OUTP-96-74P hep-th/9612251 December, 1996 D-brane Configurations and Nicolai Map in Supersymmetric Yang-Mills Theory 7 9 9 1 n a I. I. Kogana, G. W. Semenoff b,1 and R. J. Szaboa,1 J 5 1 a Department of Theoretical Physics University of Oxford 2 1 Keble Road, Oxford OX1 3NP, U.K. v 1 5 b Department of Physics and Astronomy 2 University of British Columbia 2 Vancouver, B.C. Canada V6T 1Z1 1 6 9 / h t - p e h : v i Abstract X r Wediscusssomepropertiesofasupersymmetricmatrixmodelthatisthedimensional a reduction of supersymmetric Yang-Mills theory in ten dimensions and which has been recently argued to represent the short-distance structure of M theory in the infinite momentum frame. We describe a reduced version of the matrix quantum mechanics and derive the Nicolai map of the simplified supersymmetric matrix model. We use this to argue that there are no phase transitions in the large-N limit, and hence that S-duality is preserved in the full eleven dimensional theory. 1 Work supported in part by the Natural Sciences and Engineering Research Council of Canada. The conventional understanding of the spacetime structure of string theory has drastically changed over the last few years. It has been realized recently that all ten dimensional su- perstring theories are related by non-perturbative dualities and that they can be thought of as originating, via Kaluza-Klein types of compactifications, from a single, eleven dimensional theory known as ‘M theory’ (see [1] for recent reviews). The dynamics of this theory are not yet fully understood. Some of the central objects in understanding the string dualities are non-perturbative, p-dimensionaldegreesoffreedomknownasDp-branes[2], onwhichtheend- points of strings can attach (with Dirichlet boundary conditions). The low-energy dynamics of a system of N parallel D-branes can be described by an N×N matrix model obtained from the dimensional reduction of ten dimensional supersymmetric Yang-Mills theory with gauge group U(N) [3, 4]. The large-N limit of this matrix model has been recently conjectured to describe the small distance spacetime structure of M theory in the infinite momentum frame [5]. The explicit solution of this matrix model therefore has the potential of providing a non-perturbative description of the largely unknown dynamical objects describing the short distance behaviour of the full eleven dimensional theory [6]. In this Letter we will discuss some basic properties of the supersymmetric matrix model introduced in [4, 5]. We examine a particular reduction of the model proposed in [4, 5] to static D-brane configurations with transverse SO(8) rotational symmetry. We explicitly construct the Nicolai map associated with the supersymmetry in this reduced model and use it to analyse Schwinger-Dyson equations of the matrix model. We show that the results from this analysis are consistent with other known results of the D-brane field theory, and also that this simple approach gives some insights into the structure of the full matrix model. In particular, the reduced model seems to have no phase transitions in the large-N limit and S-duality is preserved in this representation of the full eleven-dimensional theory. First, we discuss some aspects of the representation of systems of D-branes by Yang-Mills fields. Consider the gauged supersymmetric matrix quantum mechanics with action [4, 5] 1 9 2 2 9 S = dt tr D Xi − Xi,Xj +2ψαD ψ −2 ψα(γ )β ψ ,Xi (1) 2g  t t α i α β  Z Xi=1(cid:16) (cid:17) Xi<j h i Xi=1 h i   where D Y = ∂ Y −i[A (t),Y] is the temporal component of the gauge covariant derivative, t t 0 the trace is taken over unitary group indices, and we have chosen units in which the string tension is α′ = 1/2π. Here Xi(t) = [Xi (t)], a,b = 1,...,N, i = 1,...,9, are N × N ab Hermitian matrices in the adjoint representation of U(N) which are obtained as the spatial components in the reduction to (0 + 1) dimensions of a (9 + 1) dimensional U(N) Yang-Mills field A (x,t), µ = 0,...,9. They describe the collective coordinates of a system of N parallel µ D0-branes(with infinitesimal separation), and they transform under the vector representation of the rotation group SO(9) of the space transverse to the compactified eleventh dimension of 1 the underlying supergravity theory. The superpartners of the matrices Xi are the Majorana spinors ψα(t) = [ψα(t)], α = 1,...,16, which transform under the 16-dimensional spinor ab representation of SO(9), and under the adjoint representation of the gauge group U(N). The Dirac matrices γ are the generators of the spin(9) Clifford algebra i {γ ,γ } = 2δ (2) i j ij in a Majorana basis. The coupling constant g is related to the eleven dimensional compacti- fication radius R by R = g2/3l (3) P where l is the eleven dimensional Planck length. P Theaction(1)describestheshort-distancepropertiesofD0-branesinweakly-coupledtype- IIA superstring theory [4, 6]. It was argued in [5] to be the most general infinite momentum frame action with at most two derivatives which is invariant under the U(N) gauge group and the full eleven dimensional Lorentz group. It is further invariant under the infinitesimal N = 1 supersymmetry transformation δ Xi = −2εα(γi)βψ ε α β 1 1 δ ψα = D Xi(γ )α + Xi,Xj [γ ,γ ]α εβ (4) ε 2  t i β 2 i j β Xi Xi<j h i   δ A = −2εαψ ε 0 α where εα are 16 global Majorana spinor parameters. There are another set of 16 super- symmetries which are realized trivially as δε′ψα = ε′α, δε′Xi = δε′A0 = 0, where ε and ε′ are independent supersymmetry parameters. Together, these two sets of supersymmetry transfor- mations (with slight modifications) generate the full 32-dimensional 16⊕16 representation of the super-Galilean group of the eleven dimensional theory in the light-cone frame. A number of properties of M theory have been verified using the action (1) [5],[7]–[9]. It may seem puzzling at first sight that a dimensionally-reduced gauge field A yields the appropriate D-brane coordinatization. What is intriguing though is that string theoretic T- dualityisakey tothisconnection. Ifwecompactifythefirstspatialdimension ontoacircleS1 R of radius R (or more generally several dimensions onto a torus), then the angular coordinate x1 (or coordinates xi) takes values in the interval x1 ∈ [0,2πR]. The original supersymmetric Yang-Mills theory in ten dimensions describes the low-energy sector of open superstrings, and in this theory there exists other gauge-invariant variables, namely the path-ordered Wilson loop operators W[A] = tr P exp i A dxi (5) i ISR1 ! 2 whichareinvariantunderthelargegaugetransformationswhichwindaroundthecompactified direction. The connection appearing in the argument of the exponential in (5) lies in the adjoint representation of U(N). The corresponding classical gauge field orbits are topological andlieintheintervalA ∈ [0, 2π]. Foreachcoordinatexi livingonacircleS1 thereisa“dual” 1 R Ri coordinate Xi = α′A which also lives on a circle S1 but with a dual radius r = α′/R . As i ri i i we have mentioned the ten-dimensional supersymmetric Yang-Mills theory describes the low- energy dynamics of open strings with Neumann boundary conditions. Under T-duality R → α′/R the Neumann boundary conditions are transcribed into Dirichlet boundary conditions. But this means that the topological gluon degrees of freedom are converted into the D-brane fields X(t) = α′A(t) describing open strings with Dirichlet boundary conditions. Thus the ten-dimensional gluon fields describe the dual theory of the D-branes and T-duality naturally identifies the topological orbits of the Yang-Mills fields as the D-brane coordinates. The role of T-duality in the matrix model (1) has also been addressed from other points of view in [8]. A question which now arises is how to really measure the original angular coordinate xi given the gauge field coordinate Xi. The solution is to consider more complicated objects, such as the abelian Wilson loops with a winding number n ∈ Z W [A] = exp in A dxi (6) n i ISR1 ! which are associated with a single D-brane configuration. In the general case, i.e. in the case of a non-abelian gauge group, we should consider Wilson loops in different representations of U(N), but for the sake of illustration we shall discuss here only the U(1) Wilson loops (6). This simplication canbe thought ofas acompactification ofallof thetendimensions onwhich the group variables (5) are restricted to the maximal torus of the U(N) gauge group. Using these objects and T-duality one can in principle obtain the angular coordinate xi through a superposition W [A] of different Wilson loops (using harmonic analysis on unitary n∈Z n groups in the genPeral case). In fact, this construction demonstrates that the Wilson loops in the “N-theory” (i.e. ordi- nary open strings with Neumann boundary conditions) are equivalent to the vertex operators in the “D-theory” (open strings with Dirichlet boundary conditions). To see this, we ex- pand the topological gauge field configurations of the low-energy description of strings on the compactified space in Fourier modes as A (t,x) = A(ℓ)(t)eiℓRx , x ∈ [0,2π/R] (7) 1 1 ℓ∈Z X Thesedualfieldsaredescribedbya(1+1)dimensionalgaugetheorythatisalsodimensionally reduced from the ten-dimensional supersymmetric Yang-Mills theory [8]. Then the Wilson loop (6) becomes Wn = einRA(10)(t) = eipnX1(t) (8) 3 where p = n/r = nR/α′ is the momentum of the string winding mode in the compactified n direction. Thus T-duality converts the non-commuting position matrices Xi describing the D-brane configurations into gauge fields in the dual theory, and also the winding number of the topological gauge field modes into the string momentum in the compactified direction. In the general case then, we can conjecture the equivalence between the string scattering amplitudes of the D-theory defined by correlators of the vertex operators and expectation values of the Wilson loop operators in the dual N-theory2, hW [A]···W [A]i = ei~n1·X~ ··· ei~nk·X~ (9) ~n1 ~nk N D D E where ~n ∈ Z9 and we have defined W [A] = 9 W [A] in terms of the abelian Wilson j ~nj i=1 (~nj)i loops (6). In the non-abelian case, the correlatQors (9) will generalize in the appropriate way in terms of representations of U(N). We now examine the problem of obtaining an explicit solution of the matrix model (1). For this, we further dimensionally reduce the theory described by (1) to a zero-dimensional N × N supersymmetric matrix model, i.e. we ignore the time dependence in (1) and work in the Weyl gauge A = 0. This means that we are studying the model separately over each 0 constant time slice describing a static configuration of the D0-branes. This reduction can be thought of as originating by compactifying the time direction of the ten dimensional Yang- Millstheorywheretheadjointrepresentationfermionshaveperiodicboundaryconditions, and then taking the limit in which the radius of compactification vanishes. This simplification has theadvantageofeliminatingnon-localoperatorsthatwouldappearfromthetime-dependence. We shall discuss the inclusion of time-dependent fields at the end of this Letter. With this further reduction, the partition function of the model is given by the finite- dimensional matrix integral N 2 Z = dXi dψα exp tr Xi,Xj −2 ψα(γ )β ψ ,Xi (10) ab ab 2g  i α β  Z Ya,b Yi Yα  Xi<jh i Xi h i    We can expand the matrix integration variables in (10) in a basis TA of the unitary group as Xi = XiTA and ψα = ψαTA, where A = 1,...,N2 and the Hermitian U(N) generators A A satisfy 1 TA,TB = ifAB TC , tr TATB = δAB (11) C 2 h i TheintegrationovertheMajoranafermionsin(10)isGaussianandcanbeevaluatedexplicitly using the Berezin integration rules for ψα. It produces a square root of the determinant A determined by the representation of the adjoint action of Xi, and (10) becomes N2 i N 2 Z = c dXi Pfaff fABC (γ )βXi exp tr Xi,Xj (12) N D "2g i α C# 2g  Z Yi DY=1 Xi  Xi<j h i  2See [10] for another discussion of the relationship between string vertex operators and Wilson lines. 4 wherethePfaffianistakenoverboththeadjointU(N)representationindicesA,B = 1,...,N2 and the spin(9) indices α,β = 1,...,16. Here and in the following c denotes an irrelevant N numerical constant. Wewouldnowliketoexploit thesupersymmetry (4)ofthezerodimensional modeltocom- pute correlation functions of the matrix model. When the number of bosonic and fermionic degrees of freedom are the same, the supersymmetry is maximal, in that it holds even when the fields are off-shell. In the present model, the number of fermionic and bosonic degrees of freedom do not match. Normally, supersymmetry would require on-shell fields and the addition of auxilliary fields to make the number of physical boson and fermion modes equal. However, we can adjust things to match by exploiting the original interpretation of the matri- ces Xi from the dimensional reduction of ten-dimensional supersymmetric Yang-Mills theory. The latter theory can be gauge-fixed and quantized in the light-cone gauge, after which there are only eight propagating gluon degrees of freedom corresponding to the various possible transverse polarizations. Since a Majorana-Weyl spinor in ten dimensions has eight physical modes, the minimal Yang-Mills action in ten dimensions is supersymmetric without the need of introducing auxilliary fields. Thus we match the collective degrees of freedom Xi of the sys- tem of D-branes with the physical modes of the full Yang-Mills theory by reducing the target space degrees of freedom from nine to eight by setting X9 = 0 and working in the Majorana- Weyl representation of spin(9) in the matrix model above. The constraint X9 = 0 can be thought of as a light-cone gauge fixing condition in the nine-dimensional transverse space. Although not precise from the point of view of the M theory dynamics, this simplification produces a toy model that will shed light on some of the properties of the nine-dimensional theory that we started with3. With this simplification we now exploit some features of the group theory for SO(9). The Dirac generators of spin(9) in the Majorana-Weyl basis can be constructed from the reducible 8 ⊕8 chiral representation of spin(8) by decomposing the 16-dimensional gamma-matrices s c in the 8×8 block form 0 (γ )α˙ i α γ = , i = 1,...,8 (13) i (γ )β 0  i β˙   where (γ )α = (γT)α˙, α,α˙ = 1,...,8, are the Dirac generators of spin(8). The Clifford algebra i α˙ i α (2) is then equivalent to the equations (γ )α˙(γ )β +(γ )α˙(γ )β = 2δ δβ (14) i α j α˙ j α i α˙ ij α 3Some different reductions of the matrix quantum mechanics (1) have also been suggested. In [11] it was arguedthatthemodelcanbetruncatedtozerodimensionsbyaugmentingthetransverserotationalsymmetry toSO(11)andviewingthematrixmodelasthedimensionalreductionofsupersymmetricYang-Millstheoryin (10+2)dimensions. In[10]itwasarguedthattheD-branefieldtheoryassociatedwithweakly-coupledtype- IIB superstringscould be viewed as the large-N reduction of the ten dimensional supersymmetric Yang-Mills theory. 5 and similarly with dotted and undotted chiral indices interchanged. The spin(8) Dirac gen- erators can be expressed explicitly as direct products of 2×2 block matrices by (γ )α˙ = −iσ ⊗σ ⊗σ , (γ )α˙ = i1⊗σ ⊗σ 1 α 2 2 2 2 α 1 2 (γ )α˙ = i1⊗σ ⊗σ , (γ )α˙ = iσ ⊗σ ⊗1 3 α 3 2 4 α 1 2 (15) (γ )α˙ = iσ ⊗σ ⊗1 , (γ )α˙ = iσ ⊗1⊗σ 5 α 3 2 6 α 2 1 (γ )α˙ = iσ ⊗1⊗σ , (γ )α˙ = 1⊗1⊗1 7 α 2 3 8 α 0 1 0 i 1 0 where σ = , σ = and σ = are the Pauli spin matrices. The 1 2 3 1 0 −i 0 0 −1       1 0 remaining spin(9) Dirac matrix is then γ = γ γ ···γ = . 9 1 2 8 0 −1   The block decomposition (13) of the first eight gamma-matrices shows that the Pfaffian in (12) with X9 = 0 becomes squared, with the spinor part of the determinant restricted to the spin(8) chiral indices. Then we can write the partition function as 8 N2 1 8 N 2 Z = c dXi det fABC (γ )α˙Xi exp tr Xi,Xj (16) N D A,B;1≤α,α˙≤8"2g i α C# 2g  Z iY=1DY=1 Xi=1  1≤Xi<j≤8 h i  This effective reduction to eight dimensions matches bosonic and fermionic degrees of free-   dom and allows us to exploit the supersymmetry in a simple way to completely solve the matrix model. Essentially it enables us to use the triality property of the eight-dimensional rotation group, i.e. that there exists automorphisms between the vector and chiral spinor representations of SO(8). We now label the eight spatial indices i as the chiral indices α,α˙ of the spinor represen- tation, and the first eight components of the 16-component spinor field ψ as the chiral parts ψα and the last eight components as the anti-chiral parts ψα˙ in the 8 ⊕ 8 decomposition s c of spin(9) above. The static reduction of the quantum mechanical action (1) can then be written in the standard form of an N = 1 supersymmetric field theory (after integration over superspace coordinates) as 1 ∂F 2 ∂2F ∂2F S = − +ψα ψα˙ +ψα˙ ψα (17) 0 2g  ∂Xi ! A ∂Xα∂Xα˙ ! B A ∂Xα˙∂Xα! B i,A A A B A B X   where the super-potential is 1 1 F(X) ≡ (γ )α˙ tr Xk[Xα,X ] = (γ )α˙fAB XkXαXC (18) 3 k α α˙ 6 k α C A B α˙ The representation (17) can be derived using the symmetry properties of the gamma-matrices (15), the U(N) Jacobi identity fABCfADE = fADBfACE −fADCfABE (19) 6 and the SO(8) Fierz identity (γ )α(γ )β˙ = δ δβ˙ +(γ )β˙ (20) i α˙ j α ij α˙ ij α˙ where we have introduced the spinor matrix 1 (γ )β = (γ )α˙(γ )β −(γ )α˙(γ )β (21) ij α 2 i α j α˙ j α i α˙ (cid:16) (cid:17) and similarly for (γ )β˙. ij α˙ The form of the action (17) identifies the Nicolai map (i.e. the Hubbard-Stratonvich transformation for the bosonic potential tr[Xi,Xj]2) [12] of this supersymmetric field i<j theory as P ∂F 1 1 WA(X) ≡ = (γ )α˙fAB XαXC or Wab = (γ )α˙[Xα,X ]ab (22) k ∂Xk 2 k α C B α˙ k 2 k α α˙ A From (17) we see that the Jacobian factor |det[∂WA/∂Xj ]|−1 which arises in the change of k B variables X → W(X) in the partition function (16) will cancel exactly with the determinant that comes from integrating out the chiral fermion fields. The partition function is thus trivially a Gaussian Hermitian matrix integral and is formally unity, c 8 N2 Z = (2g)N64N2 dWiA e−4Ng(WiA)2 = cN (23) Z i=1A=1 Y Y The free energy logZ is thus trivially an analytic function of the coupling constant g every- where and it does not exhibit any phase transitions, even in the large-N limit. Furthermore, the correlation functions which are invariant under the supersymmetry transformations (4) can be obtained by differentiating the free energy with respect to the coupling constants of the model (in an appropriate superspace formulation). Thus any supersymmetric correlator of the model vanishes, which is just the standard non-renormalization that usually occurs in supersymmetric field theories. The existence of the Nicolai map and these implied properties of the matrix model are essentially the content of the supersymmetric Ward identities. The only observables of the matrix model which are non-trivial are those which are not supersymmetric. To examine such correlation functions, we use the Nicolai map (22) to ex- press correlators h·i of the original matrix model (normalized so that Z = 1) as free Gaussian averages hh·ii of the Nicolai field. For instance, from hhWabii = 0 we deduce h[Xi,Xj]abi = 0. i Thismeans that theclassical groundstate ofthemodel(theminimum ofthe bosonicpotential tr[Xi,Xj]2) is that wherein the D-brane coordinates commute and have simultaneous i<j Peigenvalues corresponding to definite D0-brane positions. The full matrix model, which in- corporates quantum fluctuations about the classical ground state, thus describes smeared-out D0-brane configurations in a spacetime with a non-commutative geometry [4, 5, 8]. 7 More generally, we note that the Nicolai map X → W(X) is many-to-one, so that general correlators of the X matrices can have a multi-valued branch cut structure. To see if this is the case, we use the Nicolai field to write down a set of Schwinger-Dyson equations for the matrix model. The basic identity follows from the formula for Gaussian averages of products of even numbers of the fields Wab, i g2 n Wa1b1Wa2b2···Wa2n−1b2n−1Wa2nb2n = δ δ ···δ +Π(i)[i ,i ,...,i ] i1 i2 i2n−1 i2n N! i1i2 i3i4 i2n−1i2n 1 2 2n DD EE (cid:16) (cid:17) × δ δ δ δ ···δ +Π(a,b)[a ,b ;a ,b ;...;a ,b ] a1b2 a2b1 a3b4 a4b3 a2nb2n−1 1 1 2 2 2n 2n (cid:16) (cid:17) (24) where Π contains the sum of delta-functions over all permutations of indices. The delta- functions in the indices i come from the SO(8) invariance of the reduced matrix model, k while those in the indices a ,b arise from U(N) invariance. The non-vanishing correlation k k functions of the model are those which respect both of these symmetries. As an explicit example, we set n = 2 in (24) and sum over i = i , i = i and a = b , 1 2 3 4 1 2 a = b , a = b , a = b to get 2 1 3 4 4 3 tr tr 2 (γ )α˙ [Xα,X ](γi)β˙ Xβ,X (γ )σ˙ [Xσ,X ](γk)ρ˙[Xρ,X ] = 210g4 1+ (25) N i α α˙ β β˙ N k σ σ˙ ρ ρ˙ N2 (cid:28) h i (cid:29) (cid:18) (cid:19) In the large-N limit, the expectation value of a product of invariant operators factorizes into a product of correlators. Thus at N = ∞ (25) becomes tr 2 2 Xi,Xj = 210g4 (26) N Xi,j (cid:28) h i (cid:29) On the other hand, setting i = i , i = i and the a’s and b’s equal in the same way as above, 1 3 2 4 we get the N = ∞ equation tr 2 Xi,Xj = 32g2 (27) N Xi,j (cid:28) h i (cid:29) Combining (26) and (27) together we find that the large-N invariant variance of the SO(8) operator tr[X,X]2 is trivial, N 2 tr tr 2 2 tr 2 ∆2 [X,X]2 ≡ Xi,Xj − Xi,Xj = 0 (28) N N  N  (cid:18) (cid:19) Xi,j (cid:28) h i (cid:29) Xi,j (cid:28) h i (cid:29)   Note that (28) is a stronger statement than just the large-N factorization of correlators, as it implies a non-trivial factorization over the SO(8) indices as well. This is one manifestation of the supersymmetry of this matrix model. Similar other such identities can be derived for higher-order correlators of the X-fields. In this formalism, it is also possible to treat n-point connected correlation functions of the model. 8 The forms of the correlators above (and in particular (28)) seem to suggest that all non-vanishing observables in this model are analytic functions of the coupling constant g at N = ∞. This in turn implies that the large-N limit of the matrix model exhibits no phase transitions as one continuously varies g. As the relationship with M theory dynamics is even- tually obtained in the uncompactified limit where R → ∞ [5], the Nicolai map demonstrates explicitly that this limit can be taken unambiguously since there is no variation in the ana- lytic structure of the large-N solution. Moreover, the absence of phase transitions suggests that S-duality g → 1/g is maintained in the large-N limit of the matrix model above. A more precise examination of these properties requires the inversion of the Nicolai map (22) to get X(W), which would enable one to compute arbitrary non-supersymmetric correlators of the matrix model. The problem in trying to construct this inverse map is that generally W 6= 0, corresponding to the fact that the D-brane coordinates live in a non-commutative spacetime, so that it is not possible to simultaneously diagonalize the Xi’s and find the rela- tionship between the eigenvalue models for the W and X fields. The entire non-triviality of the matrix model lies in the correlators of invariant combinations of the operator X(W). The problem of inverting the Nicolai map has been discussed from a perturbative point of view in [13], where it was also shown that this transformation is a non-polynomial functional of the bosonic fields in the ten dimensional N = 1 supersymmetric Yang-Mills theory. It would be interesting to determine this inverse map, and use it to examine the properties of the Wilson loop correlators that we discussed earlier in addition to the large-N analyticity features of general correlators of the matrix model. In any case, we have formally described a method in which one can study features of the string scattering amplitudes (9). The results described above are only precisely valid with both the elimination of the tem- poral dimension and the reduction to an SO(8) spacetime symmetry group. If we reintroduce the time dependence of the matrix variables then the Nicolai map is determined as the non- local, time-dependent functional Wi(t) = D2Xi(t)−Wi(X(t)) with Wi given in (22). Then t the partition function yields the winding number of the multi-valued Nicolai map. The above results from the reduced matrix model, such as the analyticity in the coupling constant g, and hence the S-duality in the eleven dimensional compactification, show that the simple method discussed above has the potential of providing some insights into the structure of M theory. It would be interesting to see if the reduced matrix model can describe other features, such as membrane interactions [7], of the eleven dimensional theory. It would also be interesting to determine if the Nicolai map obtained above can be used to describe any properties of the ten dimensional supersymmetric Yang-Mills theory itself. We thank O. Lechtenfeld for comments on the manuscript and N. Mavromatos for inter- esting discussions. 9