March 1997

hep-th/9703030
cern-th/97-34
thu-97/06
utfa-97/06

Matrix String Theory

Robbert Dijkgraaf
arXiv:hep-th/9703030v3 13 Mar 1997

Department of Mathematics
University of Amsterdam, 1018 TV Amsterdam

Erik Verlinde
TH-Division, CERN, CH-1211 Geneva 23
and
Institute for Theoretical Physics
Universtity of Utrecht, 3508 TA Utrecht
and
Herman Verlinde
Institute for Theoretical Physics
University of Amsterdam, 1018 XE Amsterdam

Abstract

Via compactification on a circle, the matrix model of M-theory proposed by Banks et al

suggests a concrete identification between the large N limit of two-dimensional N = 8
supersymmetric Yang-Mills theory and type IIA string theory. In this paper we collect
evidence that supports this identification. We explicitly identify the perturbative string
states and their interactions, and describe the appearance of D-particle and D-membrane
states.
Introduction
By definition, M-theory is the 11-dimensional theory that via compactification on a
circle S 1 is equivalent to ten dimensional type IIA string theory [1, 2]. The string coupling
constant gs emerges in this correspondence as the radius of the S 1 , while the particles
with non-zero KK-momentum along the S 1 are identified with the D-particles of the IIA
model. According to the matrix theory proposal put forward in [3], the full dynamics of
M-theory can be captured by means of an appropriate large N limit of supersymmetric
matrix quantum mechanics.
In the original correspondence with the type IIA string, the matrix degrees of freedom
find their origin in the collective dynamics of the D-particles [4, 5, 6]. The key new in-
gredient in the approach of [3], however, is that the large N limit effectively accomplishes
a decompactification of the extra 11th direction, which therefore should be treated on
the exact same footing as all the other uncompactified dimensions. This insight, albeit
still conjectural, provides a number of important new theoretical tools. Namely, by inter-
changing the original role of the 11-th direction with that of one of the other directions,
one in principle achieves a concrete identification of the complete non-perturbative parti-
cle spectrum of string theory in a given dimension with a relatively convenient subset of
states (namely all states that can be made up from infinitely many D-particles) of string
theory compactified to one dimension less. In this way, much of the recently developed
D-brane technology is upgraded by one dimension and has become directly applicable in
the study of M-theory compactifications.
In this paper we aim to elaborate this new viewpoint for the simplest compactification
of M-theory, namely on S 1 . Following the current approach in matrix theory [3, 7], the
compactification on this S 1 is achieved by reinterpreting the infinite dimensional matrices
X i as covariant derivatives Di (written in a Fourier mode basis) of a large N gauge field
defined on the S 1 . In the original D-particle language, this procedure in fact amounts
to applying a T-duality transformation along the S 1 -directions, thereby turning the D-
particles into D-strings. Adopting this approach, we have cast matrix theory into the
form of a two-dimensional N = 8 supersymmetric U(N) Yang-Mills theory with the
Lagrangian∗

Z !
1 i 2 T 1 1
S= tr (Dµ X ) + θ D
/θ+ gs2 Fµν
2
− 2 [X i , X j ]2 + θT γi [X i , θ] . (1)
2π gs gs

Here the 8 scalar fields X i are N × N hermitian matrices, as are the 8 fermionic fields
θLα and θR
α̇
. The fields X i , θα , θα̇ transform respectively in the 8v vector, and 8s and
8c spinor representations of the SO(8) R-symmetry group of transversal rotations. The
two-dimensional world-sheet is taken to be a cylinder parametrized by coordinates (σ, τ )
with σ between 0 and 2π. The fermions are taken in the Ramond sector, and there is no
∗
Here we work in string units α′ = 1. A derivation of (1) from matrix theory and a discussion of our
normalizations is given in the appendix.

2
 D-particle # D-string winding #
T-Duality

9-11
S-duality
FLIP

T-Duality
IIA String IIB String
momentum p+ winding #

Fig. 1: The duality diagram that relates the matrix string

theory to large N D-particle quantum mechanics and the in-
terpretation of N in the models.

projection on particular fermion number. According to the matrix model philosophy, we

will consider this theory in the limit N → ∞.
Note that the same set of fields feature in the Green-Schwarz light-cone formalism
of the type II superstring, except that here they describe non-commuting matrices [3].
Indeed, the eigenvalues of the above matrix coordinates X i are from our point of view
identified with the coordinates of the fundamental type IIA string, since, relative to the
original starting point of large N D-particle quantum mechanics on S 1 × R8 , we have
applied an 11-9 flip that interchanges the role of the 11-th and 9-th directions of M-
theory. The interpretation of N in this matrix string theory is as (proportional to) the
total light-cone momentum p+ , see fig. 1.
In the following sections we will elaborate this correspondence in some detail. We
will begin with identifying the complete perturbative string spectrum, which arises in
the IR limit of the two-dimensional SYM theory. This correspondence is based on the
exact equivalence between a (free) second-quantized string spectrum and the spectrum of
a two-dimensional SN orbifold sigma-model† . We will then proceed to study the string
interactions from the SYM point of view. Finally, we will describe how D-particles and
other D-branes naturally appear in this theory.
While this paper was being written, the preprint [10] appeared, in which closely related
results are reported. We also became aware of the earlier work [11], where the formu-
lation of nonperturbative string theory by means of two-dimensional SYM theory was
first proposed and where some of our results were independently obtained. This approach
was also anticipated in [12, 13]. Similar points of view relating strings and matrices are
advocated in, among others, [14, 15].

†
This new representation of second quantized string theory was first pointed out in [8] and elaborated
in more detail in [9].

3
The free string limit
In the interpretation of the N = 8 SYM model as a matrix string, the usual YM
gauge coupling (which has dimension 1/length in two dimensions) is given in terms of the
string coupling as gY−2M = α′ gs2 . The dependence on the string coupling constant gs can
be absorbed in the area dependence of the two-dimensional SYM model. In this way gs
scales inversely with world-sheet length. The free string at gs = 0 is recovered in the IR
limit. In this IR limit, the two-dimensional gauge theory model is strongly coupled and
we expect a nontrivial conformal field theory to describe the IR fixed point.
Rather standard argumentation determines that the conformal field theory that de-
scribes this IR limit is the N = 8 supersymmetric sigma model on the orbifold target
space
S N R8 = (R8 )N /SN , (2)
see e.g. [16]. First we observe that in the gs = 0 limit the fields X and θ will commute.
This means that we can write the matrix coordinates as

X i = Uxi U −1 (3)

with U ∈ U(N) and xi a diagonal matrix with eigenvalues xi1 , . . . , xiN . That is, xi takes
values in the Cartan subalgebra of U(N). This leads to a description of the model in
terms of N Green-Schwarz light-cone coordinates xiI , θIα , θIα̇ with I = 1, . . . , N.
The complete correspondence with a free second-quantized string Hilbert space in the
N → ∞ limit involves the twisted sectors. The only gauge invariant quantity is the set
of eigenvalues of the matrices X i . Therefore, if we go around the space-like S 1 of the
world-sheet, the eigenvalues can be interchanged and the fields xiI (σ) can be multivalued.
Specifically, we have to allow for configurations with

xi (σ + 2π) = gxi (σ)g −1, (4)

where the group element g takes value in the Weyl group of U(N), the symmetric group
SN . The eigenvalue fields xiI (σ) thus take values on the orbifold space S N R8 . These
twisted sectors are depicted in fig. 2 and correspond to configurations with strings with
various lengths.
The Hilbert space of this SN orbifold field theory is decomposed into twisted sectors
labeled by the conjugacy classes of the orbifold group SN [9],
M
H(S N R8 ) = H{Nn } . (5)
partitions {Nn }

Here we used that for the symmetric group, the conjugacy classes [g] are characterized by
partitions {Nn } of N
X
nNn = N, (6)
n

4
 0 σ 2π
Fig. 2: A twisted sector corresponds to a configuration with
strings of various lengths.

where Nn denotes the multiplicity of the cyclic permutation (n) of n elements in the
decomposition of g
[g] = (1)N1 (2)N2 . . . (s)Ns . (7)
In each twisted sector, one must further keep only the states invariant under the centralizer
subgroup Cg of g, which takes the form

s
Y
Cg = SNn × ZN
n
n
(8)
n=1

where each factor SNn permutes the Nn cycles (n), while each Zn acts within one particular
cycle (n).
Corresponding to this factorisation of [g], we can decompose each twisted sector into
the product over the subfactors (n) of Nn -fold symmetric tensor products of appropriate
smaller Hilbert spaces H(n)
O
H{Nn } = S Nn H(n) (9)
n>0

where‡
 SN

S N H = |H ⊗ .{z
. . ⊗ H} . (10)
N times

The spaces H(n) in (9) denote the Zn invariant subsector of the space of states of a single
string on R8 × S 1 with winding number n. We can represent this space via a sigma model
‡
Here the symmetrization is assumed to be compatible with the grading of H. In particular for pure
VN
odd states S N corresponds to the exterior product .

5
of n coordinate fields xiI (σ) ∈ X with the cyclic boundary condition

xiI (σ + 2π) = xiI+1 (σ), I ∈ (1, . . . , n). (11)

We can glue the n coordinate fields xI (σ) together into one single field x(σ) defined on
the interval 0 ≤ σ ≤ 2πn. Hence, relative to the string with winding number one, the
oscillators of the long string that generate H(n) have a fractional n1 moding. The group
Zn is generated by the cyclic permutation

ω : xI → xI+1 (12)

which via (11) corresponds to a translation σ → σ + 2π. Thus the Zn -invariant subspace
consists of those states for which the fractional left-moving minus right-moving oscillator
numbers combined add up to an integer.
It is instructive to describe the implications of this structure for the Virasoro generators
(i)
L0 of the individual strings. The total L0 operator of the SN orbifold CFT in a twisted
sector given by cyclic permutations of length ni decomposes as

(i)
X L0
Ltot
0 = . (13)
i ni

(i)
Here L0 is the usual canonically normalized operator in terms of the single string coor-
dinates x(σ), θ(σ) defined above. The meaning of the above described Zn projections is
that it requires that the contribution from a single string sector to the total world-sheet
translation generator Ltot tot
0 − L0 is integer-valued. From the individual string perspective,
(i) (i)
this means that L0 − L0 is a multiple of ni .
To recover the Fock space of the second-quantized type IIA string we now consider
the following large N limit. We send N → ∞ and consider twisted sectors that typically
consist of a finite number of cycles, with individual lengths ni that also tend to infinity
in proportion with N. The finite ratio ni /N then represents the fraction of the total p+
momentum (which we will normalize to ptot + = 1) carried by the corresponding string
§

(i) ni
p+ = . (14)
N

So, in the above terminology, only long strings survive. The usual oscillation states of
these strings are generated in the orbifold CFT by creation modes α−k/N with k finite.
Therefore, in the large N limit only the very low-energy IR excitations of the SN -orbifold
CFT correspond to string states at finite mass levels.
§
The fact that the length ni of the individual strings specifies its light-cone momentum is familiar
from the usual light-cone formulation of string theory [17].

6
 The Zn projection discussed above in this limit effectively amounts to the usual un-
(i) (i)
compactified level-matching conditions L0 − L0 = 0 for the individual strings, since all
single string states for which L0 − L0 6= 0 become infinitely massive at large N. The total
mass-shell condition reads (here we put the string tension equal to α′ = 1)

ptot tot
− = NL0 (15)

and we recover the mass-shell conditions of the individual strings by decomposing Ltot
0 as
in (13), via the definition of the individual p− light-cone momentum as

(i)
(i) NL0
p− = (16)
ni

(i) (i) (i)

which combined with (14) gives the usual relation p+ p− = L0 . All strings therefore
indeed have the same string tension.

Interactions
We have seen that the super-Yang-Mills model in the IR limit gives the free light-cone
quantization of the type IIA string in terms of an orbifold sigma model, that describes
the completely broken phase of the U(N) gauge symmetry. The twisted sectors of the
orbifold describe multi-string states and are superselection sectors in the non-interacting
model — the number of strings in conserved for gs = 0.
The conservation of string number and of individual string momenta will be violated
if we turn on the interactions of the 1+1-dimensional SYM theory. The idea is that by
relaxing the strict IR limit, one gradually needs to include configurations in which the
non-abelian symmetry gets restored in some small space-time region. Indeed, if at some
point in the (σ, τ ) plane two eigenvalues xI and xJ coincide, we enter a phase where an
unbroken U(2) symmetry is restored, and we should thus expect that for non-zero gs ,
there will be a non-zero transition amplitude between states that are related by a simple
transposition of these two eigenvalues¶ . In the IR SN orbifold theory, such a process will
correspond to a local interaction on the two-dimensional world-sheet, which can be seen
to correspond to the elementary joining and splitting of strings.
To see this more explicitly, consider a configuration that connects two different sectors,
labeled by SN group elements that are related by multiplication by a simple transposition.
It is easy to see that this simple transposition connects a state with, say, one string rep-
resented by a cycle (n) decays into a state with two strings represented by a permutation
that is a product of two cycles (n1 )(n2 ) with n1 + n2 = n, or vice versa. So the numbers of
incoming and outgoing strings differ by one. Pictorially, what takes place is that the two
coinciding eigenvalues connect or disconnect at the intersection point, and as illustrated
in fig. 3. this indeed represents an elementary string interaction.
¶
This observation was first made in [11].

7
 Fig. 3: The splitting and joining of strings occurs if two
eigenvalues coincide.

In the CFT this interaction is represented by a local operator, and thus according to
this physical picture one may view the SYM theory as obtained via a perturbation of
the SN -orbifold conformal field theory. In first order, this perturbation is described via a
modification of the CFT action
Z
S = SCF T + λ d2 zVint (17)

where Vint is an appropriate (e.g. space-time supersymmetric) twist operator, that gener-
ates the just described simple transposition of eigenvalues. We now claim that this joining
and splitting process is indeed first order in the coupling constant gs as defined in the SYM
Lagrangian (1). Instead of deriving this directly in the strongly coupled SYM theory, we
analyze the effective operator that produces such an interaction in the IR conformal field
theory. The identification of λ in (17) in terms of the coupling constants in (1) via

√
λ ∼ gs α ′ (18)

requires that this local interaction vertex Vint must have scale dimension ( 32 , 23 ) under
scale transformations on the two-dimensional world-sheet. We will now verify that this is
indeed the case.

The interaction vertex

It is clear from the above discussion that the interaction vertex Vint will be a twist field
that interchanges two eigenvalues, say x1 and x2 . It acts therefore as a Z2 reflection on
the relative coordinate x1 − x2 . This coordinate has 8 components, so the total conformal
dimension of the corresponding superconformal field (including the fermionic contribution)
is (1, 1). The corresponding descendent operator that one can include in the action has
dimensions ( 32 , 32 ). Note that this is an irrelevant operator, that disappears in the IR
limit. In fact, the corresponding coupling constant has dimension −1. Since the world-
sheet length-scale is inversely proportional to the string coupling gs , we immediately see

8
that this interaction linear in gs , which is what we set out to establish. In fact, we will
show that the twist field is uniquely characterized as the least irrelevant operator in the
CFT that is both space-time supersymmetric and Lorentz invariant.
Let us do this computation in a bit more detail. First we recall that in the CFT
description of the second-quantized GS light-cone string the spacetime supercharges are
given by (in the normalization ptot
+ = 1)

α 1 I XN √ I
Q =√ dσ θIα , Q =α̇
N dσ Gα̇ , (19)
N I=1

with
N
X
Gα̇ (z) = γαi α̇ θIa ∂xiI (20)
I=1

the generators of the total N = 8 world-sheet supersymmetry algebra, with similarly

defined right-moving charges.
Concentrating on the left-moving sector, let us analyze the twist fields of the Z2 twist
that interchanges two eigenvalues, say x1 with x2 and similarly θ1 and θ2 . One first goes
over to the eigenvectors x± = x1 ±x2 , θ± = θ1 ±θ2 . The Z2 acts on the minus components
as
xi− , θ−
α α
→ −xi− , −θ− , (21)
so we are essentially dealing with a standard R8 /Z2 supersymmetric orbifold.
This orbifold has well-known twist fields, which will receive contributions of the bosons
and fermions. The bosonic twist operator σ is defined via the operator product

1
xi− (z) · σ(0) ∼ z − 2 τ i (0) (22)

Since the fermions transform in spinor representation 8s , their spin fields Σi , Σα̇ will
transform in the vector representation 8v and the conjugated spinor representation 8c .
They are related via the operator products
1
α
θ− (z) · Σi (0) ∼ z − 2 γαi α̇ Σα̇ (0)
1
α
θ− (z) · Σα̇ (0) ∼ z − 2 γαi α̇ Σi (0) (23)

The bosonic twist field σ and the spin fields Σi or Σα̇ have all conformal dimension h = 12
1
(i.e. 16 for each coordinate), and the conjugated field τ i in (22) has dimension h = 1.
For the interaction vertex we propose the following space-time supersymmetric, SO(8)
invariant, weight 23 field
τ i Σi (24)

9
This twist field lies in the NS sector of the CFT, and represents an interaction between
incoming and outgoing Ramond states.k It can be written as the descendent of the chiral
primary field σΣα̇ , since (no summation over α̇)

[Gα̇− 1 , σΣα̇ ] = τ i Σi . (25)

2

This is a special case of the more general identity

[Gα̇− 1 , σΣβ̇ ] + [Gβ̇− 1 , σΣα̇ ] = δ α̇β̇ τ i Σi . (26)

2 2

The interaction vertex operator satisfies

[Gα̇− 1 , τ i Σi ] = ∂z (σΣα̇ ), (27)

2

as is clear by using the Jacobi identity and the above relation. For the full description we
also have to include the right-moving degrees of freedom.
To obtain the complete form of the effective world-sheet interaction term we have to
tensor the left-moving and right-moving twist fields and to sum over the pairs of I, J
labeling the two possible eigenvalues that can be permuted by the Z2 twist

X Z  
j
λ d2z τ i Σi ⊗ τ j Σ . (28)
IJ
I<J

This is a weigth ( 23 , 23 ) conformal field. The corresponding coupling constant λ has there-
fore total dimension −1 and the interaction will scale linear in gs just as needed, see eqn
(18).
The interaction is space-time supersymmetric. First, it preserves the world-sheet N =
8 supersymmetry representing the unbroken charges Qα̇ , since supersymmetric variations
become total derivatives and we integrate over the string world-sheet. The broken space-
time supersymmetries Qα are trivially conserved. Although the fermion fields θ− α
satisfy
Ramond boundary conditions, and therefore pick up a minus sign in the local coordinate
z when transported around z = 0, Qα is proportional to the zero-mode of the linear sum
α
θ+ , which is not broken by the twist field interaction.
It is interesting to compare the above twist field interaction with the conventional
formalism of light-cone string theory∗∗ . As is discussed by Mandelstam [18], in order to
obtain a fully SO(9, 1) Lorentz invariant interaction in the light-cone formalism, it does
k
Even though the fermion has Ramond spin structure, the supercurrent Gα̇ obeys NS periodicity.
Using N = 2 language the chiral primary fields are σΣi and σΣα̇ . After spectral flow they give Ramond
ground states.
∗∗
We thanks N. Berkovits for pointing out this relation to us.

10
not suffice to consider only the geometric joining and splitting interaction that in our
formalism is represented by the insertion of the twist field σ(0). The interaction has to
be supplemented with a further operator insertion that reads in our notation
I
dz
1 Σi ∂xi (z) σ(0) = τ i Σi (0). (29)
z 2

In comparison with the Riemann surfaces picture of interacting strings [18] one should
note that around the interaction point the usual string world-sheet is actually a double
cover of the local coordinate z of our orbifold CFT. The above operator insertion should be
tensored with a similar expression for the right-movers, giving precisely our interaction
vertex (28). We further remark that the current Σi ∂xi has a simple interpretation as
the N = 1 world-sheet supercurrent in the covariant NSR formulation of the type II
string. Indeed, in the NSR language the above operator insertion corresponds simply
to picture changing. The fact that the twist field interactions (and the higher n-point
vertices obtained through contact terms) respect the ten-dimensional Lorentz invariance
is of course highly suggestive that the matrix string will also be Lorentz invariant.

Hamiltonian formulation
It will be convenient in the following to represent the degrees of freedom in a Hamilto-
nian form, by introducing the standard conjugate variables (Πi , X i ) for the scalar fields,
and (E, A1 ), with E the 1-dimensional electric field, for the gauge fields. The Hamiltonian
H and total 1-momentum P then take the form
I I
H= dσ T00 P = dσ T01 (30)

where T00 and T01 are the components of the two-dimensional energy-momentum tensor,
given by
1  2
T00 = tr Πi + (DXi )2 + θT γ 9 Dθ
2
1 1 
+ θT γ i [Xi , θ] + 2 (E 2 + [X i , X j ]2 )
gs gs
 
T01 = tr Πi DX i + θT Dθ (31)

Here D is the YM covariant derivative along the spatial direction of the world-sheet and
the γ-matrices are taken to be 16×16 matrices. The spacetime supercharges of the matrix
string model take the form (now α denotes a non-chiral spinor and runs from 1 to 16)

1
I √ I
Q̃α = √ dσ trθα , Qα = N dσ Gα , (32)
N

11
with  α
Gα = tr θT (γ 9 E + γ 9i DXi + γ i Πi + γ ij [Xi , Xj ]) (33)

Compactification
Following the current approach in matrix theory, the compactification of the matrix
string on a torus T d is achieved by reinterpreting the infinite dimensional matrices X i
as covariant derivatives Di (written in a Fourier mode basis) of a large N gauge field
defined in these extra dimensions [3, 7]. Adopting this procedure, the matrix string
becomes equivalent to d + 2-dimensional supersymmetric gauge field theory on the space-
like manifold S 1 × T d . In this correspondence, the string coupling constant gs is identified
with square root of the volume of the torus T d (in string units), and thus the free string
limit amounts to shrinking the volume of the extra dimensions inside the T d to zero. In
general, like for the uncompactified theory, the large N limit of the gauge model needs
to be accompanied by an appropriate IR-limit in the S 1 direction to find correspondence
with the string theoretic degrees of freedom.
For the case d = 2, in which case the large N supersymmetric gauge theory lives in
3+1-dimensions, this exact rescaling was in fact considered previously in [19], see also
[16].†† As also pointed out in [20, 21], the implied equivalence of large N four-dimensional
SYM theory to type II string theory compactified on T 2 provides a natural explanation
of the SL(2, Z) S-duality symmetry of the former in terms of the T -duality of the latter.
In the gs → 0 rescaling limit of the gauge theory, this connection between S-duality and
T -duality of the resulting SN -orbifold CFT was first pointed out in [16].
In general, the maximal number of dimensions one can compactify in this fashion
is eight. In this case matrix theory becomes equivalent to 9 + 1 dimensional large N
SYM theory, whose IR behaviour should describe the type II string compactified to 1+1
dimensions. In principle one should be able to compactify one more dimension by taking
the light-cone to be a cylinder. The finite N theory may seem to be a candidate for
describing the sector of this compactification with given discrete momentum N along the
extra S 1 . Although this procedure is adequate for computing the BPS spectrum of the
theory, we suspects it gives an incomplete description of the dynamics of general non-BPS
configurations.

Fluxes and Charges

The above compactification procedure opens up the possibility of adding new charged
objects to the theory, essentially by considering gauge theory configurations that carry
non-trivial fluxes or other topological quantum numbers along the compactified directions.
††
The weak coupling limit gs → 0 was interpreted in [19] as a high energy limit, in which the typical
length scale of excitations in two of the four directions is taken to be much smaller than in the other two
transversal directions. It seems worthwhile to further elaborate this correspondence.

12
In particular, one can consider configurations with non-zero magnetic fluxes through the
various two-cycles of the compactification torus T d , and in this way one can introduce the
D-membranes. More generally, using the correspondence with the matrix theory proposal
of [3, 22], a (partially complete) list of fluxes and their type IIA interpretations are listed
below

2D SYM Theory type IIA String

I
tr E = D-particle # q0
I
tr Πi = momentum pi
I
tr DX i = NS winding wi
I
T01 = NS winding w+
I
tr [X i , X j ] = D-membrane # mij
I
T0i = D-membrane # mi+
I
tr DX [iX j X k] = D-fourbrane # Rijk+
I
tr X [iX j X k X l] = NS fivebrane # Wijkl+

Here T01 was defined in equation (31) and

 
T0i = tr EDXi + Πj [Xi , Xj ] + θT [Xi , θ]

which in the d + 2 dimensional SYM language is just the momentum flux in the i-th
direction.
The above topological charges all appear as central terms in the supersymmetry alge-
bra generated by the supercharges defined in (32). One finds [22]

{Q̃α , Q̃β } = δ αβ
{Qα , Q̃β } = (γ 9 q0 + γ i pi + γ 9i wi + γ ij mij )αβ (34)
{Qα , Qβ } = N(H + γ 9 w+ + γ i mi+ + γ 9ijk Rijk+ + γ ijkl Wijkl+ )αβ

Here we recognize the central terms corresponding to the various charged (extended)
objects that are present in the theory. In the last line, the complete right-hand side is
proportional to N, and thus these terms diverge for N → ∞ as soon as one of the central

13
charges mi+ , Rijk+ or Wijkl+ is non-zero. This corresponds to the fact that the 9th
direction is necessarily decompactified in the large N limit, and thus these configurations
represent string-like solitons with an infinite extent in the 9th direction.

D-particles
It is not immediately evident that the string, D-brane and fivebrane configurations
as defined above will indeed behave exactly as in perturbative string theory. Here we
will establish this correspondence for the case of the D-particle and the D-membrane. In
particular, we will see that they indeed give rise to a new perturbative sector of strings,
that satisfy Dirichlet boundary conditions on a corresponding codimension subspace.
First we consider a configuration with D-particle charge equal to q0 . In the SYM
language, this corresponds to a non-zero electric flux.

I
1
dσ trE = q0 . (35)
2π

This correspondence can be understood in (at least) two ways. First, the electric flux arises
upon compactification and T-duality as the KK-momentum in the extra 9th dimension.
Since this direction was used to compactify M-theory to ten dimensions, this momentum
gets interpreted as D-particle charge [1]. Alternatively, as indicated in fig. 1, after a
T-duality and an S-duality, we can map our IIA strings to the D-strings of the type
IIB theory. As shown in [5], fundamental type IIB strings attach to these D-strings by
creating a non-zero electric flux on the SYM theory that described the D-string world-sheet
dynamics. Inverting the above duality transformations, these fundamental IIB strings get
identified with the D-particles in the IIA setup.
The simplest classical configuration that carries such an electric flux q0 is

E = 1q0 ×q0 . (36)

The presence of the electric flux will break the gauge group as

U(N) → U(N −q0 ) × U(q0 ). (37)

The U(N − q0 ) sector, that does not carry an electric flux, represents the type IIA strings
in the background of the D-particles. In the large N, gs → 0 limit this gives the usual free
closed string spectrum. The U(q0 ) sector will describe the D-particle degrees of freedom
and we will now examine it in more detail.
We have a similar breaking of the permutation group symmetry from SN to SN −q0 ×Sq0 .
The symmetric group Sq0 describes the statistics of the D-particles. The Hilbert subspace
that carries the electric flux q0 will decompose (at least for weak string coupling) in twisted

14
 D-particle bound state

single D-particle

type IIA string

Fig. 4: If we put a flux in a finite rank subgroup of

the SYM gauge group, the corresponding eigenvalues give
“short strings” that represent the D-particles and their bound
states. The “long strings” in the remaining sector describe
the closed type IIA strings in the background of the D-
particles.

sectors labeled by partitions of q0 . These sectors have an interpretation as all the possible
bound state configurations of the q0 D-particles.
The eigenvalues of the U(q0 ) part of the matrices X i can depend on the world-sheet
parameter σ and thus a priori seem to describe strings. However, we would like to inter-
prete them as D-particles. An important point is that in the large N limit we have to
keep the total D-particle charge q0 finite. This implies that the strings in the U(q0 ) sector
become short strings with infinitely massive oscillations. These short string oscillations
will therefore decouple at large N, leaving only their constant modes. These constant
modes describe the positions of the D-particles and their various bound states. This be-
haviour should be contrasted with the eigenvalues in the remaining U(N−q0 ) sector, which
can form the type IIA strings with the usual oscillation spectrum. Typical configurations
thus consist of short strings describing the D-particles and long type IIA closed strings,
as depicted in fig. 4.
It is worthwhile to note that here we naturally arrive at the existence of D-particle
bound states, since these automatically arise as twisted sectors in the orbifold conformal
field theory. So unlike for the original matrix model based on large N supersymmetric
quantum mechanics, the existence of bound states is not an assumption but a direct
consequence of well-established facts!
An electric flux of the form (36) will give a contribution H = q0 /gs2 to the SYM
Hamiltonian. In the untwisted sector, that describes q0 free D-particles, the usual mass-
shell relation gives D-particle masses m = 1/gs , which is the expected result. Similarly,

15
 D-particle

type IIA string

Fig. 5: In perturbation theory the closed string can get glued

to the D-particle.

P
in a twisted sector that describes bound states of ni D-particles, with ni = q0 , we find
that the bound states have masses mi = ni /gs . This is implied by the mass-shell relation,
quite similarly as in our discussion of the free strings, since the contribution to the total
L0 is of the form
q0 X p2i
Ltot
0 = + (38)
gs2 i ni

In particular, for a maximally twisted sector with one cycle of length q0 we find H =
q0 /gs2 + p2 /q0 which implies m = q0 /gs .
It should be emphasized that, as seen from eqn (38), the D-particles states just de-
scribed in fact have very small momentum in the light-cone direction. Formally, they
have p+ = nNi → 0 in the large N limit. It is clear, however, that one can give non-zero
longitudinal momentum to the D-particles by attaching them (i.e. the short strings) to a
long string.
Do these particles indeed interact with the type II strings as D-objects? In particular,
we would like to see that the theory contains a sector of long strings, which attached to
these particles in the usual fashion. The natural candidate for such configurations are
twisted sectors in which (in spite of the symmetry breaking due to the presence of the
electric flux) two different kind of eigenvalues are transposed, corresponding to respectively
the long type IIA strings and the short string that represent the D-particles. In other
words, the topology of the eigenvalues of the X i fields in this sector is incompatible with
topology of the eigenvalues of the (non-abelian) electric field E, in that the direction of the
E field in the Lie algebra must necessarily point outside of the Cartan subalgebra specified
by the X i fields (and vice versa). In the IR limit this would however be energetically
unfavorable, unless the short and long eigenvalue (that are transposed along the S 1 )
coincide. Where this happens the unbroken gauge group gets locally enhanced to U(2),
and this allows the eigenvalues to cross without much loss of energy, see fig. 5. In this way,
the long string eigenvalue indeed stays glued to the short string eigenvalue representing
the D-particle.

16
 The ground state of such a bound state between a long and a short string represents
the D-particle moving with non-zero p+ . A quantitative verification of this picture is as
follows. A long string of total length n occupies a U(n) subgroup. If the string is in its
ground state, all xI essentially coincide and thus the U(n) gauge symmetry is unbroken.
If the configuration further carries an electric flux q0 , this flux can lower its worldsheet
energy by spreading into this unbroken gauge group‡‡ . The flux then takes the form of
a flux qn0 inside the diagonal U(1), together with an opposite ’t Hooft type electric flux
q0
n
inside the SU(n). The energy of this flux is now smaller by a factor of n, in exact
accordance with the fact it now corresponds to a D-particle with finite p+ = n/N.
These arguments are admittedly somewhat qualitative. It would be useful to perform
a more quantitative study that supports the presented physical picture.

D-membranes
D-membranes are configurations with non-zero values for the topological charge

mij = tr[Xi , Xj ]. (39)

Let us briefly recall the construction of such configurations in the matrix model of M-
theory and the resulting correspondence with the membrane world-volume theory. In
U(N) one can find two matrices U and V , such that
2πi
UV = e N V U.
Any hermitian matrix X can then be written as
X
X= xnm U n V m
nm

and this expansion can be used to associate to X the function

X
x(p, q) = xnm e2πi(np+mq)
n,m

The two coordinates (p, q) then become identified with the membrane surface. After im-
plementing this transformation within the Hamiltonian of the large N SQM, one recovers
the light-cone gauge membrane world-volume Hamiltonian [23, 24, 3]. In the matrix
theory philosophy, this membrane configuration only needs to use a part of the total
U(N) gauge group. In particular, if we consider a sector of the theory with a membrane
configuration in the 7-8 direction, we can decompose the total U(N) matrix as
X 7 = p + x7 (p, q) + X̃ 7
X 8 = q + x8 (p, q) + X̃ 8
X i = xi (p, q) + X̃ i
‡‡
We thank T. Banks for suggesting this picture.

17
Here the first two terms describe the classical membrane and its transverse fluctuations,
while X̃ i denotes the remaining part of X i that commutes with the fluctuating membrane
background.
We can now copy the same procedure in the matrix string theory. In this case the fields
X becomes σ dependent, and thus at first sight the membrane coordinate fields x(p, q)
also acquire this additional dependence. However, just as for the D-particles, the presence
of the membrane configuration breaks the total U(N) gauge symmetry to a subgroup. In
particular, this means the part of the eigenvalues that are occupied by the membrane
can no longer be permuted. The p, q dependence does not allow such an action. So, in a
similar fashion as in the case of the D-particle these eigenvalues necessarily correspond to
short strings that can only by in their ground states. Therefore the functions xi (p, q) do
not acquire a σ-dependency.
One should be able to analyze the interactions of the type IIA strings with the D-
membrane quite explicitly along the lines outlined in the discussion of the D-particles.
Again we expect that the “long” strings can now attach to “short” strings that constitute
the membrane, analogous to the way depicted in fig. 5. However, in this case both end of
the open string should be able to travel independently on the D-membrane world-volume.
We leave a more precise analysis of these interactions for further study.

Acknowledgements
We would like to thank S. Bais, T. Banks, N. Berkovits, S. Das, F. Hacquebord, L.
Motl, J-S. Park, W. Taylor, and B. de Wit for useful discussions. This research is partly
supported by a Pionier Fellowship of NWO, a Fellowship of the Royal Dutch Academy of
Sciences (K.N.A.W.), the Packard Foundation and the A.P. Sloan Foundation.

Appendix
In this appendix we derive the matrix string Hamiltonian starting from the matrix
theory formalism of [3]. In particular we wish to identify the precise dependence on the
string coupling constant. For a similar derivation see [11].
Starting point is the matrix theory Hamiltonian, written in eleven-dimensional Planck-
ian units with ℓp = 1 (we ignore numerical prefactors)
 
H = R11 tr Π2i + [X i , X j ]2 + θT γi[X i , θ] . (A.1)

We now compactify the 9th dimension on a circle of radious R9 . After the usual T-
duality we can identify X 9 with the covariant derivative R9 Dσ , where the coordinate σ
runs from 0 to 2π. The conjugate momentum will be identified with the electric field E
via E = R9 Π9 . This gives the Hamiltonian (where i = 1, . . . , 8 now labels the transverse

18
coordinates)
Z
R11 dσ  2
H = tr Πi + R92 (DXi )2 + R9 θT Dθ
2π R9
1 2 i j 2 T i

+ E + [X , X ] + θ γ i [X , θ] . (A.2)
R92
−1/2
One can rescale the coordinates as X i → R9 X i to find
Z
R11 
H = dσ tr Π2i + (DX i)2 + θT Dθ
2π
1 2 i j 2 1 T i

+ (E + [X , X ] ) + θ γ i [X , θ] . (A.3)
R93 3/2
R9
Conventionally, M-theory is related to type IIA string theory via the compactification of
3/2
the 11th direction, which relates the string coupling constant gs to R11 . To arrive at the
matrix string point of view, however, we now √ interchange the role of the 9th and the 11th
direction by defining the string scale ℓs = α′ and string coupling constant gs in terms
of R9 and the 11-dimensional Planck length ℓp

R9 = gs ℓs , ℓp = gs1/3 ℓs , (A.4)

or equivalently gs = (R9 /ℓp )3/2 . From this we obtain the final result in string units ℓs = 1
R11 Z 
H = dσ tr Π2i + (DX i )2 + θT Dθ
2π
1 1 T 
+ 2 (E 2 + [X i , X j ]2 ) + θ γi[X i , θ] . (A.5)
gs gs
In this convention, R11 normalizes the light-cone momentum p+ via

p+ = N/R11 (A.6)

and becomes infinite in the large N limit. The normalization chosen in the main text
corresponds to total light-cone momentum p+ = 1, so that R11 = N and the mass-shell
relation reads p− = H. In addition we have absorbed this factor of N into the definition
of the world-sheet time coordinate, so that the above Hamiltonian H is related to the
world-sheet time generator H = L0 + L0 of the CFT as H = NH.

References
[1] E. Witten, “String Theory Dynamics in Various Dimensions,” Nucl. Phys. B443
(1995) 85–126, hep-th/9503124.

19
 [2] J. Schwarz, “The Power of M-theory,” Phys. Lett. 367B (1996) 97, hep-th/9510086.

[3] T. Banks, W. Fischler, S. H. Shenker, and L. Susskind, “M Theory as a Matrix

Model: A Conjecture,” hep-th/9610043.

[4] J. Polchinski, “ Dirichlet-Branes and Ramond-Ramond Charges,” hep-th/951001.

[5] E. Witten, “Bound States of Strings and p-Branes,” Nucl. Phys. B460 (1996) 335–
350, hep-th/9510135.

[6] U.H. Danielsson, G. Ferretti, and B. Sundborg, “D-particle Dynamics and Bound
States,” Int. J. Mod. Phys. A11 (1996) 5463–5478, hep-th/9603081;
D. Kabat and P. Pouliot, “A Comment on Zerobrane Quantum Mechanics,” Phys.
Rev. Lett. 77 (1996) 1004–1007, hep-th/9603127;
M. R. Douglas, D. Kabat, P. Pouliot, and S. H. Shenker, “D-branes and Short Dis-
tances in String Theory,” Nucl. Phys. B485 (1997) 85–127, hep-th/9608024.

[7] W. Taylor, “D-brane Field Theory on Compact Spaces,” hep-th/9611042.

[8] R. Dijkgraaf, E. Verlinde and H. Verlinde, “BPS Spectrum of the Five-Brane and
Black Hole Entropy,” Nucl. Phys. B B486 (1997) 77–88, hep-th/9603126.

[9] R. Dijkgraaf, G, Moore, E. Verlinde, H. Verlinde, “Elliptic Genera of Symmetric

Products and Second Quantized Strings,” Commun. Math. Phys. to be published,
hep-th/9608096.

[10] T. Banks and N. Seiberg, “Strings from Matrices,” hep-th/9702187.

[11] L. Motl, “Proposals on Nonperturbative Superstring Interactions,” hep-th/9701025.

[12] T. Banks, lecture at Jerusalem Winter School, Jan 5, 1997.

[13] S. Sethi and L. Susskind, “Rotational Invariance in the M(atrix) Formulation of Type
IIB Theory,” hep-th/9702101.

[14] N. Ishibashi, H. Kawai, Y. Kitazawa, A. Tsuchiya, “A Large-N Reduced Model as

Superstring,” hep-th/9612115.
M. Li, “Strings from IIB Matrices,” hep-th/9612222.

[15] P.-M. Ho and Y.-S. Wu, “IIB/M Duality and Longitudinal Membranes,”
hep-th/9703016.

[16] J.A. Harvey, G. Moore, and A. Strominger, “Reducing S-Duality to T -Duality,”

Phys.Rev. D52 (1995) 7161, hep-th/9501022.
M. Bershadsky, A. Johansen, V. Sadov, C. Vafa, “Topological Reduction of 4D SYM
to 2D σ–Model,” Nucl. Phys. B448 (1995) 166, hep-th/9501096.

20
[17] M.B. Green, J.H. Schwarz, E. Witten, Superstring Theory (Cambridge University
Press, 1987).

[18] S. Mandelstam, “Lorentz Properties of the Three-String Vertex,” Nucl. Phys. B83
(1974) 413–439; “Interacting-String Picture of the Fermionic String,” Prog. Theor.
Phys. Suppl. 86 (1986) 163–170.

[19] H. Verlinde and E. Verlinde, “QCD at High Energies and Two-Dimensional Field
Theory,” hep-th/9302104.

[20] L. Susskind, “T Duality in M(atrix) Theory and S Duality in Field Theory,”

hep-th/9611164.

[21] O. J. Ganor, S. Ramgoolam, and W. Taylor, “Branes, Fluxes and Duality in M(atrix)-
Theory,” hep-th/9611202.

[22] T. Banks, N. Seiberg, and S. Shenker, “Branes from Matrices,” hep-th/9612157.

[23] B. de Wit, J. Hoppe, and H. Nicolai, “On the Quantum Mechanics of Supermem-
branes,” Nucl. Phys. B305 (1988) 545.
D. Fairlie, P. Fletcher, and C. Zachos, Phys. Lett. 224B (1989) 101–107.

[24] P. K. Townsend, “The Eleven-Dimensional Supermembrane Revisited,” Phys. Lett.

B350 (1995) 184–187, hep-th/9501068.

21