VUB/TENA/01/05 PreprinttypesetinJHEPstyle. -PAPERVERSION hep-th/0108169 The non-abelian Born-Infeld action through order α03 by Paul Koerber(cid:3)and Alexander Sevrin Theoretische Natuurkunde, Vrije Universiteit Brussel Pleinlaan 2, B-1050 Brussels, Belgium E-mail: koerber, [email protected] Abstract: Using the method developed in hep-th/0103015, we determine the non-abelianBorn-InfeldactionthroughO(α03). WestartfromsolutionstoaYang-Mills theory which de(cid:12)ne a stable holomorphic vector bundle. In D-brane context this corre- sponds to BPS con(cid:12)gurations in the limit of small background (cid:12)elds. Subsequently we investigate its deformation away from this limit. Through O(α02), a unique, modulo (cid:12)eld rede(cid:12)nitions, solution emerges. At O(α03) we (cid:12)nd a one-parameter family of al- lowed deformations. The presence of derivative terms turns out to be essential. Finally, we present a detailed comparison of our results to existing, partial results. Keywords: D-branes. ∗Aspirant FWO Contents 1. Introduction The bosonic massless degrees of freedom of an open string ending on a flat Dp-brane are a U(1) gauge (cid:12)eld, associated to excitations of the string longitudinal to the brane, and neutral scalar (cid:12)elds, describing the fluctuations of the brane in the transverse directions [1]. For slowly varying (cid:12)elds, i.e. ignoring derivative terms, the e(cid:11)ective action for these massless degrees of freedom is known through all orders in α0. It is the d = 10 supersymmetric abelian Born-Infeld theory, dimensionally reduced to p+1 dimensions [2], [3]. Once several, say n, D-branes coincide, the gauge group gets enhanced from U(1) to U(n), [4]. In leading order in α0, the e(cid:11)ective action is precisely the d = 10 supersym- metric U(n) Yang-Mills theory dimensionally reduced to p+1 dimensions. The exact structure of the full e(cid:11)ective action remains an elusive puzzle. Two complications arise in comparison with the abelian case: (cid:15) Because all (cid:12)elds transform in the adjoint of U(n), an ordering prescription is needed. (cid:15) There is no covariant notion of a slowly varying (cid:12)eld. In other words, higher order derivative terms have to be included. The calculation of open superstring amplitudes allows for a direct determination of the e(cid:11)ective action. This approach lead to (cid:12)rm results through order α02 [5], [6], [7]. An elegant proposal for the e(cid:11)ective action which reproduces the direct calculations was made in [8]: the part of the action which does not contain any derivatives acting on the (cid:12)eldstrengths is the Born-Infeld action and the ordering ambiguities are (cid:12)xed by imposing a symmetrized trace prescription. However, in [9] and [10], it was shown that this does not correctly capture all of the D-brane dynamics. In addition, a direct calculation of the O(α03) [11], see also [12], gave non-vanishing results, contradicting [8]. A direct calculation at higher orders becomes technically very involved, so one is forced to develop alternative approaches. One of these, motivated by the results in [9] and [10], uses the mass spectrum as a guideline which resulted in partial higher order results through order α04 [13]. 1 The problem at hand possesses a lot of supersymmetry: there are 16 linearly and 16 non-linearly realized supersymmetries. This rises the hope that requiring the de- formations of the Yang-Mills action to be supersymmetric will severely restrict the possibilities, perhaps even leading to a unique all order result [14]. Recently it was shown that supersymmetry indeed fully determines the action through O(α02) includ- ing fermionic and derivative terms [15]. The presence of both linear and non-linear supersymmetry would suggest the existence of an explicitly κ-invariant formulation. Despite an attempt at lower order [16], this remains an open problem [7]. A related, very original approach was recently developed in [17]. Starting from the N = 4,d = 4supersymmetric Yang-Millstheory, thebosonicpartofthee(cid:11)ective action through O(α03) was calculated. If one assumes that the supersymmetric deformation of the Yang-Mills action is unique, then this calculation should yield the non-abelian Born-Infeld action. The only drawback of this method is that it is limited to four dimensions. A very di(cid:11)erent approach was launched in [18]. Starting point was the existence of a particular class of solutions to Yang-Mills generalizing the usual instantons in four dimensions. Thesesolutionsde(cid:12)nestableholomorphicbundles[19]. InthecontextofD- branephysics, suchsolutionscorrespondtoBPSsolutionsintheweak(cid:12)eldlimit[20]. In [18],deformationsofsuchsolutionswereanalyzedintheabelianlimit. Arbitrarypowers of the (cid:12)eldstrength were added to the Yang-Mills action. Subsequently it was required thatstableholomorphicbundles, orsomedeformationthereof,stillprovidessolutionsto the equations of motion. Surprisingly this approach leads to a unique deformation: the abelian Born-Infeld action. While the holomorphicity condition remains unchanged, the stability condition acquires higher order corrections as well. An obvious question which arises in this context is whether the method sketched above leads to similar restrictions in the non-abelian case. The analysis in [18] used the explicit assumption that only slowly varying (cid:12)elds appeared. In other words, the Yang- Mills action was deformed by adding arbitrary powers of the (cid:12)eldstrength to it, but termscontainingderivatives ofthe(cid:12)eldstrengthwere excluded. Inthenon-abeliancase, there is no covariant notion of acceleration terms. Indeed, in the abelian case, it is not hard to (cid:12)nd a rescaling of the coordinates and the gauge (cid:12)elds such that a limit exists where the (cid:12)eldstrength remains invariant but its derivatives vanish. Such a limit does not exist in the non-abelian case. However, one might still try to repeat the analysis of [18]under thesameassumptions. Whendoingthis, wefoundnocontributionatO(α03), but we encountered an obstruction at order α04 (i.e. order F6). This clearly showed the need to include derivative terms as well in the non-abelian case. For alternative arguments we refer to the introduction of [12] (see also [11] and [17]). As a (cid:12)rst test, we analyze the deformation of the non-abelian Yang-Mills action through order α03. At this order partial results were obtained before [11], [12], [17]. A clear full answer is however still lacking. The present method shows a major drawback oncederivativetermsareallowed: thenumberoftermswhichcanpotentiallycontribute 2 to the action increases dramatically with each order in α0. An additional di(cid:14)culty is that, because of partial integration, Bianchi identities and the [D,D](cid:1) = [F,(cid:1)] identity, manyrelationsbetweenvarioustermsexist. Inordertodealwiththisinane(cid:14)cientway, we wrote a program [21] in Java, an object oriented language based on the syntax of C, which classi(cid:12)es ata given order inα0 theindependent terms intheaction, calculates the resulting equations of motion and analyzes the deformations of the stability condition. Finally it takes care of (cid:12)eld rede(cid:12)nitions as well. This paper is organized as follows. In the next section we briefly review the relevant solutions in Yang-Mills. In sections 3 to 5 we study the deformations through order α03 and systematize our method. In the last section we analyze our results and confront them with previously known results. The (cid:12)rst appendix explains our conventions and notations. Appendix B gives the equations of motions at order α02 and appendix C a base for the lagrangian and the stability condition deformation at order α03. 2. The leading order: stable holomorphic bundles in Yang-Mills In leading order, the e(cid:11)ective action of n coinciding Dp-branes is the supersymmetric U(n) Yang-Mills theory in ten dimensions1, Z (cid:18) (cid:19) 1 i S = d10xtr − F Fµν + ψ(cid:22)D/ψ , (2.1) µν 4 2 dimensionally reduced to p + 1 dimensions. The Majorana-Weyl spinor ψ transforms in the adjoint representation of U(n). In this way we get as world-volume degrees of freedoma U(n) gauge(cid:12)eld inp+1dimensions, 9−pscalar (cid:12)elds andthe16components of ψ. In thepresent paper, we willignore thetransversal scalars andthefermionic degrees of freedom as they do not seem to give additional information. Furthermore, we make one important assumption: instead of restricting ourselves to d = 10 or less, we will require our analysis to hold in any even dimension! In this way we avoid relations which exist in particular dimensions. OurstartingpointisaU(n)Yang-Millsactioninanevendimensional, flatEuclidean space2, 1 L = tr F F . (2.2) (0) 4 µ1µ2 µ2µ1 In complex coordinates, the equations of motion, D F = 0, read, ν νµ 0 = Dα¯Fαβ¯+DαFα¯β¯ = Dβ¯Fαα¯ +2DαFα¯β¯, (2.3) 1We ignore an overallmultiplicative constant. 2As the metric is +1 in alldirections, we simplify the notationby putting allindices down. Unless stated otherwise, we sum over repeated indices. Furthermore, the lagrangianeq. (2.2), should still be multiplied by an arbitrary coupling constant −1/g2. 3 where we used the Bianchi identities in the last line. One sees that con(cid:12)gurations satisfying Fαβ = Fα¯β¯ = 0, (2.4) and X gαβ¯Fαβ¯ = Fαα¯ (cid:17) Fαα¯ = 0, (2.5) α solve the equations of motion [19]. Eq. (2.4) de(cid:12)nes a holomorphic bundle, while eq. (2.5) guarantees the stability of the bundle [22]. We will alternatively call the latter equation the Donaldson-Uhlenbeck-Yau condition, henceforth abbreviated DUY condition. Restricting to dimensions less than 10, these solutions are BPS con(cid:12)gurations of D-branes in the limit where 2πα0F is small. When p = 2, the BPS conditions are recognized as the standard instanton equations. Note that constant magnetic back- ground (cid:12)elds which satisfy the conditions eqs. (2.4) and (2.5) can be reinterpreted, after T-dualization, as BPS con(cid:12)gurations of Dp-branes at angles [23]. In the next sections, we will investigate order by order in α0 the most general deformation of eq. (2.2). So we will add at each order the most general polynomial in the (cid:12)eldstrength and its covariant derivatives, each term with an arbitrary coupling constant. Subsequently we will demand that con(cid:12)gurations of the form eqs. (2.4) and (2.5) solve the deformed equations of motion. As it turns out this will (cid:12)x, modulo certain (cid:12)eld rede(cid:12)nitions, the coupling constants. Simultaneously, we will have to deform the stability condition eq. (2.5) as well. Concerning the deformation of eq. (2.5), we require it to be such that it fully determines F i.e. it should be such that αα¯ F only appears at the left-hand side of the equation. αα¯ 3. Learning from the low orders 3.1. The α01 corrections The most general deformation of the Yang-Mills action, L , at the next order is given (0) by L +L with, (0) (1) L = 2πα0l3 tr (F F F )+2πα0l3 tr ((D D F )F ), (3.1) (1) 0,0,0 µ1µ2 µ2µ3 µ3µ1 1,1,0 µ3 µ1 µ1µ2 µ2µ3 with l3 and l3 arbitrary constants. This expression takes into account partial 0,0,0 1,1,0 integration, Bianchi and [D,D](cid:1) = [F,(cid:1)] identities, as we will study in more detail in section 4. In order to simplify our notation, we will put 2πα0 = 1 from now on. In complex coordinates, one (cid:12)nds that the equations of motion following from eq. (3.1) read as (cid:0) (cid:1) (cid:0) (cid:1) 0 = Dβ¯Fαα¯(cid:0)−3 l03,0,0(cid:1)Dβ¯Fα1α¯2 Fα2α¯1 +3 l(cid:0)03,0,0Fα1α¯2(cid:1)Dβ¯Fα2α¯1 + (cid:0)2 l13,1,0 Dβ¯Fα1α¯1(cid:1)F(cid:0)α2α¯2 −2 l13,1,0Fα1(cid:1)α¯1 Dβ¯Fα2α¯(cid:0)2 + (cid:1) 4 l13,1,0 +3 l03,0,0 Dβ¯Dα¯1Dα1Fα2α¯2 −3 l03,0,0 Dα1Dβ¯Dα¯1Fα2α¯2 , (3.2) 4 where we used eq. (2.4) and the Bianchi identities. Almost all terms vanish when implementing eq. (2.5), leaving only the second and third term. At this order the only allowed deformation3 of eq. (2.5) is, 3 0 = F +d F F . (3.3) αα¯ 0,0,0 α1α¯2 α2α¯1 One sees immediately that the equations of motion are solved provided we do not deform eq. (2.5), i.e. d3 = 0 and we put l3 = 0. This eliminates the second and 0,0,0 0,0,0 the third term in eq. (3.2). The remainder of the equation of motion is satis(cid:12)ed by virtue of eq. (2.5). The surviving second term in L can then be eliminated by a (cid:12)eld (1) rede(cid:12)nition, A −! A −l3 D F , (3.4) µ µ 1,1,0 ν νµ which exhausts the (cid:12)eld rede(cid:12)nitions at this order. Concluding the O(α0) deformation of both the Yang-Mills action and the stability condition vanish, which is consistent with direct calculations. 3.2. The α02 corrections At the next order the most general deformation of the Yang-Mills lagrangian reads as L = L +L , (3.5) (0) (2) where L is given in eq. (2.2) and L is (0) (2) (cid:0) L = tr l4 F F F F +l4 F F F F + (2) 0,0,0 µ1µ2 µ2µ3 µ3µ4 µ4µ1 0,0,1 µ1µ2 µ2µ3 µ4µ1 µ3µ4 4 4 l F F F F +l F F F F + 0,1,0 µ1µ2 µ2µ1 µ3µ4 µ4µ3 0,1,1 µ1µ2 µ3µ4 µ2µ1 µ4µ3 4 4 l (D D F )F F +l (D F )F (D F )+ 1,2,1 µ4 µ1 µ1µ2 µ3µ4 µ2µ3 1,2(cid:1),6 µ1 µ1µ2 µ2µ3 µ4 µ3µ4 4 l (D D F )(D D F ) . (3.6) 2,1,23 µ4 µ1 µ1µ2 µ4 µ3 µ2µ3 This is the most general deformation at order α02 where we used the Bianchi, par- tial integration and [D,D](cid:1) = [F,(cid:1)] identities. Both the deformation of the stability condition, eq. (2.5) and the contribution to the equations of motion at this order are explicitly given in appendix B. As an illustration, we analyze the equations of motion in some detail. It is clear that the equations of the type e4 and e4 are satis(cid:12)ed provided 0,0,s 0,3,s 3 3 d4 = d4 = 4l4 = 2l4 = −8l4 = −16l4 , 0,0,0 0,0,1 0,0,0 0,0,1 0,1,0 0,1,1 4 l = 0. (3.7) 1,2,1 3We remind the reader that we view the deformed stability condition as an expression for Fαα¯. So while additional deformation terms of the form Fα1α¯1Fα2α¯2 or (Dα¯1Dα1Fα2α¯2) are dimensionally allowed, they are excluded by our ansatz. 5 The contributions of the type e4 vanish, provided 1,0,s 3 4 4 d = d = 0, (3.8) 1,0,4 1,0,5 holds. The remainder of the equations of motion now trivially vanishes when applying eq. (2.5)4. We can (cid:12)x one more parameter. Initially we had two choices: we could choose the overall multiplicative constant in front of the action and we can (cid:12)x the scale of the (cid:12)eldstrength F. These two arbitrary constants are (cid:12)xed by choosing the conventional factor 1/4 in front of the leading term in the action and by putting l4 = 1/24. 0,0,0 Having done this we (cid:12)xed the deformation of the action completely modulo the coupling constants l4 and l4 . However, we still have to consider (cid:12)eld rede(cid:12)nitions. 1,2,6 2,1,23 The most general (cid:12)eld rede(cid:12)nition relevant at this order is A ! A +f3 (D F )F +f3 F (D F )+ ν ν 0,1,0 µ1 µ1µ2 µ2ν 0,1,1 µ2ν µ1 µ1µ2 3 3 f F (D F )+f (D F )F + 0,1,2 µ1µ2 µ1 µ2ν 0,1,3 µ1 µ2ν µ1µ2 3 f (D D D F ), (3.9) 1,0,1 µ1 µ2 µ1 µ2ν where again we took Bianchi identities and the [D,D](cid:1) = [F,(cid:1)] relation into account. The coupling constants transform as follows under the (cid:12)eld rede(cid:12)nitions, l4 ! l4 , 0,s ,s 0,s ,s 2 3 2 3 l4 ! l4 −f3 +2f3 +f3 , 1,2,1 1,2,1 0,1,2 1,0,1 0,1,3 l4 ! l4 +f3 +f3 −f3 −f3 , 1,2,6 1,2,6 0,1,1 0,1,2 0,1,0 0,1,3 l4 ! l4 +f3 . (3.10) 2,1,23 2,1,23 1,0,1 Taking into account that we have to keep l4 = 0, we (cid:12)nd that the three (cid:12)eld rede(cid:12)- 1,2,1 nitions f3 −f3 = −l4 +2l4 , 0,1,1 0,1,0 1,2,6 2,1,23 f3 −f3 = −2l4 , 0,1,2 0,1,3 2,1,23 f3 = −l4 , (3.11) 1,0,1 2,1,23 eliminate the derivative terms in eq. (3.6). This leaves us two (cid:12)eld rede(cid:12)nitions, (cid:0) (cid:1) 1 A ! A + f3 +f3 f(D F ),F g+ ν ν 2 0,1,0 0,1,1 µ1 µ1µ2 µ2ν (cid:0) (cid:1) 1 f3 +f3 fF ,(D F )g, (3.12) 2 0,1,2 0,1,3 µ1µ2 µ1 µ2ν which will not play any further role in this paper as they only become potentially relevant at order α04 in the action. Note that it is quite remarkable that certain terms 4Of course these terms will contribute at order α(cid:48)4 as a consequence of the deformation eq. (B.2). See section 4, step 7.b and 7.c for a systematic approach. 6 which are removable through (cid:12)eld rede(cid:12)nitions, the l4 term in casu, can get (cid:12)xed by 1,2,1 our method. Summarizing, through order α02, the lagrangian is given by L = L + L , with (0) (2) L given in eq. (2.2) and (0) (cid:18) 1 1 L = tr F F F F + F F F F − (2) µ µ µ µ µ µ µ µ µ µ µ µ µ µ µ µ 24 1 2 2 3 3 4 4 1 12 1 2 2 3 4 1 3 4 (cid:19) 1 1 F F F F − F F F F . (3.13) µ µ µ µ µ µ µ µ µ µ µ µ µ µ µ µ 48 1 2 2 1 3 4 4 3 96 1 2 3 4 2 1 4 3 Con(cid:12)gurations satisfying eq. (2.4) and 1 1 0 = F + F F F + F F F , (3.14) αα¯ α α¯ α α¯ α α¯ α α¯ α α¯ α α¯ 6 1 2 2 3 3 1 6 1 2 3 1 2 3 solve the equations of motion. This result is again fully consistent with direct calcula- tions. 4. Systematizing our method After studying the low order cases, we are ready to put together the calculational scheme. Although this scheme can be slavishly followed at higher orders, the calcu- lations itself will become extremely lengthy. So they were carried out by a computer program, written specially for the task at hand. The language of choice was Java, which as a modern object oriented programming language proved to us more user- friendly than the in physics more commonly used C or Fortran. In this section we will discuss what it does, and leave the implementational details for what they are. More details as well as the source code will be given in [21]. Roughly put, the program will construct the most general lagrangian and deforma- tion of the DUY condition at each order in α0. Subsequently, we impose that (cid:12)eld- strength con(cid:12)gurations satisfying eq. (2.4) and the generalized DUY condition solve the equations of motion. This generates a set of equations, since the coe(cid:14)cient of each independent term in the equations of motion has to be zero. From these conditions we can (cid:12)x the coe(cid:14)cients of the lagrangian as well as the coe(cid:14)cients of the DUY deformation. Having said this, we can delve into the technical details. The program distinguishes 4 kinds of terms at each order: their properties are listed in table 1. For each of these types, the program will have to: 1. Calculate all possible terms, using as building blocks antisymmetric (cid:12)elds F and covariant derivatives. These terms have a priori arbitrary coe(cid:14)cients, which are labelled according to the classi(cid:12)cation scheme of appendix A. 2. Calculateallpossibleidentities betweenthoseterms: thesearethepartialintegra- tion identities (only for the lagrangian), the Bianchi identities and the identities of the type [D,D](cid:1) = [F,(cid:1)]. 7 3. Solve those (linear) identities and thus separate the linearly dependent terms from the linearly independent terms, forming a base. The program proceeds by eliminating one term out of each equation. Of course there is still the freedom of choosing the term to eliminate and this will often give rise to some priority rule. Sometimes a term will carry some coe(cid:14)cient information. Upon elimination, this information will be transferred to the other terms in the equation. An example: suppose T1,T2,...Tn are terms in an equation of motion, which reads: Xn j c T = 0. (4.1) j j=1 Term Tj is said to carry its coe(cid:14)cient c . Now, if Ti is eliminated from the j identity: X i j T = d T , (4.2) j j6=i it has to transfer his coe(cid:14)cient c to the other terms and the equation of motion i becomes: X j (c +d c )T = 0. (4.3) j j i j6=i Field DUY Equations Properties Lagrangian rede(cid:12)nitions deformation of motion Group trace yes no no no implieda Free index no yes no yes Complex no no yes yes coordinatesb Type of Bianchi Bianchi Bianchi Bianchi identities PIc, DDFd DDFd DDFd DDFd Used L F D E Symbolse l f d e aSo there is cyclic symmetry bSee appendix A. cPI: partial integration identities. dDDF: identities of the type [D,D](cid:1)=[F,(cid:1)]. eThe curly type is used for denoting all terms at a certain order, the small type in the labelling of the individual terms (as explained in appendix A). Table 1: Properties of the different types of terms. 8 Because of the arbitrariness in choosing a base, many lagrangians will in fact be equivalent. When comparing with the results in the literature [17] [11], we will have to express their terms in our base. Also, as an extra complication, we have to take account of the fact that the coe(cid:14)cients of some terms in the lagrangian may change when applying a (cid:12)eld rede(cid:12)nition and allow for them to di(cid:11)er [6]. We call the latter from now on FR changeable terms. After these initial remarks we state the algorithm, which has to be repeated order per order5 in α0: 1. Construct all possible terms in the lagrangian and all identities among those. 2. If we want to know which terms are FR changeable, we must put in quite a lot of extra work. To the solution for the lagrangian already found at lower orders, apply a (cid:12)eld rede(cid:12)nition: A ! A +F , (4.4) µ µ µ where F is a linear combination of all possible independent (cid:12)eld rede(cid:12)nition µ terms of the appropriate lower orders. Observe how the coe(cid:14)cients of the terms in the lagrangian at the present order change. To clarify what we mean, we study the simplest case in detail. At order α00, the most general (cid:12)eld rede(cid:12)nition is: F = f2 D F , (4.5) (0),ν 0,0,0 µ1 µ1ν and, as we have already used in section 3.1, the coe(cid:14)cient change of term l3 1,1,0 (see (3.1)) becomes 3 2 (cid:1)l = f . (4.6) 1,1,0 0,0,0 Foreach FRchangeable term, we say thatthe termcarries6 hiscoe(cid:14)cient change. The other terms, we will call empty i.e. carrying nothing. 3. Calculate all independent terms in the lagrangian. When eliminating an FR changeable term out of a certain equation, its coe(cid:14)cient change must be trans- ferred as explained, because we want to know how the remaining terms transform under a (cid:12)eld rede(cid:12)nition. So all the other terms in that equation become FR changeable. To minimize the amount of FR changeable terms we end up with, we use the following priority rule: eliminate empty terms (cid:12)rst. If we did not fol- low this rule the FR changeability would quickly spread among all independent terms, although many coe(cid:14)cient changes would be dependent7. 5We will call the order in F n, so the order in α(cid:48) is n−2. 6An alternativewayoflookingatthis, is thatthe field redefinitiontermsforma dualvectorspace, because the choice of a lagrangianterm and a field redefinition term produces a (fractional) number, i.e. the coefficient change. More precisely, the field redefinition terms are only a subspace of the dual vector space and we want to make this clear by an appropriate choice of base of the original space. 7Note that our “rule of thumb” doesn’t guarantee that all remaining coefficient changes are inde- pendent, but certainly most of them. 9
Description: