All Possible Symmetries of the S Matrix* Sioxny Coma} avo Jaxvany Maxourat Lymay Laboratory of Phics,Harsord University, Cmbridye, Masochncels (Received 16 March 1967) We prove a new theorem on the impossibility of combining space-time and internal symmetries in any ‘uta trivial way. The theorem isn improvement on known results in that it applicable to ininite param. ter groups instead of just to Lie geoups. This improvement ie gained hy wsing information about the S ‘atria; previous investigations used only information about the single-pactile spectrum. We define © syxi- ‘etry group ofthe S matic as group af unitary operators which tum one-ptsticle states nto one-particle State, transform many-particle states a8if they were tensor prosicts, and commute with the S mats. Let Grbea connected symmetry group of the § matrix, and lt the fllowing five conditions hold: () contains 4 subgroup locally isomorphic to the Poincaré group. (2) For any AT>O, there are only a finite numberof ‘one-particle states with mass es than Mf. (3) Elastic sattring amplitudes are analytic funetons of + and, insome neighborhood of the piysial region. (4) The S matrix fs nontiva in the sense that any tro one particle momentum eigenstates scatter (ato somthing), except perhape at olated vals of. (6) The gen- rators of G waitin a integral operators in momentum space, have distributions for ther kernels, Then, we show that G is neeesarily locally isomorphie to the divect product ofan Internal symmetry group and 25 JULY 1967 ‘the Poinearé group I INTRODUCTION INTTIL a few years ago, most physicists believed that the exact or approximate symmetry groups of the world were (locally) isomorphic to direct products, of the Poincaré group and compact Lie groups. This ‘world-view changed drastically with the publication of the first papers on SU(6)!; these raised the dazzling possibility of a relativistic symmetry group which was not simply such a direct product. Unfortunately, all attempts to find such a group came to disastrous ends, and the situation was finally settled by the discovery of set of theorems* which showed that, for a wide class of Lie groups, any group which contained the Poincaré group and admitted supermultiplets containing finite numbers of particles was necessarily a direct product. However, although these theorems served their polemic purposes, they are in many ways displeasing: ‘Their statements involve many unnatural and artificial, assumptions, typically concerning the normality of the translation subgroup. Even worse, they are restricted to Lie groups—this despite the fact that infinite- parameter groups have been proposed in the literature. ‘The theories based on these groups were destroyed not by general theorems but by particular arguments ‘Typically, these arguments showed that these groups do not allow scattering except in the forward and back- ward directions Thus, if one accepts the usual dogma on the analyticity of scattering amplitudes, they allow no scattering at all. ‘The purpose of this paper isto tie up these loose ends. We prove the following theorem: Let G be a connected Work supported in part by the U. S, Air Force Ofice of Seine Retearch undet Contract AF (6i8)-1380, ‘fAltred P Sloan Research Fellow } National Sdence Foundation Postdoctoral Fellow. Fin Sita, Phys: Rev. 186, 11736 (1064); Ps Grsey and L. aati Ph. Rev eters 8, 27 86 5: Coleman, Phys, Rev. 185, 151262 (1065); S. Weinberg bi 139, HS07 og) E. he and Saute Anh Heat Poincaré 2, 167 (1068), ‘Jordan, Phys. Rev. 140, B766 (1065. 159 symmetry geoup of the $ matrix, which contains the Poincaré group and which puts a finite number of particles in a supermultiplet, Let the S matrix be né trivial and let elastic scattering amplitudes be analytic functions of s and #in some neighborhood of the physical region. Finally, let the generators of G be representable as integral operators in momentum space, with kernels that are distributions, Then G is locally isomorphic to the direct product of the Poincaré group and an i temal symmetry group. (This is a loose statement of the theorem; a more precise one follows below.) We believe that all of the assumptions in this theorem are physical, except for the last one, which, although weak, is ugly. We hope that it can be el sufficiently careful analysis; to date we have been unable to do so. ‘We emphasize that our theorem has application only to groups which are symmetries of the S matrix. There fore it has nothing to say about symmetry groups arising from the saturation of current commutators; these ‘groups generate symmetries of form factors only. ‘The remainder of this section contains a precise state- rent of the theorem and some remarks about its impli- cations. Section IT contains the proof. Although at times this attains mathematical levels of obscurity, we make no claim for corresponding standards of rigor. ‘A. Statement of the Theorem We begin by briefly reviewing some of the funda- mental definitions of scattering theory. The Hilbert space of scattering theory, 3, is an infinite ‘of subspaces, RATHI a 00 js called the m-particle subspace. Tt is a subspace (determined by the generalized exclusion principle) of the direct product of » Hilbert spaces, each isomorphic to %%, The S matrix S is a unitary operator on 3. A unitary operator U on 3¢ is said to be a symmetry 12st1252 s transformation (of the S matzix) if: (1) U turns one- particle states into one-particle states; (2) U acts on Imany-particle states as if they were tensor products of one-particle states; and (3) U commutes with S. A sroup of such transformations is said to be a symmetry group (of $). S is said to be Lorentz-invariant if it possesses a symmetry group locally isomorphic to the Poincaré group P. In this case we may introduce a basis for 50°, consisting of the plane-wave states la)yp), where p is the four-momentum of the state, 1s the spin index, and a labels the particle type (tech” nically, the irreducible subspaces of ® under the action of P)s The numberof irreducible representations of P contained in 3¢% is celled the number of partile ope ‘A-symmetzy transformation is said to be an internal symmetry transformation if it commutes with P. This implies that it acts only on particle-type indices, and has no matrix elements between particles of different four-momentum or different spin. A group composed of such transformations is called an infernal symmetry group. ‘The T matrix is defined in the usual way: in) PyT. ® Equation (2), of course, defines T only between states of the same four-momentum. The scattering amplitudes are the matrix elements of 7. It will be convenient for some of our subsequent arguments to define a subset of 300, called . 1 is the set of all single-particle states whose momentum-space wave functions are test functions—that is to say, in- finitely differentiable functions with compact support. We are nov in a position to state the main result of our investigation: Theorem: Let G be a connected® symmetry group of the $ matrix, and let the following five conditions hold : 41, (Lorentz invariance.) G contains a subgroup lo- cally isomorphic to P. 2, (Particleiniteness.) All particle types correspond. to positive-energy representations of P. For any finite 1M, there are only a finite numberof particle types with sass less than If. 3. (Weak elastic analyticity.) Blasticscattering am- plitudes are analytic functions of center-of-mass energy, 5, and invariant momentum transfer, fin some neigh Vorhood of the physical region, except at normal thresholds “4 (Occurrence of scattering.) Let |p) and |’) be any two oneparticle momentum eigenstates, and let |,p") be the two-particle state ‘ade from these. Then T\pp)0, ® «Spin bars for deen gent ate conesed by te wus conventions of EP. Wigher, Ann. Math, 40,149 (1939). TMConnected” means arose connested th the Weak operator topology. COLEMAN AND J MANDULA 159 except perhaps for certain isolated values of s, Phrased briefly, at almost all energies, any two plane waves scatter, 5, (An ugly technical assumption.) The generators of G, considered as integral operators in momentum space, have distributions for their kemels. More precise) ‘There is a neighborhood of the identity in G such that every clement of G in this li ‘one-parameter group g(0). Further if x and y are any’ two states in , then 1d = (a,g(O9)= (Ay), @ ia exists at £=0, and defines a continuous* function of 2 and 9, linear in y and antilinear in 2.7 Then, Gs locally isomorphic to the direct product of an internal symmetry group and the Poincaré group. B. Remarks 1, Note that we do not assume that G is a finite- parameter group. Note also that the theorem possesses a trivial corollary a result reminiscent of the famous theorem of O'Raifeartaigh®: All the particles in a G supermultiplet have the same mass. This is surprising, for it is known that there exist infinite parameter groups which do not obey O'Raifeartaigh’s theorem. We are able to reject these groups because our assump- tions are stronger than those of O'Raifeartaigh; our analysis shows that these groups can not be symmetry group of a nontrivial S matrix. 2, Lorentz invariance is a necessary condition for there are many examples of Galilean-invariant spin- independent theories," for which the corresponding ‘theorem does not hold. 43, Particlefiniteness is also necessary, in order to exclude the well-known infinitesupermultiplet theo- ries, most intensively investigated by Fronsdal and collaborators." 4. The analyticity assumption is somewhat surpris ing in this group-theoretical context. However, something that most physicists believe to be a property of the real world, and we will use it continually in our proof, If itis eliminated completely, the theorem is not true; groups are known which are not direct products, but which do allow scattering, although only in the forward and backward directions “Conn sane contnuon in the ual Schwa) topology for tet fanetons Waltrnatively, a may be thought of a fois dun! spe, SLs Gacaiigh, Phos. Rev, Lattrs 14 S75 (196): ys. ‘Rey. 198, BIOS (1968; jst Tiel. Phys. Acta 39, 309 1960); T-Segal J. Funetional Anal ts be published). SRE amar, Progr Thegt ye (yoo 37,198 (96 WEL P. Wigner, Phys Rev. Sh, 100, (93 A sine oni he Tenet of P Bua and G, Fromadal, High nergy Pasoies ond BlementryPares (Gncernatonal Atomic Bsargy Agency, Vienna, 1900), ica foneton from159 5, Some form of nontriviality is certainly required, for if $=1, the group ofall unitary transformations on 4 satisfies all the other conditions of the theorem. However, assumption 4 is much stronger than SL We believe itis possible to derive our assumption from much weaker assumptions, essentially equivalent to SrA, with the aid of somewhat stronger analyticity conditions (in particular, the crossing relations); but ‘wo have not yet completed this investigation. 6. Our fifth assumption is, as we have said, both technical and ugly. Tt is necessary for our proof be- cause we are physicists, and accustomed to the manipu- lation of infinitesimal’ generators. Since there is, in ‘general, no analytic-vector theorem for infinte-param- eter groups, a special assumption is needed to justify such manipulations. We feel that the assumption we have chosen is a weak one; for example, it does not imply that © is in the domain of all the group genera- tors, However, we have no doubt that more competent analysts will be able to weaken it further, and perhaps even eliminate it altogether. 7, Tt has sometimes been suggested that nonunitary bounded linear operators which commute with the S matrix might be of physical interest "These opera- tors, of course, would not represent symmetries in the usual sense, but they still might restrict scattering amplitudes in interesting ways. ‘The proof for the unitary case also works here, with the addition of a few minor supplementary arguments. For the reader interested in this ease, we give these arguments, where they are required, as footnotes. 8. The theorem is not true if, in the conclusion, local isomorphism is replaced by giobal isomorphism. For ‘example, it is possible to construct theories in which only particles of halfintegral spin have half-integral hhypercharge. The symmetry groups of these theories are locally, but not globally, isomorphic to SU()@P. 9, Finally, if there are only a finite number of par- ticle types, the theorem implies that the intemal sym- retry group has compact closure, However, if there are an infinite number of particle types, spread out along the mass spectrum, this need not be so. For example, every particle type could have associated with it an independent particle-number convervation law. ‘The internal symmetry group would then be a direct product of an infinite number of factors, each isomorphic to UC). This is not @ compact group. II, PROOF OF THE THEOREM ‘We begin with some trivial remarks about the genera- tors of G, defined by Eq. (4). We may readily extend, 8S, Coleman (Ref. 2). Lipkin, Phy. Rev, 189, 1633 (1965). has also been suagested that invariance under ah algebra with theprenan i ler nia ca ot xt be of physical terest, [Mato and D. Sterohermer, Bh ey" eters 15, 954 (1005; 16, 1188 (1966) Our theorem also haa in this case, with the obnfous recasting of our assumptions into purely lgebraie language. ALL POSSIBLE SYMMETRIES OF S MATRIX 1253 our definition of the generators to two-particle states; indeed, if 1/@xs and yy» are two-particle states in HDD, it is easy to show that, 1d iat OrByx, COLnOxD=Or@y» ALxD) = Onde) orad+ Grad nde). a) ‘The fact that G commutes with S implies that** (SD:@y]45Tn@2D=O@yAln@x). Sb) Lorentz invariance implies that if A is any distribution obeying Eqs. (5), then so is U(A,o)*AU(A,a) where U(A,e) is, a8 usual, the unitary operator representing the element (A,a) of P. Likewise, any convergent sum or integral of distributions obeying (5) will also obey (5). We denote by @ the set of al distributions obeying Eq. (9)! ‘As any reader of Dirac knows, it is sometimes con- venient to speak of distribution as fit were a function, We will follow this practice, and define Arear(fp)=(aX'p|Aladp), © ‘8 distribution in momentum space. Sometimes we will suppress the indices, and speak of A(p',p), a matrix- valued distribution, ‘We are now ready for the first stage of the proof. Let J be a test funetion, with support in a region Rin p space, and let f be the Fourier transform of j. Tt is easy fo show that the integral fesvrunavanionra © converges, and defines a distribution in @. Since U (10) |adp)=e**"*|a,2,p), ® then. LAG P)=10-P)AG'P)- @ ‘Thus f-A only has matrix elements between states ‘whose supports in p space are such that they may be connected by a vector in R. ‘Now, by our partile-finiteness assumption, the mo- ‘mentum support of one-particle states is restricted to a countable set of mass hyperboloids. Tt requires only trivial algebra (which we leave to the reader) to show that if Ris sufciently small, and does not contain the zero vector, there will be regions on these hyperboloids such that, if any vector in R is added to any vector in these regions, © vector is obtained which is on none of the hyperboloids. Thus, any state in whose support lies within these regions must be annihilated by f-4. The analyicty of the scattering amplitude guarantees that the lefthand sie of Iq. (Sb) is well dened, “iNroie that we do nst aseet that (ie Lie algebra, nor that ever) elem of @ is obtained by differentiating sone Barameter soup.1254 Tic. 1. The #4 plane in p space for particles of lowest mass. ‘The operator j-A ‘anata al states whose momentum sup- ports tre outside the shaded hands, Figure 1 depicts the hyperboloid of lowest mass for a particular situation. R has been chosen to be @ small sphere surrounding a vector lying in the 2 direction. One-particle states whose supports lie outside the shaded bands are annihilated by f-A. There is one band for every mass hyperboloid. For the sake of clarity, we will construct all our subsequent arguments for’this situation; they may easily be extended to the general ‘Let x be a state in S¢® orthogonal to all states anni- hilated by f-A. Let p be a momentum in the support of #. By the arguments above, p must lie in one of the shaded bands. Let 9, p,q, be points as shown, chosen such that ptqnytd. (10) Let us define, as usual s=(p+q) and (o-Ps (at) and lets be chosen so that it lies below the threshold for production of pairs of the nest-heaviest patil. ‘Now, let us construct an inital tworparticle state by puting one particle in the state + and another particle in an arbitrary state with momentum support loalized about 4. Likewise, let us construct a final two-particle State by putting one particle in an arbitrary state with support localized about p” and the other in an arbitrary State with support localized about ¢. The S.matrix element between these two states must e zero, because GA obeys Eqs. (5). This implies that the elastic seat ‘ering amplitude vanishes for 3 and fas defined by Ea. (Qi); if the particle with momentum p iin a state determined by the wave function of = evaluated at p. However, by making «rotation in the rst frame of p— that is to say, by transforming g and 9” in accordance wrth such ® rotation—and realizing that the whole {Argument goes through without change in this ease, we can eliminate this lst restriction, and deduce that for 8. COLEMAN AND J. MANDULA 159 at least one particle type a, a8 scattering is zero at these values of s and t for any particle 8. But now we can choose a slightly different configura tion, and change the values of § and £ continuously. ‘Thus we deduce that scattering vanishes for a range of sand f. By analyticity, this means that af scattering vanishes everywhere, This is in contradiction with the nontriviality of the S matrix, Thus, z=0, and f-A annihilates all states with support on the lowest mass hyperboloid. Now we can go on, inductively, to the hhigher-mass hyperboloids. By simple repetition of the arguments above, we finally conclude that fAso. (2) Since f is an arbitrary test function whose support does not include the zero vector, this implies: Lemma 1: The support of A(p',p) is restricted to the set p=p". In particular, this lemma implies that A can not connect states on different mass hyperboloids; there- fore, neither can G. We thus have a generalization of O’Raifeartaigh’s theorem" for syrametry groups of the S matrs. Tt is a well-known result of distribution theory that a distribution whose support is a point is a finite sum of derivatives of é-functions. In our case, this means that on each mass hyperboloid, A may be written as a differential operator. Oi course, since A acts only on functions with support on the byperboloid, this dif- ferential operator does not involve differentiation with respect to all four components of p, but with respect to the three components tangent 10 the hyperboloid. ‘That is to say, on each mass hyperboloid, A is a pol nomial in the tangential differential operator: Vi=0/Ap.— me “pba (a3) (We adopt the summation convention for Greck i dices.) The next lemma isa restatement of these trivial ‘consequences of lemma 1 in terms of equations. Lemma 2: Any clement of & may be written in the form as) where the A“ are matrix-valued distributions. Also, 1 i goin hig Le na fabio ei Raa Oa Rovpuplisiale seat alee ie Se esc ca cn Cate cpa iat TESS an Hee engl teens Root ce Se ccruatce dary nee TRG HE SEARS Ore ea ak PASTS