Lecture 2

Eucledian formulation of Wightman QFT

David Kazhdan

2.1 Analytic continuation of Wightman functions. In lecture 1 we were working

with Minkowski space-time V . However it is often desirable to be able to work with
Euclidean (Riemannian rather than pseudo-Riemannian) space-time as well. In this lecture
we explain what data over the Euclidean space-time E = Rd corresponds to a QFT in
Wightman axioms.
More precisely, let VC = V
C , VnC = Vn
C , and consider the Euclidean ane space
E = iR  Rd 1  VC .
Going between QFT data on Minkowski space-time V = R  (R)d 1 and the one on
E = iR  (R)d 1 is often called Wick rotation.
The plan is to obtain Euclidean counterparts of Wightman functions by analytically con-
tinuing Wn to an open domain in VnC and then restricting the obtained analytic functions
to E .
We need some notations. Let GC be complexication of G. We set: T = V iV+  VC
(the backward tube); Tn = (T)n  VnC (the extended tube); Ten = GC (Ten)  VnC (the
permuted extended tube) ; Jn = Ten \ Vn. Recall that we identied Vn = V n+1, so we have
an action of the symmetric group n+1 on Vn.
Proposition.
a) Assume d > 2. Then for any  2 n+1 the subset Jn \ (Jn)  Vn is open and
nonempty.
b) If d > 2 then for any  2 n+1 the subset Ten \ (Ten) is connected.
c) Put E = iR  Rd 1  VC ; En = E 


 E  VnC . Also let En0  En be
the complement to all diagonals, and: E n = En=
(E ); E 0n = En0 =
(E ). Then
[ (Ten)  En0+1.
 2n+1
1
Proof. We start with the following result.
Lemma. (Jost's theorem) Jn = f(v1 ; : : : ; vn ) 2 Vn j8 (1; : : : ; n) 2 Rn+ f0g: (1 v1 +



+ nvn) 2 Vspaceg.
Proof of the Lemma. Assume that (v1 ; : : : ; vn ) 2 Jn. Then for some g 2 GC :
g(v1); g(v2 ); : : : ; g(vn ) 2 T, hence g(1 v1 +


+ nvn) 2 T for (~) 2 Rn+ f0g. Let
P P
us write k vk =  i, where  2 V+. Then ( k vk )2 = ()2 ()2 2i(; ). The
P
L.H.S. is real, hence (; ) = 0, and ()2  0, i.e. ( k vk )2 < 0.
Let us prove the converse statement. So assume that the (convex) cone generated by
v1; : : : ; vn does not intersect V + [ V . Then this cone can be separated from V by a
hyperplane tangent to V  . This means that there exist  2 V + and  2 V such that:
()2 = ( )2 = 0; (; vk ) < 0; (; vk ) < 0 for k = 1; : : : ; n and (; w) > 0 for w 2 V+;
(; w) > 0 for w 2 V . Then (;  ) 6= 0, and it is easy to show that there exists g 2 GC
such that g() = i, g( ) = i and g is identity on the subspace orthogonal to h;  i.
This g satises the condition g(vk ) 2 T for k = 1; : : : ; n.
Proof of the Proposition. a) It is obvious that this subset is open. So we must show
it is nonempty, i.e. we are to nd a point v = (v1 ; : : : ; vn ) such that (v1 ; : : : ; vn) and
(v(1); : : : ; v(n)) satisfy the conditions of the last lemma. We can choose two linearly inde-
pendent vectors `1; `2 2 Rd 1  V (since d > 2) and put v = (`1; : : : ; `1) +  1 (`2; : : : ; `2).
Then it is immediate to check the conditions.
b) It is enough to show that (Ten) \ Tn is connected for any .
We have a standard
Lemma. Generic element of GC can be written as g = g0
gR , where gR 2 G and g0
preserves an orthogonal decomposition of VC into sum of subspace of dimension not greater
than 2 dened over R. Here by generic we mean an element belonging to an open (in fact
real Zariski open) dense subset.
Let us now nish the proof of b). Fix  2 n+1 and consider S = f(g 2 GC ; v 2
VnC ) j g((v )) 2 Tng. Take a point (g; v) 2 S .
2
 Moving (g; v) slightly inside S we can achieve that g is generic as in the last lemma. We
can now take g0 , gR as in the Lemma. Obviously (g0 ; gR 1(v)) and (g; v) lie in one connected
component of S . Set v0 = gR 1(v).
Let V0  V be the unique block in the decomposition corresponding to g0 on which
( ; ) is not negative, and let  : VnC ! (V0C )n be the orthogonal projection. One
sees immediately that for v 2 Tn the segment connecting v with (v) lies in Tn. Also 
commutes with g0 and ; it follows that this segment also lies in  1 (g0 1(Tn)), hence
(g0 ; v0 ) and (g0 ; (v0 )) are connected inside S .
If dim(V0 ) = 1 then g0 jV0 = Id. So in that case we have (v0 ) 2 Tn, ((v 0 )) 2 Tn,
which is only possible if  = 1 or  = w0 is the \long" permutation (w0(i) = n + 2 i).
But in these two cases (Ten)  Tn, so there is nothing to prove anyway.
Suppose now that dim(V0) = 2. Set G0 = Spin(V0 ), G0C = Spin(V0
C ), Tn0 = V0 +
iV0+. Then G0C =G0 is the circle S 1, and one can show that for  6= 1; w0 the set fg 2
G0C =G0 j g((Tn0 )) \ Tn0 6= ;g coincides with S 1 f1; 1g. For given g the set g((Tn0 )) \ Tn0
is convex and hence connected; thus the set SV0 := f(g 2 G0C ; v 2 Tn0 j g((v)) 2 Tng
has 2 connected components. It is easily seen that these 2 components are permuted by
conjugation by an element of G, hence they lie in the same connected component of S .
But we have shown that any point in S is connected to a point in (StabV0  V0 ) \ S , and
hence to a point in SV0  S . (Here we use that the action of g0 on V0 has determinant 1,
because otherwise it has eigenvalues 1, 1, which does not happen for generic g.)
Since G transitively permutes 2-dimensional subspaces of V of signature (1; 1) we see
that S is connected, which implies the statement.
c) For v 2 VC let us write v = (v0 ;~v ), v0 2 C ; ~v 2 C d 1 . It is clear that for
(v1 ; : : : ; vn+1) 2 En0+1 there exists g 2 SO(n + 1), such that (gvi)0 6= (gvj )0 for i 6= j .
Then we can permute (v1 ; : : : ; vn+1) so as to get: (vi )0 > (vj )0 for i > j . But this implies
that the corresponding vector in V n+1 lies in Ten.
Now we are ready to prove
Theorem. Let d > 2. Then Wn extends to an analytic function on 2[n+1 (Ten). These
3
functions are n+1 and GC -equivariant. (The n+1-equivariance is understood as usually
in the \super" sense.)
Proof. Since F(Wn) is supported in V+ it follows that Wn extends to an analytic function
on T (which will be denoted by the same letter). Since Wn is G-equivariant, this analytic
function is G-equivariant.
To ensure that Wn extends to a single-valued analytic function on Te, which then is
automatically GC -equivariant, it is enough to check that Wn(x) = Wn (g(x)) for g 2 GC ,
x 2 Tn and g(x) 2 Tn.
If g 1Tn \ Tn 6= ; then also g 1T \ T 6= ;. Let us take v 2 g 1T \ T. It is easy
to see that for v 2 T the set fh 2 GC jh(v) 2 Tg is connected. Hence there exists a
neighborhood U  Tn \ g 1Tn of the point (v; : : : ; v) such that g 1 lies in the connected
component of 1 of fh 2 GC jh(U )  Tg. From G-invariance of Wn it follows that for xed
x 2 VnC the function Wn(g(x)) dened on some open subset of GC is locally constant.
Hence Wn(x) = Wn(g(x)) for x 2 U . Since U is open in the connected set g 1Tn \ Tn we
can conclude that Wn(x) = Wn(g(x)) for all x 2 g 1Tn \ Tn, so analyticity of Wn in Te is
proved.
Now by Proposition a) for any  2 n+1 the set  1(Te) \ Te contains an open non-empty
subset of Vn . Moreover, from the Jost's theorem it follows that

 1 (Te) \ Te \ Vn  (Vspace)n :
But Wn (Vspace)n is n+1 (super)equivariant. Hence Wn and  (Wn) agree on a neighbor-
hood of the set of real points in  1 (Te) \ Te, and hence on the whole of  1 (Te) \ Te by
Proposition b). So the collection of functions  (Wn ) for all  2 n give a well-dened an-
alytic function on 2[ (Ten), which is obviously n+1-(super)equivariant. The theorem
n+1
is proven.
We denote the restriction of these functions to En0 by Sn. We also put Sn =  (Sn)
where  : En0 ! En0 is the projection.
Denition. The functions Sn are called Schwinger functions.
4
 From the theorem it follows that Sn is a real-analytic SO(n) n E and n equivariant
function.
Remark. The theorem (and its proof) relies only on weak Wightman's axioms. Hence
Schwinger functions can be constructed from a set of Wn 2 eS0n which satisfy weak Wight-
man axioms.
Remark. The assumption that d > 2 is not essential in the Theorem. A method which
allows to treat also the case d = 2 and furhter extend the analyticity domain (the so called
\edge of the wedge Theorem") is explained e.g. in [Streater-Wightman], x3.3.
Remark. It was explained that QFT has two formulations: Minkowski and Euclidean.
The experience says that it is more convenient to work technically with Euclidean picture,
though the basic notions are naturally formulated in Minkowski one.
Remark. The analytic Schwinger functions on n+1(Ten) seem to be the correct object of
consideration. So it is natural to ask whether they can be further extended to some larger
domain.
The domain n+1(Ten) is not holomorphically convex.
It may be worthwhile to describe its convex hull and look at the extension of Schwinger
functions to it.
2.2 Euclidean formulation of Wightman QFT. Recall that decomposition E =
(iR)  Rd 1 is xed. Let  denote re ection at the hyperplane f0g  Rd 1; put: E + =
(iR>0)  Rd 1; En+ = E +  E + 


 E +  En. Let En denote the space of Schwartz
sections of RCnjEn , and E0n  En be the subspace of sections with support in En0 .
Claim.
a) Schwinger functions satisfy the following properties:
" i)" Sn is a real-analytic Pb-invariant and n (super) invariant real valued function
on En0 where Pb = Gb n E , Gb = Spin(E );
If moreover Sn come from a strong (as opposed to weak) QFT then they satisfy:
5
 " ii)" Sn is of moderate growth; i.e. Sn denes a tempered distribution on E0n .
" iii)" Re ection positivity.
Put: E+n = fF 2 E0n jsupp F  En+g. Also let E+ = E+n , Ee = E0n, and Se = Sn 2 (Ee)0 .
Then for F 2 E+ we have: Se ((F )  F )  0.
" iv)" Let En be the space of Schwartz sections of RCnjEn , and let E+n  En be the sub-
space of sections with support in En+ 1  E n 
= En 1. For r 2 R+ and a dierential
operator D consider the seminorm on E+n dened by:

jf jr;D = sup d 1 )n
j(1 + jxj2)r=2 DF(f )(x)j
x2(R >0 R

(here for x 2 V; x = (x0 ; ~x) we put jxj2 = x20 + j~x2 j). Then Sn denes a distribution
on E+n which is continuous with respect to the seminorms j jr;D .
b) The construction of x2.1 provides a bijection between Wightman QFT's and collec-
tions of functions Sn satisfying the conditions i){iv) above.
Proof. See [Osterwalder-Schrader(1975)].
It should be mentioned that this theorem is not really useful, because the property iv)
of the list is impossible to check in practice.
A more useful list is obtained by replacing the property iv) by the following axiom:
Axiom iv0 ). (Linear growth condition) There exist ; ; r 2 R>0 such that

jSn(f )j < (n!) jf jnr

for all f 2 E0n where the seminorm jf jnr is given by jf jnr = sup0 j(1 x2 )r=2 Df (x)j and
x2En ;D
D runs over the set of dierential monomials of degree less then nr.
This property is more tractable, but it is more restrictive than iv).
Some remarks are in order. For Schwinger functions constructed from a (strong) Wight-
man QFT the re ection positivity property is equivalent to the positivity property (5) of
Wightman axioms. However, to formulate it we do not have to assume anything about
the behavior of Sn on diagonals. (Note that if F1; F2 2 E+ then (F1 )  F2 automatically
6
vanishes on diagonals, so (F1 )  F2 2 Ee). So it can be formulated for Schwinger functions
of weak QFT. It is not clear how to restate this property directly in Minkowski picture
(which is more intuitive from the physical point of view) for weak Wightman QFT.
2.3 Schwinger functions and measures on the map-spaces. In fact, the following
picture should stand behind the formalism of Schwinger functions. Assume we have a
(quasi) measure  on E0 (where E is the space of Schwartz sections of RC E ), such that if
F1; : : : ; Fn 2 E and supp(Fi )\supp(Fj ) = ; for i 6= j then the function  ! (F1 ) : : : (Fn )
is measurable. Then we can dene a distribution in (E0n )0 for any n by requirement that its
R
value on F1  : : :  Fn equals (F1 ) : : : (Fn ). This distribution should coincide with Sn;
E0
thus Sn describes n-th moment of  . If the measure  exists, it is uniquely determined
by its moments Sn . However it does not necessarily exist for given Sn coming from a
Wightman QFT.
So working with Sn is a way to talk about the properties of a measure when the measure
itself is not dened.
The formulation of QFT involving a measure on E0 has another serious drawback. In
general one is interested in eld theories where the elds are not sections of a vector bundle,
but rather maps from a manifold to another manifold (no linear structure on the target
space). Then we can not even say on what space the measure  is dened. Indeed, in
the linear case the measure lives on the space of generalized sections of a vector bundle.
But there is no notion of a generalized map from a manifold to another manifold. The
formalism of Schwinger functions also does not work in this setting: to dene moments of
a measure one needs linear structure on the underlying space.
The diculty can be overcome if d = 1. In this case for a linear QFT the measure
is actually dened and is concentrated on the subspace of continuous sections. So one
gets a reasonable theory considering the space of continuous maps from V = R to M and
introducing a measure on this space. This is illustrated in the next two examples.
Example 1. (Wightman QFT for d = 1) In this case QFT is equivalent to usual quantum
mechanics. Precisely, let us put: H = L2(R), let A 2 C [u; @u ] be a Hamiltonian. Let A be
7
of the form A = dud22 + r(u), where r(u) ! +1 for u ! 1. Then A has the smallest
eigenvalue, and the corresponding eigenspace is one dimensional. Assume that the smallest
eigenvalue of A is 0. Let
be the corresponding eigenfunction. Also put: D = S(R), then

2 D  H. Finally, let ub 2 End(D) be the operator of multiplication by the function

f (u) = u. We dene the action of P = R on H by t ! eiAt and put:
Z
(f ) = f (t)eiAt ube iAt dt:

Let us describe the corresponding measure on S0 (R). We start with Gaussian measure 0
with covariance ( ( dtd )2 + m2) 1 , where m2 > 0 is such that r(u) m2u2 > 0 for juj  0.
(So for f 2 S we have F(0)(f ) = e(( (d=dt)2+m2) 1f;f ) .) We look for  of the form  = a0
where a is a function. It is enough to dene a on a set of measure 1. We dene a by the
formula:
1
R
(r((t)) m2(t)2 )dt
a() = e 1

Using Sobolev theorem one can show that 0 is essentially concentrated on the set of such
continuous functions  that the above integral converges. Hence the measure  = a0 is
correctly dened.
The fact that the measure  and the eld operators dened above give the same set
of Schwinger functions follows from the so-called Feynman-Kac formula, see Gawedzki's
lecture 1, x1.3.
Example 2. (Abstract QFT for d = 1) Let M be an arbitrary compact Riemannian
manifold, H = L2(M ),
is the Laplacian, A =
+ r(m). In this case one still can
describe a measure  on the space of continuous maps Map(R; M ) (one can get rid of
the compactness assumption by introducing requirements on asymptotic behavior of r;
Example 1 is a particular case of this situation).
The measure can be characterized as follows. Let t1 ; : : : ; tk be points in R, and U1 ; : : : ; Uk 
M be open subsets. Put XUt11;:::;t
;:::;Uk = ff 2 Hom(R; M )jf (ti ) 2 Ui for i = 1; : : : ; k g. Let
k

Kt (m1; m2 ), m1; m2 2 M , t 2 R>0 be the fundamental solution of the dierential equation

8
df = Af . Then  is the unique measure, such that:
dt
Z kY1
t 1 ;:::;t
 (XU1 ;:::;Uk ) =
k Kti+1 ti (mi ; mi+1)dm1 : : : dmk :
U1 :::Uk i =1

But there is no easy way to translate it to a Wightman-type picture.

We nish by a technical variation on the denition of Wightman functions which will
be used later.
2.4 Time-ordering. Let 0 2 SO(1; d 1) be re ection at the hyperplane Rd 1  V
(change of time direction). Then 0 can be lifted to an element  2 GC , but not to an
element of G. Hence the analytic continuations of Wn we constructed in section 1 are
-invariant, though Wightman functions themselves are not.
We will now present a variant of the denition of Wn which does not have this drawback.
Namely, the time-ordered Wightman function WTn is given by: WTn (v) = "lim !+0 n
S (v +
i"v). For pr(v) 2 Jn 1 we have: WTn (v) = Wn(v), where pr is the projection Vn ! V n.
The function WTn is  and n-invariant (unlike Wn).
Like a usual Wightman function, WTn is lifted from a unique function WnT on V n.
Let us describe W2T for the free scalar QFT of mass m; we will denote it by
F ( ; m2),
where F stands for Feynman. It can be expressed through functions
+( ; m2) = F(m+ ),
and
( ; m2) = F(m ).
Claim. We have:
a)
F (v; m2)=
(v; m2) for v 2 V ;
F (v; m2)=
+(v; m2 )=
(v; m2) for v 2 V;


b) (  + m2)(
F ( ; m2)) = 2i 0 , i.e. W2T is the fundamental solution of  + m2 ;
c) F(
F ( ; m2)) = 2i(p2+1m2 +i0) def
= "lim 1
!+0 2i(p2 +m2 +i")
(this is a well-dened distribu-
tion).
[Just to remind how such things look, we recall that x+1i0 and x 1i0 are well-dened dis-




tributions on R and we have: 21i x 1i0 x+1i0 = 0 and 12 x+1i0 + x 1i0 = x1 p.v. , where

1
 def lim R" 1

R
1 f (x)dx + 1 f (x)dx.]
x p.v. ; f = "! +0 1 x " x
9