Wightman axioms.

David Kazhdan Introduction. In this series of lectures we will present Wightman axioms and show how this set of axioms can be used to derive deep and unexpected results on behavior of quantum eld theories. These axioms constitute a consistent mathematical theory which can serve for a mathematician as a gateway to the strange world of quantum eld theories. In spite of the failure to describe the structure of gauge theories Wightman axioms are very useful for developing of an intuition on quantum eld theories. 1.0 Setup and notations. Let V be a pseudo-Riemannian manifold of signature (1; d 1). In these lectures we will only consider the case V = R1;d 1 with constant quadratic form. We denote by ( ; ): V V ! R the corresponding quadratic form and will write v2 instead of (v; v) for v 2 V . We will sometimes use decomposition V = R Rd 1, where R (time) and Rd 1 (space) are orthogonal subspaces equipped with constant positive (respectively negative) de nite metrics. This decomposition is xed throughout the lectures. We denote: Vspace = fv 2 V jv2 < 0g; and V space is the closure of Vspace. The complement V V space consists of 2 connected components, V+ and V (where V R Rd 1). For m 2 R we put Om = fv 2 V jv2 = m2; v 6= 0g. Let dv be the standard measure on V and let m = Res O+ v2 dvm2 be the invariant volume form on O+ . m m Let G = Spin(V ) be the double (universal for d > 3) cover of connected component of the group SO(1; d 1). The semidirect product P = G n V is called the Poincare group. It acts on V by isometries. We denote by 2 G the nontrivial element of the kernel of the map G ! SO(1; d 1). We will use the quadratic form to identify V with V , but we will distinguish between vectors and covectors by reserving the letters v, w, etc. for vectors, and p, q, etc. for covectors. Note that the cone V+ is self-dual (i.e. fp 2 V j(p; v) > 0 8 v 2 V+g = V+). For a function f on V we will denote by F(f ) its Fourier transform; so we have: 1 Z e i(p;v)f (v)dv: F(f )(p) = (2 )d=2

Lecture 1.

Let U : V ! Aut(H) be a unitary representation of the vector group V in a Hilbert space H. De nition. Representation U is called positive if support of the corresponding spectral measure lies in V + . If U : P ! Aut(H) is a unitary representation, then we say that U is positive if U V is positive.

For a G-module H we will write H = H even H odd, where H even (respectively H odd) is the space of invariants (respectively antiinvariants) of . We will endow H with the structure of superspace given by this decomposition. Let R = Reven Rodd be a nite dimensional real representation of G. Since P acts transitively on V and stabilizer of 0 is G, we can associate to R a P -equivariant vector bundle R on V . R Let SR = SR even Sodd be the space of Schwartz sections of this bundle. (Note that our bundle is trivialized equivariantly with respect to translations, and SR is identi ed with the space of smooth rapidly decreasing R-valued functions on V .) We will often abbreviate S = SR . Finally we need some preparations on signs. We say that a complex superspace H = H even H odd is a superHermitian space if it is endowed with an even R-bilinear pairing ( ; ) : H H ! C such that ( x; y) = (x; y) = (x; y) and (x; y) = ( 1)jxj jyj(y; x). Note that since ( ; ) is even we have (x; y) = 0 for x 2 H even ; y 2 H odd. H is called superHilbert space if (H even ; ( ; )) and (H odd; i( ; )) are Hilbert spaces. For T 2 End(H ) the (super)adjoint operator T is de ned by: (Tx; y) = ( 1)jxj jT j (x; T y). An operator is superHermitian symmetric if T = T . We will often omit the word super. Now we are ready to formulate:

1.1 Wightman axioms. Data:

1) A nite dimensional real representation : G ! Aut R. 2) A positive unitary representation U : P ! Aut(H) in a (super)Hilbert space H. 3) A dense P -invariant subspace D H, and a P -invariant vector 2 D, k k = 1. 4) A linear map : SR ! End(D), called the eld map such that:

) is P -equivariant where P acts on End(D) by conjugation. ) (f ) is superHermitian-symmetric. ) For any 1 ; 2 2 D the map f ! h 1 ; (f ) 2 i from S to C is continuous (S = SR is equipped with the Schwartz topology). ) The space D is generated by vectors (f1 ) (f2 ); : : : ; (fn ) , for f1 ; : : : ; fn 2 S. ") (space-locality) If f1 ; f2 2 S are such that for any v1 2 supp(f1 ), v2 2 supp(f2 ) we have (v1 v2 ) 2 Vspace (such f1 , f2 are called space-like separated), then (f1 ); (f2 )] = 0 where ; ] stands for supercommutator. Remark. It is often assumed that the space of V -invariants in H is 1-dimensional (uniqueness of vacuum). Such set of data satisfying the above properties is called QFT in Wightman axioms (or simply Wightman QFT).

V (n times); Rn = R n (this is a bundle on Vn). The space of Schwartz sections of Rn will be denoted by Sn. For n 1 the Schwartz theorem and the property ( ) imply the existence of unique tempered C -valued distribution Wn 2 S0n such that

1.2 Wightman's functions. Let us denote: Vn = V V

Wn (f



fn ) = h ; (f1 ) (f2 ) : : : (fn ) i

for f1 ; : : : ; fn 2 S. These Wn are called Wightman functions, or correlation functions of the theory. (We will use the word \function" for a section of an appropriate vector bundle when it causes no confusion). 1.3 Reconstruction of QFT from Wightman functions. We are going to show that a QFT is uniquely determined by its Wightman functions. So we assume that a representation of G and a collection of distributions Wn 2 Sn are given, and that Wn are Wightman functions of some Wightman QFT ( ; H; D; ; ). We want to reconstruct H; D; and . Let us put e = n Sn, W = n Wn 2 e0 (where we use the standard notation L0 for the S S topological dual of a topological vector space L). The space e carries an algebra structure: S f1 f2 = f1 f2 . For f 2 Sn we de ne f 2 Sn by: f (v1 ; : : : ; vn ) = f (vn ; : : : ; v1) (the sign is determined by the super-rule as usually). Consider a bilinear form f ; g on e S given by: ff1 ; f2 g = W(f1 f2):

Claim. ff ; f g = h (f ) ; (f ) i.
1 2 1 2

Proof. Clear.

We can recover H as the Hilbert space completion of (e C ; f ; g0 ), where f ; g0 S is the unique Hermitian scalar product on e C , such that f ; g0 e = f ; g. Then S S we can take D to be the image of the obvious map e C ! H and de ne as the image of S 1 2 e under this map. Since P acts on e preserving f ; g this action de nes an action S S of P on H. Finally, we get an algebra map : e ! End(D). S Before we formulate the next Claim let us x some notations. Let : V ! Vn be the diagonal embedding, and V n = Vn = be the quotient space. Since Rn is P -equivariant it de nes a bundle Rn on V n. Let Sn be the space of Schwartz sections of Rn. For f 2 Sn let f 2 Sn be its averaging over (V ), and let pr be the dual map on distributions. We identify V n!Vn 1 by: (v1; v2 ; : : : ; vn) ! (v2 v1; : : : ; vn vn 1):

Claim. Wn satisfy the following properties: 1) Wn is P -invariant for all n. In particular, Wn is (V )-invariant; hence Wn = pr (Wn ) for unique Wn 2 S0n. 2) supp F(Wn) (V )n . 3) Wn(f ) = Wn (f ). 4) W(v ; : : : ; vi ; vi ; : : : ; vn ) = W(v ; : : : ; vi ; vi ; : : : ; vn) if vi vi 2 Vspace. (Here the sign comes from the Z=2 grading on the space of test function Sn = Sn Sn .) 5) W(f f ) 0.
+ 1 1 +1 1 +1 +1 +

Proof. 1) and 3){4) are clear. Let us prove 2). We have to check that Wn (f ) = 0 if for some i supp(F(f )) lies in the domain f(v1; ::; vn ) 2 Vnjvi 2 V V +g. For such f we can nd g 2 Sn+1 with g = f satisfying: supp F(g) f(v1 ; : : : ; vn+1) 2 Vn+1jv1 + + vi+1 2 V V +g. So we have to show that for such g: Wn (g) = 0. We can assume g = g1 g2 for g1 2 Si+1; g2 2 Sn i , and supp F(g1) f(v1 ; : : : ; vi+1 ) 2 Vi+1jv1 + + vi+1 V V +g. We have Wn(g) = h (g1 ) ; (g2) i. But the spectral support of (g1 ) is contained in the set fv1 + + vi+1j(v1 ; : : : ; vi+1 ) 2 supp F(g1)g. On the other hand, it is contained in V + by the positivity condition. Hence, it is empty, and (g1 ) = 0 ) Wn (g) = 0. Remark. By a standard theorem from analysis property 2) is equivalent to the following: 20 ) Wn is a boundary value of an analytic function on Vn 1 + i(V+)n 1 . It is not hard to see the following Theorem. The above constructions provide a bijection between the following sets of data: a) ; H D 3 ; satisfying Wightman's axioms; b) representation of G and a collection of distributions fWng satisfying properties 1) { 5) above. Remark. Sometimes one considers theories satisfying weaker axioms. Namely, let S0 Sn n be the subspace of sections which vanish on diagonals together with all partial derivatives. A QFT satisfying weak Wightman's Axioms is (a representation together with) a collection of Wn 2 (S0 )0 satisfying 1), 20), 3), 4) above. n Note that an essential property 5) can not be formulated if Wn are de ned on S0 only n because f f does not lie in S0. 1.4 Spin-statistics Theorem. This is a statement that our convention on the parity of the basic superspace R is the only right one. More precisely, we have: Theorem. Consider the Wightman axioms with the sign in the axiom of space-locality changed. Any theory satisfying this modi ed set of axioms is trivial. (The eld map = 0.) Proof. Let us take a 2-dimensional subspace U V of signature (1,1). Let 0 be the element of SO(d 1; 1) which acts by 1 on U and by 1 on the orthogonal complement;

let be an element in the complexi cation of Spin(d 1; 1) which projects to 0 . Then 2 = is the spin element. Also note that the eigenspaces of in the space of R C are de ned over R. If acts trivially in R then this is obvious because is real; otherwise acts by 1 and we use the identity = to see that the i eigenspace of is invariant under the conjugation. In the next lecture (x2.2) we will show that Wn extends to an analytic GC -equivariant function on a certain open domain in the complexi cation V n C . In particular W2 is e analytic in the domain T := fv j v2 62 R 0g. This analytic function is also denoted W2 . e Let now r 2 R be a eigenvector; let v be a point in U C \ T. We claim that hr r; W2 (v)i = 0. Indeed we have hr r; W2 (v)i = h (r) (r); W2 ( v)i = spin(R) hr r; W2 ( v)i where spin(R) is the eigenvalue on R; here the rst equality follows from -equivariance, and the second is obvious from 2 = . On the other hand by our assumption hr1 r2 ; W2(v)i = spin(R) hr2 r1 ; W2( v)i e for v 2 V space. Since V space T, we get hr r; W2(v)i = 0 for v 2 V space, and hence for e all v 2 T. R For any real valued Schwartz function f on V we have h (f r) ; (f r) i = hf r f r; W2 i = R R hF(f ) r F(f ) r; F(W2)i = F(f ) F(f ) hr r; F(W2)i. This shows that the distribution hr r; F(W2)i is nonnegative. Now let us take v 2 V +. Then we get 0 = hr r; W2(iv)i =

exp( v; p) hr r; F(W2)i :

Since exp( v; p) is positive this implies that hr r; F(W2)i = 0, hence ( (f r) )2 = 0, so the theory is trivial. We nish by some elementary properties of Wightman functions, and their relation to properties of the eld map. 1.5 Mass spectrum of a theory. Let us denote +(p; m2 ) = F( O+ )(p). m Claim. For any scalar eld theory : S(V ) ! End(D) there exists a positive measure d on 0; 1) such that: Z W2 (p) = + (p; m)d (m) :
Proof. We have to show that F(W2) = d (m) O+ for a positive measure d . Since O+ m m is an orbit of the group G it is obvious that any G-invariant distribution supported in tO+ = V+ can be written down in such form for some distribution . In order to see that m is positive we have to show that F(W2) is positive. For any f 2 S we have W2 (f f ) = h (f ) ; (f ) i 0. Hence for g = F(f ) we have F(W2) (g g) 0. We know that F(W2) is concentrated on the anti-diagonal and

F(W )(p; p) = F(W )(p)
2 2


Now note that f 2 S i g = F(f ) 2 SC and g( p) = g(p). Thus for any such g which means that F(W2) is positive. De nition. The mass spectrum of a QFT is support of the measure d . The mass of the theory is the in num of its mass spectrum with 0 removed. 1.6 Asymptotics of Wightman functions. The next statement is a baby example of relation between analytical behavior of Wightman functions and properties of the eld map. Estimates of this kind will be used essentially in the lecture on scattering theory (see lecture 4, x4.3, Proposition 1 and Claim 1).

F(W )(g g) = F(W )(g g) 0
2 2


a) Uniqueness of vacuum in a QFT is equivalent to the equality:

lim Wn(v1 ; : : : ; vi ; vi+1 + a; : : : ; vn + a) = Wi (v1 ; : : : ; vi)Wn i (vi+1 ; : : : ; vn) !1
being true for all i, n and a such that a2 < 0. p b) Assume that mass of the theory is m0 > 0. Then W2 (v) e m0 v2 as v2 ! 1. Proof. a) For any f1 ; : : : ; fn 2 S we have

where pr is the projection on HV . The statement follows. b) We have

where T a stands for translation by a. From the positivity condition it follows that the weak limit lim U (T a) exists as a bilinear form (i.e. all matrix coe cients converge), and !1 equals to the projection on HV . Hence lim h (fi ) : : : (f1 ) ; U (T a ) ( (fi+1) : : : (fn ) )i = hpr( (fi ) : : : (f1 ) ); pr( (fi+1 ) : : : (fn ) )i !1
1 Z
0 1

Wn (f ; : : : ; fi ; T a(fi ); : : : ; T a(fn )) = h (fi ) : : : (f ) ; (T a(fi )) : : : (T a(fn )) i = h (fi ) : : : (f ) ; U (T a )( (fi ) : : : (fn ) )i
1 +1 1 +1 1 +1

W2 ( v ) =



F(W2)(p)ei(v;p) dp =
1 Z

d (m)


O+ m

e(iv;p)d m (p)

where m is the invariant measure on O+ . For v 2 Rd m const +

f0g it can be rewritten as:

d (m)




ei(v;q)dq= m2 q2



(Remember that q2 = (q; q) is negative for q 2 Rd 1 f0g). The integrand is analytic in the region Re(m2 q2) > 0; it also decreases as (~)2 ! 1 p 0 for xed p0 , where p = (~; p0) is decomposition into space and time components. Hence we p p can shift the contour of integration from Rd 1 to C d 1 where is Rd 1 + im0 v=p v 2 squeezed slightly towards Rd 1 in a small neighborhood of the singular point im0 v= v2 to avoid the singularity. Then the estimate becomes obvious. We will usually assume that uniqueness of vacuum holds for the theory under consideration.