ICTP/98/1, hep-th/98 Bound States Of Type I D-Strings 8 E. Gava, 1 9 9 INFN and ICTP, Trieste, Italy, 1 J.F. Morales, 2 n a SISSA, Trieste, Italy, J 9 K.S. Narain and G. Thompson3 1 ICTP, P.O. Box 586, 34014 Trieste, Italy 1 v 8 2 1 1 Abstract 0 8 Westudytheinfra-redlimitoftheO(N)gaugetheorythatdescribesthelowenergy 9 modesofasystemofN typeID-stringsandprovidesomesupporttotheconjecture / h that, in this limit, the theory flows to an orbifold conformal theory. We compute t - the elliptic genus of the orbifold theory and argue that its longest string sector p e describes the bound states of D-strings. We show that, as a result, the masses and h multiplicitiesoftheboundstatesareinagreementwiththepredictionsofheterotic- : v type I duality in 9 dimensions, for all the BPS charges in the lattice Γ . (1,17) i X r a 1e-mail: [email protected] 2e-mail: [email protected] 3e-mail: narain,[email protected] 1 Introduction The putative duality between Type I string theory and the SO(32) heterotic string theory [1, 2] requires the existance of Type I D-string bound states. The way to see this is to begin by compactifying on a circle of radius R in the X9 direction. On the H heterotic side the electric charge spectrum in the nine dimensional theory sits in the lattice Γ . This lattice arises on taking into account the charges associated with (1,17) the gauge fields of the Cartan subalgebra of SO(32), the G component of the metric µ9 and the B component of the Neveu-Schwarz antisymmetric tensor. In particular, the µ9 states carrying N units of B charge correspond to the fundamental heterotic string µ9 wrapping N times around the X9 circle. This particular charge corresponds to a null vector in Γ and scrutiny of the partition function shows that the multiplicity of (1,17) such BPS states, which arise at the level one oscillator mode of the bosonic sector, is 24. The general BPS spectrum, with all the electric charges, is given by 1 kN P2 = N 1, (1.1) R − 2 − where P is a vector in Γ , N is the bosonic oscillator number and k is an integer, (0,16) R related to the Kaluza-Klein momentum p carried by the state in the following way: 9 1 1 p = (k+B.P + B2N), (1.2) 9 R 2 H where Bi are the holonomies in the Cartan subalgebra of SO(32) in the X9 direction (Wilson lines). The multiplicity is then given by the N ’th oscillator level in the parti- R tion function η−24. Furthermore, the mass m in the string frame is given by NR H m = p + . (1.3) 9 α′ (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) Ontheotherhand,theTypeItheoryi(cid:12)srelatedto(cid:12)theheteroticbyastrong-weakduality. The coupling constants and metrics are related by 1 GH λ = , GI = MN. (1.4) I λ MN λ H H The duality relations imply that a heterotic state labelled by N, k and P is mapped to a type I state with Kaluza-Klein momentum p and mass m, in the string frame, given 9 by: 1 1 p = (k+B.P + B2N) (1.5) 9 R 2 I NR I m = p + , (1.6) 9 α′λ (cid:12) I (cid:12) (cid:12) (cid:12) where RI is the type I radius along t(cid:12)(cid:12)he X9 dire(cid:12)(cid:12)ction. The Neveu-Schwarz two-form, B , of the heterotic string is mapped to the Ramond- MN Ramond antisymmetric tensor field, BR , of the Type I string. Consequently, the MN 1 winding modes of the fundamental string on the heterotic side are mapped to the D- stringwindingmodesontheTypeIside. Duality, therefore, predictsthataBPSsystem of N D-strings, in the Type I theory, each of which is wrapped once around the circle, should have a threshold bound state with multiplicity equal to 24. Furthermore, when there are other charges turned on, the multiplicities of the resulting D-string threshold bound states should reproduce the heterotic multiplicity (1.1). Our aim in this letter is to establish that this is indeed the case. We begin by reviewing the effective world volume gauge theory of N type I D-strings and argue that in the infrared limit, this gauge theory flows to an orbifold conformal field theory [3, 4, 5], in analogy with a similar phenomenon in the type II case [6, 7, 8, 9, 10]. We compute the elliptic genus of this conformal field theory and show that the twisted sector cor- responding to the longest string reproduces the results expected from heterotic-type I duality. We also argue that the threshold bound states arise precisely in the longest string sector. 2 Type-I D-Strings We begin by recalling theform of the world volume action which describes the low lying modes of a system of N D-strings in the type I theory: 1 S = Tr d2x F2+(DX )2+g2([X ,X ])2 − 4g2 I I J Z 16 16 +ΛD/Λ+SD/S + χ¯aD/χa+gΛΓI[X ,S]+ χ¯aBaχa. (2.1) I a=1 a=1 X X The fields transform in various representations of the gauge group O(N). X and S transform as second rank symmetric tensors, while Λ and χ transform in the adjoint and fundamental representations respectively. There is an SO(8)R, R symmetry group, under which X, S, Λ and χ transform as an 8 (this is the I label), an 8 , an 8 and a V S c singlet, respectively. The χ transforms under the SO(32) in the vector representation with χa and χ¯a denoting the positive and negative weights. The Λ and χ are negative chiral (right-moving) world sheet fermions while the S are positive chiral (left-moving) fermions. Finally Ba are the background holonomies (i.e. Wilson lines on the 9-branes) in the Cartan subalgebra of SO(32). The Yang-Mills coupling g is related to the type I string coupling via g2 = λ /α′. The vev’s of the X fields, appearing in the action I above, measure the distances between the D-strings in units of √α′λ . This fact will be I important later, when we compare the spectrum with that of the heterotic theory. Geometrically the fields appear in the following fashion [2, 11]. The above action arises as the Z projection of the corresponding theory in the type II case. Recall that in 2 the type II situation a system of N branes has a U(N) symmetry. Write the hermitian matrices as a sum of real symmetric matrices and imaginary anti-symmetric matrices. The Z projection, for type I D-strings, assigns to the world volume components of the 2 2 gauge field the anti-symmetric matrices, that is, it projects out the real symmetric part and so reduces the gauge group to O(N). On the other hand, the components of the gauge field in the transverse directions, X, have their imaginary part projected out and so are symmetric matrices transforming as second rank symmetric tensors under the O(N). The diagonal components of the X give the positions for the N branes. The trace part, which we factor out, represents the center of mass motion. The χ carry the SO(32) vector label, as they are the lowest modes of the strings which are stretched between the 9-branes and the D-strings. As mentioned earlier, the winding mode N of the fundamental heterotic string is mapped, via duality, to N D-strings on the Type I side that wind around the X9 circle. Since the former appears as a fundamental BPS state with multiplicity 24, we should findthat the system of N D-strings in type I theory admits threshold boundstates with multiplicity 24. Inotherwords,intheO(N)theory,thereshouldbe24squareintegrable ground states that are 10 dimensional N = 1 vector short BPS super-multiplets. That every groundstate appearswith8bosonicand8fermionicmodes, necessary to formthe N =1shortvectorsupermultiplet,follows fromthethefactthattherearezeromodesof the O(N) singlet free field S, describing the center of mass motion (which has not been included in the action (2.1)). The remaining part of the O(N) theory described in the action (2.1), therefore, must have, 24 bosonic normalizable ground states as predicted by the heterotic-type I duality. In other words, we want to show that the Witten Index for the above theory is 24. More generally, in the heterotic theory, we can also turn on other charges, namely the one associated with Kaluza-Klein modes that couple to G and the ones associated µ9 with the Cartan subalgebra of the SO(32) gauge group. These charges can also be excited in the system of N D-strings in the type I theory. Indeed,one can includestates carrying a longitudinal momentum along the string and thereby generate Kaluza-Klein momentum. Similarly, onecan generate SO(32) quantumnumbersby suitablyapplying χmodes. Theinformationaboutthemultiplicities ofstates carryingtheseextracharges will be contained in the elliptic genus of the above theory (2.1). Since the Witten index and, more generally, the elliptic genus do not depend on the couplingconstant, wecantakealimitwhichismostconvenientforourpresentpurposes. We will consider the infra-red limit of the theory, as it has been conjectured in [3, 4, 5], that in this limit the theory flows to a (8,0) orbifold superconformal field theory. This is in analogy with a similar conjecture for a system of type IIB D-strings [8]. In the following we give some support to this conjecture by, first, gauge fixing (2.1) and then by performing a formal scaling which yields the orbifold theory directly. 3 3 Type II and the IR Limit Before discussing the type I theory we make a digression on the type II theory that will prove useful later. Our aim here is to show that with a prudent choice of gauge one can simplify matters considerably. This prepares the way for taking the large coupling limit in a fashion that is, to a large extent, controlable. We will work with a little more generality than is really required and begin with an analysis of D dimensional Yang-Mills theory reduced to d dimensions. D dimensional vector labels are denoted by M,N,..., those in d dimensions are denoted by µ,ν,... and those in the remaining D d (reduced) dimensions by I,J,.... To make contact with the type II D-brane − world volume theories one sets D = 10. The D dimensional Yang-Mills theory has a ‘potential’ of the form, g2 tr [A ,A ]2. (3.1) M N − 4 Minimising, in any dimension d, we learn that we are interested in the fields that live in the Cartan subalgebra. Decompose the Lie- algebra, g, of the gauge group as g = t k, ⊕ wheretis thepreferedCartan subalgebraand kis its ortho-complement. It makes good sense, therefore, to perform a non-canonical split, t k A = A +A , (3.2) M M M where the superscripts indicate the part of the Lie-algebra that the fields live in. Before proceeding we need to gauge fix. Given the splitting of the algebra, it behoves us to choose the ‘background field’ gauge1 DM(gAt)gAk = 0, (3.3) M which preserves the maximal Torus gauge invariance. The ghosts come in as trCkD (gAt)DM(gA)Ck+ tr g2Ck[[Ak ,Ck]t,AMk]. (3.4) M M We choose a Feynman type gauge with a co-efficient chosen to give the most straight- forward analysis namely we add tr 1 DM(gAt)Ak 2 (3.5) −2 M (cid:16) (cid:17) to the action. With this choice the potential becomes g2 tr [At ,Ak ]2+..., (3.6) − 2 M N 1AtthispointtheconnectionisgA,whichexplainssomeofthe,whatappeartobe,spuriousfactors of g. We are gauge fixing the ‘canonical’ gauge field and not the one scaled by g so that the BRST transformations are Q(gAM)=DM(gA)C and QC =C2. 4 k where the ellipses indicate higher order terms in A and which, directly, will be seen M to be irrelevant. We now perform the following sequence of scalings on the fields appearing in a N = 1 super Yang-Mills theory in D dimensions k 1 k k 1 k k 1 k A A , ψ ψ , C C . (3.7) M → g M → √g → g2 On a torus, Td, with periodic boundary conditions on all the fields appearing, this scaling has unit Jacobian. We can now take the g limit. The action, in this limit → ∞ is: 1 1 S = tr ddx F (At)2+ψt/∂ψt [At ,Ak ]2 − 4 MN − 2 M N Z +ψkΓM[At ,ψk]+[Ck,At ][Ck,At ]. (3.8) M M M All the fields in the k part of the Lie-algebra can be integrated out and clearly give an overall contribution of unity to the path integral. Thus, we are left with a free, supersymmetric, system of Cartan valued fields. By invoking the Weyl symmetry that is left over, one finds that the target space of the theory is (R(D−d)r Tdr)/W, where × r is the rank of the group and W is the Weyl group. Fixing D = 10, gives us the Type II world volume theories of parallel D branes. The limit just described, the strong coupling limit in the gauge theory, when d = 2 and D = 10 (the D-string) gives rise to the orbifold conformal field theory as in [8]. 4 Type I and the IR Limit The flat directions of the potential in this case require mutually commuting matrices t once more. We denote those X’s, with a slight abuse of notation, by X (for example one may choose these to be diagonal). A convenient way to proceed is to start with the (complexified) SU(N) Lie algebra and to split it into a Cartan subalgebra t and into positive and negative roots, k+ and k−, respectively, that is, k = k+ k−. The ⊕ Z projection means that, in this basis, the world volume gauge fields are proportional 2 to the anti-symmetric (imaginary part of k) generators, m− = k+ k−, while the X’s − are proportional to the symmetric generators, t and (real part of k) m+ = k+ + k−. With these identifications the bosonic parts of the type I and type II theories coincide. We choose the same gauge fixing as in the type II theory, now restricted to the m− directions, g∂µAm− +g2[Xt,Xm+]= 0 (4.1) µ 5 and we scale the fields in a similar way, that is 1 1 1 Am− Am−, Xm+ Xm+, Λm− Λm−, µ → g µ I → g → √g 1 1 Sm+ Sm+, Cm− Cm−. (4.2) → √g → g2 The remaining fields Xt, St, Cm− and χ are unchanged. As before the Jacobian of these scalings is unity if we take periodic boundary conditions for the fermions S and Λ. Thereis no such requirement on the χ. Consequently the g limit may besafely → ∞ taken. The action now takes the form S = tr ddx 1 ∂ Xt 2+St/∂St 1[Xt,Am−]2 1[Xt,Xm+]2 − 2 µ I − 2 I µ − 2 I J Z (cid:12) (cid:12) (cid:12) (cid:12) 16 16 +Λm−ΓI[Xt,(cid:12)Sm+](cid:12)+[Cm−,Xt][Cm−,Xt]+ χ¯aD/χa+ χ¯aBaχa.(4.3) I I I a=1 a=1 X X Formally, since the χ fields are chiral, only the right moving part of the gauge field is coupled to it and one can perform the integral over the left moving part of the gauge field which sets the right moving part to zero. Hence, on integrating out the massive t t modes, one would be left with a completely free theory of the massless modes X , S and χ. The determinant factors would then, at least formally, cancel between the fields of various statistics. However, the above cancellation of the determinant factors is a bit quick. If correct, it would imply that even if we had started with an anomalous theory we would end up, in the limit, with a well defined superconformal field theory. For example, this would seem to be the case if we simply ignored the χ fields altogether. The point is that each fermionic determinant appearing is anomalous. These determinants, when defined in a vector gauge invariant way, involve extra quadratic terms in the gauge field. The presence of these would mean that the functional determinants would not cancel, since the gauge field contribution would not be Det(Xt)2. Happily, the condition that the theory be anomaly free means that the total sum of these extra pieces is zero and this is exactly what is required to make our formal argument above work. On including the center of mass one gets N of the X’s and S’s, each transforming as a 8 and 8 of SO(8) respectively and N χ’s each transforming as a fundamental of V S SO(32). ThefieldcontentislikethatofN copiesoftheheteroticstringinthelight-cone gauge with an effective inverse tension ′ ′ α = αλ . (4.4) eff I The condition (4.1) does not completely fix the gauge, there are still discrete transfor- mations which leave the action invariant. There is the permutation group S which N permutes the N copies of (X,S,χ) and which has the interpretation of permuting the 6 N D-strings. There are also O(N) transformations which leave invariant X and S but which act non-trivially on the χ’s by reflection giving rise to a ZN. The full orbifold 2 group is therefore the semidirect product S ⋉ZN. N 2 5 Orbifold Partition Function We are interested in calculating the elliptic genus of the orbifold conformal theory. In this case the S fermions have periodic boundary conditions on the world sheet torus. The elliptic genus for our conformal field theory will be zero due to the fact that the center of mass S will have zero modes in all the twisted sectors as it is orbifold group invariant. However, the zero modes of the center of mass S precisely give rise to the 8 bosonic and 8 fermionic transverse degrees of freedom that fill out the 10-dimensional N = 1 vector supermultiplet corresponding to a BPS short multiplet. Our goal here is to calculate the multiplicities of these BPS states, that are clearly governed by the elliptic genus of theremainingconformalfield theory thatdescribes therelative motions of the D-strings. This means that we need to consider only those twisted sectors that have at most the zero modes of the center of mass S. Let us briefly review how the orbifold elliptic genus is computed [12, 13]. Each twisted sector of the orbifold corresponds to a conjugacy class of the orbifold group. A general element of the group G= S ⋉ZN can bedenoted by (g,ω) whereg S andω ZN. N 2 ∈ N ∈ 2 First let us identify the twisted sectors where the S’s have no other zero modes besides the center of mass one. For this it is sufficient to consider the action of the elements of S since the ZN part does not act on S fields. A general conjugacy class [g] in S N 2 N is characterized by partitions N of N satisfying nN = N where N denotes the n n n { } multiplicity of the cyclic permutation (n) of n elements in the decomposition of g as P [g] = (1)N1(2)N2 (s)Ns. (5.1) ··· Inthe[g]-twisted sector thefieldssatisfy theboundarycondition: (X,S,χ)(σ+2πR ) = I g(X,S,χ)(σ) where σ is the coordinate along the string. In each twisted sector one mustproject by the centralizer subgroupC of g, which takes g the form: s C = S ⋉ZNn, (5.2) g Nn n n=1 Y where each factor S permutes the N cycles (n), while each Z acts within one Nn n n particular cycle (n). In the path integral formulation this projection involves summing over all the boundary conditions along the world-sheet time direction t, twisted by elements h of C . We shall denote by ([g],h) the twisted sector with twist g along the g σ direction and twist h along the t direction. We will now show that if [g] involves cycles of different lengths, say (n)a and (m)b with n = m, then the corresponding twisted sector does not contribute to the elliptic 6 7 genus. To see this, we note that there are now at least two sets of zero modes for S, which can be expressed, by a suitable ordering of indices, as (S + S + S ) 1 2 na ··· and (S + S ), where the two factors (n)a and (m)b act on the two sets of na+1 na+mb ··· indices in the obvious way. These zero modes survive the group projection because the centralizer of [g] does not contain any element that mixes these two sets of indices with each other, thereby giving zero contribution to the elliptic genus. Thus we need only to consider those sectors with [g] = (L)M where N = LM. The centralizer in the case where [g] = (L)M is C = S ⋉ZM. From the boundary g M L condition along σ it is clear that there are L combinations of S’s that are periodic in σ. By suitable ordering, they can be expressed as Sk = L(k+1) S for k = 0,...,M 1. i=Lk+1 i − These zero modes have to be projected by the elements h in the centralizer C . In P g particular, when h is the generator of Z S C , it acts on the zero modes Sk M M g ⊂ ⊂ by cyclic permutation. It is clear, therefore, that only the center of mass combination M−1Sk is periodic along the t direction. Hence, this sector contributes to the elliptic k=0 genus. More generally any h = (e,f) C = S ⋉ZM will satisfy the above criteria P ∈ g M L provided e = (M) S and f is some element of ZM. The number of such elements h ∈ M L is (M 1)! LM. − × Inparticular whenN isprimethesectors thatcontribute totheelliptic genusare: (1,h) where h Z and ([g],h) with [g] = (N) and h in the corresponding centralizer Z . N N ∈ In the following we shall refer to these two types of sectors as the shortest and longest string sectors, respectively. The full orbifold group G is specified by an element of S (discussed above) together N with an element of ZN that acts on the χ’s. Let us denote a general element by (g,ǫ) 2 where g S and ǫ ZN. Now S acts as an automorphism in ZN by permuting ∈ N ∈ 2 N 2 the various Z factors. We denote this action by g(ǫ). Then the semi-direct product 2 ′ ′ ′ ′ is defined in the usual way: (g,ǫ).(g ,ǫ) = (gg ,ǫg(ǫ )). Twisted sectors will now be labelled by a conjugacy class in G. The relevant sectors, for the elliptic genus computa- tion, as discussed above, are the conjugacy classes [g] in S of theform [g] = (L)M with N N =LM. One can easily verify that the various classes in G are labelled by ([g],ǫ) with ǫ = ǫ .ǫ ...ǫ where each ǫ is in the quotient subgroup of ZL by its even subgroup. 1 2 M i 2 Combining this with the condition that we have found for h we may conclude that all the ǫ ’s must be equal (i.e. all ǫ ’s must be either even or odd element of ZL) in order i i 2 for such h to exist in the centralizer of ([g],ǫ) in G. In this case the centralizer is the group of elements of the form (h,α) where h C and α ZN satisfies ǫh(ǫ) = αg(α). ∈ g ∈ 2 The number of independent such α’s is 2M and therefore the order of the centralizer of ([g],ǫ) is M!LM2M. We now proceedto computethe elliptic genus tr( 1)Fe TH +2πiRτ1Pσ whereH and − − P are the Hamiltonian and the longitudinal momentum. The computation for general σ L and M is quite tedious, therefore we will only describe it for the longest string sector (i.e. M = 1 and L = N). The centralizer C = Z and consists of elements of the g N 8 form h = gs for s = 0,1,...,N 1 and as a result their action is obtained by modular − transformations τ τ +s from the h = 1 sector. Thus we can restrict ourselves to 1 1 → h = 1. The eigenvalues of [g] are ωr for r = 0,1,...,N 1 where ω = e2iπ/N. As − a result the N copies of X’s and S’s come with fractional oscillator modes that are shifted by r/N in units of 1/R . The left moving part of the non-zero mode partition I function cancels between X’s and S’s. The zero modes appear for the center of mass (i.e. r = 0): the zero modes of S give rise to the usual 8 bosonic and 8 fermionic degrees of freedom filling out the BPS vector supermultiplet, while the zero modes of the left and right moving X’s give a factor τ−4. The right moving X’s, upon taking 2 the product over r, give 1/η(qN1 )8, upto a zero point shift, where q = exp(2πiτ) with τ = τ +i T τ +iτ . 1 2πRI ≡ 1 2 To includethecontribution of theχ’s wemustspecify thegroupelements inZN as well. 2 There are two possible ǫ’s that come with [g]: the even and odd element of ZN. First, 2 let us consider the situation when the Wilson lines B are set to zero. By taking the a product of all the eigenvalues of the twist, these for odd N, give rise, respectively, to 1 the Neveu-Schwarz and Ramond sectors of the SO(32) fermions, with q replaced by qN (again upto zero point shifts). For even N, on the other hand, the Neveu-Schwarz and Ramond sectors appear for ǫ odd and even, respectively. Furthermore the centralizer contains two elements with h = 1 namely α = 1. These two choices give rise to the ± usual GSO projection. Finally, one can compute the zero point shift for the right moving X’s and χ’s and the result is that one actually gets 1/N times the right moving part of the heterotic SO(32) 1 partition function with q replaced by qN. Including also other elements h= 1, the final 6 result is: N−1 1 1 Z(ωsqN1 ), (5.3) τ4N 2 s=0 X where Z(q) is the right moving part of the SO(32) heterotic partition function: Z(q) = 1 q21P2, (5.4) η(q)24 PX∈Γ16 with Γ is the spin(32)/Z lattice. 16 2 For the sector with [g] = ((L)M,ǫ), we can again repeat the above steps. Recall that in this case there are only two possible ǫ that give non zero contribution to the elliptic genus. These are given by ǫ = ǫ .ǫ ...ǫ , with all ǫ either an even or odd element of 1 2 M i ZL. As described above, the order of the centralizer C is M!LM2M, while the number 2 g ofelements h C thatgiverisetonon-zerotraceis(M 1)!LM2M, andthereforethese g ∈ − are the relevant elements for the computation of elliptic genus. However, not all the h’s ′ of this form give different traces. Indeed, if h and h are in the same conjucacy class in C , they will give the same trace. It is easy to verify that the number of elements in the g centralizer Cˆ inC ,forarelevanth, is2ML = 2N. Asaresult,thenumberofelements h g 9