeys the
relation:
{ i , j } = 2g ij id,
where g ij is the inverse metric. On a static manifold, the operator 0 has a
coordinate free meaning. Then, a simple calculation gives:
(
)
X
{ 0 , /} = 0 ,
j ej = 0.
j

We have the following theorem due to [25]:3

3

The current arXiv version of this paper (version 2) incorrectly assumes that i/ is selfadjoint for all complete M . The published paper includes the condition on c(r).

38

Theorem 9.4. Let M be a complete, static Riemannian manifold with the

structure described above. Let E M be a holomorphic Clifford bundle
i
with
R Clifford connection , and let be as defined above. Suppose that
dr/c(r) = in the sense of the previous section, so that i/ is essen0
tially self-adjoint. Then for all smooth sections of compact support in + ,
we have
h 0 , (/ m)1 i,
where this is the inner product in L2 (E).
Proof. For convenience, let 0 and v = (/ m)1 . Let A() =
h 0 , (/ m)1 i, which we wish to prove is positive. Then
Z


h/v, vi + mhv, vi .
A() = h/v, vi mhv, vi =

Where the second equality comes from the fact that {, /} = 0. Then, using
the fact that (/ m)u = = 0 on , we have
Z
Z




A() =
h/v, vi + mhv, vi + hu, (/ m)ui =
h/u, ui hu, /ui .

We saw in the course of proving Proposition 9.2 that for smooth sections , ,
we have


div X = h/, i + h, /i ,
where X is the vector field defined by hX, vi = h, v i for v 1 M . Taking
= v and = v, we have
Z
A() =
div X dV,

where hX, vi = hv, v uiE .

Let n
F 1/2 x 0 denote the unit normal vector of that points into .
Let = F 1/2 dx0 denote the dual one form. The divergence theorem gives
Z
Z
A() =
div XdV = hX, idS.

On , we have

hX, i = hu, uiE = h 0 (u), F 0 (u)i.

Thus
Z
A() =

h 0 (u), F 0 (u)idS 0.

39

10

Conclusion

In this paper, we have constructed a theory of reflection positive Euclidean

fields for all complete, static Riemannian manifolds M with compact spatial
hypersurfaces. This reflection positivity allows us to analytically continue the
Euclidean theory to a Lorentzian theory. The representations of G0lor that
are constructed in this way are at the heart of mathematical quantum field
theory. This exposition provides the rigorous foundation for studying physics
at imaginary time. The techniques described here underlie the foremost efforts
to construct a mathematical theory of particles.
Reflection positivity, as defined in the present paper, has applications
throughout physics and mathematics. In the final section, we used the example of the Dirac operator to illustrate the richness of the reflection positive
forms available over a Riemannian manifold. Our treatment of reflection positivity is sufficient to understand its use by A. Jaffe et al. to construct representations of the Heisenberg algebra on a Riemann surface [22] and its use by
V. Pestun to compute the partition function for supersymmetric Yang-Mills on
the four-sphere [36]. Surveys of material outside the scope of this exposition
include M. Biskups paper on reflection positivity in statistical mechanics [4]

and P. Jorgensen and G. Olafssons

papers on reflection positivity in representation theory [27, 28]. The latter, in particular, are excellent further reading
for mathematicians.

11

Appendix A: Cited theorems

Bochner-Minlos theorem
This theorem characterizes a measure d on the continuous dual of a nuclear
space. It is due to Salomon Bochner and Robert Minlos. See, for instance,
[13, IV.4]:
Theorem 11.1 (Bochner-Minlos). Suppose that V is a nuclear space and V 0 is
its continuous dual. If d is a regular
Borel measure on V 0 with total weight 1,
R i(f
then its generating functional S = e ) d satisfies the following properties:
1. Continuity (in the Frechet topology of V )
2. Positive definiteness: for all ci C, fi V ,
0

N
X

ci cj S(fi fj ).

i,j=1

3. Normalization: S(0) = 1.
Conversely, given a functional S defined on V and satisfying the previous three
properties, then S is the inverseR Fourier transform of a unique regular Borel
measure d with normalization d = 1.
40

Stones theorem
In the mathematical physics literature, the name Stones theorem is abused
to refer to several theorems that imply the existence of self-adjoint generators
for one-parameter semigroups. The eponymous theorem is:
Theorem 11.2 (Stones theorem). Let U (t) be a strongly continuous oneparameter unitary group on the Hilbert space H. Then there is a unique selfadjoint operator A on H such that U (t) = eitA for t R.
A reference is [37, Theorem 6.2]. Stones theorem is used in the current
exposition to prove Theorem 6.7.
Another theorem that sometimes goes under the same name is:
Theorem 11.3. Let T (t) be a contraction semigroup of self-adjoint operators
on the Hilbert space H. Then there exists a unique positive self-adjoint operator
A on H such that T (t) = etA for t 0.
A reference is [37, Proposition 6.14]. This result is used in the current
exposition to prove Theorems 5.10 and 5.12.
The theory of symmetric local semigroups described in Section 6.2 gives
another set of conditions under which a semigroup has a self-adjoint generator.

12

Acknowledgments

Thank you to Professor Arthur Jaffe for the guidance in the three years that
I have known him at Harvard, for his constant assistance in exploring mathematical physics, and for his support during the process of writing this thesis.
Thank you to the other members of Professor Jaffes Fall 2012 reading course
in mathematical field theory: Dmitri Gekhtman, Nikko Pomata, Clay Cordova, Alex Lupasca, and Roberto Martinez. Thank you to Vasily Pestun for
his discussion regarding [36]. Thank you to my family: my mom, my dad,
Catherine, and Thayer.
Thank you to my mathematics and physics professors at Harvard. This
thesis has benefited from every course that I took with them. In the mathematics department they are Noam Elkies, Joe Harris, Michael Hopkins, Peter
Kronheimer, Curtis McMullen, Wilfried Schmid, and Shlomo Sternberg. In the
physics department they are Howard Georgi, Arthur Jaffe, Erel Levine, Misha
Lukin, Matthew Schwartz, Andrew Strominger, and Cumrun Vafa. Thank you
to Steve Carlip of UC Davis, Erel Levine of Harvard, and Sergei Tabachnikov
of Penn State, who have advised me in research projects. Thank you to my
academic advisers Peter Kronheimer, David Morin, and Clifford Taubes for
their support.

41

References
[1] Dmitri V. Alekseevsky, Andreas Kriegl, Mark Losik, and Peter W. Michor.
Reflection Groups on Riemannian Manifolds. Annali di Matematica Pura
ed Applicata, 186(1):2558, January 2007.
[2] Feliks Aleksandrovich Berezin. The Method of Second Quantization. Academic Press, 1966.
[3] Nicole Berline, Ezra Getzler, and Mich`ele Vergne. Heat Kernels and Dirac
Operators. Springer, 1992 edition, February 2004.
[4] M. Biskup. Reflection Positivity and Phase Transitions in Lattice Spin
Models. Methods of Contemporary Mathematical Statistical Physics, pages
186, 2009.
[5] Isaac Chavel. Eigenvalues in Riemannian Geometry, Volume 115, Second
Edition. Academic Press, 2nd edition, December 1984.
[6] Paul R. Chernoff. Essential self-adjointness of powers of generators of hyperbolic equations. Journal of Functional Analysis, 12(4):401414, April
1973.
[7] Piotr T. Chrusciel. On analyticity of static vacuum metrics at nondegenerate horizons. Acta Physica Polonica B, 36(1):1726, 2005.
[8] J. Feldman and K. Osterwalder. The Wightman axioms and the mass gap
for weakly coupled 43 quantum field theories. Ann. Physics, 97:80135,
1976.
[9] Carl H. Fitzgerald. On analytic continuation to a schlicht function. Proceedings of the American Mathematical Society, 18(5):788792, 1967.
[10] J. Frohlich. Verification of Axioms for Euclidean and Relativistic Fields
and Haags Theorem in a Class of P ()2 -Models. Ann. Inst. Henri
Poincare, Sect. A, v. 21, no. 4, pp. 271-317, 21(4), January 1974.
[11] J. Frohlich. Unbounded, Symmetric Semigroups on a Separable Hilbert
Space are Essentially Selfadjoint. Advances in Applied Mathematics,
1(3):237256, September 1980.
[12] J. Frohlich, K. Osterwalder, and E. Seiler. On Virtual Representations
of Symmetric Spaces and Their Analytic Continuation. Annals of Mathematics, 118(3):461489, November 1983.
[13] Israel Moiseevic Gelfand and N. Ja Vilenkin. Applications of Harmonic
Analysis. Academic Press, New York, 1964.
[14] James Glimm and Arthur Jaffe. Positivity of the 43 Hamiltonian.
Fortschr. Phys., 121:327376, 1974.
42

[15] James Glimm and Arthur Jaffe. A Note on Reflection Positivity. Letters
in Mathematical Physics, 3(5):377378, September 1979.
[16] James Glimm and Arthur Jaffe. Quantum Physics: A Functional Integral
Point of View. Springer-Verlag, New York, 2nd edition, 1987.
[17] James Glimm, Arthur Jaffe, and Thomas Spencer. The Wightman axioms and particle structure in the P ()2 quantum field model. Annals of
Mathematics, 100(3):585632, November 1974.
[18] H. Mller zum Hagen. On the analyticity of stationary vacuum solutions
of einsteins equation. Mathematical Proceedings of the Cambridge Philosophical Society, 68(01):199201, 1970.
[19] Harish-Chandra. Discrete series for semisimple Lie groups. II. Acta Mathematica, 116(1):1111, December 1966.
[20] M. R. R. Hoole, Arthur Jaffe, and Christian D. Jakel. Quantization domains. Planned for publication.
[21] Arthur Jaffe. Quantum Theory and Relativity. Contemporary Mathematics, 449:209, 2008.
[22] Arthur Jaffe, Slawomir Klimek, and Andrzej Lesniewski. Representations
of the Heisenberg Algebra on a Riemann Surface. Communications in
Mathematical Physics, 126(2):421431, December 1989.
[23] Arthur Jaffe and Gordon Ritter. Quantum Field Theory on Curved Backgrounds. I. The Euclidean Functional Integral. Communications in Mathematical Physics, 270(2):545572, March 2007.
[24] Arthur Jaffe and Gordon Ritter. Quantum Field Theory on Curved Backgrounds. II. Spacetime Symmetries. arXiv:0704.0052, March 2007.
[25] Arthur Jaffe and Gordon Ritter. Reflection positivity and monotonicity.
Journal of Mathematical Physics, 49(5):05230105230110, May 2008.
[26] Arthur Jaffe and Edward Witten. Quantum Yang-Mills Theory. The
Millennium Prize Problems, pages 129152, 2006.

[27] Palle E. T. Jorgensen and Gestur Olafsson.

Unitary Representations of
Lie Groups with Reflection Symmetry. Journal of Functional Analysis,
158(1):2688, September 1998.

[28] Palle E. T. Jorgensen and Gestur Olafsson.

Unitary representations and
Osterwalder-Schrader duality. In Proceedings of Symposia in Pure Mathematics, volume 68, pages 333402, 2000.
[29] Tosio Kato. Perturbation Theory for Linear Operators, volume 132.
Springer Verlag, 1995.
43

[30] Abel Klein and Lawrence J. Landau. Construction of a Unique SelfAdjoint Generator for a Symmetric Local Semigroup. Journal of Functional Analysis, 44(2):121137, November 1981.
[31] Abel Klein and Lawrence J. Landau. From the Euclidean Group to the
Poincare Group via Osterwalder-Schrader Positivity. Communications in
Mathematical Physics, 87(4):469484, December 1983.
[32] Edward Nelson. The Construction of Quantum Fields from Markov Fields.
J. Funct. Anal., 12:97112, 1973.
[33] A. E. Nussbaum. Spectral Representation of Certain One-Parametric
Families of Symmetric Operators in Hilbert Space. Transactions of the
American Mathematical Society, 152(2):419429, 1970.
[34] K. Osterwalder and R. Schrader. Axioms for Euclidean Greens functions.
Communications in Mathematical Physics, 31(2):83112, 1973.
[35] K. Osterwalder and R. Schrader. Axioms for Euclidean Greens functions
II. Communications in Mathematical Physics, 42(3):281305, 1975.
[36] Vasily Pestun. Localization of Gauge Theory on a Four-Sphere and Supersymmetric Wilson Loops. Communications in Mathematical Physics,
313(1):71129, July 2012.
[37] Konrad Schm
udgen. Unbounded Self-Adjoint Operators on Hilbert Space.
Number 265 in Graduate texts in mathematics. Springer, New York, 2012.
[38] Raymond F. Streater and Arthur S. Wightman. PCT, Spin and Statistics,
and All That. Princeton University Press, December 2000.
[39] Robert S. Strichartz. Analysis of the laplacian on the complete riemannian
manifold. J. Funct. Anal, 52(1):4879, 1983.
[40] Michael Eugene Taylor. Partial Differential Equations 2, Qualitative
Studies of Linear Equations. Springer, New York [u.a.], 2011.
[41] A. Uhlmann. Some Remarks on Reflection Positivity. Czechoslovak Journal of Physics B, 29(1):117126, January 1979.

44