LPT 93-3 September 1993 MINI-INSTANTONS Vincenzo Branchina a and Janos Polonyi b Laboratory of Theoretical Physics 4 Department of Physics 9 9 Louis Pasteur University 1 67087 Strasbourg Cedex France n and a J CRN Strasbourg 7 67037 Strasbourg Cedex France 1 1 v 9 7 0 1 0 4 9 / h t - Submitted to Nuclear Physics B p e h : v ABSTRACT i X It is shown that the inhomogeneous saddle points of scale invariant theories make r the semiclassical expansion sensitive on the choice of non-renormalizable operators. a In particular, the instanton fugacity and the beta function of the two dimensional non-linear sigma model depends on apparently non-renormalizable operators. This represents a non-perturbative breakdown of that concept of universality which is based on low dimensional operators. a On leave from Catania University, and INFN Sezione di Catania, Italy b On leave from L. E¨otv¨os University and CRIP, Budapest, Hungary 1 I. INTRODUCTION The ultraviolet divergences of Quantum Field Theories is an evergreen subject for new discoveries and frustration. On the one hand, it motivated the introduction of the renormalization [1] [2] which subsequently proved to be an essential notion, independently of the divergences. It was used to uncover how effects due to different interactions build up as the function of their length scale. This kind of thinking led to the application of universality in field theory [2], [3]. It states that it is enough to consider renormalizable models because the non-renromalizable coupling constants decrease as we move towards the infrared direction and become negligible in the regime where the measurements are performed. On the other hand, the original naive hope that the physics at the ultraviolet cut-off decouples from the observations as the cut- off is removed turned out out be unjustified by the isolation of the anomalies [4]. The somewhat unfortunate name for the survival of the cut-off effects in renormalized models came from the surprising and apparently unrelated manner this phenomenon arises in different regularization schemes. After decades of intensive investigations one is still uncomfortable in this subject since the true physical origin and the necessity of the anomalies has not yet been understood. But at least the class of models where the anomalies respect renormalizability and universality has been established by means of involved algebraic and topological methods. Another observation concerning the importance of the cut-off effects has been made for non-asymptotically free models. If the coupling constant is large at the cut- off then non-perturbative phenomena may generate novel infrared behavior. Consider say QED as the cut-off is removed [5]. This is certainly a non-perturbative problem due totheLandau-pole andwe areleft forspeculations only. Suppose that thecoupling constant of massless QED is so large at short distances that the electron-positron pair is bound with negative energy. These small bound states then condense in the vacuum and lead to the dynamical breakdown of the chiral symmetry. The loss of the chiral symmetry is obviously a long distance phenomenon. Thus the bound state formation at the cut-off influences the infrared behavior of the model. The methods to incorpo- rate bound state effects are not powerful enough so the same phenomenon was later established by means of the renormalization group [6]. The description of the role of an operator in the renormalization process starts with the power counting arguments. It was suggested that we should include the anomalous dimension in these considera- tions since from the point of view of the scaling behavior there is no difference between the canonical and the anomalous dimensions. The power coupling argument applied with anomalous dimensions which were computed in the weak coupling expansion and becameextended tothestrong coupling regimebrought asurprise: theoperatorswhich control the dynamical breakdown of the chiral symmetry appear renormalizable. The pile-up of the graphs in the ultraviolet regime observed in the solution of the renor- malization group generates those coupling constants which are needed for the chiral 2 symmetry broken vacuum. The net result is that some new operators turn out to be relevant in describing the physical content of the theory. We present another cut-off phenomenon which generates new important coupling constants in addition to the perturbatively renormalizable ones. The novel feature of thisphenomenon isitssimplicityandthatitoccursinasymptoticallyfree theories. Itis believedthat thesmallnessofthebarecoupling constant inasymptoticallyfree theories renders all cut-off effects perturbative. It will be shown below that this expectation is not always fulfilled. We found new operators mixed in the renormalization of such scale invariant theories which possess nontrivial saddle points. As an example consider the saddle point approximationof QCD. The conventional approach is based on the renormalized perturbation expansion [7], [8]. One first isolates the extrema of the renormalized action, 1 1 S [A] = d4xF2 = d4xL[A], (1.1) r 4g2 µν 4g2 r Z r Z the instantons. In the next step the one-loop quantum correction is computed. It is ultraviolet divergent so counterterms S [A] are needed. Fortunately it turns out ct that for instantons which are locally flat the counterterms are the same as in the zero winding number sector so the renormalization procedure remains unchanged . We should not forget that this procedure is rather formal, the application of the renormalized perturbation expansion is usually difficult to justify. The mathematically solid framework is that of the bare cut-off theory described by the well defined path integral, D[A]e−Sb[A]. (1.2) Z which includes the bare action, 1 S = d4xL[A] = S +S . (1.3) b 4g2 r ct b Z (1.2) is an integral with large but finite dimension. We are interested in the limit when its dimension diverges in order to arrive the renormalized theory. The bare perturbation expansion is obtained by expanding (1.2) in the non-quadratic parts of S [A] = S [A]+S [A], b 0 1 1 D[A]e−S0[A]+S1[A] = D[A]e−S0[A] Sn[A]. (1.4) n! 1 Z Z X The small parameter of the expansion is the bare coupling constant, g . As the cut-off b is removed the bare parameters must be rescaled in order to keep the physical content ofthetheorycut-offindependent. Aftertheexpansiononerecognizessomecancellation between different contribution in the sum of (1.4). After taking these cancellations into 3 account we arrive to a new expression, the renormalized perturbation expansion which looks as if it had come from the expansion of S [A] according to a small parameter of r the renormalized action, g the renormalized coupling constant. r But this is not quite right. The more precise statement is that one first expands S [A] and then retains only the ‘finite’ parts of the graphs. The ‘divergent’ pieces are r canceled by the counterterms. The problem is that there is no such action which would generate only the finite contribution in the perturbation expansion. In order to prove the existence of the renormalized perturbation expansion one first has to establish the reliability of the bare perturbation expansion. It is only after this latter step that one can take the left over ’renormalized’ part of the perturbation expansion seriously. One would think that the renormalized perturbation expansion is reliable in QCD. In fact, the one-loop relation 1 1 µ = 2β ln , (1.5) g2 g2 − 0 Λ b r where µ and Λ are the subtraction scale and the cut-off, respectively, indicates that the counterterms are suppressed compared to the renormalized action, 1 µ2 S [A] = d4xL[A] 2β ln d4xL[A], (1.6) b 0 g2 − Λ r Z Z for the sufficiently large region µ 1 < e2β0gr2. (1.7) Λ Alas, there is a problem with this argument inthe nonzero winding number sector. According to the renormalized perturbation expansion the tree level step, the selection of the saddle points of D[A]e−Sr[A]+Sct[A] (1.8) Z is done by the help of the renormalized part, S [A] and the counterterms are included r perturbatively only at the one-loop level. Since the theory is scale invariant the renor- malized action has no scale parameter and the saddle points form a degenerate family with respect to the scale transformation. The scale invariance is broken by the quan- tum corrections, the well known phenomenon of the dimensional transmutation. In the contrary, the scale invariance is broken already at the tree level in the bare theory. This seems rather unusual in the dimensional regularization scheme where the usually employed minimal subtraction hides the cut-off in the computations. But the less for- mal, non-perturbative regularization schemes reveal the scale dependence of the bare theory. 4 Consider first the regularization by adding higher order derivatives to the action, 1 C C S [A] = d4xF + 2D2 + 4D4 + F , (1.9) b µν 4g2 Λ2 Λ4 ··· µν Z (cid:26) b (cid:27) where D stands for the covariant derivative. Though these higher dimensional opera- tors do not remove all the UV divergences from the theory they serve as an example for our purpose. The new terms play no role in the physical content of the theory according to the perturbation expansion. This is because each insertion of the new vertices bring the coefficient n µ C , (1.10) n Λ (cid:18) (cid:19) where µ is the characteristic energy scale. The same scale dependence generates ultra- violet divergences and may make the theory non-renromalizable. The instanton with scale ρ has the action 1 1 1 S = 8π2 +C O +C O + . (1.11) i 4g2 2 ρ2Λ2 4 ρ4Λ4 ··· (cid:26) b (cid:18) (cid:19) (cid:18) (cid:19) (cid:27) It is not surprising that the scale degeneracy of the instantons is split in the bare action since after all it contains a scale parameter, the cut-off. The situation is similar in lattice regularization, 2π2 S = +O(ρ2a2), (1.12) i g2 b where the lattice spacing is denoted by a. In the systematical saddle point expansion of the bare theory the scale dependence appears already at the tree level in a man- ner radically different from the one-loop scaling violation of the minimal subtraction scheme. It is certainly true that the magnitude of the ‘counterterms’, i.e. the cut-off dependent terms of the action is negligible compared to the renormalized one but this is not enough now. It is not enough because the renormalized perturbation expansion has an infinite degeneracy at the tree level withrespect to the scale transformation. An arbitrarilysmall perturbation which liftsthis degeneracy may have substantial effect in this situation. Similar phenomenon can be observed in the quantum Hall effect where degeneracy of the Landau levels is lifted in a very specific manner by the electrostatic interaction. The weak scale dependence of the instanton action generated by the cut-off is enough to destabilize several saddle points or make only some of them stable. When C > 0, C = C = 0 (1.13) 2 4 6 ··· is chosen then the instantons grow without limit. For C < 0, C > 0, C = 0, (1.14) 2 4 6 ··· 5 there is a stable saddle point, called mini-instanton with the size parameter ρ¯ = O(1/Λ). On the lattice the instanton action tends to zero for small instantons which ‘fall through the lattice’. Thus depending on the choice of the lattice action we may or may not have stable saddle points. The tree level instanton action is sensitive to the irrelevant i.e. non-renormalizable operators present in the action. Since the regularization is achieved by the introduction of some specific irrelevant operators, different regularizations lead to different saddle point structure. For large saddle points the presence of the cut-off lead to small changes. But for small saddle points whose size is comparable with the cut-off the non-renormalizable operators are important. By an appropriate choice of the non-renormalizable coupling constants the scale degeneracy of the renormalized action is split in such a manner that the stable saddle points will be at the scale of the cut-off. In this case the saddle point, being close to the cut-off, reflects the physics of short distances. One recognizes the analogy with strong coupling QED at this point. There the strong attraction at short distances led to the formation of small bound states. These bound states populate the vacuum and lead to long distance effects. In QCD the (weak) interactions at short distances lead to new saddle points. Can the modification of the instanton action at short distances change the physics at large distances ? It is difficult to answer this question before actually completing the saddle point expansion. At the first moment one would think that the presence of instantons with small size only will certainly modify the long range structure. The disappear- ance of the degeneracy corresponding to the dilatation transformation raises even the hope of an infrared consistent instanton gas description of the vacuum. But this is not necessarily so. The integration over the dilatation mode can be performed by the collective coordinate method as usual. Then for a given value of the instanton size the collective coordinate keeps the saddle point stable for the Gaussian integration. As we integrate over all values of the instanton size we should ultimately recover the contri- bution of large saddle points. This is the same as in the renormalized perturbation expansion. We are now confronted with the question whether the small or the large instantons dominate the integral over the scale parameter. The action of the small stable instanton in the case (1.14) is smaller than the asymptotic value achieved for large size. Thus the small instanton contribution is enhanced. We believe that as long as the integration over the scale parameter is done only in the ‘safe’ region, ρ << Λ−1 , (1.15) QCD far from the confinement radius the effect of the small instantons will be felt. The coupling constants generated by them might be responsible for the restoration of the center symmetry which is usually broken by the gauge fixing condition and crucial for the understanding of the long range features of the vacuum in the framework of the renormalization group [9]. One might argue that such a strange dependence of the physics on certain higher dimensional operators is not important if one uses the lagrangian which does not 6 contain them from the very beginning. This is certainly true mathematically but in the real world we always have irrelevant operators in our theories. Denote the theory we investigate by T . Unless it is the Theory of Everything there is always a 1 higher energy unified theory, T , which gradually becomes relevant as the energy of 2 the observations is increased. Thus the renormalized trajectory does not approach the ultraviolet fixed point of T but rather turns to another direction, given by T . The 1 2 interactions of T generate vertices when expressed in terms of the degrees of freedom 2 of T . These effective vertices represent non-renormalizable irrelevant operators in 1 T since otherwise the renormalized trajectory would run into the fixed point of T . 1 1 Thus the ‘seeds’ of the new physics are always hidden in the irrelevant operators of the theories we used to describe the high energy experiments. Expressing it in mathematical terms: the theory T formally regulates the ultraviolet region of T and 2 1 the ultraviolet regulators are always irrelevant operators. Thus we have irrelevant operators in a given regularization scheme even if they do not appear in the formal, unregulated expression of the lagrangian. In order to demonstrate the above peculiarities of the semiclassical expansion we computed the weight of the one-instanton sector in the two dimensional sigma model. This model is usually defined by the action 1 S[φ~] = d2x(∂ φa)2, (1.16) µ 2f Z where the three component field φa, a = 1,2,3 is subject of the constraint 3 φaφa = 1. (1.17) a=1 X Since there is no other renormalizable operator with the same symmetry the action (1.16) is believed to cover the universality class of the model. It will be shown that this is not what happens. (1.16) is usually investigated by means of the weak coupling expansion. Since there are instanton solutions one has to include their contributions as well. It turns out that in the nonzero winding number sector some naively non- renormalizable operators play role in the evolution of the coupling constant, f. Weextendtheaction(1.16)byadding higherdimensional terms, such as(∂ φa)2n, µ n = 2,3 to the lagrangian, 1 3 C 5 C S[φ~] = d2x 1 2(∂ φa)2 + 4(∂ φa)2(∂ φa)2 (∂ φa)2, (1.18) 2f − 4 Λ2 µ 64 Λ4 µ µ µ ! Z where Λ is the euclidean momentum cut-off. The sign of the quartic term is fixed in (1.18) by anticipating the choice of the non-renormalizable operators which actually 7 influence the physical content of the model. The numerical coefficients were chosen for later convenience. Though there are non-renormalizable operators in (1.18) the theory is still renor- malizable. There are no overall divergences generated by the new pieces. In fact, each additional derivative is compensated by a factor of 1 and the insertion of the new ver- Λ tices into a graph does not increase the overall degree of divergence. The issue of the overlapping divergences is much more complex and one has to use the O(3) symmetry [10]. It can be proven by induction [11] that in each order of the loop expansion the ultraviolet divergences can be eliminated by the fine tuning the coupling constant f for any action which reduces to (1.16) classically as the cut-off is removed. Naturally the constants C become nontrivial functions of the cut-off as one leaves the vicinity n of the ultraviolet fixed point. It is important to bear in mind that the proof of the renormalizability holds only for the expansion around a homogeneous configuration. When the background around which we expand is inhomogeneous then the situation is more involved. The new vertices in (1.18) leave the overall divergence unchanged only if the characteristic length scale ρ, of the background field configuration is far from the cut-off, ρΛ >> 1. The generalization of the inductive proof for the overlapping divergences for the case of inhomogeneous background field is not known and one expects some problem there, too, when ρΛ = O(1). Consider a physical quantity, P, defined by the help of the scale parameter µ with dimension of energy. In case of a Green function µ can be taken to be the typical energy scale of the external legs. For a bulk quantity µ−1 is the characteristic length scale of the system. The observable depends on the coupling constants and the cut-off, P = Λ[P]P˜(f,C µn/Λn), (1.19) n ˜ where [P] is the dimension of P in units of energy and P is a dimensionless function. If the scales of the observable and the cut-off are far, µ/Λ 0, then dependence of P ≈ on the non-renormalizable coupling constants is negligible. In this paper the ratio Z 1 P = , (1.20) Z 0 will be computed where Z denotes the path integral for the field configurations with n winding number n. Since (1.20) is a bulk quantity its dependence on C should be n negligible for large enough value of the ultraviolet cut-off, Λ, ˜ P = P(f) = F(Λ /Λ), (1.21) σ where Λ is the cut-off independent lambda-parameter of the model. This is not what σ happens in our case, the sensitivity of the ratio(1.20) onC survivesthe removal of the n cut-off. For an observable computed on a homogenous or slowly varying background 8 fieldthedependence onthenon-renromalizablecouplingconstant, (1.10),issuppressed. But the ratio (1.20) is saturated by the fluctuations around a mini-instanton and the cut-off effects survive the renormalization. Since P must be finite as the cut-off is removed our computation can be used to determine the cut-off dependence of the coupling constant f as C are kept constants. n We find that the beta function obtained in this manner contains the parameters C , n ∂ Λ f = β (C ,C )f2 +O(f3), (1.22) 0 2 4 ∂Λ with 1 1 β (C ,C ) = . (1.23) 0 2 4 −2π 2C2 1 2 − C4 It is natural that the presence of the additional coupling constants in the theory mod- ify the beta function. What is remarkable in this result is that the dependence on C n remains unsuppressed over all the range of the perturbation expansion, µ > Λ . Any σ dependence on C should appear through the combination (1.10) according to the per- n turbation expansion on flat or slowly varying background field. But the beta function contains such dimensionless ratio of the coupling constants in which the suppression factor of (1.10) is cancelled. Thus the ‘irrelevant’ coupling constants C are not at all n irrelevant when instantons are present. Additional non-perturbative effects are expected to show up for C = O(1). Since n the lagrangian contains new, non positive definite terms the self-dual saddle points are not necessarily stable. If they become unstable for certain choice of C then we find n a phase transition at those values of the coupling constants. Even if the saddle points remain stable the mini-instanton action can be negative. For such a choice of the C n the mini-instantons condense and the saddle point expansion ceases to be reliable. The theory becomes similar to frustrated spin models and the weak coupling expansion is rendered useless despite the presence of the small f. The organization of the paper is the following. In Section II. the tree level ap- proximation to the path integral is discussed. The one-loop quantum fluctuations are considered in Section III. Section IV. contains the derivation of the beta function for the unit winding number sector. The renormalization group in the dilute instanton gas approximation is considered qualitatively in Section V. Finally our conclusion is presented in Section VI. II. TREE LEVEL The saddle point approximation for the path integral D[φ~] δ(φ~2(x) 1)e−S[φ~], (2.1) − Z x Y 9 starts with the selection of the extrema of the action S[φ], on the tree level. In order to find the saddle point of the action (1.18) which satisfies the constraint (1.17) we introduce the Lagrange multiplier field λ(x) and use the action 1 3 C 5 C Sλ[φ~] = d2x 1 2(∂ φ~)2+ 4(∂ φ~)4 (∂ φ~)2 +λ(x)(φ~2 1) , (2.2) 2f − 4 Λ2 µ 64 Λ4 µ µ − " # Z (cid:16) (cid:17) δS(λ)[φ~] at the tree level computation. The equation of motion, = 0, for the constrained δφ~(x) system is then 3 C 15 C 1 2(∂ φ~∂ φ~)+ 4(∂ φ~∂ φ~)2 ∂ ∂ φ~ +2λ(x)fφ~ = 0. (2.3) − 2 Λ2 µ µ 64 Λ4 µ µ µ µ (cid:16) (cid:17) The Lagrange multiplier field is eliminated by the help of (1.17), 1 3 C 15 C λ(x) = − 1 2(∂ φ~∂ φ~)+ 4(∂ φ~∂ φ~)2 ∂ φ~∂ φ~, (2.4) 2f − 2 Λ2 µ µ 64 Λ4 µ µ µ µ (cid:16) (cid:17) and the equation of motion is written in the form 3 C 15 C 1 2(∂ φ~∂ φ~)+ 4(∂ φ~∂ φ~)2 ∂ ∂ ∂ φ~∂ φ~ φ~ = 0. (2.5) − 2 Λ2 µ µ 64 Λ4 µ µ − µ µ − µ µ (cid:16) (cid:17)(cid:16) (cid:17) Notice that due to the particular choice of the irrelevant operators in (1.18), one can factorize out the equation of motion of the usual lagrangian, (1.16) in (2.5). Then the one-instanton solution to (2.5) is the well known Belavin-Polyakov solution [12] which can be conveniently parametrized in the following way, 2ρ(x x ) (1) 0 φ = − 0 ρ2 + z z 2 0  | − |  φ(2) = 2ρ(y−y0) (2.6)  0 ρ2 + z z 2  | − 0| z z 2 ρ2 (3) 0 φ = | − | − .  0 ρ2 +|z −z0|2    Here (x ,y ) are the center ofthe instanton coordinates, ρ is the instanton size. Ac- 0 0 tually the solution contains six arbitrary parameters related to the symmetry of the action : two for the translational invariance, one for the dilatational invariance and three related to rotations in the internal space. In (2.6) we have displayed only three of them ( translational and dilatational invariance). The fourth parameter, which is related to the rotation around the third axis in internal space (which we orient for convenience in the direction orthogonal to the two dimensional space-time) and mixes 10