RucRIvED: May #1, 1998, ACCEPTED: October 5, 1998 Massless picture, massive picture, and symmetry in the Gaussian renormalization group Christian Wieczerkowski Institut fiir Theoretische Physik I, Universitét Miinster, Withelm-Klomm-Strofe 9, D-48149 Minster E-mail: wieczerQuni-menster.de ABSTRACT: We study transformation properties of momentum space renormalization group transformations. The renormalization group transformations are composed of an integral over a high momentum field with a scale transformation of a low momentum field. Asan example, we consider perturbations of a massless real euclidean free field ¢ with a smooth momentum space regulator. We show that this renormalization group admits two equivalent formulations culled the massless picture and the massive picture. In the massive picture, the renormulization group is shown to have a symmetry. The symmetry consists of global (space-time independent) scale transformations of the field composed with a certain (gaussian) integral. We then translate the symmetry back to the massless picture. The relation between the symmetry and the notion of an anomalous dimension is briefly discussed. KEyworbs: Renormalization Regularization and Renormalons, Field Theories in Lower Dimensions, Stochastic Proce:Contents Be . Introduction . Renormalization group - massless picture 2.1. Covariance Ch, 2.2. Dilatation D, 2.3. Transformation R(Ci, Di) 2.3.1. Physical meaning of R(C,, D1) 2.3.2. Semi-group property of R(C., Di) 24. Generator R(O,D) Renormalization group — massive picture 3.1. Gaussian fixed point 3.2. Transformation Q(C) 3.3. Substitution Z(¢) = Z,(#) 2(4) 3.4. Covariance c, 3.5. Dilatation d, 3.6. Generator R(, d) . Symmetry — massive picture Al. Symmetry pair (1,41) 4.2. Commutator of R(ex,di) and R(qu, 6x) 4.3. Composed transformation 4A. Differential pair (7,4) . Symmetry — massless picture 3.1. Al, BY, C!, and C” for the symmetry 3.2. A! BI, C!, and C" for the composed transformation 5.3. Generators in the massless picture .» Anomalous dimension 7 6.1. Fixed point (2(4),7) 6.2. Discussion of (2. ($),) 6.3. emodel 6.4. Differential equation for u,(d) Caowmwowsan 10 10 11 gn 12 12 12 13 13 14 14 15 15 16 16 17 18 18 18 19 191. Introduction Momentum space renormalization group transformations [1] huve a wide range applica- tions, both in statistical physics and in euclidean quantum field theory. We will restiric OUr attention to a basic class of models, to perturbations of a massless one component real scalar field ¢ on euclidean space-time RY. The momentum sp: e renormalization group comes in two equivalent versions, a differential type and a discrete type. The dynamical equation is a functional differential equation in the former and a functional integral in the latter case. We will display our main formulas parallel in both ver- ions. As a very successfull application, we mention the elegant proofs of perturbative renormalizability (2, 3]. A key idea of the renormalization group is to sort models into universality classes. The universality classes are understood in terms of fixed points and their invariant manifolds, which are propertis properties of a specific bare action. But there are as many renormalization groups as there are ways to regularize a field theory. Therefore, a natural problem is to search for relations among different renormalization groups. Although this question is central in of the transformation rather than the understanding of universality, it has received little attention in the literature. See however [4]. The subject of this paper is a particular relation, the symmetry associated with global scale transformations. Although this paper will remain on a formal level, [ will borrow convenient notation from mathematical renormalization theory, as for instance [5]. Since it may be unusual in field theory, I will briefly explain my notation together with the setup. Transformations of the following kind are tailor made te study perturbations of free massless fields. We consider effective field theories, with a unit ultraviolet cutoff, defined by an effective action $(¢) = T(¢) + V(¢), which consists of a kinetic term. T() and an interaction V(¢) The kinetic term is a quadratic form a i. Te) = 5 (+. Ge 2) (1.1) and is given by a free propagator Cz,. We build the ultraviolet cutoff into Cx2 We will keep Cyp fixed. The renormalization group will transform only V(@). Any action can be written in this form, at the expense that the interaction V(¢) will in general contain another quadratic form 1 é a Vee) = 5 fata f aan Wles.2y) (e.) $lee) + higher versions, (1.2) which corrects Cy. Of course, it will in general also contain higher vertices, for instance a ¢*-interaction. To avoid confusion on an important point, let me emphasize that the interaction is understood as an effective action and will in general include non-local 1 We use dimensionless units. The field ¢ is always scaled such as to have a unit ultraviolet cutolf 2 The subscript x. indicates that Cue is a massless propagator without, infrared cutoff.verti Enective actions are generated rom bare actlous by lierated renormalization group transformations. The Schwinger finctions of this effective field theory are generated by the functional integral [De axp(-S(¢) + (4 4)) exp (W(J)) = Soe oe) =" D6 exp 5) where ($,J) = [d?x9(z) d(x). The theory is understood to be in infinite volume? If Cay is massive, and provided that V(¢) is well bebaved,! then (1.3) can be analyzed by standard technology. The renormalization group comes into the play when Co. (23) is massless. It reduces the computation of (1.3) to an infimite sequence of massive integrals. In perturbation theory, the Schwinger functions are computed as sums of connected Feynman diagrams built with the propagator C,. and all the vertices V,(zi,..- ,Zn) contained in V(¢). Although we do not intend to perform a perturbation expansion, we will think of V(¢) asa perturbation of the free theory, defined by Cae. The following notation, which is widely spread in the mathematical literature, is adapted to this point of view. The formal Lebesque measure D¢ can be combined with the quadratic action to a gaussian measure” diga() = det(@xCn)"? D6 ex {5 (+ Z-#) } (13) on field space {¢: R? + R}. The propagator Coe is called the covariance of djic,, (¢)- Furthermore, we can trade W(J) for the generating function Y() of the connected Schwinger functions, amputed by C... It is given by WUD) = 5 (Ca) -V Coal) 0.9) All information in W (J) is also contained in Y(®). This new generating function has the path integral representation Sdn) exp(-V(6 + 4)) ol YO) = Tae, (8) exa(-V 0) Finally, we will abbreviate Z(¢) = exp(—V(¢)) and speak of Z(g) as a perturbation® of the gaussian measure. To compute the Schwinger functions of the effective field an *Think of (1.3) as @ lat improved action. 4¥(@) should be approximately local. ¥(¢) should alsa be steble. A less formal definition of the gaussian measure is given by its Fourier transform e field theory on an infinite unit lattice with a renormalization group [teeth extn) = on {-} u.cns)} a4 rather than (1.4). *2(¢) will be normalized such that, Z(0) = 1.theory, the effective Interaction has to be convohited with a massless gaussian Measure. The idea of the renormalization group is to decompose this problem into digestible pieces. The first step is to decompose C., into two parts, a massive short distance part Ci), and a massless long distanc: for gaussian measures, (1.7) then becomes a double integral, one integral over short distance fluctuations and another integral over long distance fluctuations. Consider the following particular decomposition, which uses a dilatation of the remainder. Thanks to the convolution formula field. Let: Cao(z,y) he decaying like a certain power of |z — y|. Say, Cao(z,y) is of the order O (|z— y[?-2-*) at |z— y| 4 00. We assume a unit ultraviolet cutoff, built in a8 a smooth momentum space regulator, such that C..(z,y) is a regular function of |z —y| at short distances. In this situation, the following renormalization group [3] is custom built to study the properties of (1.7). Let D, be the dilatation operator (whose transposed is) given by DE(f)(z) = L4-°* f (2). We decompose Ore into Ch = Cn — DE Cy, Di (1.8) The dilatation operator is made such that the power like asymptotic tails of Coo and its scaled version cancel each other. The result is a massive propagator C,(z,y) which is of the order O (F (42) at |c —y| + 00 with a rapidly decreasing function F(t). (We will choose an exponential cutoff where F(#) = For O,(2,y) to be short ranged, the asymptotic of Oxo(z,y) and its scaled version (D! C,, D1)(z,y) = L?-0-« (2, 2) have to agree. This condition determines the scaling dimension in the dilatation operator D,. One cannot decompose Coo with an anomalous dilatation, unless one gives up that C,, be short ranged. The convolution formula for gaussian measures allows us to do the integral (1.7) in two steps. We have that — [Abaveove bine) J Ate, (Onign) ZEB + drow + ign) © Pditn,c.q0¢ (Pine) J bite, Gaian) 2 (Piow + Pion) exp(-V(8)) (19) The short distance in momentum. spi covariance C, has a unit ultraviolet cutoff and an infrared cutoff 1/L The long distance covariance has an ultraviolet cutoff 1/Z and no tof. Iterating this decomposition, one obtains a sequence of integrals over vale If similar in the sense that all fluctuation integrals obtained by iterating (1.9) are scaled versions of a single integral. Thus by rescaling ¢iow with D,, we obtain the transformation _ Sane.(9) Z(DE) +9) J dye, (9) 2(9) of the perturbation Z(¢@). Ik defines a representation of the semi-group of scale transfor- mations with scale factor L > 1 on a suituble space of perturbations Z(¢). (For L <1, C1, becomes negative.) Furthermore, it can be shown to perform a dilatation of the fully infrared «1 fluctuations on longer and longer s ‘The special decomposition (1.8) is R(Ci, DiNZ\O) (2.19)integrated potential U(®). Various applications of renormalzation transtormations of this kind are reviewed in [6]. The subject of this paper is to study general properties of renormalization group transformations of the form (1.10). The subject is not to compute the renormalization group flow for a particular model. To save space, I find it convenient to abbreviate (1.10) by (Cz, Dz). The meaning of this abbreviation is as follows. With the pair of operators (C,,,Dz), where C,, is a short ranged covariance, and where D,, is a dilatation, I associate the renormalization group transformation (1.10). The usefulness of this shorthand relies on the fact that the composition of two transformations of the form (1.10) is another transformation of this form. We will show that the transformation given by (C,, D,) is equivalent to the trans- formation given by another pair (c;,d,). The equivalence is of the form ze) &2 2g) e0)| or) (1.1) Z(|L) <— 2(¢lE) QU) where Q(K.)(2)(¢) = Ze(¢) 2(4) and Z,(¢) = eo HOS") is a short ranged gaussian fixed point of R,(C,,Dz). The new pair (cz,,d,) defines a scale decomposition Cos = Coe di + c1, of w covariance tac. The two formmlations will be called the massless picture and the massive picture respectively. They ure well known in the hierarchical (or ultra-local) approximation [7]. In the massive picture, both ¢, and tq are short ranged. Therefore, the scaling dimension of dy is variable. In the massive picture, we will show the transformation defined by yet: another pair (7,4.), which defines a globul (space-time independent) scale decomposition of Css, with 6,, the multiplication by a global factor L—1, is a symmetry. Renormalization group and symmetry together spread a perturbation 2(¢) to a surface ae) ws oh, L) wou| [onan (1.12) 2651) + a9, 2) Rye Bus) We will then use the equivalence to translate the symmetry back to the massless picture. The composition of a renormalization group transformation, which scales by a factor L, and a. symmetry transformation, which scales by a factor L?, defines an anomalous ale decomposition of eso. Inspired by the construction of a non-trivial gf, fixed point in the planar ap- proximation [8] and beyond it [5], we consider the renormalization group given by Cu = F(-A,e) (-A)-*+8, where € is a small parameter, and where F(—A,¢) is a certain cutoff function. We will argue that the two cases e = 0 and ¢ # 0 are different. onWe will brielly discuss the Telatlon between the anomalous dimension 7 and the sym- metry applying the concepts of [9] to this case, The anomalous dimension of the #4, fixed point is by definition zero. 2. Renormalization group — massless picture We consider renormalization groups of transformations R(C., Dz), defined by a pair le, with Z > 1. Our starting point is a particular pair (Cz, D1), which defines the renor- malization group in the massless picture. of operators (Cz, Dz), where C, is a covariance,’ Dz, is a dilatation, and L is a sc 2.1. Covariance Cz, We start from a scale decomposition of the form (1.8). Let C,, be the covariance operator, acting as a convolution O,(f)(z) = f d?y O..(2,y) f(y) in real space, whose kernel given by the fourier integral cute) = f gird Gio, (23) where Cz(p) is a scaled difference of propagators Gi) = ao (2.2) Here 2(p) is a cutoff function, and eis small parameter (possibly zero). At € = 0, the free theory corresponding to (2.1) is simply a regularized massless free field. Ate #0, it is a statistical field theory with long ranged kinetic term. We choose a euclidean invariant regulator 2(p) = F(p?). The function F(#) could be a general monotonously decreasing function, which takes values between F(0) = 1 and lim, 4. F(s) = 0. For definiteness, we choose the smooth regulator, given by the incomplete [function (2.3) For ¢ = 0, tt becomes an exponential cutoff: (p) = e7?*. The advantages of this particular cutoff function are apparent from the integral representation Gio) - Se (24) Therefrom, the real space kernel is immediately computed as a gaussian integral and shown to decay exponentially. It reads _ Ses CLz,¥) = 7 The name covariance will be used as synonym for a free propagator.The covariance C;, is thus a two sided regularization of (—A)~**?, with unit ultraviolet cutoff and infrared cutoff 71 (in units of mass). We included a parameter e for the sake of generality, and to distinguish it from another parameter, the anomalous dimension nto appear below. We will assume that D+ > 2, so that (2.5) remains integrable in the infrared limit L + oo. 2.2. Dilatation D, When the infrared cutoff is removed, (2.5) becomes a massless covariance Ox, = lim oc Ci, with a unit ultraviolet cutoff. In this limit, (2.4) simplifies to any, _ FO-3”) P= aD are It follows that Cyo(z,y) decays powerlike at |z — y| + 00. Indeed, (2.5) becomes (2.6) 7 (4 1 ey) Coo(t,y) = (2.7) (an) P(A §) EP Conversely, C, is the difference of Czy and a scaled version of Cp. Equation (2.2) reads Gi(p) = Cal) — 29-* Gy (Ep). In real space, we therefore find that Oulu) = Cue(e¥) — P-?~* One (FF) (28) To save space, we will use the operator notation Cz, = F(—A)/(-A)!-#. Then (2.8) becomes C, = Cx — DY Og Di, where D, denotes the dilatation operator Dilf\(2) = P+ fiz), (2.9) and where Dj’ denotes its transposed Di(Ae) = LPF #(F)- (2:10) In other words, D, represents a dilatation by the scule factor Z on function space {f: RY — R}, with scaling dimension o = 14 25.8 To summarize, we have a pair of operators (C,,, Dz), depending on a scale factor L, where 1. Dy is. dilatation, with D) =1, D, Dy =Dyy, and 2. Cy is a short ranged covariance, with Cz = Co — DE Cun Di. In this situation, we say that (CL, Dz) defines a scale decomposition of Cop with respect to Di. "The dilatations by scale factors L > 1 form a semi-group.4.3. Transformation (0, 4.) As explained in the introduction, we associate a renormalization group transformation R(C1, D1) with the pair (Ci, Di). It consists of a gaussian convolution with respect: to C,, composed with a dilatation with respect to Dz: _ Lape, (9) ZDEG) + 4) Faye, (@) 2) Here dyic, (4) denotes the gaussian measure with mean zero and covariance C, defined by fdjia,(4) &!) = eFGC..2_ The transformation acts on perturbations Z(¢), which will be normalized such that Z(0) = 1. R(Ci, Di)(Z)(®) (2.11) 2.8.1. Physical meaning of R(C., Dz) Admit a brief detour around the physical meaning of (2.11). We consider effective quan- tum field theories defined by the covariance Cs and an interaction V(¢) = —log Z(¢). Their physical content: is coded in the generating function Lina $) Z(G) oF? J Ape0 (8) 208) where 2(6) * is given by a convolution of Z(d) with dite,,(d), divided by a normaliza- tion constant: = FU) 2106), (2.12) — {dite 9) ZG + 4) 20)= Fie. (8) 2) * (2.13) The transformation R(C,,D,) performs a dilatation of Z(5): We have that vga, _ Lalla) R(Cis Di ZV +8) BOE) = Fai, (@) Ca, DZ) The transformation R(C:, Dz) thus enables a study of the behavior of Z(6) under Di, without having to perform the convolution with respect to djsg,,(¢), which is difficult becuse Czy is long ranged. If R(C,, Dz)(Z,)(8) = 2,(6) then 2(D!(6)) = 2(6). In other words, we rediscover the well known fuct that the correlators become dilatation invariant at a fixed point of the renormalization group. (2.14) 2.3.2. Semi-group property of A(C,, Dz) We are lead to consider the following structure. We have a pair of operators (A, B) (with certain properties). With this pair, we associate a functional transformation RUA, By (e) = Het 2 (2.13) *2(®) equals exp(—V()) and is the generating function of the free propagator amputated Schwinger functions.The particular pair (A, 6) = (0,1) gives the identity transiormation. since (©), 14) = (0,1), we have that R(Ci, D1) = 1. The composition of two transformation (2.15) is another transformation of this kind. Let (A,B) and (A!, B') be two pairs of operators. A standard computation with gaussian integrals reveals that R(A, B) R(A', B)) = R(B™ AB+ A’, BB"). (2.16) Equation (2.16) can be interpreted as a product (A, B)(A!, BY) = (BY! AB +A!, BB’) of (A,B) and (A',B'). From the two properties Di Dur = Diu and Cr = Coo — Di Coq Dz, 3h follows that (Cz, Dz) (Cx, Dus) = (Curr Durr). Therefore, the trans formation R(C,, Dz) satisfies the semi-group property R(Ci, Di) RC, Du) = R(Cii, Div). (2.17) Of course, this does not come as a surprise. I induded this computation to empha size that the semi-group property can be traced back to the algebra of the pair of operators defining the renormalization group. Below, we will meet different pairs of operators which all satisfy this same algebra. Therefore all of them define semi-groups of renormalization transformations. 2.4. Generator R(C, D) The above formulas have a differential pendant. The discrete transformation R(C,, D,) associates with a given initial perturbation Z(¢) a renormalization group trajectory Z(4|L) = R(C,Dz)(Z)($). Due to the semi-group property, Z(@|L) is the solution of the diserete flow equation R(C;,D,)(Z)(¢|L") = Z(4|L L’) to the initial condition Z(d|1) = 2(¢). The discrete flow equation implies a continous flow equation for Z(@|L), the renormalization group differential equation 2 agit) = MC DZVEI), (2.18) where R(C,B) = 2 explic luer®(Ci, Dz) is the generator of the renormalization group. Its form is apr ble a Coa) a0 CG, and BY = 2|,_ Df. See (10, 1]. The second term in (2.19) dition Z(0) = In practical computations, Z(@) is inconvenient due to its bad locality properties, There one prefers to work with the interaction V(¢|L) = —log Z(@|L). Equation (2.19) A a vnere C= 2, serves to maintain the normalizationimmediately Impies that 3 ran fife) (0.8) na0- a 2 (;p¥ Oe GVA) - —BEYV CE) (2.00) where E(C)(V(-|L)) denotes the (field independent) normalization constant seven =} [Ge #) ven — = (rene Avew)| (2.21) o=0 Indispensable intermediate volume cutoffs are here treated casually. From (2.20) one also deduces a flow equation for the 1PI-vertex functions. We will not repeat this calculation here. The bottom line is that with every scale decomposition (C,,D,) is associated an exact renormalization group differential equation. For the instant pair (C., D.), the differential version (C, D) is explicitely given by (2.22) together with (2.23) Notice that the differential pair (C, D) is independent of L. This is a characteristic feature of the rescaled renormalization group with. To summarize, the renormulization group associated with the pair of operators (C1, D1) comes in two equivalent formulations. In its di version, one deals with difference equations. In its continuous version, one deals with differential equations. 3. Renormalization group — massive picture In this section, we show that the renormalization group defined by the pair (C1, Dz) is equivalent: to a renormalization group defined by another pair (¢., di). 3.1. Gaussian fixed point The transformation R(C., D1) has a gaussian fixed point! Z,(¢) = 7 2@"5"9), given x)/(-A)*"4. For the particular cutoff function (2.3), it has the integral 107, (#) satisfies R(Cr,D1)(Z.)(6) = Z,(¢) for all L > 1. 10representation a (an)? T (1 §) Inspecting (3.1), in follows that K, is short ringed. The fourier transform of (8.1) is indeed regular at zero momentum: K,(p) = 1/T (2— £) + O(p"). More generally, we have a line of fixed points given by K, +const./(—A)!-#. All of them are long ranged, Klay) = (3.0) except for the special point with const. = 0. 3.2, Transformation Q(C) The multiplication of perturbations with gauss functions and the renormalization group transformation satisfy the following commutation law. Let C be an operator. Let Q(C) be the functional operator given by QO Z)(O) = HH 2(g). (3.2) Thus Q(C) multiplies a perturbation with a Gauss function. The two transformations R(A, B) and Q(C) conspire to the identity RCA, B) Q(C) = Q(B" LTO B™) RILA, BL), (3.3) where £7? = 14 AC71, assuming that A = A, C = C", invertibility of A, B, and C, and existence of the gaussian convolutions. In other words, to commute a gauss function through the renormalization group transformation requires to compute its renormalization group image. 3.3. Substitution Z(¢) = Z,(¢) z(¢) 2 Let us write the perturbation Z(@) as a product of the fixed point Z,(¢) and a new perturbation z($). Substitute Z(¢) = Z, (4) 2(6) to obtain RCL, Di) (4 2)(9) = Z(G) Reus di )(2)(4) » (3.4) with a new pair (¢;,,d,) of operators, defining the massive picture of the renormalization group. They ure given by wa=LiC1, &=Dili, (3.5) where £, is the operator £, = Ky/(K,+ C1). Notice that C, and K, commie. £1 is explicitely given by iy — 2c _ pon KO) Li(p) = Tein = 2 ‘ Eda (3.6) Think of 2;(p) as a p-dependent correction to the scale factor in the field dilatation. At p? = 0 it yields an extra factor E-?+¢ to the dilataton, while it becomes one as p+ oo. Remarkably, we obtain another renormalization group transformation of the same kind as before. But the covariance and the dilatation have changed. 113.4. Covariance ¢,, Like C,, the covariance «, is short ranged. It is again diagonal in momentum space. From (3.6), it follows that ~ Eon 1-R(P)_ Rp) — R(ZP) q = £.(p) Cie) = —a, “~a- 3.7) 20) = Lite) Ge) = ty ae an The new covariance and the old covariance both vanish ai L = 1. Therefore, the corresponding renormalization group transformations become the identity at L = 1 as they should. On the other side, (3.7) turns into (1-2) RO) wy when Z — oo. Recall that C,, is a massless covariance with unit ultraviolet cutoff. In CODLTASt, Coo is a massive (short ranged) covariance with twosided unit eutofis. Ep) = = (0-2) Ge), (3.8) 3.5. Dilatation dj, From (3.6), it follows that £; = 1 and consequently d, = 1, in analogy to D, = 1. Furthermore, d;, satisfies the composition law 4, diy = duis. The pair (e,,d,) can be shown to be a scale decomposition of the covariamce Coq with respect to d,: We have that ch = ta — di ceeds. Therefrom, we conclude that R(ez, dz) satisfies the semi-group properties R(q, d)) = 1 and Re, ds) Rew dur) = Rlerw daw). (3.9) The reasoning is completely analogous to the massless case, since the algebraic prop- erties of the pair (c1,d1) are identical to those of (Cz,D1). We conclude that the transformation (3.4) is an equivalence between two representations of the renormal- ization group, a massless picture and a massive picture. The attributes massless and massive refer to the underlying scale decomposed covariance. 3.6. Generator A(é, d) To compute the renormalization group differential equation in the massive picture, all we have to do, is to compute the differential pair (¢,d) and to replace (C, D) by it in (2.22) and (2.23). The differential covariance and dilatation pick up an extra term £ as compared to the massless picture. The extra term is 3 al & VeXtP) £o) = =|) Lip) = ee, 3.10 @) = 57| fio) = Rete (3.10) With our cutoff function (2.3), the extra term becomes & _ —2 (pyri a? _ oe “= Gg) Raat For om) 12In the special case when € = U, 1t becomes the generating Nunctlom ol the Bernal numbers, namely L(p) = —2p?/(e?* — 1). In the massive picture, we thus have a differential pair of operators (é,d), given by fio) = [oy 41425405] A. eas The change from the massless picture to the massive picture is thus particularly simple in the differential formulation of the renormalization group. Finally, we remark that d, = Dy, £1, Dz! Dy, with (LDP @) = Eig = REM, (3.4) 1— RP) Which of the two pictures is better suited, depends on the kind of perturbation one is considering, whether it is functionally closer to the trivial fixed point or to the gaussian fixed point. In the hierarchical approximation for instance, the massive picture has proved to be better suited for an investigation of the non-trivial scalar fixed point in three dimensions [7]. 4. Symmetry — massive picture In this section, we show that the renormalization group defined by the pair (¢,,d,) has a symmetry, the transformation defined by yet another pair (71, 4.)- 4.1, Symmetry pair (7,42) Let L be a scale, with L > 1. Let 7, and 4, be the operators given by n= (L-L7) , d= 57. (4.0 The operator dz performs a global (space-time independent) multiplication by Z7*. Obviously, (-7z,,4:,) is a (global) scale decomposition of cz. with respect to d,, namely Yb = Coe — 8F toe Sy. It follows that the transformation R(y,,4,) satisfies the semi- group properties R(j, 61) = 1 and Ry, bt) Rw Fv) = Rows, du)- (4.2) 134.2. Commutator of it(c,,@,) and Myx, 01.) The two one parameter families of transformations R(c,,d,) and R(yy,4,,) commute: For all L > 1 and LE’ > 1, we have that Rex, du) Ryu, dur) = Ryu, 67) Ren, da) (4.3) This property defines a symmetry of R(c, dz). An analogous construction can be made in the massless picture. The point here is that the symmetry covariance is short ranged in the massive picture. To prove (4.3), we compute (cus dn) (yur, Ser) = (88 en, Bas yas de Bar) (4.4) and compure the result with (M1 Bu) (Cus de) = (dE wy de + 6x, Su da) « (4.3) The two dilatations commute, d, 4, = dy dz. The combined covariances are Senda tw =< (LN (cx — dl tm a.) + G@ - wy) = cm — (LY? df cme (46) and dB ous dy ben = @ 4 wy) tag, + O59 — AE Cag hy = toe — (LI)? di exe di, (47) These two covariances are identical. It follows that Rez, dz) and R(qu/,du) indeed commute. 4.3. Composed transformation To obtain a better understanding of the symmetry transformation, we can study the composition of a renormalization group transformation and a symmetry transformation. E i; i Hi g This composition turns out to be given by yeti another pair. Let L’ = L3, where q parametrizes the symmetry transformation. (A special value of 7 will be identified with the anomalous dimension.) Then we have that Rex 4s.) RY,9,9,9) = Alcuin dan) > (4.8) where dig = di d,g = Lids (4.9) Cig = Coe — LE Con dy, = Cog — dig Coe dig - (4.10) In other words, the composition of the renormalization group and the symmetry is a renormalization group built from an anomalous scale decomposition (¢19,di,9) of Coc- 144.4. Differential pair (7, 4) The generator of the symmetry is particularly simple. It is given by the differential pair of operators (4,4) with 7 = 2¢.0 and § = —1. The generators of the renormalization Jo(es) They satisfy [r,u] = 0, as can be checked directly. Renormalization group and symme- group und the symmetry are r= (5 é nie try associate with a perturbation z(¢) a two parametric orbit 2{GIE,U)) = Ren, du) Rw, 1-)(2)(8), (4.13) whose dependence on L is governed by (4.11), and whose dependence on Z! is governed by (4.12). Indeed, (4.13) satisfies the two differential equations LA AIL L) = r(M@lL, 1) —2(9L,1) rl2)(01E, U) (4.14) and LA AO|LL) = wleV(|E 2) AGL, 1) wle\ (02,2). (4.3) The interaction o(@|Z, L') = —log z(@|L, L’) obeys an exponentiated version of (4.14) and (4.15), amalogous to (2-20). We leave it to the reader, to write ik down explicitely. 5. Symmetry — massless picture In this section, we transform the symmetry R{7y,,4,) to the massless picture. Recall that the change of pictures is the transformation Z(¢) = Q(K,}(z)(@) and that R(C1, Di} Q(K,) = Q(B.) Rex, di) - (3-1) To translate R(y,, 4.) to the massless picture, we have to conjugate with Q(K,). From (4.8) and it, follows that RUA, B) QC) =Q10) @ (-0 with A’ = £ A, B= BL, C! = (BOY (C+ A) BO, where £ = 1/(1 +407). In the massless picture, the symmetry transformation therefore consists of two parts. One part is the transformation given by a pair of operators. The second part is the multiplication with a certain Gauss function. This term is essentially a kinetic term as we will see.a1. A’, , OC, and ©" tor the symmetry Let A, B, and C be given by A = (1— £7?) og, B = L7}, and G = —K,. It follows that £=1/(1—(1—Z7*) x) and thus (1-179) 0 A= ay’ 64) , I ms 2 Py 65) 2 C= (@ ae =) Coe 5 (5.8) and 1 _ C-C'_ Q-£7)(-A)"3 (7) oO” OG ~ 1-1 Ey The symmetry transformation R(-,41) thus has the following form in the massless picture. It maps Z() to a symmetry orbit (agro) Jdta(d) Z(BY"(®) + 9) Sia) 24) . Remarkably, the fiction G¥(p)-? = (1 — L-) ()'-#/(1 — (1 — £77) 8(p)) is an analytic function of p? at the origin only if e = 0. A straight forward calculation ctly that the transformation (5.8) satisfies the properties of a semi-group 2(B|1,L) = e7# (5.8) confirms di and that it commutes with the renormalization group. 5.2. A', B', C!, and C" far the composed transformation The composition of » symmetry transformation, which scales by a factor L', and a renormalization group transformation, which scales by a factor L, is computed analo- gously. Let A = Cy, —(L)-2 df tmp d,, B = (Ed, and C =—K,. Then (6.5) where ¥2(p) = X(Ep). It follows that yo x x2) = (PO xe 5 “Cart 1-G-@)u (5.1) 7 wy ; Be Dig G-@)7?)x.” (64) C= cae ter (a+6-Wy)x}, G2) 16and 1 1-(P)? inte eat) For L = 1, one reproduces the formulas (3.4), (5.5), and (5.6) for the pure symmetry transformation. Notice that (3.13) is independent of Z, as it should. Equation (5.8) generalizes to »y = echtergrey [ie (0) 2B" (B) +8) - FOIE) =e Fea 8) ZB) (6.14) 5.3. Generators in the massless picture We cam compute the generators of the double dow (3.14) for the interaction V(@|Z, L’) = —log Z(4|L, E’) from V(@IL, L’) = 3(® a®) - —log / djsar(@) eo” (ryt) + log / dpa(d) oo, (5.15) where A!, B', and 2; are given by (5-10), (5-11), and (5.13) respectively. We find two flow equations. The first flow equation is identical with (2.20). It describes the flow in the renormalization group direction. The second flow equation is v2 Vigit24 = i Ayintg) + a te ; iy 2 ) + Picts z)| V(@IL, )- “2 Ga" Ly, res Vib, 1) - — where BVCLLY = 5[( Frm gVOld ”)- - (AV OHA vn2))| Gan $0 It describes the flow in the symmetry direction. Both together, plus an initial condition, are equivalent to the integral (5.15). The novelty as compared to the massive picture is the appearance of a kinetic term. 176. Anomalous dimension 7 This section contains a brief discussion of fixed points with anomalous dimension based on the following hypothesis, adapted from [9] and [4]. See also the recent discussion in [11, 12] and references therein. 6.1. Fixed point (z,(4),7) We define a fixed point 2(¢) (in the massive picture) with anomalous dimension 7 as a fixed point of R(cz,d:) R(y,4,4,4), the composition of renormalization group transformation R(c., di), which scales by a factor L, and a symmetry transformation R(y,9. 4,4), which scales by a factor L3. 6.2. Discussion of (z,(¢),7) z,(¢) 1s a fixed point of the renormalization group transformation R{c,,dz) only if 1 = 0. In this case, there are two possibilities. Either 2,(6) is also a fixed point of R(w, du), so to speak a double fixed point, or one has a line of fixed points 24(6|Z4) generated as asymmetry orbit from an arbitrary representative z,(@). This is the case for the trivial fixed point. The case with anomalous dimension q is a generalization he invariant line of the renormalization group rather than a line of fixed points, where the renormalization group ucts on the invariant line by an inverse symmetry transfor- mation. A fixed point 2(¢) with non-zero anomalous dimension 7 is not a stationary renormalization group flow. (There remains the possibility to declare a composttion of the renormalization group and the symmetry to be a new renormalization group.) eof. There one has an The appropriate framework to analyze such a situation is to divide the space of per- turbations into symmetry orbits and to consider the renormalization group on orbit space. A natural possibility is to consider one representative on each orbit by imposing a renormalization condition, which breaks the symmetry. Such a condition could be to demand the pre factor of the kinetic term in the interaction to be a given number. In the presenc of asymmetry, the spectrum of a fixed point should include a marginal operator, the direction of the symmetry. A principal problem is the following. To make the fixed point problem. tractable, of interac one has to truncate the space ions. Ideally, the truncation would be such that the symmetry leaves invariant the space of truncated interactions. Because then, the orbit construction can be done on the truncated space. If however the truncation breaks the symmetry, the topic of anomalous dimension is buried in the no-truncation mit, which is usually very difficult to analyze. The idea forwarded in [9] for this situation is to truncate in such a way that a marginal operator survives truncation. In other words that the truncations is such that at least differentially the symmetry persists. The local approximation to the renormalization group unfortunately breaks 18ine symmetry, wherelore te concept of anomalous dimension loses lis Meaning In toe local approximation, and in particular in the hierarchical renormalization group. 6.3. e-madel Consider the fixed point constructions [8] and [5] for the e-model in this light. Their onstructions work directly in the massless picture. When ¢ is non-zero, the situation is the following, If one restricts the renormalization group to vertices, which are C2 functions of the momenta, then the symmetry can be ruled out because it generates a non-C* kinetic term. In this situation, the authors consider an invariant subspace of the renormalization group, which is not invariant under the symmetry. Remarkably, the authors su to construct a non-trivial fixed point in the C*-subspace. (For the planar approximation, this is perhaps not so surprising because the planar approxima- tiom itself also breaks the symmetzy.) Such a non-trivial fixed point is by definition a fixed point with anomalous dimension zero. Tf on the other hand one wants to investigate flows modulo the symmetry (as one has to in the case of non-zero anomalous dimension) then one cannot restrict the flow to C™-vertices in this model. Whether the é-model has another fixed point in this bigger space, I do not know. If this is the case, it would be very interesting to determine (e). In the most interesting case, the three dimensional theory with ¢ = 0, the situation is different. There the symmetry preserves the property of analyticity in the momenta. Therefore it cannot be ruled out. 6.4. Differential equation for 1,(¢) ure takes the form of the follow- The fixed point problem (6.1) in the massive pic ing renormalization group differential equation for the fixed point interaction u,(¢) = — log z.(4), 6 : 6 a ealgg) + ([d*- 3] egg) | (OlE~ ne] Zymld))} = BE em), (61) where Aatte—nenl duty] (6.2) ip ¢ e=0 A natumil problem is to find good trmeation schemes for (6.1) and (6.2) and to de- termine 7 by the condition that the spectrum of u,(¢) have a marginal operator. An investigation of various schemes will be presented elsewhere. 19Recall the generators (4.11) and (4.12) of the renormabzation group and the sym- meti Tes] ively. we neglect the trivial normalization constants, the fixes Lolnt ry respectively. If he trivial normalizati , the fixed poi problem becomes a linear eigenvalue problem r(e)(8) = —F (2) (0 (6:3) for ru? The meaning of the anomalous dimension is thus to be an eigenvalue. Equa- tion (6.1) is just an exponentiated and thus non-linear form of (6.3). The analogous equations in the massless picture follow immediately from (5.15) and (5.16). Acknowledgments [ would like to thank Andreas Pordt and Peter Wittwer for helpful communications on the anomalous dimension 7, and Bernd Gehrmann for computational support. References [1] K. Wilson and J. Kogut, The renormalization group and the © expansion, Phys. Rep. 12 (1974) 75. [2] J. Polchinski, Renormalization and effect. Lagrangians, Nucl. Phys. B 21 (1984) 269. [3] G. Gallavotti, Renormalization theory and ultraviolet stability for scalar fields nia renor- malization group methods, Rew. Mod. Phys. 5% (1985) 471. Id] F. Wegner, The Critical State, General Aspects. Phase Transitions and Critical Phenom- ena, Vol. 6, C. Domb and M. §. Grocn ads., 1976. [5] D. Brydges, J. Dimock and T. R. Hurd, A non-Gaussian iced point for f* in 4 —e dimensions, MP-ARC/96-681. [6] G. Benfatto and G. Gallavotti, Renormalization Group, Vol. 1 of Physics Notes, Prinoc- ton University Pross, 1995. [7] H. Koch and P. Wittwor, On the renormalization group transformation for scalar hier- archical models, Comm. Math. Phys. 188 (1991) 537. [8] G. Felder, Constonction of a non-trivial planar field theory with ultraniolet stable fired point, Comm. Math. Phys. 102 (1985) 139. 9] T.L. Bell and K.G. Wilson, Finite lattice a; wwimations to renormalization groups. Wilson, pres group: Phys. Rew. B 11 (1975) 3431. [10] J. Glimm, A. Jaffo, Quantum physics, a functional point of view, Springer-Verlag, 1987. 11] J. Comellas, Polehinski equation, reparemetrization invariance, and the derivatine ex- equation, repo 7 pension, Nucl. Phys. B 509 (1998) 662 [hep—th/9705129v9]. [12] T.R. Morris and M.D. Turner, Derivative expansion of the renormalization group in O(N) scalar field theory, Nucl. Phys. B 509 (1998) 63 [nep-th/9704202]. 20