A onesemester class about gauge/gravity duality (often called AdS/CFT) and its applications
Fall 2008
LECTURE NOTES
1. What you need to know about string theory 2. D3branes at small and large coupling 3. AdS/CFT from scratch, no strings 4. What holds up the throat? 5. How to defend yourself from a supersymmetric field theory 6. About the N = 4 SYM theory 7. Big picture of correspondence; strings and strong interactions 8. 't Hooft counting 9. Scale and conformal invariance in field theory 10. CFT in D > 2 (cont.) 11. Geometry of AdS 12. Geometry of AdS (cont.); Poincar patch; wave equation in AdS 13. Masses of fields and dimensions of operators; BFallowed tachyons 14. How to compute twopoint correlators of scalar operators 15. Preview of realtime issues; twopoint functions in momentum space (cont.); more on lowmass2 fields in AdS 16. Threepoint functions, anomalies; expectation values 17. Wilson loops 18. Wilson loops (cont.) 19. Pointlike probes of the bulk; Baryons and branes in AdS; 'Nonspherical horizons' 20. Brief survey of other examples of the correspondence (M2, M5, D1D5, Dp, branes at singularities); a model of confinement 21. Confinement (cont.): how to measure the spectrum, proof of mass gap 22. Black hole mechanics, classical and quantum 23. AdS black holes and thermal gauge theory; equation of state, free energy and stress tensor 24. Hawking effect for interacting field theories and BH thermodynamics 25. AdS black holes and thermal gauge theory: PolyakovSusskind loop, screening, quasinormal modes, Hawkingpage transition
8.821 Lecture 01: What you need to know about string theory
Lecturer: McGreevy Scribe: McGreevy
Since I learned that we would get to have this class, Ive been torn between a) starting over with string perturbation theory and b) continuing where we left o. Option a) is favored by some people who didnt take last years class, and by me when Im feeling like I should do more research. But because of a sneaky and perhaps surprising fact of nature and the history of science there is a way to do option b) which doesnt leave out the people who missed last years class (and is only not in the interest of the lazy version of me). The fact is this. It is actually rare that the structure of worldsheet perturbation theory is directly used in research in the subject that is called string theory. And further, one of the most important developments in this subject, which is usually called, synecdochically, the AdS/CFT correspondence, can be discussed without actually using this machinery. The most welldeveloped results involve only classical gravity and quantum eld theory. So heres my crazy plan: we will study the AdS/CFT correspondence and its applications and generalizations, without relying on string perturbation theory. Why should we do this? You may have heard that string theory promises to put an end once and for all to that pesky business of physical science. Maybe something like it unies particle physics and gravity and cooks your breakfast. Frankly, in this capacity, it is at best an idea machine at the moment. But this AdS/CFT correspondence, whereby the string theory under discussion lives not in the space in which our quantum elds are local, but in an auxiliary curved extradimensional space (like a soupedup fourier transform space), is where string theory comes the closest to physics. The reason: it oers otherwiseunavailable insight into stronglycoupled eld theories (examples of which: QCD in the infrared, hightemperature superconductors, cold atoms at unitarity), and into quantum gravity (questions about which include the blackhole information paradox and the resolution of singularities), and because through this correspondence, gauge theories provide a better description of string theory than the perturbative one. The role of string theory in our discussion will be like its role in the lives of practitioners of the subject: a source of power, a source of inspiration, a source of mystery and a source of vexation. The choice of subjects is motivated mainly by what I want to learn better. After describing how to do calculations using the correspondence, we will focus on physics at nite temperature. For what I think will happen after that, see the syllabus on the course webpage. Suggestions in the spirit described above are very welcome. 1
ADMINISTRIVIA please look at course homepage for announcements, syllabus, reading assignments please register I promise to try to go more slowly than last year. coursework: 1. psets. less work than last year. hand them in at lecture or at my oce.
pset 0 posted, due tomorrow (survey).
2. scribe notes. theres no textbook. this is a brilliant idea from quantum computing. method of
assigning scribe TBD. as you can see, I am writing the scribe notes for the rst lecture.
3. end of term project: a brief presentation (or short paper) summarizing a topic of interest. a list
of candidate topics will be posted.
goal: give some context, say what the crucial point is, say what the implications are.
try to save the rest of us from having to read the paper.
(benets: you will learn this subject much better, you will have a chance to practice giving a talk
in a friendly environment)
Next time, we will start from scratch, and motivate the shocking statement of AdS/CFT duality without reference to string theory. It will be useful, however, for you to have some big picture of the epistemological status of string theory. Todays lecture will contain an unusually high density of statements that I will not explain. I explained many of them last fall; experts please be patient. What you need to know about string theory for this class: 1) Its a quantum theory which at low energy and low curvature 1 reduces to general relativity coupled to some other elds plus calculable higherderivative corrections. 2) It contains Dbranes. These have YangMills theories living on them. We will now discuss these statements in just a little more detail.
0.1
Pick a background spacetime M, endowed with a metric which in some local coords looks like ds2 = g (x)dx dx ; there are some other elds to specify, too, but lets ignore them for now.
1
Consider the set of maps X : worldsheet target spacetime (, ) X (, ) where , are local coordinates on the worldsheet. Now try to compute the following kind of path integral I [DX(, )] exp (iSws [X..]). This is meant to be a proxy for a physical quantity like a scattering amplitude for two strings to go to two strings; the data about the external states are hidden in the measure, hence the . The subscript ws stands for worldsheet; more on the action below. An analog to keep rmly in mind is the rst quantized description of quantum eld theory. The FeynmanKac formula says that a transition amplitude takes the form x(2 )=x2 eiSwl [x( )] = x2 , 2 x1 , 1 (1)
x(1 )=x1
Here x : worldline target spacetime ( ) x ( ). can be written as a sum over trajectories interpolating between specied inital and nal states. Here x( ) is a map from the world line of the particle into the target space, which has some coordinates x . For example, for a massless charged scalar particle, propagating in a background spacetime with metric g and background (abelian) gauge eld A , the worldline action takes the form Swl [x] = d g (x)x x + d x A where x x. Note that the minimal coupling to the gauge eld can be written as A d x A =
wl
where the second expression is meant to indicate an integral of the oneform A over the image of the worldline in the target space (The expression on the LHS is often called the pullback to the worldline. Those are words.). A few comments about this worldline integral. 1) The spacetime gauge symmetry A A + d requires the worldline not to end, except at a place where weve explicitly stuck some nongaugeinvariant source (x1 , x2 in the expression above). 2) The path integral in (1) is that of a onedimensional QFT with scalar elds x ( ), and coupling constants determined by g , A . To understand this last statement, Taylor expand around some point in the target space, which lets call 0. Then for example A (x) = A (0) + x A (0) + ... and the coupling to the gauge eld becomes d (x A (0) + x x A (0) + ...) 3
so the Taylor coecients are literally coupling constants, and an innite number of them are encoded in the background elds g, A. After all this discussion of the worldline analog, and before we explain more about the , why might it be good to make this replacement of worldlines with worldsheets? Good Thing number 1: many particles from vibrational modes. From far away, meaning when probed at energies E Ms , they look like particles. This suggests that there might be some kind of unication, where all the particles (dierent spin, dierent quantum numbers) might be made up of one kind of string. To see other potential Good Things about this replacement of particles with strings, lets consider how interactions are included. In eld theory, we can include interactions by specifying initial and nal particle states (say, localized wavepackets), and summing over all ways of connecting them, e.g. by positionspace feynman diagrams. To do this, we need to include all possible interaction vertices, and integrate over the spacetime points where the interactions occur. UV divergences arise when these points collide. To do the same thing in string theory, we see that we can now draw smooth diagrams with no special points (see gure). For closed strings, this is just the sewing together of pants. A similar procedure can be done for open strings (though in fact we must also attach handles there). So, Good Thing Number Two is the fact that one string diagram corresponds to many feynman diagrams. Next lets ask Q: where is the interaction point? A: nowhere. Its in dierent places in dierent Lorentz frames. This is a rst (correct) indication that strings will have soft UV behavior (Good Thing Number Three). They are oppy at high energies. The crucial point: near any point in the worldsheet (contributing to some amplitude), it is just free propagation. This implies a certain uniqueness (Good Thing Number Four): once the free theory is specied, the interactions are determined. The structure of string perturbation theory is very rigid; it is very dicult to mess with it (e.g. by including nonlocal interactions) without destroying it. 2
BUT: Only for some choices of (M, ) can we even dene I. To see why, lets consider the simplest example where M = IRD1,1 with g = diag(1, 1, 1, 1...1). Then the worldsheet action is 1 d2 X X . (2) Sws [X] = 4 Here = 1, 2 is a worldsheet index and = 0...D 1 indexes the spacetime coords.
1 The quantity 4 has dimensions of length2 . It is the tension of the string, because it suppresses congurations with large gradients of X ; when it is large, the string is sti. This is the mass scale
2 Why stop at onedimensional objects? Why not move up to membranes? Its still true that the interaction points get smoothed out. But the divergences on the worldvolume are worse. d = 2 is a sort of compromise. In fact, some sense can be made of the case of 3dimensional base space, using the BFSS matrix theory. They turn out to be already 2d quantized! The spectrum is a continuum, corresponding to the degree of freedom of bubbling o parts of the worldvolume. Its not entirely clear to me what to make of this.
Crudely rendered perturbative interactions of closed strings. Time goes upwards. Slicing this picture by horizontal planes shows that these are diagrams which contribute to the path integral for 2 to 2 scattering. Tilting the horizontal plane corresponds to boosting your frame of reference and moves the point where you think the splitting and joining happened.
(2) is the action for free bosons. We can compute anything we want about them. The equations of motion 0 = Sws give the massless 2d wave equation: X 2 2 0 = + X = + X
To study closed strings, we can treat the spatial direction of the worldsheet as a circle + 2, in which case we can expand in modes: + X (, ) = ein + ein . n n
nZ Z nZ Z
For open strings, the boundary conditions will relate the rightmoving and leftmoving modes: . We can quantize this canonically, and nd [ , ] = nn+m , n m [ , ] = nn+m , [, = 0. n m ]
This says that we can think of n<0 as creation operators and n>0 as annihilation operators. Let 0 be annihilated by all the n>0 . The zeromodes 0 , 0 are related to the centerofmass position of the string. The worldsheet hamiltonian is and momentum p Hws = p2 + n n a,
n>0
Here comes the problem: consider the state BAD =0 0 for n > 0. Because of the commutator n [0 , 0 ] = n, we face the choice n n BAD
2
<0
or
EBAD < 0.
Neither is OK, especially since EBAD becomes arbitrarily negative as n gets larger. This kind of problem is familiar from QED or YangMills theory, where the time component of the gauge eld presents with very similar symptoms. We need to gauge some symmetry of the theory to decouple the negativenorm states. One way to do this is to gauge the 2d conformal group. Were going to talk a lot more about conformal symmetry soon, so Ill postpone even the denition. There are fancy motivations for this involving 2d gravity on the worldsheet that we are not going to talk about now. Of course, like any gauge symmetry, this symmetry is a fake, and one could imagine formulations of the theory wherein it is never introduced. One necessary condition arising from this is that the worldsheet QFT needs to be a conformal eld theory (CFT). This is not the CFT of the AdS/CFT correspondence. The generators of the conformal group include, in particular, the hamiltonian, so we must impose 0 = Hws phys as a constraint, like Gauss law in QED. Acting on a state with N oscillator excitations, this equation takes the form 0 = p2 + N 1 p2 + m2 , which looks like an onshell condition for a particle in minkowski space. This tells us the targetspace mass of a string state in terms of its worldsheet data. Now we can build physical states3 . The result looks like STRING STATE 1 1 0 1 0 1
{ [ } ]
MASS2
2 m2 = Ms
0
m2 = 0 m2 = 0
1 1 0
m2 = 0
2 m2 = nMs
11 ...n1 n 0 1 n
...
...
(where [..] is meant to indicate antisymmetrization of indices, {..} is meant to indicate symmetriza tion of indices). The scalar is called the dilaton. For open strings, we would have, among other states, STRING STATE 0 1 FIELD MASS2
vector A m2 = 0
3 Theres one important constraint I didnt mention, called levelmatching, which says that the number of right moving excitations N and the number of leftmoving excitations N should be equal. This follows from demanding that there be no special point on the worldsheet.
The worldsheet gauge symmetry leads directly to the spacetime gauge symmetries under which e.g.
A A + d(0) and the twoform B B dx dx B + d(1) . Theres a general theorem that a massless symmetric tensor like g which interacts must couple to stress energy, and hence is a graviton. Using the machinery built from the above picture, this mode can be seen to participate in nontrivial scattering amplitudes, which indeed reduce to those computed from GR at low energies. So string theory, however unwieldy this description, is a quantum theory of gravity.
0.2
Almost every string theory result goes through the spacetime eective action: Sst [g, B, ...] dD x ge2 R(D) + ()2 + H 2 + ... where H dB i.e.H [ B] , R(D) is the Ricci scalar made from g , and more about the dots below. This action is correct in two a priori dierent ways. 1) It reproduces scattering amplitudes of the uctuations g , B , ... computed by the world sheet path integral, at leading order in E 2 , where E is say the biggest energy of a state involved in the scattering event. That is to say, that the eective action is the leading term in a deriva tive expansion, where the higherdimension operators are suppressed by the appropriate number of powers of ; the coecients of the corrections are denite and computable. 2) To see the second way of thinking about Sst , consider a string propagating in a background of g , B ... Its action is 1 d2 g g (X) X X + B (X) X X + (X)R(2) + ... (3) Sws[X] = 4 Here, is an auxiliary metric on the worldsheet that we can introduce for convenience using the gauge symmetry. This is a 2d QFT with elds X . The bosons X are free is the spacetime is at, and the other elds are trivial. Otherwise, they interact, and the coupling constants are determined by the background elds, like on the worldline above. If the target space is nearly at, we can treat these interactions perturbatively. The eective , i.e. strength of interaction, is determined by R(D) . In general, these coupling constants will run with scale. A necessary condition for this QFT to be a CFT is that the beta functions for all the coupling constants vanish. It turns out that the leadingorder beta functions for g , B , are exactly the equations of motion arising from varying the spacetime eective action Sst !
We can learn some more simple lessons from the worldsheet action (3).
1) The worldsheet metric is innocuous4 . If you tried to take it seriously and give it some dynamics,
you would try to add the EinsteinHilbert term,
(2) R . Sws = But this term, which is already present, with coecient determined by the spacetime zeromode 0 of the dilaton eld , is topological it is the Euler characteristic of the worldsheet. On a worldsheet with h handles (genus h) and b boundaries, it evaluates to (2) R = () = 2 2h b. Letting gs ln 0 , this means that the contribution to the sum over worldsheets of worldsheets with h splittings and joinings is proportional to
2h2 gs .
Thus gs is the quantity which determines how much strings want to interact, the string coupling. It is the in spacetime. This means that the eective action above is also the leading term in an expansion in powers of gs . What I wrote above is the treelevel eective action. A more correct expression is 2 Sst [g, B, ...] dD x ge2 R(D) + ()2 + H 2 + ... 1 + O( ) 1 + O(gs ) 2) Recall that on the worldline with coupling d A x = A, gauge invariance under A A + d(0) implied that the worldline was not allowed to end randomly, and that the worldline was a monopole source for the gauge eld A. The coupling of the string to the AS tensor can similarly be written as a minimal coupling: dd X X B = B.
X()
The gauge invariance ((1) is an arbitrary, wellbehaved oneform in the target space) implies that the worldsheet cannot end just anywhere. More on this soon. It also means that fundamental strings are charged under the AS tensor, which is usually called the NSNS Beld, or KalbRamond eld. [To be continued] B B + d(1)
Recall:
Last time I was showing you how primitive is our description of string theory. Specically, we are
reduced to doing a path sum over worldsheet embeddings. To decouple timelike oscillators, we
gauged the worldsheet conformal group. I told you that this leads to a theory of quantum gravity,
which reduces to GR plus stu at low energies.
The spacetime eective action which summarizes the scattering amplitudes computed by the world
sheet path integrals is
2 Sst [g, B, ...] dD x ge2 R(D) + ()2 + H 2 + ... 1 + O( ) 1 + O(gs ) Here H = dB, meaning H = [ B] , and H H = H H To study other backgrounds (which are only weakly curved), we can try to study a worldsheet action in background elds (a nonlinear sigma model): 1 Sws[X] = d2 g g (X) X X + B (X) X X + (X)R(2) + ... (1) 4 A few observations about this which we made in the rst lecture: conformal invariance constrains form of BG elds to solve spacetime EOM the dilaton term implies that the contribution to the path sum of a worldsheets with more handles are weighted by more factors of gs = e Fstrings charged under Beld, and cannot end at some random point in spacetime. The specic objection to the ending of the string is the consistency of the Gauss law for the Beld. This arises from its equation of motion, which in the presence of a fundamental string looks like: 0= Sst = d H + D2 (string); B 1
H H +
string
with respect to B. Integrating this (D2)form over a (D2)ball containing the string, we get the integrated gauss law: H = Q1 ,
S D3
the number of strings inside the (D3)sphere (4 is equal to one in this discussion). If the string ended suddenly, the answer here would depend on which choice we made for the interior of the S D3 . problems: 1. bosonic string has a tachyon, a boson eld with mass2 < 0. this is a real instability 2. conformal anomaly: under a scale transformation on a curved worldsheet, D 26 (2) R Lws T 2 3. perturbation expansion does not converge. worse than QFT. e1/gs . Problem 2 is solved by string compactication: just study backgrounds where the dimensions you dont like are too small to see. This is an interesting thing that were not going to talk about. To x problem 1: superstrings. The RNS description of superstrings is obtained from bosonic strings by changing the worldsheet gauge group to superconformal group no tachyon, still graviton, B, dilaton AND spacetime fermions (which exist) sometimes spacetime susy Dc = 10. optimal for susy (will explain later) The supersymmetric ones are the most interesting because we know their stable vacua. There
are ve avors of 10d supersymmetric superstrings: IIB, IIA, type I, het E8 , het SO.
There duality relations between all of them: they are descriptions of dierent probes of dierent
states of the same theory.
We will focus on the type II theories.
They are very closely related to each other. At low energies, they reduce to type II supergravity,
with 32 supercharges.
A few words about the lowenergy spectrum of type II. In addition to the elds which they share in common with the bosonic string, and the spacetime fermions, the type II strings (also type I has these) contain various antisymmetric tensor elds, called RamondRamond (RR) elds. They are like Maxwell elds and like the Beld. They dier from the Beld in that strings are neutral under the associated gauge symmetry. Type IIA has RR potentials of every odd degree (and evendegree eld strengths G = dC): C1 , C3 , C5 , C7 , C9 . They are related in pairs by EM duality (like F E = 4 F B of E&M in 4d): e.g. the 1form potential is the magnetic dual of the 7form potential: dC1 = 10 dC7 . 2
Type IIB has evendegree potentials, again related in pairs by EM duality. The middledegree
5form eld strength is selfdual.
(actually all of this is determined by supersymmetry.)
The simplest duality relation, namely Tduality, is worth mentioning here if only to convince you
that the two type II theories are not rival candidate theories of everything or something silly like
that.
[Tduality interlude:
compactication of string theory on a circle is described on the worldsheet by making some of the
elds periodic.
Duality of 2d eld theory relates strings on big circles to strings on small circles (in units of ).
The spectrum of strings on a circle is made up of a KK tower of momentum modes, and another
tower of winding modes. Tduality interchanges the interpretation of these two towers. The basic
idea is just that as the original circle shrinks, the winding modes of the strings start to look more
and more like the continuum momentum modes on the big dual circle.
In the type II superstring, this is accompanied by an interchange of the two avors IIA and IIB.
To relate the RR potentials: A RRpotential with an index along the circle being dualized loses
that index; one without an index along the circle gains an extra index. ]
(Dbranes) re problem 3: the few insights beyond perturbation come from the following set of observations. The eective action (derived by expanding around particular solutions like at space) can be used to nd other solutions. If these solutions have small curvatures and eld gradients (in string units) and if the dilaton is uniformly small, we can trust the leadingorder eective action to tell us that it will really be a solution of string theory. In general, we dont know a worldsheet description of most such solutions; the ones where the RR uxes are nonzero are particularly intractable from the worldsheet point of view. One particularly interesting family of solutions are analogs of the ReissnerNordstrom black hole solution of 4d EinsteinMaxwell theory, but which carry RR charge. An object which couples minimally to a p + 1form potential (i.e. which carries the associated monopole moment) has a p + 1dimensional worldvolume p+1 , so that we can have a term in the action of the form Sst Cp+1 = d 1 ...d p+1 1 ..p+1 1 X 1 ...p+1 X p+1 C1 ...p+1 .
p+1
In at space, the stable congurations of such objects are just at innite pdimensional planes in spacetime. These objects have a nite tension, and hence these are not niteenergy excitations above the vacuum; they are dierent superselection sectors. Nevertheless, they exist (they can preserve half the supersymmetry of the vacuum) and are interesting as well see. If the space contains compact factors, we can imagine wrapping such objects on compact submanifolds and getting particlelike excitations. 3
1 The tension of these RR solitons is proportional to gs . They are heavy at weak coupling, (though 2 not as heavy as ordinary GR solitons, which have tension gs ) and are not made of a nite number of fundamental string excitations. This means that the euclidean versions of these objects can contribute the mysterious eects of order eSeucl e1/gs .
At weak coupling, it turns out that these objects have a remarkably simple description. We said earlier that the coupling to the NSNS Beld meant that string worldsheets cant end just anywhere. A Dpbrane is dened to be a (p + 1)dimensional locus where strings can end. The open strings which end there have tension and hence their light states are localized on the brane. These provide a string theory description of worldvolume degrees of freedom. We saw above that open string spectra contain vector elds; these worldvolume degrees of freedom very generally include a gauge eld propagating in p + 1 dimensions. The physics of this vector eld is gauge invariant. [It would be more interesting if it werent. This could be that this is just a result of the fact that we used the gauge principle in constructing the string theory.] So the leading term in the derivative expansion of the eective action for the gauge eld is 1 S = dp+1 x 2 tr F F + O( ) gY M Because this action comes from interactions involving worldsheets with one boundary, it should go 22hb = g 1 and therefore g 2 2 p3 ; like gs s Y M gs . In p + 1 dimensions, gY M has dimensions of length . these are made up by powers of Strings have two ends. Each end can be labelled by which brane it ends on. This means that the string states are N N matrices, where N is the number of Dbranes. The ends of the strings are charged particles (quarks and antiquarks). We should try to understand how the ending of the string is consistent with the Beld gauge symmetry. To see this, we use the fact that the end of the string is a charged particle to guess that the coupling of the worldsheet to a background worldvolume gauge eld should take the minimal form SA = A where is the boundary of the worldsheet. The variation of the bulk worldsheet action under the Beld gauge transformation B B + d is 1 1 Sws = d = . 4 4 We can cancel this if we allow A to vary by A A 4 . This means that only the combination F 4 F + B, where F = dA is invariant under the Beld gauge symmetry. Hence F must enter the eective action in this combination. To see how this story is modied by the presence of the brane, we need to think about the depen dence of the Dbrane action on the Beld. Given our previous statement that the Dbrane action contains L F F , it must actually be L F F. 4
B 8
Dpbrane
B8
string
This means that in the presence of a fundamental string ending on a Dbrane, the spacetime action for the NSNS Beld is 10 B F + O(B 2 ) d x H H + B+2 S[B] =
everywhere F1 Dp
The equation of motion is then 0 = d H + 8 (F 1) + 10(p+1) (Dp) F ; Integrating this over an 8ball B8 thats pierced by the string gives H = n1
S7
where S 7 = B8 is the boundary of the 8ball and n1 is the number of strings. If we integrate instead over B8 which is not pierced by the string, it must instead go through the Dbrane, as in the gure. Integrating the EOM for B over B8 gives H + F. 0=
S7
B8 Dp
The rst term here is the same as before, and is n1 . But this gives a consistent answer because S 7 Dp F is equal to the number of strings ending on the brane, by the gauss law for the worldvolume gauge eld. Basically, the worldvolume gauge ux carries the string charge away.
Note that IIA has Dbranes of these dimensions: D0, D2, D4, D6, D8 The D0brane is just a particle, charged under the RR vector eld; The D6 is the associated magnetic charge. These are interchanged under Tduality with the branes of type IIB. while IIB has 5
I forgot to say in lecture that these objects dened by new sectors of open strings can be shown
to be charged under the RR tensor elds (and in fact saturate the Dirac quantization condition on
the charge), and have tension 1/gs times the right power of to make up the dimensions. The
gravitational backreaction of a stack of N parallel such Dbranes is controlled by GN T gs N .
When this quantity is small, the description in terms of open strings is good; when this quantity
is large, the backreaction is important and the better description is in terms of the gravitating
soliton. More on this soon.
8.821 F2008 Lecture 03: The decoupling argument; AdS/CFT without string theory, a discovery with hindsight
Lecturer: McGreevy
Today: 1. Backreaction and decoupling 2. A bold assertion 3. Hints, lore, prophesy Recall from last time that: Dbranes are subspaces where open strings can end. Nomenclature: Dpbranes have p spatial dimensions. Light open strings are localized at the brane, and represent the uctuations of the brane. Dbranes carry RR charges (which saturate Dirac quantization condition; this is necessary since we have both electric and magnetic charges). This can be seen by computing the amplitude for Dbranes to emit RR gauge bosons; it is described by worldsheets as in the gure below. The world volume theory on the brane is a YM theory.
Dp
Figure 1: The disk amplitude for the emission of a closed string by a Dbrane.
The disk amplitude for the emission of a graviton gives the tension of the Dbrane and is proportional to 1 2+2h+b (1) gs = . gs 1
The backreaction on the geometry from N coincident Dbranes is determined from Einstein equa tions 1 brane 2 1 gs N = gs N = , (2) R Rg = 8GN T 2 gs where we used the fact that
2 GN closed string coupling gs .
(3)
At small , the backreaction is negligible and the physics is decribed by closed strings everywhere (these are excitations of empty space) plus open strings (YM + F 4 + ) localized on the branes (these are excitations of the branes) in at space. At large , the Dbranes will gravitationally collapse into an RR soliton (black hole with the same charge). For p > 0, the black hole is not a point particle, but is also extended in p spatial dimensions.
The case of p = 3 The 3brane will ll 3 + 1 dimensions (e.g., x=0,1,2,3 ). We can put the brane at y 1 = = y 6 = 0, 2 i.e, the R3,1 is at a point in the transverse R6 . Let r 2 = yi be the distance from the brane, then the metric of the brane will take the form
R6
ds2 = and
1 H(r)
6 2 dx dx + H(r) dyi ,
R3,1 i=1
(4)
L2 , dy 2 = dr 2 + r 2 d5 , r4 where H(r) 1 for large r and L is like an ADM mass. The RR F 5 satises F5 = N. H(r) = 1 +
S 5 at constant r
(5)
(6)
The solution looks like an RN black hole (Figure 2). Far away (large r with H(r) = 1), the solution will asymptote to R9,1 . Near the horizon (r L with H(r) = L2 /r 4 ), the metric will take the form of an AdS5 S5 / r2 r2 dr 2 (7) ds2 = 2 dx dx + L2 2 + L2 2 d5 . / L r r
AdS5 S5
The ultralow energy excitations near the brane cant escape the potential well (throat) and the stu from innity cant get in. The absorption cross section of the brane goes like 3 L8 . 2 (8)
R3,1 r
Raduis is constant L
Figure 2:
The above conclusion can be reached in another way by noting that for low , the wavelength of the excitations will be too large to t in the throat which has xed size. Comparison of low energy decoupling at large and small ( = N gs ) Large Throat states = IIB strings in AdS5 S5 + Closed strings in R9,1
related by variation of , i.e, dual
Matching of parameters The parameters of the gauge theory are g2 = gs (we will see that it is really a parameter in N = 4 YM YM) and the number of colors N . By Gausss law F5 = N (quantized by Dirac). (9) F5 =
S5 S5
The supergravity equations of motion relate L (size of space) and N (number of branes) as follows:
5 5... R = GN F... F ,
(10) (11)
where
2 GN gs 4 , 5 5... and F... F N 2 .
Dimensional analysis then says R L8 . This gives L4 = gs N = (t Hooft coupling). 2 In terms of gravitational parameters, this says
2 GN gs ( )4 ,
(12)
L = N 1/4 GN GN
1/2
1 N2
A picture of what has happened here which can sometimes be useful is the Picture of the Tents. We draw the distance from the branes as the vertical direction, and keep track only of the size of the longitudinal and transverse directions as a function of this radial coordinate. Flat space looks like this: (Hence, tents.) The picture of the nearhorizon limit looks instead like:
R r
6 3,1
R
3,1
3,1
Figure 3: Left: Flat space. Right: AdS5 S 5 . Now its the Minkowski space IR3,1 which shrinks at r = 0, and sphere stays nite size.
A Bold Assertion
Now we back up and try to understand what is being suggested here, without using string theory.
We will follow the interesting logic of:
Reference: HorowitzPolchinski, grqc/0602037
Assertion: Hidden inside any nonabelian gauge theory is a quantum theory of gravity.
What can this possibly mean??
Three hints from the Elders:
1) At the least it must mean that there is some spintwo graviton particle, that is somehow a
composite object made of gauge theory degrees of freedom. This statement seems to run afoul of
the WeinbergWitten nogo theorem, which says:
Theorem (WeinbergWitten): A QFT with a Poincar covariant conserved stress tensor T
e forbids massless particles of spin j > 1 which carry momentum (i.e. with P = dD xT 0 = 0).
GR gets around this because the total stress tensor (including the gravitational bit) vanishes by
S the metric EOM g = 0. (Alternatively, the matter stress tensor which doesnt vanish is not
generalcoordinate invariant.)
4
Like any good nogo theorem, it is best considered a sign pointing away from wrong directions. The loophole in this case is blindingly obvious in retrospect: the graviton neednt live in the same spacetime as the QFT. 2) Hint number two comes from the Holographic Principle [t Hooft, Susskind]. This is a farreaching consequence of black hole thermodynamics. The basic fact is that a black hole must be assigned an entropy proportional to the area of its horizon in planck units (much more later). On the other hand, dense matter will collapse into a black hole. The combination of these two observations leads to the following crazy thing: The maximum entropy in a region of space is its area in Planck units. To see this, suppose you have in a volume V (bounded by an area A) a conguration with entropy A S > SBH = 4GN (where SBH is the entropy of the biggest blackhole ttable in V ), but which has less energy. Then by throwing in more stu (as arbitrarily nonadiabatically as necessary, i.e. you can increase the entropy) you can make a black hole. This would violate the second law of thermodynamics, and you can use it to save the planet from the Humans. This probably means you cant do it, and instead we conclude that the black hole is the most entropic conguration of the theory in this volume. But its entropy goes like the area! This is much smaller than the entropy of a local quantum eld theory, even with some UV cuto, which would have a number of states Ns eV ( maximum entropy = ln Ns ) Indeed it is smaller (when the linear dimensions are large!) than any system with local degrees of freedom, such as a bunch of spins on a spacetime lattice. We conclude from this that a quantum theory of gravity must have a number of degrees of freedom which scales like that of a QFT in a smaller number of dimensions. This crazy thing is actually true, and AdS/CFT is a precise implementation of it. Actually, we already know some examples like this in low dimensions. One denition of a quantum gravity is a generallycovariant quantum theory. This means that observables (for example the eective action) are independent of the metric: 0= Sef f = T . g
We know two ways to accomplish this: 1) Integrate over all metrics. This is how GR works. 2) Dont ever introduce a metric. Such a thing is generally called a topological eld theory. The bestunderstood example is ChernSimons gauge theory in three dimensions, where the variable is a gauge eld and the action is SCS = tr A dA + ...
M
(where the dots is extra stu to make the nonabelian case gauge invariant); note that theres no metric anywhere here. With either option (1) or (2) there are no local observables. But if you put the theory on a space with boundary, there are local observables which live on the boundary. ChernSimons theory on some manifold M induces a WZW model (a 2d CFT) on the boundary of M . We will see that the same thing happens for more dynamical quantum gravities. 3) A beautiful hint as to the possible identity of the extra dimensions is this. Wilson taught us that a QFT is best thought of as being sliced up by length (or energy) scale, as a family of trajectories of the renormalization group (RG). A remarkable fact about this is that the RG equations for the 5
behavior of the coupling constants as a function of RG scale z is local in scale: zz g = (g(z)) the RHS depends only on physics at scale z. It is determined by the coupling constant evaluated at the scale z, and we dont need to know its behavior in the deep UV or IR. This fact is not completely independent of locality in spacetime. This opens the possibility that we can associate the extra dimensions raised by the Holographic idea with energy scale. Next we will make simplifying assumptions in an eort to nd concrete examples.
Today 1. Finish hindsight derivation 2. What holds up the throat? 3. Initial checks (counting of states) 4. next time: Supersymmetry selfdefense Previously, we had three hints for interpreting the bold Assertion that we started with, that hidden inside any nonAbelian gauge theory is a quantum theory of gravity.
1. The WeinbergWitten theorem suggests that the graviton lives on a dierent space than the Quantum Field Theory. 2. This comes from the holographic principle. The idea, motivated by black hole thermodynam ics, is that the theory of gravity should have a number of degrees of freedom that grows more slowly than the volume (nonextensive) suggesting that the quantum gravity should live is some extra (more) dimensions than the QFT. 3. The structure of the Renormalization Group suggests that we can identify one of these extra dimensions as the RGscale.
In order to make these statements a little more concrete, we will make a number of simplifying assumptions. That means we will not start by writing down the more general case of the corre spondence, but we will start with a special case.
1. There are many colors in our nonAbelian gauge theory. The motivation for this is best given by the quote:
You can hide a lot in a largeN matrix.
1
Shenker The idea is that at large N the QFT has many degrees of freedom, thus corresponding to the limit where the extra dimensions will be macroscopic. 2. We will work in the limit of strong coupling. The motivation for that lies in the fact that we know some things about weakly coupled Field Theories, and they dont seem like Quantum Gravity. This constrains the manner in which we take N to be large, since we would still like the system to be interacting. This means that we dont want vector models. We want the theory to have as much symmetry as possible. However there is a theorem that stands in the way of this.
Theorem (ColemanMandula) : All bosonic symmetries of a sensible Smatrix (nontrivial,
having nite matrix elements) belong to the Poincare group.
We are going to use all possible loopholes to circumvent this theorem. The rst lies in the word bosonic and the second one in sensible. 3. SuperSymmetry (SUSY) (a) It constrains the form of the interactions. This means that there are fewer candidates for the dual. (b) Supersymmetric theories have more adiabatic invariants, meaning observables that are independent of the coupling. So there are more ways to check the duality. (c) It controls the strongcoupling behavior. The argument is that in nonSUSY theories, if one takes the strongcoupling limit, it tends not to exist. The examples include QED (where one hits the landau Pole for strong couplings) and the Thirring model, which is exactly solvable, yet doesnt exist above some coupling. (d) Nima says its simpler (http://arxiv.org/abs/0808.1446). The most supersymmetric theory in d=4 is the N = 4 SYM and well be focusing on this theory for a little while. It has the strange property that, unlike other gauge theories, the beta function of the gauge coupling is exactly zero. This means that the gauge coupling is really a dimensionless parameter, and so, reason number (e) to like SUSY is... (e) SUSY allows a line of xed points. This means that there is a dimensionless parameter which interpolates between weak and strong coupling. 4. Conformal Invariance: The fact the coupling constant is dimensionless (even quantummechanically) says that there is scale invariance. The Smatrix is not nite! Everything is soft gluons.
Scale invariance applies to both space and time, so X X ,where = 0, 1, 2, 3. As we
said, the extra dimension coordinate is to be thought of as an energy scale. Dimensional anal
z ysis suggests that this will scale under the scale transformation, so z . The most general
vedimensional metric (one extra dimension) with this symmetry and Poincare invariance is of the following form:
z ds2 = ( L )2 dx dx + dz 2 2 L z2
. Let z =
z familiar form (by change of coordinates) = ( L )2 dx dx + dz2 L2 , which is AdS5 . It z turns out that this metric also has conformal invariance. So scale and Poincar implies e conformal invariance, at least when there is a gravity dual. This is believed to be true more
L Lz ds2
generally, but there is no proof. Without Poincar invariance, scale invariance denitely does e not imply conformal invariance. We now formulate the guess that a 4dimensional Conformal eld Theory is related to a theory of Gravity on AdS5 Note: On a theory of gravity spacetime is a dynamical variable, where you specify asymp totics. This CFT denes a theory of gravity on spaces that are asymptotically AdS5 . Why is AdS5 a solution? 1. Check on PSet 2. Use eective eld theory (which in this case just means dimensional reduction) to elaborate on my previous statement that the reason why this is a solution is that the ux holds it up. To do this lets consider the relevant part of the supergravity action, which is S = D d x[ GR(D) Fg Fg ] As an example consider the case where D=10 and g=5 (note that we are working in dimensions (D) where GN = 1 = 1/5! = 2 = 4). [Note: The reference for this is Denef et al hepth/0701050] Lets make an Ansatz (called FreundRubin) The metric is some spacetime (x) plus a qsphere of radius R, so that ds2 = g dx dx + R(x)2 Gmn dy m dy n and furthermore for the ux though the qsphere we have S q Fq = N To nd the form of this part of the metric as a result of the ux, we integrate over the qsphere and get the action in Ddimensions, which will contain a term which is GGm1 n1 ...Gmq nq (Fg )m1 ...mq (Fg )n1 ...nq Sq We should then think of R(x) as a moduli eld. Now we have to evaluate the following a integral S q GR(D) = gRq [R + R2 + ...] We will now go to something called the FRAME GAME To make the Ddimensional EinsteinHilbert term canonical, we can do a Weylrescaling of E the metric, meaning to dene a new metric g = (x)g .
So the new metric is some function of the spacetime variables times the old metric.
So gE  = d/2 g , RE = 1 R.
Well pick to absorb the factor of R, so gE RE = d/21 gR(d) = Rq gR(d) , = R2q/d2 . N We know that the ux is the integral over the qsphere, so S q F = N N Rq , so the 2 n ux term of the potential is Vf lux (R) S q F F = V ol(S q ) gF = Rg d/2 gE ( Rq )2 At small R the rst term in the eective potential dominates, while for R large, the eective potential approaches zero from below. In between there is some minimum. To nd the actual value of this we take 0 = V (Rmin ) N 2 Rmin . So Vf lux (R)
N2 R
1 . R
The fact that the potential goes to zero at large R is unavoidable. It is because when R is large, the curvature is small and ux is dilute. So all terms in the potential go away. 4 For special values of and , we get that N Rmin = L4 3
This endeavor is called ux compactication and it is the way to nd vacua of string theory without massless scalars. Furthermore, we can add some eective action for scalars with m2 V (Rmin ), which R 1 looks like this : S = 2 dd x g e (RE 2)R . The cosmological constant is determined as = Vef f (Rmin ) < 0
The fact the the cosmological constant exists, means that the metric is not at.
The equation of motion is the following: 0 = R =
2d d2 S g 1 R = g ( 2 R )
dr 2 + dx dx
r2
R ( 2 ) 2d
We know the solution for AdSd , which has the metric ds2 = L2
Now I will describe the right way to compute curvatures, which is to use what is called tetrad or
vielbein or CartanWeyl method.
There are three steps
Cartan 1 demands that they are covariantly constant and is formulated as follows: dea = ab eb , where the is called the spin connection.
1 1 L For example der = 0 and e = r2 dr dx = L er e = L e = r
From that you can get the Ricci tensor, which appears in the equation of motion, by contracting some indices R = R Lets do some examples: 1 R = r r = L2 e e
Rr = d r = 1 r e e L2
1 R = L2 ( ) r 1 Rr = L2
R =
d g L2
2 2d
d L2
d(d+1) 2L2
Counting of degrees of freedom [SusskindWitten, hepth/9805114] The holographic principle tells us that the area of the boundary in planck units is the number of degrees of freedom (maximum entropy). 4
Both sides are equal to innity, so we need to regulate our counting somehow.
The picture we have of AdS is a collection of copies of Minkowski space of varying size and the
3 boundary is the locus where they get really big. The area is dened as A = R3 gd3 x = R3 d3 x L3 r This is innite for two reasons: from the integral over x and from the fact that r is going to zero. Lets regulate the Field Theory rst. We put the thing on a lattice, introducing a low distance cuto and we put it also in a box of size R. The number of degrees of freedom is the number of 3 cells times N 2 , so Nd = R3 N 2 . To regulate the integral we write before, we stop not at r = 0 but 3 L3 3 L3 we cut it o at r = . A = d x r3 r= = R3 Now we can translate this
L2 G10 N
N 2 GN =
L3 R3 (5) GN 3
(5)
GN L5
(10)
Nd =
A (5) GN
Todays lecture will teach you enough supersymmetry to defend yourself against a hostile supersymmetric eld theory, should you meet one down a dark alleyway. Topics will include 1. SUSY representation theory, including basic ideas regarding the algebra and an explanation of why N = 4 is maximal (a proof due to Nahm) 2. Properties of N = 4 Super YangMills theory, such as the spectrum and where it comes from in string theory. Before plunging in, we might be inclined to wonder: why is supersymmetry so wonderful? In addition to the delightful properties that will be explored in the remainder of this course, there are various reasons arising from pure particle physics: for example, it stabilizes the electroweak hierarchy, changes the trajectory of RG ows such that the gauge couplings unify at very high energies, and halves the exponent in the cosmological constant problem. These are issues that we will not explore further in this lecture.
1.1
Poincare group
Recall that the isometry group of Minkowski space is the Poincare group, consisting of translations, rotations, and boosts. The various charges associated with the subgroups of the Poincare group are quite familiar and are listed below Charge P M 1 Subgroup R3,1 SO(3,1) (SU(2) SU(2))
Translations Rotations/boosts
Luckily the rotation/boost part of the algebra SO(3,1) decomposes (at least in the sense of the sloppy representation theory that we are doing today) into two copies of SU(2). We are all familiar with representations of SU(2), each of which are labeled by a halfinteger si 0, 1/2, 1, ...; the product of two of these reps gives us a rep of SO(3,1):
(s1 , s2 )
(0, 0)
(1/2, 1/2)
(1, 1)
(1/2, 0), (0, 1/2)
(1, 0), (0, 1)
Having mastered this group, we wonder whether we could possibly have a larger spacetime symme try group to deal with, leading us to the idea of enlarging the Poincare group by adding fermionic generators!
1.2
Supercharges
Lets call these fermionic generators QI ; is a twocompnent Weyl spinor index belonging to the (1/2, 0) spinor representation above; essentially this simply denes the commutation relations of Q with M , telling us that it transforms like a spinor. I labels the number of supersymmetries and thus goes from 1 to N . The total number of supercharges is thus equal to the number N multiplied by the dimension of the spinor representation and so in four dimensions is 4N . The adjoint of this operator is called QI (QI ) . We note that [P , Q] = 0 and the anticommu tators of Q with itself are given by I J I QI , QJ = 2 P k Q , Q = 2 Z IJ (1) Here Z IJ is antisymmetric in its indices and thus exists only for N > 1; playing with the SUSY algebra assures us that it commutes with everything and so is just a number called the central charge. Note that this algebra is preserved under the action of a U (N ) called (inexplicably) an Rsymmetry.
I QI UJ QJ I UJ U (N )
(2)
Whether or not the diagonal phase rotation of this U (N ) survives in the full quantum theory depends on dertails of the theory. Finally, the current jR of this symmetry does not commute with the supercharges [jR , Q] = 0, meaning dierent elements of the same supermultiplet can have dierent Rcharges.
1.3
Next, let us construct representations of this algebra, starting with massless states. Consider the Hilbert space H of a single massless particle, and pick a frame where the momentum is P = (E, 0, 0, E), E > 0. Now using (1) and standard formulas for ( 0 = I2 and i are the Pauli matrices) we see that 4E 0 I I Q , QJ = J (3) 0 0 Focusing on the bottom left corner of this matrix, we see that QI , (QJ ) = 0 QI 2 = 0, 2 2 2 (4)
for all  H. If we assume this zeronorm state QI  is actually the zero state (which is the 2 case in a unitary Hilbert space), then we conclude that half of the Qs annihilate every state in the space, implying also that Z IJ = 0 in this subspace. We are left with the QI s; note each QI 1 1 lowers the helicity by 1/2 and each QI raises the helicity by 1/2. Finally, if we normalize the Q1 s 1 appropriately we get I Q1 Q1J I , = J (5) 2 E 2 E From now on we will suppress the 1 subscript. This is simply the algebra of a fermionic oscillator, which we know how to deal with: start with a highesthelicity state that is annihilated by QI h = 0 and hit it with QI s to build the Hilbert space: h, QI1 h, QI1 QI2 h, ... QI1 QI2 ...QIN h (6) There are a total of 2N states because (as pointed out by T. Senthil), each operator QI may or may not be present in the string of Qs. Now we wonder: how big can N be? A theorem by Nahm answers this question. A useful lemma (related to the WeinbergWitten theorem discussed previously) is: A quantum eld theory without gravity cannot contain massless states with helicity h > 1. This means that we cannot apply Q too many times, as eventually we will end up with a state with helicity more negative than 1. This is a constraint on N and allows us to summarize all possibilities in the following (nite) table, in which each entry is the number of states with that given helicity in the corresponding supermultiplet: Helicity 1 0 1 2 1
1 2
N = 1 gauge 1 1 1 1
N = 1 chiral 0 1 (1+1) 1 0
N = 2 gauge 1 2 (1+1) 2 1
N = 4 gauge 1 4 6 4 1
Note that if N > 4, then we will necessarily have states with helicity greater than 1. Thus the maximal supersymmetry for a nongravitational theory in d = 4 is N = 4, corresponding to 16 supercharges. If gravity is allowed, the corresponding bound on helicity is h < 2 N < 8, and thus we can have N = 8 supergravity with 32 supercharges in four dimensions. 3
1.4
It is time to move on to massive states, which turn out to have larger representations. Following a procedure analogous to that above, we boost to the restframe of the particle where we have P = (M, 0, 0, 0). In that case (1) becomes
I J I J I Q
, (Q
) =
2M
Q
, Q
= 2 Z IJ (7)
J It is now necessary to introduce some unfortunate notation. We take a deep breath and block
diagonalize Z IJ so that it takes the form
0 Z1
Z1 0 IJ (8) 0 Z2 Z = Z2 0 ... To keep track of this structure we split the index I into two parts I = (I, I), where I runs over each of the subblocks in the large matrix I = 1, 2, ...Floor(N /2) and I = 1, 2 and runs over the two entries in each subblock. We now form the following linear combinations of Q s: 1 (I =2,I ) I (I =1,I ) 0 Q
=
Q Q
(9)
2
These supercharges obey the anticommutation relations I J I Q
, Q
=
(M ZI ), J
(10)
where the are correlated on all sides of this equation. Once again, a unitary representation implies that the righthand side of this expression should be greater than zero, leaving us with the celebrated BPS bound: M

ZI  (11)
Helicity 1 0
1 2
N =1 gauge 1 2 1
N =1 chiral 0 1 2
N = 2
gauge
1
4
6
N = 2
BPS gauge
1
2
1
N = 2
BPS hyper
0
(1+1)
(2+2)
N = 4
BPS gauge
1
4
6
Here all states will be present with their CPT conjugates of opposite helicity. Note that the massive gauge multiplet contains precisely the same states as (massless chiral + massless gauge) multiplets. Compare this to the familiar nonsupersymmetric construction via the Higgs mechanism of a massive vector boson from (massless scalar + massless gauge) bosons. It should not be surprising that the supersymmetric version of this is called the SuperHiggs mechanism.
N = 4 Spectrum
We now turn to a closer examination of the spectrum for the maximally supersymmetric N = 4 case, which can be formulated with any gauge group (but we will restrict ourselves to U (N )). Here the eld content is (compare with the massless N = 4 gauge multiplet in the table): 1. Gauge eld A : = 0...3 is a Lorentz vector index. 2. Weyl spinors I : I = 1...(N = 4), a spinor index. 3. Scalars X i : here i = 1...6, as will be explained shortly. As the gauge eld is necessarily in the adjoint of the gauge group, all of these elds are N N matrices in the adjoint of U (N ). Note that the Rsymmetry is SU (4) (the diagonal U (1) is broken, for reasons which we will presumably learn later), which acts on the fermions I in the fundamental of SU (4). To understand the action on the scalars, it is useful to know that SU (4) SO(6) (12)
The fundamental of SU (4) is the spinor representation of SO(6). As we obtained the Xs by the action of two antisymmetrized Qs on the (SU (4) singlet) A , we expect it to transform in the antisymmetric tensor representation of SU (4). This works out to be the standard vector 6 of SO(6), which acts on the i index on X i . If we stray into string theory briey, we see that this theory is also the worldvolume theory on a stack of N D3 branes. This gives us another way to understand the spectrum. Imagine placing a single D3 brane in R3,1 . This breaks 10 (total) 4 (along the brane) = 6 translational symmetries, and so we expect to have 6 Goldstone bosons in the worldvolume theory of the D3 branesthese are the X i s! Similarly, the full tendimensional theory had 32 supercharges (from N = 2 supergravity in ten dimensions), but we only see 16 supercharges remaining in the N = 4 theory. To understand where they went, consider the supercharges Q: {Q, Q} P 5 (13)
Those Q whose anticommutators generate translations along the 6 broken translational symmetries are broken, and this leaves us with 16 unbroken supersymmetries. By Goldstinos theorem, this breaking results in 16 massless fermions or goldstinos (goldstini ?), which can be rearranged into the 4I s. Finally, we note that another way to produce this theory is to start with maximal superymmetry in d = 10 without gravity. This is a pure N = 1 gauge multiplet, and contains 16 supercharges; we are no longer allowed to break these supersymmetries with D3 branes (since we need all of them!). What we can do is consider eld congurations that are independent of the 6 extra dimensions and simply integrate over them in the 10 dimensional action. This procedure is called dimensional reduction and will give us a 4dimensional action containing precisely the eld content of N = 4 SYM.
8.821 F2008 Lecture 06: Supersymmetric Lagrangians and Basic Checks of AdS/CFT
Lecturer: McGreevy September 24, 2008
We are on our way to talking about really awesome things about supercool stu. Before we get there, though, we need to develop some very powerful technology. To that end, today we will talk about 1. SUSY Lagrangians and a whirlwind tour of the beauties of superspace. 2. more on N = 4 SYM. 3. Back to the Big Picture: Some basic checks of AdS/CFT
Looking past this lecture, we will be talking about strings from gauge theory next.
Recall that the eld content of N = 4 SYM is a vector A , gaugini I=1...4 , and six scalars X i , all in the adjoint of the gauge group. The Lagrangian density (which is completely determined by the amount of SUSY, up to two parameters, (gY M , )), is
L = +
1
2 gY M 6 i<j
/ tr F 2 + (DX i )2 + iD
i tr(F F ) 8 2
A Note: There are obviously indices and gamma/sigma matrices suppressed ALL over the place (Lorentz vector, Lorentz spinor and SO(6) vector/spinor, supersymmetry). If you want to put the indices in, either leave that as a fun exercise, or check out Weinberg, volume 3. As an example, F + F .
1.1
A Superspace Detour
The N = 4 SYM Lagrangian is an example of a (highly) supersymmetric Lagrangian. So far, I just told you what it was and that it was SUSY invariant (something you could sit down in the privacy of your oce and check, if you wanted). Itd be nice, though, if there were some sort of a machine that one could crank to generate supersymmetric lagrangians. That crankable machine is superspace. To understand why superspace is useful, we should think about why elds are useful for representing translationally invariant Lagrangians in ordinary QFT. One reason is that the representations of the translation group on the elds are particularly simple
(3)
Wed like to introduce a supereld (that comes with its own supercapitalization) (x, ), which is now a function of spacecoordinates x and superspace coordinates , that well represents transla tional invariance AND supersymmetry
(x, ) = eiP x+iQ (0, 0) where the Q s are the operators that generate supersymmetry transformations. Now, in QFT, one can automatically get translationally invariant actions
(4)
S=
dd x L(, )
(5)
dd x dN s L((x, )) 2
(6)
is supersymmetric as long as L = 0. Here s denotes the smallest (real) dimension of the spinor representation in d dimensions and N , is, as usual the number of supersymmetries. Then N s is just the number of real supercharges. For example for N = 1, d = 4, N s = 4, because either a Weyl or Majorana spinor (the minimum in four dimensions) has four real components.
1.1.1
In lecture 5 we made a big deal about special representations of supersymmetry which are killed by some of the supercharges. Such multiplets have correspondingly special properties in superspace. Consider a eld which satises
These multiplets, which are BPS (half of the supersymmetries annihilate them) are generally called chiral multiplets. Sometimes they are actually chiral (in the sense of the Lorentz group), but often times they are not. This follows a long tradition in physics of calling things other things which they are not. These multiplets are functions of only half of superspace as
(x, , ) = ei(Q+Q) (x, 0, 0) = (x, , 0)
(8)
Ok, well this equation is not exactly correct (as Senthil pointed out; really the RHS should be (y x + i, )), but it is morally correct: the s are functions of half of superspace. Because of this, it is possible to add terms to the Lagrangian density that are integrated over only half of superspace and maintain supersymmetry:
L =
d2 W () +
d2 W ()
(9)
Here, we are explicitly working in d = 4, N = 1 superspace. These terms are supersymmetric, as long as W is a holomorphic function of , W = 0. With this constraint, W is a function we are free to choose, and is known as the superpotential. Two examples of a superpotential are The second line of equation (1). Here, we refer to the fact that this line can be written in d = 4, N = 1 superspace (where we pick out a particular N = 1 subgroup from the N =4). In a certain sense, one of the s and F can be thought of as comprising one N = 1 chiral multiplet, while the remaining three s and six s can be thought of as another three chiral multiplets. The second line of equation (1) is a superpotential for these three chiral multiplets. On pset 2 you will have a chance to think about this more precisely. 3
The gauge kinetic terms and third line of equation (1) can be thought of as coming from the N = 1 superpotential d2 tr( ) (10)
where the appearing is the superpartner of F ( is a spinor index), and is a complexied coupling constant g4i + 2 . Here should be thought of as a supereld whose lowest 2 YM component is the gaugino. The supereld expansion contains a = . . . F term. Multiplying two together, one gets 1
2 gY M
d F
d2 tr( )
(11)
The rest of the gauge kinetic terms and theta angle term can be understood similarly.
1.2
Whats the big deal with this newfangled superpotential? Well, as we said before, the superpotential has to be holomorphic in order for supersymmetry to be preserved. We can take this line of reasoning one step furtherwe can think of promoting the couplings to dynamical superelds (whose lowest component vevs are just the constant couplings). Then, the superpotential must be holomorphic, also, in the couplings. This statement is uncomfortably powerful. For example, it implies that if SUSY is not broken, the form of radiatively generated corrections to the superpotential are severely constrainedthey must be holmorphic in the elds and the couplings. For example, one could never generate a term in the superpotential that was a function of both and , W = W ( ). This leads to many nonrenormalization theorems in supersymmetric eld theories, one example of which is for the function of supersymmetric gauge theories. This says that
Y M = 1 loop + nonperturbative
(12)
This makes sense because we know the theta angle cannot appear in the beta function perturba tively. However, since the eective gauge coupling function must be holomorphic in , the only perturbative contribution to the beta function can be O( 0 ), i.e., the one loop contribution. For N = 4 SYM, one can go home and calculate (for pset 2) that the one loop contribution to the beta function is identically zero. In this theory, the nonperturbative terms can be understood as coming from instantons. However, as it turns out, they do not correct the beta function, but do add higher derivative terms elsewhere in the action1 . Hence, in this theory the beta function is zero, even nonperturbatively, and the theory really is scale invariant.
1
tr[X i , X j ]2
(13)
In general, supersymmetry is unbroken i V = 0, and so there is a moduli space of supersymmetric vacua, labeled by vevs of X which satisfy M = {X[X i , X j ] = 0, i, j}/{gauge symmetry} (14)
We divide out by the gauge symmetry, since vacua related by gauge rotation are physically equivalent. We can x this gauge symmetry and satisfy the moduli space condition by picking a basis such that X i=1,...,6 = diag(xi , . . . xi ) 1 N (15)
where N is the number of colors. Hence, the moduli space is just an 6 N dimensional space given by these xs. At a generic point in this moduli space xi = xi , and so the gauge bosons that commute n m with each X i are all just proportional to the X i s themselves, and so, at a generic point, the gauge group is broken from U (N ) to the U (1)N subgroup generated by these. Since we have broken to a whole bunch of copies of electromagnetism (with some charged scalars and spinors), this moduli space is known as a Coulomb branch. Since the vevs of the Xs give a mass scale, not only does a generic point in this moduli space break the gauge symmetry, but it also spontaneously breaks the dilation, and hence, the superconformal symmetry. The higgsing from U (N ) to U (1)N gives W bosons with a mass derived from this mass scale mab = xa xb  = Z ab W (16)
The W bosons are BPS objects with respect to the stillunbroken N = 4 supersymmetry. This must be true since we said in lecture 5 that the N = 4 BPS massive multiplet (what we have created by higgsing) has the same number of degrees of freedom as the N = 4 massless multiplet (what we had before higgsing). The massless N = 4 gauge multiplet carries its food around with it. The Z above are just the central charges of these BPS objects. Recall also that this story has a Dbrane interpretation: N = 4 SYM with gauge group U (N ) is the worldvolume theory of N (coincident) D3 branes. We can think about separating these parallel N D3 branes. Since they are parallel, and sit in a ten dimensional spacetime, 5
there are six coordinates which label their positions: xi where i runs over the six transverse a dimensions and a runs over N , the number of D branes. Actually, to make sure the mass dimensions of xi , match with those above, we write the separation between brane a and b as a
i yab = xi xi  a b
(17)
Recall, however, that when the D branes were coincident, we could interpret the lowest (massless) string states which started on one D brane and ended on another as massless gauge bosons for the U (N ) gauge theory. As we separate the branes, it is no longer the case that strings stretched between branes are massless: they are stretched. Therefore, the gauge bosons are acquiring a mass! In this context, the Higgs mechanism is just pulling a stack of coincident branes apartwhich is kind of superawesome. One can calculate the mass of these W bosons, and not surprisingly they match with the picture above 1 yab = xa xb  (18) The BPS property means that this relation must be true for any value of the string coupling. mab = (string tension) x (length) = W 2. Sduality (a.k.a. MontonenOlive) Sduality says the following absurd thing 1 [N = 4 with gauge group G, coupling ] = [N = 4 with gauge group L G, coupling ] (19) where L G is the dual gauge group, more on that in a second. As with many such dualities, it says that these two theories are completely equal, though it might be hard to gure out how to map the observables of one onto observables of the other. The dual group is dened such that the weight lattice of L G is equal to the dual of the weight lattice of G w (L G) = (w (G)) For example, (20)
L L
SU (N ) = SU (N )/ZN
L
The L is for Langlands. Really, Sduality is a subset of a larger duality that the theory enjoys: it is invariant under shifts of the theta angle, + 2. These two transformations are usually called S and T :
2 S ( = 0) : gY M
16 2 2 gY M
(22) (23)
a + b , ad bc = 1 (24) c + d There are many checks of this duality. For example, on the Coulomb branch, the theory has BPS solitons whose spectrum matches that of the (perturbative) vector multiplet of the dual group. Under an S transformation, the BPS solitons and vector multiplets switch. 3. Superconformal Symmetry Supersymmetry and conformal symmetry combine (like Voltron) to form a larger spacetime symmetry group, superconformal symmetry. In particular, the commutator of K (which generates special conformal transformations) and Q (which generates supersymmetry trans formations) is nonzero (indices omitted) [K, Q] = S (25)
where S is a fermionic symmetry generator. So, although supersymmetry gave us 16 fermionic symmetry generators, superconformal symmetry enlarges that to 32 fermionic symmetry gen erators.
2. Perturbations The rst claim each linearized supergravity perturbation corresponds to some N = 4 gauge invariant operator. For now we focus on local operators, combinations of elds taken at the same spacetime point. Since the operators are gauge invariant, they involve a trace, or prod uct of traces over SU (N ) indices. Well focus on single trace operators (the reasons for which will be clear in a future lecture). We can organize operator representations of the superconformal algebra by starting with operators that are so called superconformal primaries. These operators O are annihilated by both K and S [K, O] = [S, O] = 0 (26)
We can then get other operators in the representation by acting with the operators Q and P . Since, for a superconformal primary [S, O] = 0, it cannot be written as [Q, another operator]. Glancing at (2), this suggests that all superconformal primaries are built out of the Xs. In fact, since commutators appear on the RHS of (2) only symmetric combinations of the Xs will be superconformal primaries. Thus we are looking at operators of the form Oi1 ...il T r(X {i1 . . . X il } ) (27)
On the supergravity side, elds (perturbations) on AdS5 S 5 can be expanded in spherical harmonics on the S 5 . For example, using x as coordinates on AdS5 and y on the S 5 (with 2 y = 1), any eld (x, y) can be expanded as (x, y) =
l
l (x)Y l (y)
(28)
and the spherical harmonics can be written as Y l (y) = Ti1 ...il y i1 . . . y il P y2 =1 (29)
The correspondence says that we should identify the spherical harmonics with superconformal primaries as Ti1 ...il y i1 . . . y il Ti1 ...il T r(X {i1 . . . X il } ) (30)
in a way that will be made precise in not too future lectures. This is enough to organize the whole spectrum of supergravity perturbations. Since we know which supergravity elds correspond to superconformal primaries, we can get descendant operators by acting with Q and P . Similarly, we can get the supergravity perturbations that correspond to these operators by acting with the corresponding symmetries in AdS5 S 5 .
Lecturer: McGreevy
September 25, 2008
In particular, for scalars, m2 L2 = ( 4) [recall that = L4 /2 ]1 AdS AdS Last time we discussed the KaluzaKlein (KK) harmonics of 10d supergravity (SUGRA) elds (= massless string modes). These correspond to superconformal primary operators of the N = 4 theory according to: (x, y) = (x)Y (y)
On the left hand side, this is just the mass of a KK harmonic on a space of size L, i.e. an eigenvalue of the laplacian on the 5sphere. On the right hand side, we have an operator made from a product of l scalar elds. A 4d scalar has engineering dimension 1, so the dimension of the product is l in the free theory, at least. (It is actually true for all by the BPS property.) The above can be considered as the highest weight states. Then, using the fact that the supersymmetry algebras match on the two sides, we can ll out multiplets on both sides by acting with the generators. In this way, the correspondence of all SUGRA modes in the N = 4 theory can be found (an intimidating table can be found in EDHDZF, hepth/0201253, p.50).2 Now consider some observables of a QFT (well assume Euclidean spacetime for the now): O1 (x1 )O2 (x2 ) On (xn ) We can write down a generating functional Z[J]:
L(x) L(x) + JA (x)OA (x) L(x) + LJ (x)
A
Z[J] = e
LJ
CF T
where JA (x) are arbitrary functions (sources) and {OA (x)} is some basis of local operators. The npoint function is then given by:
n
On (xn ) =
ln Z
Jn (xn ) J=0
=???
gs 0 L2 /
eSSUGRA
EOM, b.c. depend on J
As an example of the application of this formula, one can compute the 3point function of supergravity states. It turns out that the results agree exactly with N = 4 SYM (at leading order in 1/N ). A priori this is a surprising statement, since the eld theory calculation is done by pertur 2 bation theory at small = gY M N , while the gravity calculation is meant to work at large . This happens because all the supergravity states correspond to BPS (= chiral primary) operators, and hence their 3point functions actually do not depend on . This specialness of the states weve been able to map so far may worry you: how do we describe the rest of the eld theory operators using the bulk theory? To approach this question, we take a small detour.
There are several longsuspected connection between strings and gauge elds.
2.1
For hadron with largest spin J for a given mass, J = m2 (J) + const. Regge trajectory
which is precisely the form of the spectrum of the vibration modes of a spinning quantum relativistic string in RD .
2.2
We know that the gauge eld in QCD is in the conned phase (as opposed to the Coulomb phase, see Fig. 1(a),(b) for illustration). Consequently, the eld lines bunch together and behave like a tensionful ux tube. More concretely, consider the area law for Wilson loop: Tr Pei
R
= e Area()tension
Here denotes the worldsheet as shown in Fig. 1(c). This is an order parameter for connement and indicates that ux tube has a nite tension if and only if the gauge theory is in the conned phase. The above suggests that we may try to use ux tubes as microscopic variables5 , which should behave as strings in R3,1 . However, a result from Polyakov suggests that this is problematic. Polyakovs result is this: in quantizing the string worldsheet with metric , the transformation e() is a gauge symmetry and hence e = 0. There is however a conformal anomaly,
5 It will turn out that a string description does not require connement. The loophole is that one may consider funny shaped worldsheets.
loop
worldsheet
(a)
(b)
(c)
Figure 1: Flux tubes in QCD: (a) meson in conned phase, (b) meson in Coulomb phase, i.e. a dipole (c) the world sheet of ux tube for computing Wilson loop.
which generates kinetic term of in e . Hence should be treated as an extra dimension. For QFT in d 1, the procedure for quantizing is known (Liouville), and this is really the correct idea (this leads to what is sometimes called old matrix models, including the c = 1 matrix quantum mechanics). In d = 4 this result (so far) serves as more inspiration for our extradimensional picture..
t Hooft Counting
The most explicit evidence that gauge theory leads to string theory comes from t Hooft counting. Consider a (any) quantum eld theory with matrix elds6 b=1,...,N (for concreteness well take the a=1,...,N matrix group to be U (N )), and consider a Lagrangian of the form: 1
2 gYM
L Rescale = gYM , L Tr
Tr ()2 + 2 + 3 + 4 + . . .
2 ( )2 + 2 + gYM 3 + gYM 4 + . . .
Now consider propagators. It is convenient to adopt the double line notation, in which oriented index lines follow conserved color ow, so that, for propagator7 :
a a d a d c b = c b c
6
c d
By matrix eld we mean that their products appear in the Lagrangian only in the form of matrix multiplication, e.g. (2 )c = b c .
a a b 7 a d a c b c Had we been considering SU (N ), the result would be a d c b b d /N 2 =
a c a b b a a d
2 gYM
gYM
c b b c c d
2 Now consider the t Hooft limit in which N and gYM 0 but which = gYM N = const. Is this limit classical/free? The answer turns out to be no. The loophole is that even though the coupling goes to zero, the number of modes diverges.
N2
N 2
3 N 2
(a)
(b)
(c)
2 gYM N = N 0
Figure 3: Nonplaner graph that contributes to the vacuumvacuum amplitude. To see this more concretely, consider vacuumvacuum diagrams (see Fig. 2 and 3 for illustration). In Fig. 2 we have a set of planer graphs, whose contributions take the general form n N 2 . However, there are also contributing nonplaner graphs, such as the one in Fig. 3, whose contribution does the take that general form. Every doubleline graph specics a triangulation of a 2dimensional surface . There are two ways to construct the explicit mapping: Method 1 (direct surface) Fill in index loops with little plaquettes. Method 2 (dual surface) (1) draw a vertex in every index loop and (2) draw an edge across every propagator. These constructions are illustrated in Fig. 4 and 5. Comparing the resulting surface and the corresponding contribution, it can be seen that the contri bution of a graph with h handles is proportional to N 22h n . Thus, the partition function (which is also the vacuumvacuum amplitude) must take the form: 5
S2
T2
(a)
(b)
Figure 4: Direct surfaces constructed from the vacuum diagram in (a) Fig. 2a and (b) Fig. 3.
S2
Figure 5: Dual surface constructed from the vaccum diagram in Fig. 2c. Note that points at innity are identied.
ln Z =
h=0
N 22h fh ()
Introduction
Today well continue our discussion of large N scaling in the tHooft limit of quantum matrix models. In particular, well review the N counting for vacuum diagrams and learn how to do N counting for correlators. Our perspective on the large N limit continues to be one of convenience. It is one of several assumptions that enable the simplest realization of holography and gauge/gravity duality. Next time well continue talking about simplifying assumptions and turn our attention to conformal invariance.
Lets recall our starting point. We were thinking about matrix eld theories, eld theories in which all the elds are N x N matrices in addition to whatever other labels they may have. In particular, we write our theory schematically in terms of one big eld which we think of as potentially including scalars , gauge elds A , and fermions all of which are N x N matrices. The schematic Lagrangian is 1 L = 2 Tr ( V ()) (1) g where V () 2 + 4 + ... and all products should be thought of as matrix products. This schematic representation is meant to convey two pieces of information. First, that the Lagrangian is invariant under U (N ) rotations which act on the matrix elds by conjugation: U U 1 . Second, that the elds are scaled so that the coupling g 2 sits out in front of the entire Lagrangian. In particular, we have to very careful about eld normalizations when doing large N counting for correlation functions. The tHooft limit is the limit where N and g 2 0 while = g 2 N remains xed. is the tHooft coupling. Note that we are adopting a dierent normalization condition than in the previous lecture on large N counting (there we used the eld = (1/g) which had kinetic terms with the usual eld theory normalization). With our current normalization the propagator goes like g 2 = N . The propagator in the double line notation appears in Fig. 1. Remember that for the purposes of 1
Figure 1: Propagator, /N
large N counting we are instructed to think only about counting powers of N, so we will suppress positions of elds, spin indices, etc in our analysis. In a generic matrix theory there will be three point (Fig. 2) and four point (Fig. 3) interaction vertices. Both types of vertices come with factors of 1/g 2 = N/ from the overall 1/g 2 multiplying the Lagrangian. A general diagram consists of propagators, interaction vertices, and index loops, and gives a contribution no. of prop. no. of int. vert. N N no. of index loops . (2) diagram N For example, the planar diagram in Fig. 4 has 4 three point vertices, 6 propagators, and 4 index loops giving the nal result N 2 2 . We learned last time how we to associate a triangulated surface with a Feynman diagram in two ways. The rst way is to let the interaction vertices of the diagram be the vertices of the triangu lation. In this case the index loops of the diagram bound pieces of surface. The alternative is the dual triangulation in which we place vertices in the middle of the index loops of the Feynman dia gram. Two vertices are connected if their respective index loops share a propagator in the Feynman diagram. Dont be confused by multiple uses of the word vertex. There are interaction vertices of various kinds in the Feynman diagrams and these correspond to vertices in the triangulation only in the rst formulation. If E = no. of prop., V = no. of int. vert., and F = no. of index loops then we showed last time that the diagram gives a contribution N F E+V EV . The letters refer to the rst way to triangulate
Figure 4: This diagram consists of 4 three point vertices, 6 propagators, and 4 index loops
the surface in which interaction vertices are triangulation vertices. Then we interpret E as the number of edges, F as the number of faces, and V as the number of vertices in the triangulation. In the dual triangulation there are dual faces F , dual edges E , and dual vertices V . The relationship , V = F , and F = V . In the dual formulation we between the original and dual variables is E = E found that a diagram contributes N F E+V EF . Its not a coincidence that the powers of N agree in both formulations. The exponent = F E + V = F E + V is the Euler character and it is a topological invariant of two dimensional surfaces. In general it is given by = 2 2h b where h is the number of handles (the genus) and b is the number of boundaries. Note that the exponent of , E V or E F is not a topological invariant and depends on the triangulation (Feynman diagram). Before we continue with the analysis its worth pointing out that there are other large N limits one might want to consider. The motivation for the tHooft limit comes in part from thinking about radiative corrections which grow with N and decrease with g. This suggests some sort of balancing act might be possible which keeps the theory nontrivial in the large N limit. We encode this scaling by xing = g 2 N as we make N large. Other interesting possibilities are provided by critical points of the matrix model. Suppose when g = gc the matrix model is at a critical point. Then there may be a sensible large N limit in which we x g gc  N as we make N large. One example of this is matrix quantum mechanics with an inverted oscillator potential. Back to the tHooft limit. Because the N counting is topological (depending only on ) we can sensibly organize the perturbation series for the free energy in terms of surfaces! Because were computing only vacuum diagrams for the moment, the surfaces were considering have no boundaries b = 0 and are classied by their number of handles h. h = 0 is the two dimensional sphere, h = 1 is the torus, and so on. We may write the free energy as ln Z =
h=0
22h
=0
,h
h=0
N 22h Fh ()
(3)
were the sum over surfaces is explicit. Now we can see some similarities between this expansion and perturbative string expansions. By writing N 22h = (1/N )2h2 we can identify the string coupling gs = g()/N where we can allow some function of in addition to the pure N dependence. Because of the proliferation of coupling 2 constants, lets rename the g 2 in our matrix model gY M in anticipation of the case of most interest 2 to us when the matrix model is a gauge theory. The tHooft coupling is now written = gY M N . 3
If 1/N plays the role of the string coupling gs (so that in the large N limit joining and splitting of strings is suppressed) then its reasonable to ask what plays the role of the worldsheet coupling. It turns out that we can think of as a sort of chemical potential for edges in our triangulation. Looking back at our diagram counting we can see that if becomes large then diagrams with lots of edges are important. Thus large encourages a smoother triangulation of the world sheet which we might interpret as fewer quantum uctuations on the worldsheet. More formally, we expect a relation of the form 1 which encodes our intuition about large suppressing uctuations. This story is very general in the sense that all matrix models dene something like a theory of two dimensional quantum gravity via these random triangulations. The connection makes even more sense when we remember all the extra labels weve been suppressing on our eld . For example, the position labeling where the eld sits plays the role of embedding coordinates on the worldsheet. Other indices (spin, etc.) indicate further worldsheet degrees of freedom. We may think of the matrix model partition function as giving the thermodynamics of the string theory in the sense that it provides the functions Fh entering the free energy. However, the microscopic details of the theory are not so easily discovered. In particular, its not at all clear what microscopic action we should integrate over moduli space to obtain the functions Fh . As a nal check on the nontriviality of the theory in the large N limit, lets see if the world sheet coupling will generically run in a matrix model. In d = 4 the one loop behavior is basically 3 gY M = g gY M N = gY M and this implies the beta function for is = = 2gY M N g 2 . Thus can still run in the large N limit and the theory is nontrivial. If d = then 3 the beta function for g now looks like g gY M N + 4d gY M and we nd that 2 + 4d 2 2 which is still nite in the large N limit.
Correlation functions
Lets now consider how to calculate correlation functions of operators, and in particular, how to do large N counting for such correlation functions. We will consider only single trace operators, operators O(x) that look like O(x) = c(k, N )Tr(1 (x)...k (x)) (4)
of O (to determine the constant c(k, N )) we will adopt the convention that OO c N 0 where the subscript c stands for connected. This convention is nice because states like Ovac will have nite norm (given by the correlation function) in the large N limit.
and which we will often abbreviate as Tr(k ). There are two annoying complications now. We must be careful about how we normalize the elds and we must be careful about how we normalize the operator O. The normalization of the elds will continue to be such that the Lagrangian takes the form L = g21 L = N L which L() containing no explicit factors of N. To x the normalization
YM
To use our convention to determine c(k, N ) we need to know how to insert single trace operators into the tHooft counting. We do this by associating a new vertex with each single trace operator. This vertex has k legs where k propagators can be attached and looks like a big squid. An example of such a new vertex appears in Fig. 5 which corresponds to the insertion of the operator Tr(6 ). 4
For the moment we dont associate any explicit factors of N with the new vertex. Lets consider the example Tr(4 )Tr(4 ) . We need to draw two four point vertices for the two single trace operators in the correlation function. How are we to connect these vertices with propagators? The dominate contribution actually comes from disconnected diagrams like the one shown in Fig. 6. The leading disconnected diagram has four propagators and six index loops and so gives a factor 4 N 2 N 2 . One the other hand, the leading connected diagram shown in Fig. 7 has four propagators and four index loops and so only gives a contribution 4 N 0 . The fact that disconnected diagrams win in the large N limit is general and goes by the name large N factorization. It says that single trace operators are basically classical objects in the large N limit OO O O . As an aside, another convenient way to draw the connected diagram in Fig. 7 is shown in Fig. 8 where we have deformed the two four point operator insertion vertices so that they are ready for contraction making it easier to draw propagators.
The leading connected contribution to the correlation function is independent of N and so OO c c2 N 0 . Requiring that OO c N 0 means we can just set c = 1. Note that the MAGOO review normalizes their single trace operators with an explicit 1/N which is wrong. Having xed the normalization of O we can now determine the N dependence of higher order correlation functions. For example, the leading connected diagram for O3 where O = Tr(2 ) is just a triangle and contributes a factor 3 N 1 N 1 . In fact, quite generally we have On c N 2n for the leading contribution. The states Ovac are called glueball states, and they correspond to excitations of the theory that are free at large N. These states also make sense from the string worldsheet point of view where the operator O is like an integrated vertex operator. We now want to study the generating function of correlation functions of single trace operators. Consider a compound operator like A OA where each OA Tr(kA ) is a single trace operator with kA N . This is a multitrace operator because it is the product of single trace operators. We can write a generating function for correlation functions like A OA using the operator S = N A A OA where the overall factor of N is meant to normalize S in the same way as the Lagrangian. Remember that were suppressing all indices including spacetime locations of elds meaning there may be integrals involved in the proper denition of S. The generating function is dened as (5) Z = eF = eS and when calculating Z we must remember that each OA acts just like an extra vertex with the normalization we dened above. As a simple check of the AdS/CFT correspondence we would like to determine the large N scaling of F . The naive answer is just that F N 2 because there are N 2 degrees of freedom, however this counting may fail for example when the theory is in a conning phase. Let us therefore assume the theory is not in a conning phase and calculate the large N behavior of F more carefully. In the free theory we do indeed have F N 2 just by our naive counting of degrees of freedom. What about in tHooft limit? We may lump the interactions in L in with the operators in S since they are all single trace with an overall constant set by N. Then we can write the generating function as Z = e N O
f ree
eN
f ree
(6)
because in the tHooft limit we have seen that single trace operators behave as if they are classical. Using our previous formula for the N dependence of correlation functions we nd that O N so that F N 2 because of the overall factor of N.
Figure 9: A quark vacuum bubble and a quark vacuum bubble with gluon exchange
Lets check this result using the GKPW formula relating the partition functions of eld theory and supergravity Z = eS = exp (SSU GRA ). (7) 5 We can determine the N dependence of SSU GRA G1 d x gR by scaling the action in terms 5 N of L the AdS scale. In ve dimensions the Newton constant has units of length cubed, and so the L3 correct scaled form of the action is SSU GRA G5 Sdimensionless . Using the fact that L3 /G5 N 2 N N we discover that SSU GRA scales in the same way as F from the eld theory. This fact follows from the relation L4 /( )4 = which we found by studying the RR soliton, from the ten dimensional 2 2 Newton constant in string units G10 = gs ( )4 , and from the denition = gY M N = gs N . The N assumption of a nonconning ground state on the eld theory side translates into the absence of additional scales in the bulk spacetime that would invalidate our scaling argument.
Simple generalizations
We can generalize the analysis performed so far without too much eort. One possibility is the addition of elds, quarks, in the fundamental of U (N ). We can add fermions L q D q or 2 . Because quarks are in the fundamental of U (N ) their propagator consists bosons L D q of only a single line. When using Feynman diagrams to triangulate surfaces we now have the possibility of surfaces with boundary. Two quark diagrams are shown in Fig. 9 both of which triangulate a disk. Notice in particular the presence of only a single outer line representing the quark propagator. We can conclude that adding quarks into our theory corresponds to admitting open strings into the string theory. We can also consider meson operators like qq or q k q in addition to single trace operators. Another direction for generalization is to consider dierent matrix groups such as SO(N ) or Sp(N ). The adjoint of U (N ) is just the fundamental times the antifundmental. However, the adjoint representations of SO(N ) and Sp(N ) are more complicated. For SO(N ) the adjoint is given by the antisymmetric product of two fundamentals (vectors), and for Sp(N ) the adjoint is the symmetric product of two fundamentals. Also, lines in the double line formalism no longer have arrows. As a consequence the lines in the propagator for the matrix eld can join directly or cross and then join as shown in Fig. 10. In the string language the worldsheet can now be unoriented, an example being given by a matrix eld vacuum bubble where the lines cross giving rise to the worldsheet RP2 . 7
Everything weve said for large N matrix models can be contrasted with the story for large N vector models. The dierence between these two classes of models is that large N matrix models have N 2 degrees of freedom while large N vector models have only N degrees of freedom. This changes the scaling of the coupling in the large N limit. The Lagrangian of a large N vector model might look something like L = N V ( /N ) where the whole second term has the form N V m2 v /N ( )2 . This form of the Lagrangian amounts to many more interactions than in the large N matrix model. The coupling gv = v /N 1/N (v is not the tHooft coupling!) goes to zero much faster than the coupling in the matrix model gY M 1/ N . Because of the dierent scaling, the only interactions in the standard large N vector model are cactus diagrams. This means that modulo some self energy corrections the theory is essentially free. There also dont appear to be any strings.
As a preview of coming attractions, lets see what matrix models can tell us about the appearance of string theory in gauge theory. Recall that we said every SUGRA perturbation in AdS5 S5 corresponds to some single trace operator in N = 4 SYM. In particular, we found a correspondence between symmetric scalar KK modes in SUGRA and chiral primary operators in the N = 4 theory. However, this map from SUGRA modes to N = 4 operators is not onto, in fact it represents a very small subset of the N = 4 operators. We would like to know what the rest of the N = 4 operators correspond to in the gravity theory. A clue comes from looking at single trace operators like O = Tr(XXXXXY XXDXX...Y X) (8)
which we think of as a string of J letters (with 1 J N ). Because of the cyclic property of the trace, this string of letters looks a lot like a sequence of operator insertions on a closed string!
8.821 F2008 Lecture 09: Preview of Strings in N = 4 SYM; Hierarchy of Scaling dimensions; Conformal Symmetry in QFT
Lecturer: McGreevy October 8, 2008
Continuing from the previous lecture we substantiate the correspondence between the operators in N = 4 SYM theory with the excitations of some string theory. We saw that the primary chiral operators (i.e. primary operators in the conformal eld theory which also belong to the short multiplets of the supersymmetry) correspond to the SUGRA modes but we didnt discuss what happens when the operators in the gauge theory are in the long multiplet. Thus consider a more generic operator (residing in a long supersymmetry multiplet) of the matrix QFT in the large N limit O(x) = tr(XXXXY XXY....Y X) (1) We consider the limit N J 1 where J is the number of entities in the above product. Cyclicity of the trace implies that the structure of the above operator has the symmetry of a closed loop. In fact the above operator corresponds to creation operator for excited string states with X s and Y s the elds living on the worldsheet. Further, in the large J limit, J corresponds to the angular momentum of the string excitations and in fact this relation could be used to reconstruct the string theory. Having sketched the correspondence between the various operators of the SYM guage theory with SUGRA modes or string excitations, let us organize this knowledge to get a better perspective on when it could be useful.
The mass of the SUGRA mode is given by m2 GRA = 1/L2 . Using the AdSCFT corre SU AdS spondence m2 GRA L2 = ( 4) where is the scaling dimension of the corresponding SU AdS chiral primary operator in the SYM theory. This implies that SU GRA N 0 0 . Moving on to the excited string states m2 /L2 string states 1/ AdS where 1/ is string tension. Thus the corresponding scaling dimension equals string states N 0 1/4 . Finally, there are Dbranes in the string theory which correspond to baryonic states in the
2 m2 Dbrane 1/ gs
Clearly, the scaling dimension for the dierent states (SUGRA, strings, Dbranes) has distinctive dependence on N and . This hierarchy of s is what a QFT needs to have a weakly coupled gravity description without strings. For example the limit removes all states except SUGRA modes.2 There are various forms of AdSCFT conjecture. For example, one may believe that it only holds in the limit , N (yielding classical gravity) or perhaps also at nite and N limit (when one obtains classical strings on small AdS radius). Remarkably, all evidence till date points to a much stronger statement that it holds for all N and all .
This is a worthy subject in itself and study of conformal invariance is relevant for understanding both UV and IR limit of various QFTs and also for the worldsheet theory of the strings in the conformal gauge. As you may already know (and we recapitulate it below) that conformal symmetry in two dimensions is very special due to existence of innite number of conserved currents. Therefore it pays to understand which aspects of a conformal eld theory (CFT) are particular to D = 2 and which apply to any dimension D > 2. We have already seen some of the constraints due to the requirements of Lorentz invariance and SUSY on the QFTs and expectedly, requirement of conformal invariance makes it even more constrained. This is also the last stop in our ruthless program to evade the loopholes in the ColemanMandula theorem. Some of the useful references are the book Conformal Field theory by Di Francesco, Mathieu and Senechal and the articles by Callan, Ginsparg and portions of MAGOO (links posted on the course webpage).
2.1
We dene CFT by a list of operators and their Green functions which satisfy certain constraints described below. Succintly, a CFT is a theory with symmetries generated by conformal group . Obviously, it behooves us to dene conformal group which we do now. Please note that the route we follow is not a good one for nonrelativistic CFTs.
The power of here depends on the specic geometry of the Dbrane in question, i.e. which of the dimensions of the bulk the Dbrane is wrapping. 2 It is worth noting here that gauge theories in general also contain nonlocal operators such as Wilson loops: tr P ei
R
C
where C is some closed curve in the space on which the eld theory lives. It can be thought of as the phase acquired by a charged particle dragged along the specied path by an arbitrarily powerful external force. The connection between these operators and strings turns out to be quite direct, as one might expect from the relationship between charges on Dbranes and the ends of open strings.
Recall that isometry group of a spacetime with coordinates x and metric g is the set of coordinate transformations which leave the metric ds2 = g dx dx unchanged. Conformal group corresponds to a bigger set of transformations which preserve the metric up to an overall (possibly positiondependent) rescaling: ds2 (x)ds2 . Thus Poincare group (which is the group of isometries of at spacetime) is a subgroup of the conformal group with (x) = 1. Conformal transformations could also be seen as the successive application of a coordinate transformation x x , g g which preserve ds2 = g dx dx followed by a Weyl rescaling which takes x x so that ds2 is not preserved. Conformal transformations preserve the angles between vectors, hence the name: v w cos( ) v v w w
cos() =
(2)
Consider the spacetime IRp,q with a constant metric g = and lets look at the piece of innitesimal coordinate transformations connected to the identity:
x x + (x)
(3) (4)
( + )
Note that above there are no terms with derivative(s) of metric since the metric is constant. The requirement of conformal invariance implies that
( + ) = f (x) 2 = d
(5) (6)
2 where the equality f (x) = d is obtained by taking trace of on both sides of the eqn. 5. Applying an extra derivative on 5 gives
2 + (1 ) ( ) = 0 d
(7)
Pausing for a moment, we note that d = 2 is clearly special since = 0 implies that any holomorphic (antiholomorphic) function (z) (()) of complex coordinates z = x1 + ix2 would z correspond to a conformal transformation. A little more thought leads to the result that conformal algebra is innite dimensional in two dimensions. Applying one more partial derivative on this equation and applying on eqn. 5 and using the both equations thus obtained together gives (2 d) = 3
(8)
Contracting with implies = 0 = and thus (x) is at most a quadratic function of x. Lets organize (x) by its degree in x:
Degree zero: Translation : (x) = a Degree one: Dilatation : (x) = x Degree two: Special Conformal : (x) = b . x x xx b. (12)
Rotation : (x) = x with =
(9)
(10) (11)
The corresponding generators for the above transformations can now be immediately written down:
= i(x x )
Special Conformal : C = i( . 2x . ) xx x The commutation relations for the above generators are obtained as:
[P , M ] = i( P P )
[C , P ] = 2i( D M )
The rst four equations mean that P and C transform as vectors under Lorentz transformations while D is a scalar and M is a rank2 tensor. The next two equations mean that P and C act as raising and lowering operator for the eigenvectors of dilation operator D. 4
Interestingly, recombining conformal generators in suitable way reveals the very important and remarkable result that the conformal group in IRp,q is isomorphic to SO(p + 1, q + 1)!. First we do simple counting to check that the number of generators match correctly. The total number of generators of conformal group in d(= p + q) dimensions are d (translation) +d(d + 1)/2 (rotation) +d (special conformal) +1 (dilatation) which sum up correctly to (d + 1)(d + 2)/2, the number of generators for SO(p + 1, q + 1). To see the exact correspondence, dene the following generators:
J J d J d+1 Jd+1 d
= M 1 (P C ) = 2 1 = (P + C ) 2 = D
where {0, 1, ...d 1}. These new generators follow the SO(p + 1, q + 1) commutation relations: [Jab , Jcd ] = i(ad Jbc + bc Jad ac Jbd bd Jac ) (28)
where a, b, c, d {0, 1, ...d + 1}. Since the conformal algebra closes, it is possible to exponentiate the innitesimal transformations to obtain nite ones:
Translation : x x = x + a , Lorentz : x x
(x) = 1
= x ,
The last transformation (special conformal) may not seem very easy to visualize. One simpler way to rewrite its nite form is x x = + b .x . xx x
(33)
That is, it is nothing but a inversion followed by a translation followed again by inversion. As an aside, an alternative denition of the conformal group is that it is the smallest group which contains both Poincare and inversion operations.
2.2
Representation Theory
In a general QFT one is dealing with a large set of elds, their derivatives and products and it is dicult to organize various elds/operators. However in a CFT i.e. in a conformally invariant QFT there is a systematic way to organize various operators. Namely, we rst diagonalize the dilatation operator D and label the various operators by the their eigenvalues when being acted upon by D. Note that D commutes with M and hence we can still specify representations of the Lorentz algebra. Considering a spinless eld for simplicity, under a dilatation (x) (x ) = (0) which implies (34)
(35)
of special interest is a conformal primary operator transforms under conformal transformations as: /d x (x ) (x) x
(36)
where x is the Jacobian of the conformal transformation of the coordinates which is related to x the scale factor by x
d/2 x =
(37)
One must note that not every eld transforms this way even if it has denite scaling dimension. It turns out that the unitary representations of conformal algebra always have a lowest whose value is determined by the dimension of the spin. Physically this makes sense since correlators (x)(0) of the elds are not expected to diverge as x . Readers familiar with stateoperator correspondence (to be discussed in the following lecture(s)) may note that this could also be seen as a consequence of the fact that the spectrum of any unitary theory is bounded from below. Since C acts as a lowering operator for the scaling dimension (eqn. 22) , it annihilates the state lowest . This operator of lowest can be seen to be conformal primary according to the denition above (the Callan article does this somewhat explicitly). Now the strategy for organizing the whole spectrum of operators might be obvious: we label the various representations of the CFT 6
by lowest and build the complete multiplet by acting with P which acts as raising operator for scaling dimension (eqn. 21) . As a nal remark for this lecture: please note that in general primary operators are not eigen functions of the Hamiltonian H since the dilatation operator doesnt commute with H. As a consequence p p which is the Casimir of the Poincare group is not Casimir of the full conformal group.
In todays lecture well elaborate on what conformal invariance really means for a QFT. Remember, well keep working with CFTs in greater than 2 spacetime dimensions. A very good reference is Conformal Field Theory by Di Francesco et al, p95107.
Conformal Transformations
To begin, well give a few dierent perspectives on what conformal invariance is and what conformal transformations are. Our rst perspective was emphasized by Brian Swingle after last lecture. Recall from last lecture that the group structure of the conformal group is the set of coordinate transformations: x x
g (x ) = (x)g (x)
(1)
However, we should point out that coordinate transformations dont do anything at all in the sense that the metric is invariant, ds2 = ds2 , and all we did was relabel points. To say something useful about physics, we really want to compare a theory at dierent distances. For example, we would like to compare the correlators of physical observables separated at dierent distances. Conformal symmetry relates these kinds of observables! To implement a conformal transformation, we need to change the metric via a Weyl rescaling xx g (x)g (2)
which changes physical distances ds2 (x)ds2 . Now follow this by a coordinate transformation such as the one in Eq.(1), so that x x and (x)g g (x ). This takes us to a coordinate system where the metric has the same form as the one we started with, but the points have all been moved around and pushed closer together or farther apart. Therefore, we can view the conformal group as those coordinate transformations which can undo a Weyl rescaling. 1
1.1
Scale transformations
Another useful perspective on conformal invariance, particularly for AdS/CFT, is to think about scale transformations. Any local (relativistic) QFT has a stressenergy tensor T which is conserved if the theory is translation invariant. T encodes the linear response of the action to a small change in the metric: S = T g (3)
If we put in gravity, then T = 0 by the equations of motion. But lets return to QFT without gravity. Consider making a scale transformation, which changes the metric g = 2g S = T 2 (4)
where is a constant. Therefore we see that if T = 0 then the theory is scale invariant.
In fact, since nothing we said above depended on the fact that is a constant, and since our theory is a local theory, we can actually make the following transformation: g = (x)g S = T (x) (5) We conclude from the two equations above that, at least classically,
If T = 0 the theory has both scale invariance and conformal invariance. If the theory is conformally invariant, it is also scale invariant, and T = 0
However scale invariance does not quite imply conformal invariance. For more details please see Polchinskis paper, Scale and conformal invariance in quantum eld theory. The conserved currents and charges of the transformations above are: S = x T
C = (2x x x2 g )T since both S and C are proportional to T .
S0 dd x C0 dd x
(6) (7)
1.2
Geometric Interpretation
Lastly wed like to give an alternative geometric interpretation of the conformal group. The con formal group in Rd,1 is isomorphic to SO(d + 1, 2), the Lorentz group of a space with two extra dimensions. Suppose this space Rd+1,2 has metric ab = diag( + +... + )ab 2 (8)
where the last two dimensions are the extra ones. A light ray in this space can be parameterized by d + 1 dimensional coordinates x in the following way: 1 1 a = (x , (1 x2 ), (1 + x2 )) 2 2 (9)
where is some arbitrary constant. The group SO(d + 1, 2) moves these light rays around. We can interpet these transformations as maps on x , and in fact these transformations are precisely the conformal transformations, as you can check on your own. Invariants in Rd+1,2 should also be conformal invariants, for example 1 a b 1 2 = ab 1 1 = 1 2 (x1 x2 )2 . 2 (10)
This statement is not quite true because a and a are identied with the same x , so is a redundant variable. Conformal invariants actually are cross ratios of invariants in Rd+1,2 , for example 1 2 3 4 . (11) 1 3 2 4 This perspective was lifted from a paper by Callan. It may be useful if you are in the business of relating CFTs to theories in higher dimensional spacetimes, but you are otherwise free to ignore it.
We will now attempt to say something concrete about constraints on a QFT from conformal in variance, namely constraints on correlators. Consider a Greens function G = 1 (x1 )...n (xn ) (12)
where i is a conformal primary. Recall from last lecture that this means satises [D, (0)] = is the dilatation operator dened above. i(0), where is the weight and D
How does (x) transform under D ? Writing the eld as (x) = eiP x (0)e+iP x , and using the following result eiP x De+iP x = D + x P (13)
which follows from the algebra, then [D, (x)] = i( + x )(x). The Greens function thus transforms under the action of D by: G =
n (xi i (xi ) + 0D i (xi )0 + 0 i (xi )D 0. + i ) xi i=1
(14)
Assuming that the vacuum is conformally invariant, the last two terms above vanish. We will assume this in the rest of the discussion. N = 4 is a good example where there is a conformally invariant vacuum. 3
Now lets ignore what we just did with innitesimal transformations and look at the nite trans formations of the greens function:
n
n
n
x i /d
/2
G =
(15)
i i i (x ) i (x )
i i
x
x=xi
i=1 i=1 i=1
where
2 i = (1 + 2b xi + b2 x2 )2 i
Conformal invariance tells us exactly how the Greens function must transform. Therefore, up to the i factors, the Greens function can only depend on conformal invariants I made from x . What i are the conformal invariants? Assuming s are scalars, lets constrain the conformal invariants by using the symmetries of the theory: Translation invariance implies I depends only on dierences: xi xj . Therefore there are d(n 1) numbers I can depend on. Rotational invariance implies I depends only on magnitudes of dierences: rij = xi xj . There are n(n 1)/2 of these numbers. Scale invariance implies I depends only on ratios of magnitudes rij /rkl Finally special conformal transformations transform rij by
(r12 )2 = 2 r2 r12 = 1/212 1/2 . (1 + 2b x1 + b2 x2 )(1 + 2b x2 + b2 x2 ) 1 2 1 2 rij rkl rik rjl .
(16)
There are n(n 3)/2 of these cross ratios, so only n(n 3)/2 conformal invariants.1 Note that this implies there are no conformal invariants for n < 4, namely the 2pt and 3pt function. This is obvious from the fact that you cannot form cross ratios with only 2 or 3 spatial vectors.
2.1
Twopoint function
In fact lets look at the 2pt function for scalar primaries. Its spatial dependence must be completely determined by the logic above: 1 (x1 )2 (x2 ) = c12 f (r12 ) = c12
1
1
r121 +2
(17)
Ginsparg proves why there are n(n 3)/2 if you are curious.
where the rst equality follows from translation plus rotation invariance, and the second equality follows from scale transformations on both sides. Finally lets apply a special conformal transfor mation to both sides: 1 +2 1 /2 /2 1 1 2 2 1 (x )2 (x ) = c12 1 2 1/4 1/4 r12 1 2 1 (1 +2 )/4 (1 +2 )/4 = c12 2 (18) 1 +2 1 (r )12 However using Eq.(17), we already know what 1 (x )2 (x ) is. From this we conclude either 1 2 1 = 2 , OR c12 = 0 The special conformal transformation was necessary for this result. As an aside, there is an alternate shorter argument by Cardy which gives the same result without using the annoying formula for the special conformal transformation, though we still have to do one. First we note that the twopoint function 1 (x1 )2 (x2 ) depends only on x1 x2 , so we can perform a 180 degree rotation to obtain 2 (x1 )1 (x2 ), which must give the same answer. Performing a conformal transformation to the two Greens functions above gives 1
1 /2
2 /2
1 (x1 )2 (x2 ) = 2
1 /2
2 /2
2 (x1 )1 (x2 )
(19)
So either G12 = 0 or 1 = 2 . Some nal comments on the two point function: for a set of elds i with i = j = i (x1 )j (x2 ) = dij 2 r12 (20)
where dij must be symmetric by the argument of Cardy. If we assume the operators are hermitian then dij is also real, and then we can diagonalize dij . In a unitary theory, the diagonal elements of dij are norms of states, and positive. By rescaling i , we can arrive at a basis where dij = ij .
2.2
There are no conformal invariants for the threepoint function either. Using translation and rotation invariance, we can write the threepoint function as Cabc . (21) G123 = 1 (x1 )2 (x2 )3 (x3 ) = ra rb rc a,b,c 12 23 31 Scale invariance then implies all terms on the RHS transform in the same way as the LHS, so a + b + c = 1 + 2 + 3 . You can then check that special conformal transformations give the additional constraint that a = +1 + 2 3 b = 1 + 2 + 3 c = +1 2 + 3 5 (22)
or else Cabc = 0. The permutation symmetry of rij also implies Cabc is symmetric. We cant say anything more. For higher point functions, we can say even less, since there exist conformal invariants formed by the cross ratios, and any correlator contains an arbitrary function of the conformal invariants which cannot be determined by the conformal symmetry. For example, for n = 4, there are 4 1/2 = 2 invariants: 1 r12 r34 r12 r34 (23) G(x1 , x2 , x3 , x4 ) = F , i j /3 r13 r24 r23 r14 i<j rij where = i and F is undetermined function of the two invariants. The results in this section are true for D > 1, and actually for D = 2 there are extra symmetries which allow us to learn something about these undetermined functions of invariants. Theyre often hypergeometric. Also, in principle, there exist similar expressions for nonscalar elds, though John McGreevy has never caught a glimpse of one. He has glimpsed some spectacular things about Wilson loops in this context, though.
Conformal Compactications
Lets take a short detour to conformal compactications. Well nd that understanding this will be quite useful for understanding the stateoperator correspondence, and later for understanding the geometry of AdS. The general goal of such coordinate transformations is to map innite coordinates to nite ones, thus allowing us to understanding the structure of the boundary of spacetime at . In particular wed like to nd coordinates in which the metric is (x)g where the coordinates have a nite range because we can take care of the (x) with a Weyl transformation. Furthermore the presence of the factor of (x) does not aect the casual structure of spacetime spacelike vectors remain spacelike, and the same goes for lightlike and timelike.2 All this talk makes Penrose diagrams sound way more informative than they may seem. Start with the simple example of Euclidean space Rp+1 with metric ds2 = dr 2 + r 2 d2 where d2 p p is the metric on the psphere. Dene r = tan /2. Then the metric is ds2 (d 2 + sin2 d2 ) p 1 4 cos2 /2 (24)
which is the metric on the p + 1sphere with an extra conformal factor 4 cos1 /2 . This extra factor 2 blows up at = or r but we dont care! By making this coordinate change, we turned r to a nite point, thus adding a point to Rp+1 to get S p+1 . It was a onepoint compactication. Moving onto Minkowski Rp,1 with ds2 = dt2 + dr 2 + r 2 d21 , by making the coordinate change p t r = tan the metric becomes 2 ds2 = (d 2 + d 2 + sin2 d2 ) p1 1 4 cos2 + 2 cos2
2
(25)
Figure 1: Penrose diagram of Minkowski space, with angular coordinates on S p1 suppressed. The size of the S p1 shrinks at sin = 0. The analytical continuation of the nite coordinate is indicated by the dashed lines, and gives the Einstein static universe.
which looks like R S p with a conformal factor which blows up at the conformal boundary. has range [, ] and has range [0, ]. The Penrose diagram is shown in Fig(1). The reason we went though all this trouble is that a dierent subgroup of the action of the conformal group is actually obvious in these coordinates. Recall the conformal group of Rp,1 is SO(p + 1, 2). There are two interesting decompositions (subgroups) of the conformal group: SO(p, 1) SO(1, 1) where SO(p, 1) is the Lorentz group and SO(1, 1) corresponds to dilata tions D (dont worry about why). If we think about this decomposition in terms of the higher dimension space Rp+1,2 , the two subgroups act on dierent directions: ( + ) ......+ +
SO(p, 1) SO(1, 1)
(26)
The Hamililtonian is just the usual generator of time translations and has a continuous spectrum because the eld theory lives on Rp , and weve assumed its a CFT. SO(p+1)SO(2) is obvious in the conformal coordinates we discussed above. Here SO(p+1) rotates S p and SO(2) corresponds to translations. In Rp+1,2 , SO(p + 1) rotates the spatial coordinates and SO(2) rotates the timelike coordinates. SO(2) ( ) +......+ 7
SO(p + 1)
(27)
Figure 2: The state operator correpondence. The cylinder on the left represents S p . Start with an initial state  and evolve in with H . The picture on the right is the space in r coordinates where the red dashed lines are constant slices which are also constant r slices.
The translations are generated by a dierent operator than the one above. Lets call it Hglobal since is called the global time. In fact, its actually equal to 1 Hglobal = (C0 + P0 ) = J0,p+2 2 (28)
where C0 is 0 component of special conformal generator, and J0,p+2 is the generator in the space Rp+1,2 . This time spatial sections of this theory are S p which implies Hglobal has a discrete spectrum! One lesson from this is that its dicult in a CFT to agree on what the Hamiltonian is, because we have a bunch of charges to pick from. The physics looks dierent in dierent coordinates though they are of course related by conformal transformations.
StateOperator Correspondence
Consider a QFT on R S p, which means time a compact space to avoid some IR issues, and so there is a ground state separated by a gap from the rst excited state. The metric is ds2 = d 2 +d2 . p Make a (large) change of coordinates r = e , then the metric is ds2 = (dr 2 + r 2 d2 )/r 2 which is p at space with a conformal factor. 3 The statement of the stateoperator correspondence is that given an intial state on the cylinder (S p ) , we can make a conformal transformation that squashes it all to the origin on the plane (Rp+1 ). Thus it is a local perturbation on the plane, and there is a corresponding local operator O (0) at the origin. We can also go from local operators on the plane back to initial states on the cylinder: O = lim O(x)0 (29)
x0
Therefore in a CFT there is a correspondence between local operators and states. See Fig(2).
3
OK, we cheated for the sign of d 2 in ds2 but it just means using a slightly dierent coordinate transformation.
In this session, we are going to talk about the following topics. 1. We are making a few comments about CFT. 2. We are discussing spheres and hyperboloids. 3. Finally we are focusing on Lorentzian AdS and its boundary.
1
1.1
Conformal Symmetry
Weyl anomaly
Quantumly, conformal symmetry in a curved space (with even number of dimensions) could be anomalous, that is ds2 (x)ds2 could be no longer a symmetry of the full quantum theory. This anomaly can be evaluated from the following diagram with operator T inserted at the left vertex.
The conformal anomaly signals a nonzero value for the trace of the energymomentum tensor. In a curved spacetime, it is related to the curvature:
T RD/2
where R denotes some scalar contractions of curvature tensors and D is the number of spacetime dimensions; the power is determined by dimensional analysis. c (2) R (1) 12 where R(2) is the Ricci scalar in two dimensions and c is the central charge of the Virasoro algebra of the 2d CFT. Also in D = 4, the anomaly is given by T = aW + cGB where W and B are dened as 1 (2) W = (Weyl tensor)2 = R.... R.... 2R.. R.. + R2 3 GB = Euler density = R....R.... 4R.. R.. + R2 ; (3)
T =
where (M) is the Euler character of M. c and a are central charges which are proportional to the number of degrees of freedom of CFT. In D = 2 the central charge can be dened even away from critical theories as follows: c = lim z 4 T (z)T (0) .
z0
The GaussBonnet tensor is the Euler density in the sense that GB = (M) = (1)p bp (M)
M
(4)
The Zamolodchikov ctheorem says that this quantity is equal to the c dened above when eval uated at a RG xed point, and is monotonically decreasing under RG ow. In four dimensions, a longstanding conjecture of Cardy suggests that a should decrease under RG ow, but there are now some counterexamples 1 .
1.2
OPE
Any local QFT has an Operator Product Exansion (OPE). The idea is that any local disturbance is created by local operators. Consider some correlation function of local operators Oi (xi ) ; focus on one local operator O1 (x) inserted at point x and there is another one O2 (y) at y. Consider an imaginary line around them and suppose that x y < which is their distance from nearby operators. Then we can squint the local disturbance by a superposition of local operators O1 (x1 )O2 (x2 ) = cn On (y) (5) 12
n, all operators with same quantum numbers
The coecients are independent of the other operators in the correlator. The coecients are in general hard to know. In a CFT, we can plug (5) into a threepoint function and we nd that the OPE coecients are determined to be cn (x y) = 12 cn 12 x y1 +2 n (6)
1.3
Conformal Dimensions
Lets give a few examples of operators with denite conformal dimension: a) Energymomentum tensor: The conformal dimension is = D which is guaranteed by di mensional analysis. The energymomentum tensor must be coupled to g by denition, and the metric components are dimensionless. There are other more algebraic arguments for this (use the Conformal Ward identities to constrain the OPE of the stress tensor with itself). b) For a global symmetry, there is a conserved current j which has = D 1. The current can be coupled to a gauge eld in which case these is a gauge invariance that guarantees the conservation of the current, that is j = 0, and xes the dimension. In general, the conformal dimensions are very hard to know, however there are some lower bounds on them in unitary CFTs. The lower bound constrains the possible values of the conformal dimension according to their spins. A few examples of this is as follows (for a full analysis see hepth/9712074) D2 = free eld dimension 2 D1 = free eld dimension spin 1/2 operaor 2 spin 1 D 1 = conserved current dimension scalar operator In the last equation, spin 1 is meant to be in ( 1 , 1 ) representation as opposed to (1, 0). 2 2
1.4
Thermodynamics of a CFT
As a last remark on CFT we discuss the thermodynamics of a CFT. The partition function is dened as ZCF T = TrCF T (exp(H/T )) . In the thermodynamic limit, ln Z is proportional to the volume of the space. ln Z is a dimensionless quantity. Hence, we must have ln Z V T d (d is the number of spatial dimensions) in the absence of any other energy scales (such as a chemical potential for some conserved charge). The free energy then will be F = T ln Z = cV T d+1 . where this c is also somehow proportional to the number of degrees of freedom of CFT.
Exploiting simple facts about CFT we can derive some interesting results. We must regard T = 0 as an operator equation in the full quantum theory. So we might be tempted to put it inside Tr(eH/T ). The operator equation then translates into the following equation 0 = Tr(T eH/T ) = T00 Tii = E dP.
(7)
The AdS space has a constant negative curvature. It is actually the most symmetrical space with a negative curvature. The most symmetric (Euclidean) space with a positive curvature is obviously a sphere. A useful and immediate description of spheres and their metrics arises by embedding in a higher dimensional space. Below we will use the same logic to investigate the AdS space, but as a warmup we start with a sphere
d+1 S ={ x2 = L2 } IRn i d i=1
(8)
with the at metric ds2 = d+1 dx2 . Note that the dening equation of the manifold respects the i i=1 symmetries of the ambient space. That is, under the transformation xi i j xj for SO(d+ 1), the manifold will be mapped to itself. In other words, the embedding is isometric. We can solve equation (8) to nd a set of global coordinates. In two dimensions, for example, we nd the familiar spherical coordinates x1 = L cos cos , x2 = L cos sin , x3 = L sin The metric then would look like ds2 2 = L2 (d 2 + cos2 d2 ). S
2.1
Our next goal is to describe the (Euclidean) hyperbolic space in a higher dimensional space. In three dimensions this is like the familiar hyperboloid dened as {x2 y 2 z 2 = R2 } IR3 , ds2 = dx2 + dy 2 + dz 2 (9)
However, it will more useful to embed hyberbolic space in Minkowski space; this is because the locus (9) doesnt respect the SO(3) symmetries of the ambient IR3 metric. So instead, we dene ddimensional hyperbolic space to be the locus
2 Hd = {Xd+1 + d i=1
Xi2 = L2 } IRd,1
(10)
2 ds2 = dXd+1 +
d i=1
dXi2
The manifold dened by equation (10) has two disconnected branches with Xd+1 1 and Xd+1 1. Here we will only consider the connected subspace with Xd+1 1. We can easily see that this space is spacelike. The argument goes as follows. Dene vector v A = (xd+1 , x). This is a timelike vector for any x Hd , for (v A )2 = L2 . suppose that v A is the tangent vector to Hd at (xd+1 , x). Therefor we have (v A + v A )2 = L2 v A .v A = 0 which means that v A is spacelike (otherwise we have two timelike orthogonal vectors). Note that the induced metric on Hd will manifestly preserve the SO(d, 1) symmetry of the ambient Minkowski space. This space is actually homogenous, in the sense that any point can be mapped to any other by an SO(d, 1) transformation. In this case, we can choose the set of global coordinates as xd+1 = L cosh , xi = L sinh i , Using
d i=1
i 2 = 1.
i 2 = 1 and thus i di = 0, we nd the induced metric on the hyperbolic space ds2 = L2 (d2 + sinh2 di 2 ) = L2 (d2 + sinh2 d2 ) p
We can also dene the following coordinates which conformally compactify the space (i.e. map the d innite coordinate range of to a nite range in the new coordinate). Dene d = sinh from which we nd that tan(/2) = tanh(/2). The metric then would take the following form ds2 = L2 (d2 + tan2 (/2)d2 )(cosh2 )
This space is called Poincar disk and it is topologically a ball whose boundary is H = S d1 . The e distance from some point in the interior to the boundary is innite.
Lets also dene the following subset of the manifold S() = {p H, (p) < }. The curious fact is that
lim
1 Area(S()) . Volume(S()) L
We can also dene the following coordinates (UHP coordinates). Lets dene u = Xd Xd+1 , The equation (10) then becomes L = The ambient metric also becomes ds2 = dudv + dX 2 . Figure 3: The Poincar plane in the global coordinates and in the Upper Half Plane coordinates. e
2 2 Xd+1
v = Xd + Xd+1 .
2 Xd
d1 i=1
Xi2 = uv +
X 2,
We can now solve for a variable, say v, in the previous equations. Lets also redene X as x = The metric in these new coordinates is ds2 Hd du2 u2 dx2 = dudv + dx2 = L2 2 + L2 2 u 2 dz + dx = L2 z2
X u L.
(11) (12)
where in the last line, we introduced the new variable z = R2 /u. Note that z > 0. These last coordinates dene the Poincar Upper Half Plane (UHP). e
2.2
Lorentzian AdS
Now we are in a position to focus on the problem which is really of our interest. This is the Lorentzian AdSp+2 . Lets rst dene the signature that we will need in this case. The embedding space is Rp+1,2 with metric ab = diag( + + + ). The metric is then
p+1 s
ds2 = dx2 + 0
a p + 1 vector
dx2
dx2 p+2
(two times!)
{ab X a X b = L2 } IRp+1,2
(13)
Despite the appearance of two times in the denitions, this space has really only one time direction. The argument is the same as above. (Note that the vector orthogonal to the AdS in IRp+1,2 is timelike). The symmetries of the AdS space is apparent from equation (13). It has a SO(p + 1, 2) symmetry and it is also homogenous. Lets dene the following set of coordinates (called global coordinates) X0 = L cosh cos , Xi = L sinh i , Xp+2 = L cosh sin where p+1
i=1
This metric has a visible SO(2) isometry under which goes to + a. However is supposed to be the time direction. Near = 0, for example, the spacetime looks like
p+1 S1 IR
timelike
So in order to avoid closed timelike curves and to make any sense of the theory, we should unwrap , that is, we should instead use the covering space of the manifold that doesnt identify and + 2.
Today: 1. the boundary of AdS 2. Poincar patch e 3. motivate boundary value problem 4. wave equation in AdS.
(1)
1.1
Projective boundary
Take a solution V = X0 , X, Xp+2 of equation (1). Reach the boundary by rescaling X, preserving (1). Let X = X , then equation (1) becomes ab X a X b = We now take , the boundary is {ab X a X b = 0} / {X X } IRp,1 1 (4) L2 2 (3)
Figure 1: Lorentzian AdS: The leftright axis is the direction. At = 0, the S p in the lower gure shrinks to zero size (like sinh ), while the radius of the direction, depicted in the top gure, approaches a constant (like cosh ).
This relation can also be read as follows: the boundary of AdS is the set of lightrays in IRp+1,2 , modulo the rescaling. Recall that this is exactly parametrized by points in IRp,1 as: 1 1 a = X , (1 X 2 ), (1 + X 2 ) . 2 2 (5)
We used this fact earlier to make write the SO(p + 1, 2) action of the conformal group on IRp,1 in a linear way. The fact that the conformal group of IRp,1 has a nice action on the boundary of AdS is very encouraging.
2 2 X0 + Xp+2 = X 2 = 1
1.2
Let d = results in
(this variable was called squiggle in lecture). The metric in these new coordinates ds2 = cosh2 d 2 + d2 + tan2 d2 p 2 (7)
and therefore
= tanh [0, /2] (8) 2 2 The boundary is { = /2} IR S p . Note that the metric on the boundary is only specied up tan
Figure 2: The squiggle variable runs from 0 to /2 as goes from 0 to to rescaling, i.e. a Weyl transformation.
But why do we care about this boundary more than say the conformal boundary of Minkowski
space? The answer is in the next two subsections.
1.3
Massless geodesics
The massless geodesics are given by the condition ds2 = 0, which implies 0 = ds2 = L2 cosh2 d 2 + d2 cosh = d d d = d = d cosh (9)
is the time elapsed for a static observer. Whether the lightray reects o the boundary depends on the BCs. Hence: Cauchy problem problem.
1.4
Massive geodesics
The action for a massive relativistic point particle is S=m The equation of motion is S =0 X X + X X = 0 (11) ds = m g X X X = X (10)
where the second equation follows if X = s X where s is proper time. If we assume = 0 the action is S = mL d cosh2 ( )2 . You will show on problem set 3 that this has an oscillatory solution around = 0, it never reaches .
Poincar patch e
Pick out X p+1 from among the X i . This will break the SO(p + 1) symmetry of the psphere. Let L X = zx (12) X + Xp+1 = L z p+2 Xp+2 + Xp+1 = v L L2 v 2 x x = L2 z dz 2 + dx dx ds2 = L2 z2
(13)
(same cancellation as UHP). This is the metric which we showed has 1 R g R = g 2 = (p + 1)(p + 2) 2L2 (14)
NOTE: it covers part of AdS. As z , /t becomes NULL (Poincar horizon). e CLAIM: relation between Poincar patch and global time is stateoperator correspondence. e EVIDENCE: symmetries SO(p, 1) IRp+1 and SO(p + 1) SO(2).
2.1
Our goal is to evaluate e 0 O CF T eWCF T [0 ] . R Conjecture: e 0 O CF T = Zstrings in AdS [0 ], but we cannot compute it. The pratical version is the following WCF T [0 ] = lne A few comments: The supergravity description is valid for large N and large . In (15) weve made the N dependence explicit: in units of the AdS radius, the Newton constant is G1 = N 2 . ISU GRA N is some dimensionless action. 5
R 0 O
1 N2
+O
(15)
anticipating divergences at z 0, we introduce a cuto (which will be a UV cuto in the CFT) and set boundary conditions at z = . Eqn (15) is written as if there is just one eld in the bulk. Really there is a for every operator O in the dual eld theory. Well say couples to O at the boundary. How to match? We give four examples 1. Dilaton eld. Before near horizon limit, we have D3branes in IR10 ; the asymptotic value of the dilaton 2 determines the string coupling constant gs = e(x) . The YM coupling on D3s is gY M =
gs .
Changing + we get
Tr F2 + ... (16) S = 2 gs where the dots stand for all the CPeven term in the lagrangian. In conclusion we have
1 2 Zstrings [ + ] e gs
T r[F 2 ]
CF T
(17)
The dilaton couples to all the terms in the lagrangian which are CP invariant. 2. RR axion. We have that str =
i gs
i gs
2 .
Therefore (18)
3. Stress energy tensor. The tensor T is the response of a local QFT to local change in the metric. SQF T Here we are writing for the metric on the boundary. In this case g T
T . (19)
4. IIB in AdS5 S 5 . Isometry on S 5 SO(6) KaluzaKlein (KK) gauge elds SO(6)R = SU (4)R . In this case the correspondence is between these gauge elds and the Rcurrent operators
AKK a JR a
(20)
i.e. Sbdy
Aa Ja
2.2
Useful visualization
Figure 5: Feynman graphs in AdS. We do the one with two ext. legs rst
2 For this metric g = det g = L . z Since is a scalar eld we can rewrite the kinetic term as gAB A B = ()2 = gAB DA DB (23)
where DA is the covariant derivative. Thus we can use DA (gBC ) = 0 to move the Ds around the gs with impunity. By integrating by parts we can rewrite the action as S= 2 dp+2 x A g gAB B A g gAB B + g m2 2 + . . . (24)
and nally by using the Stokes theorem we can rewrite the action as S= 2 dp+1 x g gzB B 2 AdS
1 A g
g + m2 + O(3 )
(25)
where we dene =
g gAB B = D A DA .
g DA J A =
M
nA J A
(26)
i.e. it is the induced metric on the boundary surface z = . The vector nA is a unit vector normal to boundary (z = ). We can nd an expression for it nA z gAB nA nB z= = 1 1 z n= = gzz z L z (28)
In todays lecture we will talk about: 1. AdS wave equation near the boundary. 2. Masses and operator dimensions: ( D) = m2 L2 . Erratum: The massive geodesic equation x + xx = 0 assumes that the dot dierentiates with respect to proper time. Recap: Consider a scalar in AdSp+2 (where p + 1 is the number of spacetime dimensions that the eld theory lives in). Let the metric be: ds2 = L2 then the action takes the form: S[] = 2 dp+1 x g (()2 + m2 2 + b3 + ...), (2) dz 2 + dx dx , z2 (1)
CF T = extremum[  0
at z=] S[],
(3)
where S[] S[ (0 )] W [0 ], i.e. by using the solution to the equation of motion subject to boundary conditions. Now Taylor expand: 1 D W [0 ] = W [0] + d x 0 (x)G1 (x) + dD x1 dD x2 0 (x1 )0 (x2 )G2 (x1 , x2 ) + ... (4) 2 where W  =0 , 0 (x) 0 2 W G2 (x) = O(x1 )O(x2 )c =  =0 . 0 (x1 )0 (x2 ) 0 G1 (x) = O(x) = 1
(5) (6)
Now if there is no instability, then 0 is small and so is , so you can ignore third order terms in . From last time: p+2 2 2 3 dp+1 x (n.) , d x g [ ( + m ) + O( )] (7) S[] = 2 AdSp+2 2 AdS where the last term is the boundary action, n is a normalized vector perpendicular to the boundary and 1 2 = A ( ggAB B ). (8) g Now if the scalar eld satises the wave equation: (2 + m2 ) = 0, W [0 ] = Sbdy [ [0 ]],
(9) (10)
then we can use translational invariance in p + 1 dimensions, x x + a , in order to Fourier decompose the scalar eld: (z, x ) = eik.x fk (z). (11) Now, substituting (11) into (9) and assuming that the metric only depends on z we get: 1 0 = (g k k z ( ggzz z ) + m2 )fk (z) g 1 2 2 = [z k z D+1 z (z D+1 z ) + m2 L2 ]fk , L2 (12) (13)
where we have used g = (z/L)2 . The solutions of (12) are Bessel functions but we can learn a lot without using their full form. For example, look at the solutions near the boundary (i.e. z 0). In this limit we have power law solutions, which are spoiled by the z 2 k2 term. Try using fk = z in (12): 0 = k2 z 2+ z D+1 z (z D+ ) + m2 L2 z = (k z ( D) + m L )z , ( D) = m2 L2 D = 2 D 2 2 + m 2 L2 .
2 2 2 2
(17)
Comments The solution proportional to z is bigger near z 0. + > 0 m, therefore z + decays near the boundary. + + = D. Next, we want to improve the boundary conditions that allow solutions, so take: (x, z)z= = 0 (x, ) = Ren (x), 0 (18)
where Ren is the renormalized eld. Now with this boundary condition, (z, x) is nite when 0 0, since Ren is nite in this limit. 0 Wavefunction renormalization of O (Heuristic but useful) Suppose: Sbdy = where we have used dp+1 x
D
0 (x, )O(x, )
(19) (20)
z=
D L d x ( Ren (x))O(x, ), 0
(21) (22)
Ren
(x),
where in the last line we have used + + = D. Therefore, the scaling of ORen is + . Comments
1 x2
We had a second order ODE, therefore we need two conditions in order to determine a solution (for each k). So far we have imposed: 1. For z , z 0 + (terms subleading in z). Now we will also impose
1. The factor is independent of k and x, which is a consequence of a local QFT (this fails in exotic examples). 3
2. Relevantness: Since m2 > 0 = + > D, so O is an irrelevant operator. This means that if you perturb the CFT by adding O to the Lagrangian, then: S = dD x (mass)D O , (23) where the exponent is negative, so the eects of such an operator go away in the IR. For example, consider a dilaton mode with l > 0, its mass is given by (for D = 4): m2 = The operator corresponding to this is: tr(F 2 X i1 ... il ), (25) (l + 4)l . L2 (24)
with = 4 + l > D, therefore it is an irrelevant operator. Now consider a dilaton mode with l = 0: then m2 = 0, therefore, = D and hence it corresponds to a marginal operator (an example of such operator is the Lagrangian). If m2 < 0, then < D, so it corresponds to a relevant operator, but it is ok if m2 is not too negative (Breitenlohner  Freedmasn (BF) allowed tachyons with mBF 2 (D/2L)2 < m2 ). 3. Instability: This occurs when a renormalizable mode grows with time without a source. But in order to have S[] < , the solution must fall o at the boundary. This requires a gradient 1 energy that L . Note: D D 2 + m 2 L2 . (26) = 2 2 D 2 ) mBF 2 , (27) 2 then is complex, therefore we have = D/2, which is larger than the unitary bound. In this case, z decays near the boundary (i.e. in the UV). In order to see the instability that occurs when m2 L2 < ( D )2 more explicitly, rewrite (9) as a Schrodinger equation, by 2 writing (z) = A(z)(z), where we choose A(z) in order to remove the rst derivative of (z). Then, equation (9) becomes: m 2 L2 < (
2 (z + V (z))(z) = E(z),
If:
(28)
where E = 2 k2 , V (z) = /z 2 and = m2 L2 (D2 1)/4. An instability occurs when E < 0, i.e. 2 < 0 and hence eit (z) = e+t (z) grows with time. Now the claim is that V = /z 2 has no negative energy states if > 1/4. Note that the notion of normalizability here and before are related (Pset 4): 2 = dz < , (29) and S[] = dz g (()2 + m2 ) (30) 4. The formula we found before (expression (16)) depends on the spin. For a jform in AdS we have: ( + j)( + j D) = m2 L2 . (31) 4
For example, for A massless we have: (j ) = D 1 conserved, for g massless we have: (T ) = D required from CFT. (33) (32)
References
Witten, hepth/9802150 GKP, hepth/9802109
(1) (2)
p), z
where p is some point on the boundary, then the eld in the bulk [0 ] (z, x) = dD x Ren (x )Kx (z, x) z Ren (x) 0 0 solves (1).
Euclidean AdS
Recall the metric on AdS with curvature scale L in the upper half plane coordinates: ds2 = L2 dz 2 + dx2 z2
(iii) The limit of the Greens function as z 0 , i.e. the boundary is cz + 0, if x = x K(z, x; x ) cz + , if x = x 2
(recall that + > 0 for any D, m). More specically, the Greens function approaches a delta function: K(z, x; x ) const D (x x ). Clearly it has support only near x = x , but to check this claim we need to show that it has nite measure: c+ dD x K0 (, x) = dD x 2 ( + x2 )+ cD 2+ D 1 =
dD x 2+ (1 + x2 )+ 2 (+ = c (+ )
D
D 2)
We will choose the constant c to set this last expression equal to one. Hence, [0 ] (z, x) = dD x Kx (z, x)Ren (x ) 0
x + =
dD x c 2 Ren (x ) ; (z + (x x )2 )+ 0 this solves (1) and approaches Ren (x) as z . 0 The action is related to expectation values of operators on the boundary: R S
[0 ] = lne 0 O = n
2
AdS
D zz = d x g g (z, x)z (z, x)
2
z=
= dD x1 dD x2 Ren (x1 )Ren (x2 )F (x1 , x2 ) 0 0 2 where the ux factor is
F (x1 , x2 ) dD x K(z, x; x1 )zz K(z, x; x2 )
.
D z
z=
c
+ O(2 ) (x x )2
We must be careful when evaluating the cases x1 = x2 and x1 = x2 , which we do in turn. Firstly, if x1 = x2 : G2 (x1 = x2 ) = = = D + cz + D D 2 2 d xz z (x x1 ) + O(z ) (ignore by x1 = x2 ) + + O(z ) 2 z= (x1 x2 )2+ 1 c+ D+ ++ + O(2 ) 2 (x1 x2 )2+ c+ . 2(x1 x2 )2+
Good. This is the correct form for a two point function of a conformal primary of dimension + in a CFT; this is a check on the prescription. Secondly, if x1 = x2 : G2 (x1 , x2 ) = 2 D D (x1 x2 ) +
c+ + + c2 2+ D (x1 x2 )2+
1 d x 2+ (x x )2+ (x x1 ) 2
D
As 0, the rst term is divergent, the second term is nite, and the third term vanishes. The rst term is called a divergent contact term. It is schemedependent and useless. Remedy: Holographic Renormalization. Add to Sgeometry the contact term S = Sc.t. = dD x 2 D (Ren (x))2 0 2 bdy 2 = (z, x). 2 AdS,z= Note that this doesnt aect the equations of motion. Nor does it aect G2 (x1 = x2 ).
At boundary:
K (n) n Hence fk (z) = AK z D/2 K (kz) z 2 = z , as z 0.
D
a0 + a1 n + a2 n + . . . +
Introduction
In todays lecture we will continue to carry out the calculation of OO in momentum space. Along the way, we will deal with some issues that arise when we work with a real time coordinate (i.e. with a Lorentzian metric), which shows up for modes with k2 < 0. After a short discussion of which scaling dimensions of operators (namely ) are allowed from the bulk point of view, we will start our discussion on 3point functions. The following material will be relevant: 1. dHokerFreedman, hepth/0201253, 8. 2. KlebanovWitten, hepth/9905104.
We continue considering the equation, 0 = [z D+1 z (z D+1 z ) m2 L2 z 2 k2 ]fk (z) from last lecture. If k2 > 0, that is spacelike (or Euclidean), the solution would be, fk (z) = Ak z D/2 K (kz) + AI z D/2 I (kz) where = D/2 = horizon as, (D/2)2 + m2 L2 . Since each Bessel function shows denite behavior at the K (z) ekz I (z) ekz ()
we see that the regularity in the interior uniquely xes fk and hence the bulktoboundary prop agator. Actually there is a theorem (the GrahamLee theorem) exactly addressing this issue for 1
gravity elds which states that if you specify a Euclidean metric on the boundary of a Euclidean AdS (which in their formalism, would be a S D ) modulo conformal rescaling, the metric for the space inside of the S D , whose topology would be that of AdS, would be uniquely determined.1 A similar result holds for gauge elds. In contrary to this, in Lorentzian signature with timelike k2 , i.e. onshell states with 2 > k2 , there exist many normalizable solutions with the same leading z behavior. If we dene q = 2 k2 , K (iqz) eiqz (z ) so these modes behave like oscillating modes near the poincare horizon. A better basis of writing the solution (one that is more convenient in dealing with this) is, fk (z) = A1 z D/2 Y (qz) + A2 z D/2 J (qz) where the two terms behave at z 0 as z (0 ) + z + and z + respectively. These are purely normalizable. The ambiguity of the propagator for k2 < 0, is equivalent to the ambiguity in the coecient A2 . We see that now we can have dierent subleading terms in the propagator. But actually this is to be expected, as this corresponds to dierent choices of quantum states of the boundary QFT (Balasubramanian, Kraus, Lawrence and Trivedi, hepth/9808017.) This is actually exactly the same statement as saying that adding dierent homogeneous solutions to propagators in a free QFT corresponds to the correlator of dierent states. As usual, the Wick rotation from (1) the Euclidean answer K (kz) gives the timeordered Feynman propagator H (qz). One last thing we must deal with before proceeding is to dene what we mean by a normalizable mode, or solution, when we say that we have many normalizable solutions for k2 < 0 with a given scaling behavior. In Euclidean space, is normalizable when S[] < . This is because when we are thinking about the partition function Z[] = eS[] , modes with boundary conditions which force S[] = would not contribute. In real time, we say is normalizable if E[] < where, E[] = dD1 xdz hT []n =
x0 =constant
dD1 xdz hT 00 []
where is a given spatial slice, h is the induced metric on that slice, n is a normal unit vector to and is a timelike killing vector. TAB is dened as, 2 SBulk [] TAB [] = g gAB
We return to considering spacelike k in this section. The normalized solution at z = is given by the condition fk (z = ) = 1. This means that the function in x space is given at z = , so the
There is a subtlety here as the specied metric on the boundary sphere needs to be in some nite region around the round metric. Witten provides some conjectural understanding of this region in terms of the conformal coupling of the N = 4 scalars to the boundary metric on page 11 of hepth/9802150.
1
bulk to boundary propagator would be dierent from the one we have obtained in the last lecture. And indeed the position space greens function obtained, fk (z) = z D/2 K (kz) D/2 K (k)
is dierent from KW itten obtained previously. The general position space solution can be obtained as, [0 ] (x) = so the action would be, S[] = 2 = 2 = 2 dD x n dD x dD k1 dD k2 e(k1 +k2 )x 0 (k1 , )0 (k2 , )z D fk1 (z)zz fk2 (z) dD keikx fk (z)0 (k, )
We note that if you dont like functional derivatives, you may see this by calculating, O(k1 )O(k2 )
c
Now F (k) could be calculated to be, F (k) = z D fk (z)zz fk (z)z= + (k k) = 2D+1 z The near boundary behavior of K for integer is, K (u) =u (a0 + a1 u2 + a2 u4 + ) +u ln u(b0 + b1 u2 + b2 u4 + ) z D/2 K (kz) D/2 K (z) z=
where the coecients of the series ai , bi depend on . For noninteger , there would be no (ln u) multiplied in the second line. F (k) = 2D+1 z = 2D (kz)+D/2 (a0 + ) + (kz)+D/2 ln kz(b0 + ) (k)+D/2 (a0 + ) + (k)+D/2 ln k(b0 + ) z=epsilon
where (I) and (II) denote the rst and second group of terms of the previous line. 3
(I) is a Laurent series in with coecients which are positive powers of k (i.e. analytic in k at k = 0.) These are contact terms, i.e. short distance goo that we can subtract o. We can see this by writing, dD keikx (k)2m D = 2mD m D (x) x for m > 0. The 2mD factor reinforces the notion that , which is an IR cuto in AdS is a UV cuto for the QFT. The interesting bit of F(k) which gives the x1 = x2 behavior is nonanalytic at k = 0. (II) = 2 b0 2 k ln(k) 2D (1 + O(2 )), a0 b0 (1)1 = 2 for Z a0 2 ()2
The claim is that the fourier transformation of the leading term of (II) is given by, dD keikx (II) = 2(+ ) 1 2 D/2 (+ D/2) x2+
We note that for Z obtaining the previous result is a lot more transparent as O(k)O(k) k2 1 and hence in position space this would be x2 since there are no logs concerned. Also, since 2D = if we let 0 (k, ) = Ren (k) as before the operation, 0 = Ren 0 (k, ) 0 (k) removes this factor. We also see that for 0, the O(2 ) terms vanish. One last thing we must touch upon is the dierence between the prefactors obtained in this lecture and the last, namely, OO = OO W itten We claim that the l.h.s. is the correct correlator to obtain. This is discussed in dHokerFreedman, hepth/0201253; well come back to this when we talk about threepoint functions.
Now lets think about which are attainable in our setting. Weve seen that in Euclidean space, is normalizable if S[] < . This depends on the z 0 behavior of . For SBulk =
dD+1 x g(gAB A B + m2 2 )
with
and hence, gAB A B + m2 2 2 z 2 + k2 z 2+2 + m2 z 2 = (2 + m2 )z 2 (1 + O(z 2 )) in the limit z 0. Since for = , 2 + m2 = D = 0, SBulk [z ]
1 e2D 2 D
We emphasize that only the boundary behavior of is dened, and it is not assumed that it satises the equation of motion. We see that SBulk [z ] < > D/2 This does not saturate the unitary bound. We note that the calculation of SBulk [z ] can be done by looking at the boundary term, in which case we get, n z e2D
z=
which is o by the same factor in the coecient mentioned at the end of the last section, namely /. Now consider the alternative action by KlebanovWitten which is,
KW SBulk =
dD+1 x g( + m2 ) = SBulk
AdS
n z
is equivalent to
D2 2 which is exactly the unitary bound. We see that in this case both give normalizable modes for 1. Note that it is actually that gives the lowest value in the unitray bound, that is when, 2 D D2 D = + m 2 L2 = 2 2 2 2
m2 =1 D 4
The coecient of z + would be the source in this case. What we have here is a dierent boundary CFT with the same bulk action, which we have obtained by adding a boundary term to the action.
3Point Functions
Next we will talk about connected correlation functions of three or more operators. Unlike twopoint functions, such observables are sensitive to the details of the bulk interactions, and we need to make a choice. We will consider the three point functions of the scalar elds coming from the action 3 1 dD+1 x g( (i )2 + m2 2 + b1 2 3 ) S= i i 2
i=1
The arguments presented could be easily extended to npoint functions with n > 3. The equations of motion we get is, ( m2 )i (z, x) = bj k ijk i We solve this perturbatively to obtain, i (z, x) = +bijk dD x1 K i (z, x; x1 )i (x1 ) 0 dD x dz gGi (z, x; z , x ) dD x1 dD x2 K j (z , x ; x1 )K k (z , x ; x2 )j (x1 )k (x1 ) 0 0
+O(b2 3 ) 0 where Gi (z, x; z x ) is the bulktobulk propagator given by the condition, 1 ( m2 )Gi (z, x; z , x ) = (z z )D (x x ) i g so that ( m2 ) i for a source J. The rst and second terms would be obtained from the Witten diagrams, gure 1 and 2. A typical higherorder diagram would look something like gure 3. gGJ = J
i 0 (x 1)
j 0 (x 1)
Ki (z,x) G
K
i
(z,x) K
k
(z,x)
k(x 2) 0
Intro
This lecture covers: 1. 3point functions. 2. the relationship between the subleading term in the bulk eld solution and the expectation value of the dual operator. 3. bulk gauge elds.
3Point Functions
Lets consider a bulk gravity theory with 3 scalar elds. 3 1 D+1 2 2 2 Sbulk = d x g ((i ) + mi i ) + b1 2 3 2
i=1
The interaction term could be modied with some other couplings, e.g. 2 2 or (1 )2 2 , etc... ; 1 such dierences will modify the details of the following calculation. We want to solve the equations of motion perturbatively in 0 . This could be justied either by small b coupling or by small (0) boundary values i .
This is just like a Feynman diagram expansion; these Witten Diagrams also keep track of which insertions are at the boundary of AdS. D i 0 i (z, x) = d x K (z, x; x )i (x ) + b dD x dz gGi (z, x; z , x ) D d x1 dD x2 K j (z, x; x1 )K k (z, x; x2 )0 (x1 )0 (x2 ) + j k The diagram with three insertions at the boundary wont contribute to the vacuum 3point function, since its contribution to the onshell action will be at least quartic in 0 (and the threepoint function 3 is obtained by acting with 3 on the action and setting 0 = 0). The s that are running around are the weights of the primary operators for the scalar eld insertions. The Gs and Ks are just spatial propagators for our theory. G(z, x; z , x ) is the bulktobulk propagator, dened as the normalizable solution to 1 ( m2 )Gi (z, x; z , x ) (z z ) D (x x ) i g which is otherwise regular in the interior of AdS. The bulktoboundary propagator K is dened as: 1 K(z, x; x ) lim G(z, x; z , x ), n 0 z where is the boundary metric and is the outward pointing normal at the boundary. The n bulktobulk propagator and bulktoboundary propagator are related by1 : K (z, x; x ) = lim G (z, x; z , x ). z 2 D
Of course, these have actual, real live expressions that can be put in terms of hypergeometric functions. Just in case you ever need them: +1 1 G (z, x; z , x ) = c F1 , ; + 1 , 2 , 2 2 2 = c = z 2 + z 2 + (x x )2 , geodesic distance in AdS, 2zz 2 () . (2 D) D/2 ( D/2)
To compute 3point functions, we plug into the onshell action S[]. After an integration by parts, the action becomes: S[] = =
1
1 2
i=1
D+1
b x ( gi i ) 2 I +
dD+1 x g1 2 3 + c.t. II
The proof of this relation follows from Greens second identity Z Z ` ( m2 ) + (( m2 ))) = (n + (n ))
U U
with = G, = K.
The rst term (I) vanishes by properties of the bulktobulk propagator; this is true in general for n 3. This is glossed over in many discussions of this calculation. We will not prove it directly, but it follows immediately from the result in 3 below. Anyway, the bulk term is nonzero and thats what well compute. b II = 2 d xdz g
D
3 i=1
K i (z, x; xi )0 (xi ) i
So with this in hand we can nd the 3point function by functional dierentiation using the GKPW formula. R R e 0 O = e SU GRA S[0 ] O1 (x1 )O2 (x2 )O3 (x3 )0 =0 =
3
0 (xi ) i i=1
(II)0 =0
i dD xdz g K (z, x; xi )
i=1
To see the structure of the 3point function, we do some work on this integral. First, lets relabel 2 x x x x the coordinates. Let wA = (z, ), so w0 = z and w = . Well also dene (w )2 w0 + (w )2 . In this case, x0 0. With these new coordinate labels, the bulktoboundary propagator becomes: w0 Kx (w) = . (w )2 x Now we do a change of variables, inversion. Let wA = two claims: 1. dD+1 w
wa , w 2
2. Kx (w) = x 2 Kx (w ). This is how we found K. From these, we can nd the transformation of the 3point correlator. Gfrom II (xi ) = 3 Some more claims: 3
3 i=1
x 2 Gfrom II (x ) i 3 i
1. This is the correct transformation of a CFT 3point function under inversion (large conformal transformation), denoted I. 2. Translation invariance is clear Tb : x x + b . We simply redene the integration variable i i w = w b to remove the b dependence. 3. Special conformal invariance = ITb I, so thats good. 4. This must be of the required form, since rotational invariance is clear. G3 = cijk ij i>j (xi )
This is determined up to cijk . To nd cijk , need to do integral. See [DZF]. Translate 3 to 0. x Then only two denominator factors in G(xi ), and use Feynman parameters. npoint functions proceed quite similarly. The only new complication is that in general one must evaluate some Witten diagrams with both bulktoboundary and bulktobulk propagators (which dont go away).
Expectation Values
Next we make a valuable observation about expectation values & the classical eld. The reference is KlebanovWitten, hepth/9905104. The solution in response to a source 0 at the boundary (in some state) is: 0 [ ] (z, x)] lim 0 (x) + O(2 ) + + A(x) + O(2 )
z
In this case 0 is the source. The function A(x) is a normalizable uctuation, determined by the source and choice of the propagator (not just to linear order in 0 ). CAREFUL: The term O(2 ) in the rst term can be larger than the A(x) in the second term! Use caution when calculating and expanding! The claim is that: A(x) =
R 0 1 1 O(x)0 = O(x)e O CF T 2 D 2 D
0 0 OOO +
allows us to compute all other correlators. So this formula circumvents the need for the onshell action S[], and applies not just in the Euclidean case, but also in the realtime case. Well give a diagrammatic proof. The value of the bulk eld in response to some source can be represented perturbatively by the following collection of diagrams:
(z, x) =
dz dD x G (z , x ; z, x)BLOB(z , x )
So we bring all our sources at the boundary into some blob, then connect the blob to our eld using the bulktobulk propagator at (z , x ). On the other hand, by the GKPW formula, the expectation value of the dual operator takes the form:
O(x)0 =
GKPW
dz dD x K (z , x ; x)BLOB(z , x )
Now we use our fact about the relation between the bulktobulk and bulktoboundary propagators. G (z, x; z , x ) lim Putting this in, we nd the relation lim (z, x) = dz dD x
z
K (z, x; x ) z 2 D K (z , x ; x)BLOB(z , x ) 2 D
O(x)0 2 D
To really work, we need the functional derivatives to be away from the support of the sources, e.g. take the 0 to be functions. But given that, this shows our claim. i 5
Gauge Fields
This last bit is the beginning of the discussion on bulk, massive gauge elds. Consider a CFT with a conserved current Ja , e.g. for N = 4 SYM there is a U (1) SO(6). This leads to massless gauge elds in AdS. Things coupling to conserved currents are massless. See pset #4. We can imagine a term at the boundary of the form S Aa Ja .
(AdS)
Ja
= 0 = D 1 m2 = 0 A
This allows us a nice check on the 2point function of (charged) scalar operators (due to Freedman et al hepth/9804058) as follows. If under the U (1) the scalar operators transform like: O = iO, and O = iO , then the 3point function
O (x1 )O (x2 )J (x3 ) G 123
The Ward Identity for a conserved current is: 0 = [Df ields]eS O (x1 )O (x2 ) . The change in the action is: S = J (x)(x).
Putting this all in the Ward Identity will give us something interesting. Lets dene xij xi xj , as it will come in handy. D 0 = d x3 J (x3 )(x3 ) O (x1 )O (x2 )
+ (i(x1 )O (x1 )) O (x2 ) + O (x1 ) (i(x2 )O (x2 ))
This is true for a general (x), and in particular (x) = D (x x3 ). O (x1 )O (x2 )J (x3 ) = i((x13 ) (x23 )) O (x1 )O (x2 ) x3
For a CFT, the form of a two scalar and one vector 3point correlator is constrained to be of the form: 1 x13 x23 1 O (x1 )O (x2 )J (x3 ) = c 2D+2 cS (x1 , x2 , x3 ). 2 x2 D2 D2 x13 x12 x13 x23 23 6
Here, c is some constant that needs to be calculated. This gives as the divergence:
2 D/2 1 O (x1 )O (x2 )J (x3 ) = i((x13 ) (x23 ))c 2 ) x3 (D/2) (x12
On the gravity side of things, we just need to include a new bulk vector eld, and have a complex scalar minimally coupled to our gauge eld. 1 AB D+1 AB 2 2 Sbulk = d x g FAB F + g (A + iAA ) (B iAB ) + m  2 4 This is the usual charged scalar eld theory with our vertex now dependent on the metric, rather than the usual at case were used to.
The threepoint function of interest gets a contribution from a single Witten diagram:
= i
The K (w, x)s inside the parentheses are the usual scalar bulktoboundary propagators, while the KA outside is the bulktoboundary propagator for the gauge boson.
M KA (w, ) = cD x D w0 1 J M (w ) x (w )2(D1) A x xA xM M M JA (x) A 2 2 x x A 2 xA = x with xA = 2 xM x
M This JA is the Jacobian for the inversion transformation. This propagator solves the bulk Maxwell x equations and D (w ) as w0 0.
[to be continued] 7
Introduction
1. Finish OO J 2. Anomalies 3. Wilson loops
Consider a CFT with a conserved U (1) current J. The operator O is an operator in the CFT with scaling dimension that couples to a charged scalar eld in the bulk, while the complex conjugate of this bulk scalar eld couples to O . Suppose that there is a massless vector eld A in the bulk that couples to J . The minimal coupling of this charged scalar eld to the vector eld leads to a vertex (ig AB (w)), which gives the leading contribution to the threepoint function
G O (x1 )O (x2 )J (x3 ) 123 D d wdw0 AB = (ig (w)) K (w 1 ) B K (w 2 ) KA (w 3 ). x x x D+1 w w0
(1)
The doublearrow means take the derivative acting to the right minus the derivative acting to the left. K is the bulktoboundary propagator for the charged scalar eld and KA (w 3 ) is the x bulktoboundary propagator for the gauge boson. The bulktoboundary propagator for the vector eld solves the bulk Maxwell equations and goes to a delta function at the boundary (it goes to (w ) as w0 0). The Witten trick that we used to deduce the bulktoboundary propagator x for the scalar eld can also be used to obtain the bulktoboundary propagator of the vector eld. We claim that it implies:
D1 w0 J B (w ), x (w )2(D1) A x
B KA (w ) = CD x
(2)
B were CD is some normalization constant and JA is a Jacobian for inversion it is given by B B JA (x) A 2
xA xB , x2
A
(3)
x x B which can be determined from JA = x2 xA , with xA = (x )2 . This Jacobian arises because for the B Witten trick, one rst nds the bulktoboundary propagator with the singularity at innity and then performs an inversion to put the singularity at some nite value.
Using the inversion trick and doing the w integral yields G = S (x1 , x2 , x3 ) 123 ( D/2)(D/2)() . D/2 ( D/2) (4)
S is the unique combination of x1 , x2 , and x3 allowed by conformal symmetry and is given by 1 x13 x 1 23 S (x1 , x2 , x3 ) = 2D+2 2 . (5) x2 x23 xD2 xD2 x12 13 13 23 Using the Ward identity, G = i((x13 ) (x23 ))OO , x 123 3 we can determine the normalization of the twopoint function: OO = 2 D () 1 . D/2 ( D/2) x2 12
2D
(6)
+ Note the subtlety of the extra factor of 2D = + , which occurs only for twopoint functions because the dz is dicey. For higher point functions, this issue does not arise. This factor comes from the following. Recall our expression for the twopoint function: b ( + + a0 ) zz (z D/2 K (kz)) + 0 z= = b D/2 K (k) + ( a0 ) +
0
Gz =
= (diverging term +
+ b0
a0
(+ )).
(8)
b K is a Bessel function and we have used K (k) = + a0 and = D . The dependence 2 0 on k is absorbed into b0 /a0 . The divergence corresponds to some kind of contact interaction that we dont care about. Because of this divergence, it is possible for the subleading term to actually matter. This is where the coecient + comes from. Wittens positionspace calculation takes the limit z 0 before taking this ratio, so this factor does not appear in his positionspace calculation.
Scale invariance and the SU (4)R symmetry of the N = 4 Super YangMills theory are exact symmetries if the theory is dened on at space with no sources. But, if we couple the SU (4)R 2
currents to some external gauge eld A (which we can think of as the boundary value of some bulk gauge eld in AdS) or if we have an external background metric (which we may think of as the boundary value of the metric in AdS), then there can be anomalies. This is because the gauginos 4 (the s) are in the 4 = of SU (4)R . So we may have anomalies in the trace of the stress tensor, T , or in the Rcurrent. The parityodd part of the threepoint function of three Rcurrents in the N = 4 SU (N ) theory is: N2 1 Tr (5 x12 x23 x31 ) / / / idabc (9) 32 6 x4 x4 x4 12 23 31 The subscript refers to the parityodd bit of the correlator and dabc 2T r(Ta {Tb , Tc }). This threepoint correlation function implies that if we couple the Rcurrent to some background gauge eld, then the divergence of the Rcurrent will be:
Ja (x1 )Jb (x2 )Jc (x3 ) =
(10)
Now we would like to see how to compute this anomaly at strong coupling using the gravity dual. In AdS5 , Ja couples to some SO(6) KaluzaKlein gauge eld Aa . The statement of the anomaly is that the boundary symmetry, generated by the R current J, is broken by the nonzero eld strength of the background gauge eld. This global Rsymmetry is generated by gauge transformations in the bulk, Aa Aa + (D)a , where the gauge parameter does not vanish at the boundary. The eective action for Type IIB supergravity includes SKK SY M [A] + SCS [A], where SCS is a ChernSimons term: iN 2 SCS [A] = 96 2 T rA F F +
AdS5
(11)
(12)
Note that T rA F F = d5 xdabc B1 B2 B3 B4 B5 AB1 FB2 B3 FB4 B5 . The ChernSimons term arises because Type IIB supergravity breaks parity. The basic reason for this paritybreaking is that the veform eld strength is selfdual, not antiselfdual. This ChernSimons term is gauge invariant in the bulk; i.e. it is gauge invariant up to a total derivative. 2p+1 SCS T rF F, (13) A so under A A + D, the change in the action is iN 2 b c SCS = d4 x dabc a F F . (14) 384 2 AdS Here we have assumed dF = 0, or that there is no magnetic charge. Alternatively, since A couples to the Rcurrent J, Sef f = d4 xD a Ja = d4 xa D Ja , (15) so can read o the anomaly (Du J )a . Notice that the coecent matches at large N (ignore 1 in N 2 1). The 1/N 2 correction has also been matched. 3
Wilson Loops
Now we turn to correlation functions of nonlocal operators, namely the Wilson loop operator, which is an important and interesting gauge theory observable. Let C be a closed curve in spacetime and R a representation of the gauge group. The Wilson loop is: W (C) T rR P ei P denotes path ordering of the exponential: W (C) =
H
C
(16)
(i)
(17)
n=0
ti tj i=1
Physically, this describes the YangMills contribution to the propagation amplitude of an external, heavy, charged particle in the reprsentation R. It describes a heavy particle because we have xed the trajectory of the particle, C, and we are not considering uctuations of this path. If C is a rectangular path that stretches a length T in the time direction and a length L in the spatial direction, we have W (C) eT V (L) , (18) where V is the static quarkantiquark potential. This can be used as an order parameter for connement if V (L) L, we say that the theory is conning. Recall that in our discussion of t Hooftology, at large N a quark has a string ending on it. Are these the strings in AdS? To analyze this question, let us take a short detour back into the world of string theory. Consider a stack of N + 1 D3 branes sitting in some asymptotically at space. There is therefore some U (N + 1) gauge theory living on the D3 branes. Suppose N is not so big, so that the gauge theory is at small t Hooft coupling. Suppose we take one of the branes and separate it from the other N . This corresponds to giving an expectation value to the adjoint scalars and breaking the U (N + 1) down to U (N ) U (1) (Figure 1). Suppose x is the position of the single D3 brane and x is the position of the N other branes. The mass of the string which extends from the single brane to the N branes is proportional to the separation: mw x x  2 . We claim that we can describe such a state at large t Hooft coupling by placing a probe D3 brane in the throat at position zw = (2mw )1 in AdS (with N units of ux). A string stretches from this probe D3 brane down to the stack of N branes sitting at z (Figure 2). We can see that the string stretches from zw by matching the mass of this stretched string that falls into the horizon with the mass mw , which we can do because it is a BPS state and therefore we can follow its mass from weak to strong coupling. We can make our probe arbitrarily heavy, mw , by taking zw 0 (that is, to the boundary of AdS). We do this to avoid uctuations of the trajectory. This string is in the N of the U (N ) and it couples to the adjoint scalars: the stretched string pulls on the branes so it aects the X a (x) (Figure 3). It also couples to the fermions, but we ignore this
Figure 1: One of the D3 branes is separated from the other N , thus breaking the U (N + 1) down to U (N ) U (1).
Figure 2: The string stretches into the throat from the probe D3 brane at zw .
Figure 3: The string pulls on the stack of D3 branes, so it couples to the adjoint scalars.
(19)
= 1.
Now this generalizes the GKPW prescription: W [C]CF T = Zstring theory [Fstrings end on C] (20)
At small string coupling gs , contribution will be dominated by disk ampliitude. At large , the world sheet path integral is dominated by saddle point. Also, at large , we can ignore the S 5 coordinates and coupling to the 5form ux. So, in this approximation, the string theory partition function is approximated by: Zstring theory [DX]eSN G [X]+ e min(Area) (21)
(world sheet)=C
and Area[x] = dd det xA xB gAB (x) = dd L4 det xxAB = L2 AreaL=1 [x]. z4 (22)
L2 . z 2 AB
Minimum Surfaces
The expectation value of Wilson loop operators W [C] in the CFT can be computed by computing the string theory partition function on AdS with a string world sheet ending on the loop C. That is, W [C] CF T = Zstring theory [string ending on C] Further we saw that, al large and large N , the world sheet path integral is dominated by the saddle point. In the saddle point approximation, the string theory partition function is given by, Area () . Zstring theory exp extremum =C Hence, the computation of W [C] in the CFT can be mapped to nding an minimum surface in AdS. The area of string worldsheet (classical) is given by, L4 det X A X B AB Area[X] = dd det X A X B gAB (X) = dd z4 Here we have used the fact that gAB = L2 AB for Euclidean AdS. Notice that, z 2 L ln W = extremum Area () = extremum Area () =C =C L=1 L=1 Further, the area of worldsheet is innite. This divergence arises from the fact that the worldsheet goes all the way to the boundary of AdS. Let us zoom into a region close to the boundary and introduce a cuto at z = which is equivalent to the UV cuto of the eld theory. We can think of the direction along the curve C as the coordinate of the worldsheet near z = and extends along the radial direction near z = (please see Fig. 1). The area evaluated upto the cuto has the following behavior, 1 length[C] + ... A = dz d 2 det X A X B AB z C It can be seen that A diverges as 1 . Let us compare this result with Wilson loop calculation in the eld theory. In the eld theory, the Wilson loop is the trajectory of a heavy quark. Quarks couple 1
Figure 1: (a) Worldsheet ending on curve C (b) Worldsheet near the boundary.
_ _ _
Sw l = q q
m length [ C ]
Figure 2: Diagram contributing to Mass Renormalization of heavy quark to gluons and they can emit and reabsorb gluons. An example of such a process is the leading oneloop correction to the quark propagator (shown in Fig. 2). This diagram corresponds to the mass renormalization (mq ) of the heavy quark at zero quark momentum. This implies we must add mq ds to the worldline action of the quark. So, the 1 divergence in the Wilson loop can be absorbed into mass of the quark. Hence, the prescription to calculate the renormalized wilson loop is to subtract o the 1 divergence in the extremum area of the worldsheet. This prescription has a geometrical interpretation too! The divergence in the world sheet area is the area of a tube (shape of the crosssection perpendicular to the radial direction is independent of z) in AdS, whose base is the C. Subtracting the 1 divergence is equivalent to subtracting the area of this tube from the area of the minimum surface. Note that this is not true for the BPS case as, there is no mass renormalization of a BPS particle. There are two diagrams that contribute to the mass renormalization (shown in Fig .3) a diagram involving gauge boson and another diagram involving the scalars. BPS case corresponds to some correlated choice of the trajectory in spacetime and the trajectory in the S5 . If there is some constraint on these trajectories, the loop can be BPS and the two contributions cancel. On the other hand, the area of the worldsheet still has some divergences. The resolution of this puzzle
_ _ q _
+
Figure 3: Diagrams contributing to Mass renormalization of a BPS particle
Figure 4: Minimum surface ending on a straight line is that the expectation value of the wilson loop is not the area but its Legendre transformation. 1 We shall now look at some examples of curves for which we can say something about the Wilson loops.
2
2.1
Examples
C0 : Straight Line
In this case the minimal surface is a at surface that ends on the straight line (Please see Fig. 4). Itis obvious that a at surface satises the equation of motion. The area of this fat surface is 1 d . Further, note that the at surface is a surface whose crosssection doesnt depend on the radial coordinate. Hence, the renormalized Wilson loop is obtained by subtracting the area of the at surface from itself. That is,, ln W [C0 ]
ren
In the next example, we shall see how we can use this trivial result to compute the Wilson loop for a circular contour.
2.2
C1 : Circle
There are two ways by which we can nd the minimal surface  (a) solve the equation of motion and (b) make use of conformal symmetry. In this example, we will calculate the Wilson loop of a circular contour by mapping the circle to a straight line using conformal transformations. The transformation that takes a line to a circle at the boundary is given by, xi =
1
xi + bi x2 1 + 2bi xi + b2 x2
Please see N. Drukker, D. J. Gross and H. Ooguri (hepth/9904191) for a detailed discussion.
Figure 5: Minimum surface ending on a circle Let us asume b3 = 0, b4 = 0 for simplicity. Then, the above transformation maps the line C0 = {{xi }x2 = 0, x3 = 0, x4 = 0} to the circle C1 = {{xi }x2 + b2 x2 + x2 = 0, 3 = 0, x4 = 0}. The x 2 2 1 2 and the centre of this circle is located at 0, (2b )1 , 0, 0 . Let radius of the circle C1 is (2b2 ) 2 us denote the radius of the circle by a. The special conformal transformation symmetry of the boundary theory extends to an isometry of the AdS which is given by, xi + bi (x2 + z 2 ) 1 + 2bi xi + b2 x2 z z= 1 + 2bi xi + b2 x2 Using the above transformation on the trivial solution in the previous example, we nd that the solution for the minimum surface (in the parametric form) is given by, x = a2 z 2 (cos , sin , 0, 0) , a z a xi = One can check that this solves the equation of motion. The worldsheet can be parametrized by z() and ( ). The induced metric has the following form, L2 2 a2
a z 2 d 2 + 2 dz 2
ds2 = 2 z a z2 Plugging the above solution into the expression for worldsheet area (NambuGoto action) get, 2 a Area L2 a L2 a = d dz 2 = 1 SN G [X] = 2 2 0 z The 1 divergence is cancelled by the area of the cylinder with circular base (shown in Fig. 6). Hence, L2 ln W [C1 ] ren = SN G + Area of cylinder = = W [C1 ] = e 4 Note that, the expectation value of the Wilson loop is independent of a. This is a consequence of scale invariance. Further, we can see that the expectation value of this Wilson loop is not 1, even though it is related to the previous example by a symmetry. This is because, the whole contribution to the Wilson loop comes from the anomaly in conformal transformation at the point at innity. The expectation value of the Wilson loop obtained here is the large limit of the exact answer I1 ( ).2
2
2.3
QuarkAntiquark Potential
In this example we shall calculate the potential between a static quark and a static antiquark. Figure (7) shows the conguration used to calculate the quarkantiquark potential. The vertical direction in Figure (7 a) denoted by t is taken as the real time direction and nothing depends on t. The quark and the antiquark are separated by a distance r in the x direction. We may further assume that they are located at the same point in S5 . This time independent conguration at the boundary suggests that we can look for timeindependent solution for the minimum surface. The area will have divergence arising from the integral over the timedirection. This can be regulated by introducing a cuto (T ). The goal is to nd the area per unit time of the minimum surface ending on this conguration. To calculate this , it is useful to work in the static gauge, in which, = t = x and z = z(x) The induced metric in AdS in this parametrization of the world sheet is given by, 2 L2 z = ab dxa dxb ds2 = 2 dt2 + dx2 1 + z x The NambuGoto action is then given by,
T /2 r/2 1 S= dt dx(det[])1/2 2 T /2 r/2 In the above expression, the integral over time is cuto by T . The above expression simplies to, 1/2 r/2 1 + (x z)2 S = 4T dx z 2 (x) r/2 Finding, the minimum surface in this case is a classical mechanics problem in one variable. This variational problem can be solved by making use of the translation symmetry ( x + a) of the Lagrangian. There is a conserved charge (independent of x) associated with this symmetry which is given by, L H=Lz z 5
Figure 7: (a) Time Independent Boundary Conguration (b) Topview of the conguration In the above expression, L is the Lagrangian, H is the conserved current and z denotes the derivative of z with respect to x. Hence, z2 1 = H = constant 1+z2
Note that, z is an even function of x. This implies z is maximum at x = 0. Using this condition and the above equation, we can now solve for x as a function of z. The solution is given by, z z2 Hdz x= 1 H2 z 4 zmax Note that, the integrand becomes imaginary when z > 1/H. Hence, the maximum value that z can take is 1/H. Let us substitute z = yH1/2 in the expression for the integrand. The solution for x > 0 can be written as, 1 y2 x = zmax dy 1 y4 z/zmax We can now solve for zmax in terms of r using the fact that x = r/2 when z = 0. Substituting this condition in the above expression we get, 1 y2 2 3 r/2 = zmax dy = zmax [(1/4)]2 1 y4 0 Hence, we can express our solution as x = X(r, z). Substituting this solution in NambuGoto action we get, r/2 1+z2 SN G = T 4 dx z2 r/2
SN G
dz
z2
2 zmax 4 zmax z 4
In order to get the renormalized expectation value of the Wilson loop, we must subtract the area of a constant crosssection surface that ends on the boundary conguration shown in Fig . 7(a). Hence we must subtract the sum of the area of two at surfaces ending on the two straight lines from SN G . Hence, the renormalized expectation value of the Wilson loop is given by, zmax dz ln W [C =  ] ren = SN G 4T 2 z2 The above expression can be simplied to yield, ln W [C =  ] 4T
ren
2 zmax
1 /zmax
dy y2
1 1 y4
Hence, the static quarkantiquark potential V (r) is given by, 4 2 V (r) = r [(1/4)]4 The 1/r dependence (in 3+1 dimension) of this potential is determined by scale invariance. Note that in the weak coupling limit, the potential is given by, V (r) = const r
An important lesson that we learn from this calculation is that string description does not imply conning behavior. If the separation between the quark is small, then the string cannot stretch far into the bulk. Whereas if the separation is large then the string can stretch far into the bulk where the warp factor is bigger and the renormalized area is smaller. This explains why the potential is smaller when the separation is larger. In the following we shall briey discuss other related calculations.
2.4
This Wilson loop is a pair of semiinnite straight lines intersecting at an angle (shown in Fig. 8). This can be used to calculate a quantity called cusp anomalous dimension. Calculations in QFT (resummation of rainbow graphs) show that this quantity has additional lograthimic divergence proportional to the opening angle near the neighbourhood of the cusp. That is, ln W (cusp) ln(L.) where, L is the length of the loop and is the UV cuto. This quantity can be computed using the method described in this lecture. It can be shown that the renormalized area of the worldsheet ending on this cusp, has an extra lograthimic divergence (ln L/ ).
2.5
Other quantities of interest in QFT are the correlators of Wilson loops ( W (C1 )W (C2 ) , . . . ). These quantities can be expanded as an OPE in local operators. In order to compute these quantities using the AdS/CF T correspondence, we must extremize over all possible congurations of the worldsheet. For example, W (C1 )W (C2 ) = exp (FA , FB ) where FA and FB are the congurations shown in Fig. 9(a) and Fig. 9 (b). Let us now see
Figure 9: (a) Time Independent Boundary Conguration (b) Topview of the conguration how we can compute observables that are similar to the observables described above using the correspondence.
2.6
MonopoleMonopole Potential
This is the potential between two heavy monoples and it can be computed by computing tHooft loop, which is the magnetic dual of Wilson loop. In AdS, this quantity (tHooft loop) can be computed by replacing the Fstrings in example 2.3 by Dstrings. We know (from problem set 1) that the end of a D string ending on D3 branes carry magnetic charge. More generally, we can compute tHooft loops by studying Dstrings ending on tHooft loops. We must remember that the
Hence, the monopolemonopole potential is given by, N Vmono (r) r Similarly, we can compute quarkmonopole potential.
2.7
ksurface observables
We can generalize the method used to compute Wilson loops to study observables associated with a kdimensional surface specied on the boundary of AdS by computing the minimum area of the k + 1 dimesnional surface ending on it. Such observables have been studied by Graham and Witten in hepth/9901021.
Pointlike probes of the bulk  Geometric optics through Dbranes. Baryons and Branes Giant gravitons Baryon vertices in N = 4 SU (N ) gauge theory
Last time we observed that one could compute observables associated to kdimensional submanifolds
=C,k+1 AdS
Extremum
Vol()
The simplest example is the case k = 0, when C is a discrete point set. This would lead us to consider onedimensional curves in AdS. This one dimensional curve is the worldline of some point particle excitation in AdS. Let C = {a, +a} be the discrete point set and denote a curve parametrized as z( ) with z( = 0) = a and z( = 1) = +a as shown in Figure 1. The observable associated with C is clearly the two point function of some local operator O. This local operator couples to a scalar eld of mass m in the gravity theory. The usual prescription involves computing the two point function in the bulk and nding the onshell action. Since we have a free scalar eld, we can instead use a rst quantized approach. The action for a point particle of mass m whose worldline is is given by S[z] = m ds. The two point function for this particle z( = 0)z( = 1) is given by the Feynman path integral Z[a] = [dz( )] exp(S[z]1 ). In the 0 limit of large m, we have Z[a] exp(S[z]) (1) where we have used the saddle point approximation. z is the geodesic connecting the points a on the boundary.
+a
z
Figure 1: The curve AdS connecting the two points in C. The arrows on the curve indicate the orientation. We now compute Z[a] in the saddle point approximation. The metric restricted to is given by 2 ds2  = L2 (1 + z 2 )(d )2 , where z := dz/d . This implies the action is z L S[z] = d 1 + z 2 (2) z
The geodesic can be computed easily by noting that the action S[z] does not depend on explicitly. This implies we have a conserved quantity L L 1 h = z L = z 1 + z 2 z 2 L 1 z 2 = z2 2 h z2
(3) (4)
2 The above is a rst order dierential equation with solution = zmax z 2 , with z( = 0) = zmax = L/h = a. This is the equation of a semicircle. Substituting the solution back into the action gives a L dz a 2 = 2L S[z] = d 1+z (5) 2 z2 z z a 2a = 2L log + (terms 0 as 0) (6) Recall that the expectation value of a circular Wilson loop of radius a was independent of a. We argued that since we are computing this in a conformal eld theory, the only scale in the problem is a and therefore the expectation value of the Wilson loop should be independent of a. This argument fails in the case at hand because there are two scales: a and . The scale transformation is anomalous and this is manifested in the dependence of S[z]. According to the GKPW prescription, the generating functional Z[a] with the prescribed bound ary conditions corresponds to the two point function O(+a)O (a). The saddle point approxima tion of large m corresponds to large in the CFT. Therefore O(+a)O (a) Z[a] exp(S[z]) 1 a2mL (7)
This is exactly what we expect for the two point function, since m 2 L2 = (D) 2 in the large limit. Without this anomaly, the two point function of the operator O would be independent of 2
Figure 2: N Fstrings ending on a D5 brane at the origin = 0 in global coordinates of AdS 5 . The orientation of the strings is indicated with an arrow. a forcing = 0, which is impossible in a unitary CFT. This is similar to the anomaly that gives a nonzero scaling dimension of k 2 /4 to the operator exp(ikX) for the 2d free boson CFT which has classical scaling dimension 0. Graham and Witten [1] showed that such a scale anomaly exists for surface observables for any even k. This relation between correlators of large observables and geodesics in the bulk theory oers a probe of the bulk spacetime [2].
In the previous section, we considered a very heavy point particle excitation in the bulk theory. The question that naturally arises is  Who are these heavy objects? A good candidate would be the D0 brane which is extremely heavy at weak coupling, but in Type IIB there are no BPS D0 branes. This is because the IIB theory has only even form RR potentials. In the absence of a conserved charge, this object would decay into a pu of closed string states. The next best thing we could consider are Dq branes wrapping qcycles of S 5 . The only closed cycles on S 5 are the 0cycle and the 5cycle. This leaves us with only one choice  the D5 brane wrapping the S 5 . This would 6 correspond to a point particle in AdS 5 . The tension of the D5 brane is T5 = 1/(gs ls ). Naively, one 5 ) L5 /(g l 6 ) N 3/2 /L. would expect that the wrapped D5 brane would have a mass T 5 Vol(S s s However, this turns out to be incorrect. Recall (from Problem Set 1) that the action of the D5 brane includes the term SD5 Fwv C (4) = Awv F (5)
M6 M6
(8)
where M 6 is the worldvolume of the D5 brane which happens to be S 5 worldline in AdS5 . The subscript wv indicates that the elds live on the worldvolume of the D5 brane. There are N units 3
B v
Figure 3: Electric dipole moving in a magnetic eld. of ve form ux threading the S 5 that stabilize its volume. This implies that SD5 N Awv Worldline
(9)
The worldvolume gauge eld has N negative point charge sources induced by the veform ux. Since the space S 5 is compact, this implies that we must have N positive point charge sources. Recall that the fundamental string that ends on a Dbrane is electrically charged under the worldvolume gauge eld. Thus, we can cancel the net charge by attaching N Fstrings to the D5 brane. Since these Fstrings cant end anywhere, they extend all the way to innity. Figure 2 illustrates this state in global coordinates. The attached strings have a mass N/( ) dz/z 2 which is innite and proportional to N . This object is not a dynamical point particle in IIB gravity theory on AdS 5 since it is innitely massive. It does not couple to a nite operator in the CFT. Recall from or discussion of Wilson loops that strings which extend all the way to the boundary of AdS correspond to innitely massive quarks transforming in the fundamental of SU (N ) in the CFT 1 . Thus, the D5 brane with N fundamental strings going o to innity couples to a baryon vertex between N heavy, external quarks. We can regulate the mass by considering a probe D3 brane at a nite radial position. The heavy quarks in the gauge theory have nite mass proportional to the radial position. This is completely analogous to what we did for the case of the Wilson loop. Now, we turn to perturbative string theory and compute the ground state for strings stretched between the D5 and D3 brane. The worldvolume of the D5 brane is S 5 R and the worldvolume of the D3 brane is S 3 R. Consider an open string stretching from the D5 brane (left end) to the D3 brane (right end). In quantizing the open string the boundary conditions for each coordinate are either DirichletDirichlet (DD), NeumannNeumann (NN) or mixed (ND). The number of coordinates with ND boundary conditions is 8 for teh 53 string. This implies that there is a fermionic massless ground state (See Polchinski II, section 13.4). This implies that the baryon vertex is antisymmetric under exchange of heavy quark elds.
X3
X2
X1
Figure 4: A D3 brane wrapping a 3sphere inside a 5sphere. Drawn above is the case of a 0sphere 2 2 2 2 X3 = r 2 at a polar angle , embedded in a 2sphere X 1 + X2 + X3 = L2 .
Giant Gravitons[3]
Consider an electric dipole consisting of charges +q and q connected by a spring happily moving with a velocity v in a magnetic eld B as shown in Figure 3. The forces acting on the dipole are the Lorentz force and the force exerted by the spring. The equation of motion is m = k + qvB (10)
These forces oppose each other producing an equilibrium where the dipole has length = qvB/k. We can imagine the dipole moving on the surface of a twosphere with a magnetic monopole at the origin producing the magnetic eld. Now, consider a D3 brane wrapping an S 3 inside an S 5 . We dene the 5sphere as the locus 6 2 2 5 2 2 2 i=1 Xi = L . A 3sphere of radius r inside the S is specied by two equations  X 3 + X4 + X5 + i . This setup is shown in Figure 4 using a 0sphere (two points) 2 = r 2 and X +iX = X6 L2 r 2 e 1 2 2 . The S 3 is a contractible cycle in the S 5 and since the D3 brane has a tension, the inside an S minimum energy conguration is a 3sphere of radius zero. A spherical D3 brane is neutral under the RR 4form. To see this consider two dierential volume elements diametrically opposite on the 3sphere. These two elements will trace out worldvolumes with opposite orientations. Although each of these elements individually is charged under the 4form, the total charge is zero. However, the D3 brane does have a dipole moment and if it moves in a magnetic eld, the magnetic force would counter the force due to the tension and stabilize the volume of the S 3 . Consider the D3 brane shown in Figure 4 moving along the direction with a velocity . To write 5 considered as a two parameter family of down the DBI term, we rst write the metric on the S
We take the gauge group as SU (N ) and not U (N ). The U (1) factor in the CFT gauge group corresponds to the center of mass coordinate of the N D3 branes. For a discussion of this fact, see MAGOO page 58.
1
X3
X2
X1
Figure 5: The worldvolume of the D3 brane is represented as the boundary of a four dimensional space time. S 3 s as ds2 = L2rr2 dr 2 + (L2 r 2 )d2 + r 2 d2 . The D3 brane is located at xed r, with the angular 3 coordinate varying . Ignoring the world volume gauge eld on the D3 brane, the DBI action as is simply TD 3 W V det g. On computing the induced metric (only the coordinate varies with time), we have det g = r 6 (1 (L2 r 2 )2 ) det(3sphere metric). The DBI action can be written as 3 SDBI = TD3 3 r dt 1 (L2 r 2 )2 (11) The AharanovBohmChernSimons term can be written down in analogy with the case of an electric dipole(a 0sphere) on a 2sphere with a magnetic monopole at its center. L ABCS = e dt v i Ai (+) + e dt v i Ai (). The + and denote the positions of the positive and nega tive charges constituting the dipole. Let be the twodimensional manifold whose boundary is the worldline of the dipole as shown in Figure 5. If is the angular velocity of the dipole, it makes /(2) revolutions per unit time. Then, L ABCS = e/(2) A = e/(2) F = eN r 2 /L2 where N is the strength of the monopole charge. For the case of the D3 brane, let be the four dimensional manifold whose boundary is one orbit of the D3 brane around the S 5 . Then, B Vol() 2 Vol(S 5 ) r 2N 3 = 3 dr r 2 5 L5 0 4 r = N 4 L The angular momentum of the D3 brane is given by LABCS = l= S r4 m(R2 r 2 ) =N 4 + L 1 (L2 r 2 )2 6 (12) (13) (14)
2
(15)
r The eective potential for the D3 brane has the form V (r) lL2 N r 4 and has a minimum at 2 = l l 2 . The size of the D3 brane is determined by its angular momentum. In the SYM theory, r N the angular momentum maps to the Rcharge. At a radius r , the energy of the D3 brane is E(r ) = l/L which is the BPS bound for an operator with Rcharge l. The analysis above used the DBI action for the D3 brane which is valid only when the radius of the D3 brane is large compared to and this corresponds to l L. Surprisingly, the D3 brane with angular momentum around the 5sphere has the same quantum numbers as the KK mode of the graviton. In fact there is yet another BPS object with the same quantum numbers[5],[4]. This is a D3 brane wrapping a spatial S 3 in AdS5 located at a point in the S 5 . In the SYM theory there is a single BPS operator with Rcharge l, the single trace operator Tr [X l ]. Thus we have three BPS objects on the gravity side with the same quantum numbers as the KK graviton. Since we are in the large regime, one can imagine that one linear combination of these BPS short multiplets remains short, while the other two combine to form a long multiplet. Observe that the largest radius the S 3 can have is L which corresponds to a maximum angular momentum of l max = N . This surprising fact called the stringy exclusion principle matches with the result that there is no singleparticle BPS state in N = 4 with l > N . The operator O = det X has the maximum allowed l = N .
There are other vacua of Type IIB string theory with an AdS 5 factor. These would correspond to other four dimensional CFTs. For example, the space AdS 5 X 5 satises the equations of motion if X 5 is a ve dimensional Einstein manifold. An Einstein manifold is a manifold whose Ricci curvature is proportional to the metric i.e. R M N = gM N with constant. N = 4 SYM is the low energy theory living on D3 branes in at space. The transverse space to R3,1 R9,1 is R6 which is a cone over S 5 (the celestial sphere). In fact any smooth space looks like R6 locally and the low energy theory on the D3 branes only probes a neighbourhood of the geometry. So if we start with D3 branes at a point on a smooth space and look at the near horizon limit, the geometry would again look like AdS 5 S 5 . One way to get around this is to consider D3 branes at a singular point in the geometry where the neighbourhood is no longer R 6 . R6 / orbifolds: Orbifolds are manifolds except for a nite set of points which have a neigh bourhood homeomorphic to R6 / where is an element of a discrete subgroup of SO(6). The near horizon limit at large gives a geometry AdS 5 S 5 /. There are prescriptions to nd the resulting QFT. The basic idea is that we have a projection and we have to decide how acts on the branes. After that we just project onto the invariant states. Cone over X5 : Not all singularities are orbifolds. A classic example is the conifold. These spaces can have qcycles and therefore one has static baryonlike excitations of nite energy. For the case of AdS5 S 5 , we found an innite mass baryonlike excitation (D5 brane with N fundamental strings). It is also possible to deform these theories (by adding fractional branes) so that the gauge coupling runs. In the case of the N = 4 theory we have a line of xed points and the dilaton is free. 7
References
[1] C. R. Graham and E. Witten, Conformal anomaly of submanifold observables in AdS/CFT correspondence, Nucl. Phys. B 546, 52 (1999) [arXiv:hepth/9901021]. [2] L. Fidkowski, V. Hubeny, M. Kleban and S. Shenker, The black hole singularity in AdS/CFT, JHEP 0402, 014 (2004) [arXiv:hepth/0306170]. [3] J. McGreevy, L. Susskind and N. Toumbas, Invasion of the giant gravitons from antide Sitter space, JHEP 0006, 008 (2000) [arXiv:hepth/0003075]. [4] A. Hashimoto, S. Hirano and N. Itzhaki, Large branes in AdS and their eld theory dual, JHEP 0008, 051 (2000) [arXiv:hepth/0008016]. [5] M. T. Grisaru, R. C. Myers and O. Tafjord, JHEP 0008, 040 (2000) [arXiv:hepth/0008015].
Introduction
Today were going to talk more about other examples of gauge gravity duality besides the N = 4 SYM theory. The rst half of the discussion will be very survey like in which we attempt to get some feel for the possibilities beyond the N = 4 theory. In the second half we will focus on a particularly simple extension of what we know so far that will lead to a breaking of the conformal symmetry, connement, and a mass gap. Once we nish up our discussion of connement well be moving on to black hole mechanics: thermodynamics and hydrodynamics.
2
2.1
In the N = 4 case the gravity dual had the form AdS5 S 5 . A natural generalization would be to consider spaces of the form AdS5 X where X is some compact space. The AdS isometry group was responsible for the conformal invariance in the dual eld theory while the isometry group of S 5 related to the Rsymmetry in the N = 4 theory. We can consider nonspherical horizons by replacing the sphere S 5 with a nonspherical manifold X. This should correspond to a change in the global symmetries of the dual eld theory. Such nonspherical Xs arise as the locus of points equidistant from a singularity (such as orbifolds) and the dual eld theory can be obtained by studying D3branes probing these singularities. The coupling can even be made to run (changing the AdS part of the gravity dual) with the addition of uxes and fractional branes (but dont ask what this means).
2.2
Dpbranes, p = 3
Weve focused a lot on D3branes so far, but there are other branes in string theory and they have interesting world volume theories as well. Based on our experience with the D3brane, we might not be too surprised to learn that the world volume theory of a Dpbrane is the dimensional reduction of ten dimensional N = 1 SYM on a 10 (p + 1) torus with periodic boundary conditions. The result is a p + 1 dimensional YangMills theory with 16 supercharges. Now there is a very important dierence between the 3 + 1 dimensional YangMills theory and YangMills theory in any other dimension. To see this dierence remember that the YangMills action contains a term like dp+1 x 1/g2 Tr F 2 (gauge kinetic term). This term allows us to gure out the mass dimension of g2 since the dimension of A is always 1 and the dimension of F 2 is therefore 4 regardless of space time dimension. Requiring the action to be dimensionless means that g2 has mass dimension 3 p so that g2 is dimensionful whenever p = 3.
2 In terms of a running eective coupling gef f (E) = g2 /E 3p we nd that gauge theory perturbation 2 2 theory is good when gef f 1. The behavior of gef f depends strongly on dimension. When p < 3 2 2 then gef f is big in the IR, and when p > 3 we nd the opposite situation where gef f is large in the UV.
In the large N limit there exists a type II SUGRA solution with near horizon limit given by gs N 2 u7p du + dx dx + gs N up3 d2 ds2 / = 8p 7p u gs N
(1)
where d2 is the metric on a unit S 8p and runs from 0 to p for the p + 1 directions on the 8p brane world volume. When p = 3 this metric is just that of our old friend AdS5 S 5 , but in general the ten dimensional manifold doesnt have such a simple product structure. In addition to the metric we must also specify a ux condition F8p = 2N (2)
S 8p
where F8p is the eld strength of some appropriate RR form. Also, the dilaton has a nontrivial prole given by 3p 4 gs N e = gs . (3) u7p The fact that the dilaton depends on u is essentially the statement that the YangMills coupling runs when p = 3. Its important to note that both the dilaton and the curvature blow up for some values u when p = 3. When the dilaton grows large we can no longer trust perturbative string theory and when the curvature grows large corrections to supergravity become important. Conservation of evil demands that no two dierent descriptions of the system should be valid simultaneously, and indeed it was observed in hepth/9802042 that the blow up occurs precisely where another description becomes valid/useful. As an example of this important point lets consider the case of D2branes. 2
ln N
nearhorizon limit of D2s ????? strongcoupling limit of IIA strings
perturbative
SYM
ln(g eff = gp /u )
1/2
2.3
Mtheory
Mtheory also provides further examples of gauge gravity duality. Lets recall the basics of Mtheory very briey. The theory has a vacuum with eleven noncompact dimensions forming R10,1 and no coupling. The low energy limit of the theory is eleven dimensional supergravity with 32 supercharges which is the mother of all supergravity theories. The theory also has a massless 3form potential with eld strength G4 = dC3 . This eld is coupled minimally to branes in the theory called M 2branes. The dual of the 4form eld strength is a 7form eld strength G7 = dC6 which strongly suggests that the theory also has M 5branes that source G7 electrically and hence G4 magnetically. What does Mtheory have to do with string theory? Compactifying the theory on a circle of radius 11 R10 gives IIA string theory with string coupling gs (R10 MP )3/2 . This claim can be eshed out by identifying the IIA elds in terms of Mtheory objects. The KaluzaKlein gauge boson G11 becomes the RR 1form. The D0branes that source the 1form arise as KaluzaKlein RR excitations. They have a mass given by m n/R10 n/(gs which agrees with the mass of KK excitations and the result from IIA which is protected by supersymmetry. M 2branes perpendicular to the circle become D2branes in IIA, and thus the Mtheory 3form C3 becomes the IIA RR 3form when it is perpendicular to the circle. An M 2brane wrapped on the circle becomes a fundamental string, and so the Mtheory 3form when pointing along the circle becomes the NSNS 2form B2 as C3 = B2 dx10 . Mtheory is useful here in part because it provides the answer to the question, what happens to the D2branes at strong coupling? The answer according to Mtheory is that we should think about M 2branes on a large circle instead of D2branes when the gauge theory eective coupling is large. The world volume theory on the M 2branes should ow in the IR to some superconformal eld theory that lls in the blank in our diagram above.
There is one limit in particular in which we know how to make much more sense of this statement. In the limit of a large number N of M 2branes we expect a supergravity solution (in eleven dimensions) that takes back reaction of the branes into account. Taking the near horizon limit of this solution we nd the space AdS4 S 7 with S 7 G7 = 2N which describes the IR limit of many M 2branes in the gravity dual language. Without the large N limit the IR limit of the M 2brane theory is a superconformal eld theory with N = 8 and an SO(8)R R symmetry. This theory is just the IR limit of three dimensional super YangMills (which we remember can be obtained by dimensionally reducing N = 4 SYM on a circle). This theory has 7 scalars X i representing transverse uctuations of the M 2branes leading to a visible SO(7) symmetry. To make the SO(8) symmetry manifest we must somehow nd another coordinate. The belief is that this coordinate is related to the three dimensional dual of a U (1) gauge eld da = 3 d with essentially the eighth coordinate on the circle. Recently, there has been progress in formulating a Lagrangian description of the theory of M 2 branes. The theory has a discrete parameter related to a ChernSimons coupling, and indeed the theory looks like a ChernSimons theory with special matter content.
ln N
nearhorizon
limit of D2s
M theory on AdS_4 x S^7
perturbative
SYM
3d N=8 SCFT
ln(g eff = gp /u )
1/2
Figure 2: Regimes of validity of various descriptions of N D2branes, after incorporating the fact that the strongcoupling limit of type IIA is Mtheory. A similar story applies for M 5branes, where at low energy the theory becomes some superconformal eld theory in six dimensions. Very little is known about this theory. For a large number of M 5 branes some information about the theory can be obtained from black holes. The free energy goes 3 as FM 5 T 6 V NM 5 suggesting that the theory is not best formulated in terms of the matrices of SYM! It is tempting to mumble words about elds which are N N N objects, but not much progress has been made with that idea.
Connement
To motivate our picture of connement lets remember that the radial coordinate behaves like a spectrograph separating the theory into energy scales. In the case of AdS we know that z = 0 is like the UV while z is like the IR. The fact that AdS goes forever in the z direction seems to be related to the fact that the dual theory is a conformal eld theory with degrees of freedom at all energies. We can make this more precise using the warp factor W (z) which is dened to be W in dz 2 ds2 = 2 + W 2 (z)dx dx . z The warp factor for AdS is simply 1/z, and we can interpret the existence of modes at arbitrarily low energy as the statement that the warp factor has a zero at z = . More precisely, there are O(N 2 ) degrees of freedom at every energy scale in the CFT. What would we expect for a theory exhibiting connement? We might expect the theory to have a mass gap as in pure YangMills theory. On the other hand, there may be Goldstone bosons from symmetry breaking caused by the strong dynamics. Such pions would represent low energy degrees of freedom, but the important point is that we wouldnt expect the number of pion modes to scale with N (they are color neutral). Thus a conning largeN gauge theory might reasonably have a mass gap except for a few low energy modes. The absence of modes at low energy should therefore be reected in the bulk geometry and the warp factor. In particular, we would like to interpret a minimum of the warp factor as a signal of a mass gap in the dual eld theory. The simplest realization of this idea [2] is the N = 4 theory in four dimensions with one direction compactied into a circle. Call the coordinate along the circle y and identify y y + 2Ry . We give the fermions antiperiodic boundary conditions around the circle so that (y + 2Ry ) = (y). To the bosons and gauge elds we give periodic boundary conditions. Such boundary conditions are called ScherkSchwarz or thermal boundary conditions. Indeed, if the direction we compactify is Euclidean time then we are considering precisely the N = 4 theory at nite temperature. Alternatively, if we compactify some other spatial direction then we get YangMills theory in one lower dimension (for energies lower than 1/Ry ) The boundary conditions treat bosons and fermions dierently so supersymmetry is broken. The fermionic modes begin at energies of order 1/Ry because they are required to have nonzero angular momentum around the circle. The bosons also get a mass from fermions running in loops because the protection from supersymmetry is lost at low energy. All that remains therefore are the gauge elds at energies small compared to 1/R and we might expect that we have essentially YangMills theory in three dimensions at low energy. However, we should be a little careful since there are factors of oating around and because the energy scale of the bosonic and fermionic modes isnt separated from the scale at which the theory looks four dimensional again.
References
[1] N. Itzhaki, J. M. Maldacena, J. Sonnenschein and S. Yankielowicz, Supergravity and the large N limit of theories with sixteen supercharges, Phys. Rev. D 58, 046004 (1998) [arXiv:hep 5
th/9802042]. [2] E. Witten, Antide Sitter space, thermal phase transition, and connement in gauge theories, Adv. Theor. Math. Phys. 2, 505 (1998) [arXiv:hepth/9803131].
Conning geometry and mass gap (main lesson: mass gap smooth minimum of the wrap factor in the dual geometry) Black hole mechanics Last time we were talking about attempting to nd a gravity dual of a deformation to 3D YM theory (which we get from the N = 4 on a thermal circle1 ). In these compactications, the 3D YM coupling g3 is of order of the mass of the scalars g3 mX g4 N Ry (1)
where g4 N is the 4D t Hooft coupling. The 3D theory is not just YM but it also has a bunch of scalars and this obscures our intuition about the theory. We can do a similar thing to get 4D YM theory by putting ND4 branes (at low energies) on a thermal circle with a radius Ry and again Fermions will have APBCs. The ND4 theory is a gauge theory in 5D with 16 supercharges and its well dened at low energies. The boundary conditions on the thermal circle will break supersymmetry and the fermions will get 4D masses of order 1/Ry 2 2 while bosons will get masses at one loop of order m2 g4 N m2 4 /Ry . At energies E 1/Ry , X 2 2 the 4D and 5D YM couplings are related by 1/g4 = Ry /g5 which implies that mX 4 /Ry . The relation between the the 4D and 5D YM couplings comes from integrating over the extra dimension in the compactied theory. The nice thing about the above construction is that if at the UV scale (E = 1/Ry ), the t Hooft coupling 4 1, then the theory is asymptotically free and we can calculate the function perturbatively. There will be a dynamically genrated mass scale such that e1/4 1 4 QCD mX (2) Ry Ry Ry Notice that 1/Ry plays the role of a UV. Its like putting the 4D YM theory on a lattice where Ry plays the role of a lattice spacing. If we probe the theory with energies of order 1/Ry we will see a 5D theory (the UV completion).
1 Remember that thermal circle means periodic boundary conditions for Bosons and antiperiodic boundary condi tions for Fermions
Notice also that when E < 1/Ry , we will have a pure YM theory. The previous analysis was for 4 1, the gravity dual has a curvature which goes like 1/4 and we cant trust the supergravity description. The gravity dual is valid when QCD UV, i.e, when 4 is large. At large 4 the QFT isnt weakly coupled at the cuto, so there is never a regime where the usual perturbative calculation of asymptotic freedom is valid.
Back to the 3D case Despite that the 4D YM theory is a little bit cleaner in the weak coupling limit than the 3D case but the details of the gravity dual in the 3D is simpler. To nd a gravity dual we would like some solution which satises Einstein equations in the bulk with a negative cosmological constant 1 Rab gab = gab 2 (3)
1 which asymptotes at the boundary to R2 Sy . There are two solutions which satisfy these require ments
y NOTHING
zm
z=0
(a) Compactify a circle on AdS (b) The other solution which is slightly more interesting is ds2 = where f =1 L2 dz 2 , dt2 + dx2 + f dy 2 + z2 f z4 , 4 zm y y + 2Ry = (4)
(5)
Note that f has a zero at zm where the radius of the y circle shrinks to zero. For a random value of z there will a canonical decit angle. For z < zm , there is NOTHING, i.e, when the circle shrinks the space ends. We can make the solution much more well dened if we demand that there is no singularity at z = zm . Locally near z = zm , ds2 = 2 (z)dy 2 + d2 + , where, = zm (z zm ), = 2 2 zm Surface gravity (6)
(7)
This form of the metric makes it clear that the shrinking of the y circle is like a cone where the y direction is bred over the direction which shrinks to zero at z = zm . To avoid the conical decit, we chose the periodicity of y such that y y y + 21 + 2 = = and the metric becomes ds2 = 2 d 2 + d2 + (9) and = 0 is the origin of the polar coordinates. If we chose a dierent periodicity there will be a conical decit angle. To match the boundary conditions and since we know that y is periodic with a period 21 where = 2/zm , then zm = 2Ry . The topology of the boundary is R2+1 S 1 and the topology of the bulk is R2+1 D2 (z, y), where D2 (z, y) is a disk with angular coordinate y and the y circle is contractible in the bulk which implies APBCs on the bulk Fermions. Side Comments The importance of APBCs on the circle for the existence of a solution where the geometry ends is quit reminiscent of the instability of the KaluzaKlein vacuum [1] to the appearance of a bubble of nothing, which can only exist if we put APBCs on the Fermions, and is similar to the region z < zm here. We should think of the two solutions which we described as two saddle point contributions to the path integral of gravity with the specied boundary conditions. The solution which dominates is the one which has a smaller action when evaluated on the saddle point. John claims that solution (b), which has a mass gap, dominates over (a) and its free energy wins for all values of Ry . (8)
X ,t
1,2
ds2 = W (z)2 dx dx +
dz 2 . z2
(10)
Heuristically, the reason why there is a mass gap is that if we have a particle propagating in this space it will fall to this minimum value of the warp factor which acts like a gravitational potential, i.e, lowest energy states are localized around z = zm unlike the Poincar AdS case where particles e fell to the Poincar horizon. This statement will be made more precise next time. e
The spectrum of the 2+1 theory from the 2 point function Lets think about the two point function in momentum space of some gauge invariant operator such as the glueball operator O = TrF F . In a unitary eld theory, the two point function has the spectral representation 1 , (11) G2 (k) = O(k)O(k) = d() 2 k 2 2 0 where () is the density of states in the eld theory which couples to the operator O, () = O(, k)0
2
(12)
and unitarity implies that () 0. If there is a mass gap then will have support only above some min , ( < min ) = 0. (13) If there is a discrete spectrum then () =
i
ci ( Mi )
(14)
this need not be the case if there is no mass gap. Consider a bulk scalar with mass m in this conning background with the action S= g ()2 + m2 2 ,
(15)
where we kept only the quadratic part since we are interested in the two point function. For example, could be the dilaton with its zero momentum mode on the sphere (m = 0) couples to the glueball O = TrF F . To compute G2 we use the following ansatz for = k (z)eit+ikx (16)
and without loss of generality we will set k = 0. This is justied by the fact that as long as there is no instability then modes with nite k will have higher energies than modes with k = 0. Now we plug the above ansatz for in the bulk wave equation 1 0 = g00  2 k + z ( ggzz z k )m2 L2 k g (17)
In the interior, z = zm is like the origin of polar coordinates and and regularity at zm implies that z z=zm = 0 (18)
this condition replaces the condition of niteness at the Poincar horizon for the usual AdS. At the e boundary, we should impose that the solution is normalizable. This because we want to nd the spectrum of the theory in the vacuum without any sources. Remember that the general solution of the wave equation near the boundary behaves as z 0 + z + A, 4 (19)
where is the dimension of the source 0 , + is the dimension of the operator to which 0 couples, and we showed before that A O 0 . (20) If the source is small then we can use linear response (or by Taylor expanding A in 0 ) to write A O
0
+ O(2 ) G2 0 = G2 0
A , 0
(21)
the above equation says that if we nd a normalizable solution with 0 = 0 for some value of the frequency and A = 0, then the Greens function G2 will have a pole. There is a general argument that poles in G2 never arise from A [2]. The general lesson from the above discussion is that to nd poles of the Greens function, we look
for normalizable solutions of the wave equation.
The wave equation after plugging the metric takes the form (we will drop the subscript k on )
2 = z D1 z (z D1 f z ) + m 2 L2 = L z2 (22)
for some dierential operator L. This is an eigenvalue equation which can be written as a Schrdinger o 2 = M 2 for equation (Pset 4). Poles in the Greens function G2 will occur at the special values of i which L has normalizable eigenfunctions subject to the boundary condition at z = zm . Multiplying both sides of (22) with and integrating by parts we get zm zm 00 2 2 (23) dz g gzz (z )2 + m2 L2 2 > 0 dz gg  = 0 0
0 and =0 i =0
The spectrum is as if the bulk geometry were compact (in that there are discrete energy levels), but there is no zeromode since the zeromode is not normalizable in AdS. Mass gap: Without zm , we would have an oscillating solution near z = , i.e. continuous spectrum down to = 0.
2 Region of validity: If = gYM (N = 4)N 1, we have a weakly coupled 3D YM at E 1/Ry which is UV completed by N = 4 SYM on S 1 . The gravity description in this case is bad. For 1, the gauge theory is strongly coupled and the gravity description is good.
1 r << (mass gap)
z =zm
References
[1] E. Witten, Instability of the KaluzaKlein Vacuum, Nucl.Phys. B195:481,1982. [2] D. T. Son and A. O. Starinets, Minkowskispace correlators in AdS/CFT correspondence: Recipe and applications, JHEP 0209, 042 (2002) [arXiv:hepth/0205051].
We will begin to cleave some blazing swath through the set of results of classical gravity. We will see that black hole mechanics is thermodynamics. Our next goal will be to answer the question: Thermodynamics of what? First, though, lets tie up some loose ends from the last lecture.
Loose Ends
Last time we wrote down a conning solution and argued that it had a mass gap. There was the question of why this argument fails in AdS. In the conning solution is normalizable so that
2 =
(z )2
(1)
are not only greater than zero, but less than . In AdS, = z 2 J (qz), which, as z , goes as z 2 J (qz) z 2 k2 > 0 so that dz 2 = (3)
D D1 2 D
cos(z +
with q =
1 )(1 + O( )) 2 4 z
(2)
A rough denition of a black hole is that it is a region of space from which there is no escape (at least classically). We can, to zeroth order, argue the existence of black holes even without GR. If a heavy object creates a gravitational potential well, the velocity that a test particle would need to escape at distance r is vesc , where 2 GM vesc 2 . (4) r r Under some reasonable assumptions about physics, no particle can go faster than c, and so if the test particle is within rs = 2GM (the Schwarzschild radius (for the 2 we need GR)) of the mass, c2 it will not be able to escape. Thus, a BH forms if the massive object has a radius less than rs . The boundary of the region of no escape is called the event horizon. If what were interested in is just some sort of classical evolution outside of the horizon, we never have to ask what sits behind it. An important question which might annoy us: when we say escape, escape to where? This, in general, depends on the asymptotics of the space. For example, one might be tempted to say that a massive particle in AdS can never escape, since the geodesics just oscillate around the center of AdS and never reach the boundary. However, we should probably be more careful in our denition of a BH so that this case is excluded. Most of the theorems were about to talk about were proven for the asymptotically at case (i.e. not AdS), so we may have to think about how to apply them. The horizon is a global concept. To know where the horizon is, you actually need to know the whole future evolution of the spacetime. The denition of the horizon in the asymptotically at case is the boundary of the past of future null innity. It makes sense if you stare at it long enough. Really.
The Schwarzchild BH
ds2 = f dt2 +
(5)
with f 1 2GM (the emblackening factor). As r , f 1 and the space becomes r asymptotically at (Minkowskian). In this spacetime, there exists a timelike killing vector = t which generates time translations. At the horizon r = rs , however, this Killing vector becomes null as
 =
g =
f (rs ) = 0
(6)
Behind the horizon, theres a real singularity at r = 0, but the classical evolution outside the horizon is unaected. Which is nice. We can use GR even though we dont know what happens behind the horizon. This fact is so convenient that the relativists have elevated it to a grand conjecture.....
5 BH Uniqueness
In some sense, black holes are actually extremely simple. The uniqueness theorems (or no hair theorems) state that there are only a few numbers which parametrize stationary BHs (those with a timelike killing vector): its mass M , its angular momentum J, and its long range gauge charges (i.e. electric or magnetic charges), Qi . Black holes are the simplest macroscopic objects there arein this respect they are very much like elementary particles. Also, excitations above stationary solutions evolve back to the stationary solutions by ringdown via quasinormal modes (which have Im > 0). The decay is due to energy falling into the horizon. In asymptotically at space, the black holes can also radiate to innity via gravitational waves. You should think of the black hole as some equilibrium which is parametrized by a small number of state variables. If you perturb the equilibrium, it returns.
Assuming cosmic censorship, and the null energy condition (which states that Tab ka kb 0 null k), Hawkings area theorem states that during the evolution of any black hole, the area of the horizon cant decrease. The idea of the proof is in two parts. If the area ever decreases ( dA < 0), then the horizon develops crossing points at some nite later dt time. Consider a bundle of null geodesics which span the horizonthat is to say, every point on the horizon lies on some null geodesic that stays on the horizon. Consider the congruence of such a geodesics. These curves have some tangent fuctor ka = dx where is an ane parameter. The d fact that this is a geodesic with ane parameter implies that ka a kb = 0. Its useful to introduce a quantity known as the convergence of the congruence, which, roughly speaking, measures how the geodesics are spreading out, and hence of how the area is changing d ln(dA) (7) d If 0 then the area of the horizon is increasing. We will next use some GR magicthe Raychaud huri focusing equation: d 1 = 2 + 2 + Rab ka kb (8) dt 2 is something thats called the shear of the congruence. Its precise denition need not concern usall we need to know here is that 2 0. The third term in the equation is, via the Einstein equations, proportional to Tab ka kb , and hence by the null energy condition, also greater than or equal to zero. Therefore d 1 2 dt 2 Integrating, this gives 2
(9)
()
2 o
(10)
with (0) = 0 . From this we can see that if the initial 0 is positive (that is if dA < 0), then d for some nite 20 . A place where the convergence blows up is known as a crossing point or a causticit is where two geodesics cross and is a singularity in the congruence. However, this does not necessarily mean that it is a singularity in the spacetime itself. The second part of the proof of the area theorem states something like this cant happen. One can use some global properties to show that the horizon generators would have to end before , and this would imply a singularity. However, we cant quite use cosmic censorship now, because its not clear that some drunken, barely clothed singularity cannot exist. Theres some small deformation of this argument that works, though.
NB: if the denition of the convergence (7) makes you slightly uncomfortable, as it did me, theres a completely kosher, if much less illuminating denition of it on p. 217 of Wald (where it is labeled and called the expansion.) A nice derivation of Raychaduhuris equation follows, along with a similar derivation of the rst part of this proof.
There are two ways one might imagine using black holes to extract energy and, say, run your toaster.
7.1
Box Lowering
One can use a BH to extract the rest energy of a particle of mass m. Along a geodesic, a particle has a conserved energy given by E = mx ( ) , where is the timelike killing vector. If we hold the particle at a xed r, , , then the velocity vector takes the form x =  (the particle is moving  in the time direction only). Therefore the energy is E = m, which goes to m as r , and goes to zero as we go to the horizon of the black hole (as we have already discussed, becomes lightlike at the horizon). So if you carefully, quasistatically lower a box to the horizon, you can extract the rest energy, m, in the process.
7.2
Penrose Process
Rotating (Kerr) black holes have an ergosphere that lies outside the horizon for which the timelike killing vector becomes spacelike. The associated energy, therefore, is now proportional to the product of a timelike vector and a spacelike vector and may be negative. The basic idea of the Penrose process is that you send into the ergosphere some mass of energy E0 attached to a bomb. In the ergosphere, the bomb explodes and the mass fragments into two pieces, one of which falls into the horizon and has E2 < 0. By energy conservation, the other piece, which can now leave the ergosphere has E1 > E0 , more energy than it went in with. In this way you can extract the rotational energy of the BH. One can show that this process is most ecient when the area of the horizon doesnt change (something like the adiabatic limit). If the change in area is not zero, then the process is irreversible. This analogy with thermodynamics is becoming hard to ignore...
dM = dJ + dQ + 5
(11)
with M, J, Q, A the state variables mass, angular momentum, charge and area, and the angular velocity, the electrostatic potential and something known as the surface gravity, about which we will have more to say shortly. The nal term in this equation is analogous to the T dS term in the rst law of thermodynamics, although it is unclear which constants to assign to T and which to dS; well work this out in future lectures. Nevertheless the surface gravity is something like the temperature and the area is something like the entropy. Here we sketch the proof for uncharged, nonrotating black holes with = = 0. Lets consider some quasistatic process in which we throw some small bit of mass into the black hole. The ux of the conserved energy current is Tab b , where T is the stress energy tensor. Thus the ux through the horizon is
M =
Tab a kb d dA
horizon
(12)
k is a normal vector to the horizon (it is also tangent, since the horizon is null). The (unproven inlecture, though plausable1 ) claim here is that a = ka , and so
= = = = =
(13)
In the rst line we have used Einsteins equations, in the second weve used an identity that comes from playing with the focusing equation (i.e. the focusing equation expanded to rst order in the
Most of the proof is as follows (thanks go to Julian Sonner for help with this): the statement that the ka are tangent vectors to anely parametrized geodesics means that kb a = 0 (the tangents are parallellytransported and dont change length). On the other hand, a denition of surface gravity (for a nonrotating black hole; for the proof of equivalence to the one we discuss below, see Wald section 12.5 where he proves the zeroth law) is a a b = b . Putting these together gives a = Aet ka ,
1
for some constant A. Next we use the denition of tangent vector and the chain rule: ka = dxa dt dxa dt a d = = = Bet d d dt d dt B t e
and we set the initial value 0 to zero for convenience. This gives: A a = ka . B I actually dont know how to show that the constant is
A B
= 1.
small change in the mass). In the third, weve integrated by parts, and then used the denition of convergence in the fourth. For the nal line, weve used the usual shadiness that comes with manipulating dierentials in thermodynamics (i.e. no shadiness at all really). To tie up a loose end, surface gravity can be dened in a few dierent ways. One can take the near horizon limit of a black hole; the metric will have the form
ds2 = d2 2 2 dt2
(14)
for some which we call the surface gravity. Alternatively, it is the thing that satises the equation
(15)
1 4M .
Next time, well complete the laws of thermodynamics, and then well show how theyre wrong (classically) and x them with quantum mechanics.
In todays lecture well discuss the laws of black hole thermodynamics and how AdS black holes are related to nite temperature CFTs, and Koushik will give a related presentation.
Laws of Thermodynamics
Recall from last time that for a black hole Area Entropy T where was the surface gravity. The nearhorizon metric is ds2 2 2 dt2 + d2 + ... = 2 2 d 2 + d2 + ... (2)
(1)
when we go to Euclidean time it. If has periodicity +2/ then the euclidean geometry is regular. Recall the canonical ensemble thermal partition function is Zth = treH/T (3)
where eH/T propagates the system with imaginary time t = 1/iT . Thermal equilibrium is equiv alent to periodic euclidean time with period 1/T , so we identify with temperature T . The laws of (stationary) black hole thermodynamics, analogous to the usual laws of thermodynam ics, are: 0th (thermal equilibrium): is constant over the event horizon. This means temperature is constant in space and time. Thus stationary black holes are in thermal equilibrium with constant temperature. John thinks the proof of the 0th law doesnt depend on the shape of the black hole, as long as its a stationary solution.
dA (+P dV ) 8G
(4)
dJ is the change in rotational energy, dQ is the electrical energy, and 8G dA = T dS is heat exchange. This law relates the change in the energy (or equivalently mass) to changes in various properties of the black hole.
The last term describing mechanical work P dV isnt present for black holes but IS for black branes... 2nd (entropy increases): This is the area theorem for a black hole we proved last lecture, A A 0, since S = 4G . (Proof of the exact relation between S and A in a later lecture.) 3rd (absolute zero entropy): (or rather T ) cannot taken to zero in a nite number of steps. This doesnt mean that S(T = 0) = 0, but it does probably mean at T = 0 there is a minimum in entropy. These laws follow from Einsteins Equation, the energy condition we discussed last class, and assuming we have stationary black holes.
1.1
3rd law
Since we discussed the 2nd law last time, and the 0th and 1st laws are pretty convincing, we now provide some evidence for the validity of the 3rd law. First, why isnt it true that S(T = 0) = 0? Counterexamples are everywhere if you just open your eyes to them: It is well known to some people that there exist supersymmetric theories with LARGE ground state degeneracies, eQ where Q is the charge and is some power. So, S(T = 0) = ln(degeneracies) Q . The KerrNewman black hole is another counterexample. Here are some facts about the KN black hole that you can easily derive or look up: A = 4(2M (M + ) Q2 ) = M 2 Q2 J 2 /M 2 = 4/A
(5)
The black hole is extremal when = 0. (This is also the BPS bound when the black hole is supersymmetric.) If < 0 then there is a naked singularity. Note that that when = 0, T = 0 but S A = 0. As for the claims of the 3rd law, we have some anecdotal evidence. Lets consider a nonextremal KN black hole with J = 0, in other words a nonextremal RN black hole, so Q < M . (Everywhere Q is really  Q). 2
What can we do to try to make this black hole extremal? We need to throw on some charge q and mass m, such that the black hole becomes extremal, namely M +m=Q+q (6)
How, the mass m is attracted to the beautiful black hole by a force F M m/r 2 but the charge q is repulsed by a force F Qq/r 2 . Thus for the matter to fall in freely, M m > Qq. According to some mysterious algebra, this relation along with Q < M actually implies Q + q < M + m. Therefore you have to force the matter onto the black hole, which somehow adds heat and prevents you from cooling the black hole. Or you have to throw in innitesimal little bits which takes FOREVER.
Were going to use the power of AdSCFT to describe CFTs at nite temperature with black holes. In particular we mean a 3 + 1 dimensional relativistic CFT. The partition function is Z( ) = treH/T = eF/T (7)
1 with free energy F , on a space with geometry Sth 3 where the S 1 has radius 1/T, + 1/T and 3 is some 3 manifold. We can give 3 nite volume as an IR regulator.
This is a deformation of the IR physics (modes with T = EKK dont notice). For large V3 = V ol(3 ), then F = cV3 T 4 which is clear from extensivity of F and dimensional analysis.
This object goes by many names, such as planar black hole, Poincare black hole, black brane... This is a black hole in AdSD+1 , but probably many of the equations below mean D = 4. The metric is dz 2 L2 2 2 2 f dt + d + x ds = z2 f 4 z f = 1 4 (8) zm We again put the coordinates on a nite volume space, for example in box of volume V3 , x x 1/3 x + V3 , periodic BCs. Notice that if f = 1 we get the Poincare AdS metric, and in fact f only deviates from 1 at larger z representing the fact that this is an IR deformation. Whence:
It solves Einsteins equations with a cosmological constant = (D+1)(D+2) and asymptotes 2L2 to Poincare AdS, diering only in the IR region with a horizon at z = zm , xed t. Its the double Wick rotation of the conning solution with t = iythere , y = itthere . Analogous to how we got the AdS solution from the nearhorizon limit of D3branes, its the nearhorizon limit of black 3branes in R9,1 , in particular the nearextremal RR soliton with geometry: 2 dr f dt2 + d 2 x 2 2 2 ds = + H(r) + r d5 f (r) H(r) L4
H(r) = 1 + 4
r r4 f (r) = 1 H (9) r 4 (Again f = 1 gives the usual RR soliton.) Note that there also exists a black hole which asymptotes to GLOBAL AdS (with boundary S 1 S 3 ), which is known as AdSSchwarzchild which describes a CFT on S 3 at nite temperature T. Lets check out the horizon properties so we can nd the usual thermodynamic quantities were interested in. The nearhorizon metric is ds2 2 2 d 2 + d2 + L2 2 d x 2 zm (10)
where = 2/zm and the temperature is T = /(2) = 1/(zm ). Meanwhile the area of horizon is A= 3 gd x = L zm 3 V3 (11)
z=zm ,xedt
Therefore the entropy is (in the deconned phase of the gauge theory): S= (Recall from long ago that L 3 V3 N2 2 2 A = = (T )3 V3 = N V3 T 3 3 4G5 4G5 zm 2 2 =
N2 2 .)
(12)
L3 4G5
Again, we want some anecdotal evidence (at least) to support the claim that this describes a CFT in thermal equilibrium. There are some checks. One is just by checking the rst law, which relates horizon quantities such as T, S to global quantities such as the free energy F . Koushik will explain just how in his talk, as sketched below. Consider again the partition function which we claim is ZCF T = eSg = eF . Sg is the onshell gravity action for the black hole solution and is equal to Sg = SEH + SGH (+Sct ). (13)
EH means EinsteinHilbert and GH means Gibbons Hawking, SGH M dD x K is a boundary term for the action such that when we vary Sg we get the usual equations of motion. ct means 4
counterterm, as we need to subtract some divergences as r 0. You can see hepth/9902121 for some details. Thus by plugging in the AdS planar black hole solution we obtain the free energy, and consistent with the entropy calculation we did above we obtain 2 2 4 F L2 1 = = N T 4 V 16G5 rH 8 (14)
V = V3 probably. Also, zm = rH . This we claim is the strongcoupling free energy. As for the eld theory calculation it has been done at weak coupling and one obtains 4/3 times the answer at strong coupling. Lets consider also the boundary stress tensor T which couples to the induced metric on the boundary . The energy is E = V Ttt and the pressure P involves similar expressions with x, y, z components of T . This is also a straightforward calculation using the action above and one 4 4 obtains (E/V ) = 3/(2rH ) and P = 1/(2rH ) which satises E = 3P , so T = 0 as required in a 4 CFT. One can do the same for the AdS Schwarzchild black hole, where E = 3/(2rH )+3L2 /(8G). As T 0, rH but E is nonzero and in fact matches the calculation for the zeropoint (Casimir) energy for N = 4 SYM on a sphere of radius L. (There is no extra 4/3 factor because...)
Previously we have stated the third law for blackhole thermodynamics. To rephrase the result: 3rd Law of blackhole thermodynamics. Extremal blackholes are unnatural. In spite of this, extremal blackholes are beloved by string theorists, because the computations are easier. In particular, Sextr. BH (T = 0) = ln(# of ground states), and SUSY implies that this is independent of couplings and hence can be computed in the weak coupling limit. Previously we have also stated the second law, but in which we only considered blackholes alone. We now consider a more general version the second law in which both blackholes and normal stu are included: Generalized 2rd Law (GSL) of blackhole thermodynamics. [Bekenstein PRD 9 3292 (1974)] Stotal = Snormal stu + Stotal A GN 0 for physical processes.
where is a number yet to be determined. The idea behind the GSL is that it will be possible to build perpetual machines if it is false. However, there is a problem with the loweringthebox energy extraction scheme described in a previous lecture. To be more specic, suppose we lower an (innitesimal) box of entropy from to the event horizon and extract energy along the way as described in the previous lecture. When the box is at the event horizon, we know that Ebox (r = rs ) = 0, and if we let go at that point, A = 0 for the blackhole according to the rst law. Hence the GSL is violated. The other laws of classical blackhole thermodynamics also contain aws: 1
Flaw # 1. Classically a blackhole cannot radiate, so TBH = 0 ? Flaw # 3. Classically, the area of each individual blackhole must be nondecreasing. But thermodynamics should require only that Stotal 0. Thus the blackhole thermodynamics and ordinary thermodynamics are not in good parallel.1
The apparent contradictions in the blackhole thermodynamic laws can be resolved by realizing that the blackhole entropy SSek = A/ GN contains a factor of , hence SBek in the classical limit ( 0) when A and GN are xed. In particular, for the boxlowering scenario, even though the change in area A is innitesimal, it can still lead to a nite change S in entropy. Hence there is no contradiction with the GSL. Similarly, from the 1st Law dM = TBek dSBek . As not lead to a contradiction in the classical limit. 0, TBek = /8 0. Hence aw # 1 does
Also, as 0, any nite A for any blackhole will lead to S = , so aw # 3 does not lead to a contradiction in the classical limit.
Our major result is the Hawking radiation law: Hawking Radiation. A blackhole radiates like a blackbody with T = TBH = Lets rst consider some consequences before considering the explanation: A . 4 GN . 2
2. The area of individual blackhole can decrease, thus xes aw # 3 of GSL. 3. The resulting dA satises GSL for reasonable equation of state (E.O.S.). For example, con sider radiation in 4d. We have E = (1/4)aT 4 and S = (1/3)aT 3 . Treating blackhole as a 4 dE blackbody in at space, dSrad = 3 TBH , while energy conservation implies that dM = dE dE dE and hence dSBH = TBH . Summing up2 , dStotal = d(Srad + SBH ) = 1 TBH > 0. 3 4. Lets return to the boxlowering puzzle, (a) Bekenstein bound (Conjecture). Entropy in a box of linear size R must satisfy Sbox 2ER.
(b) Buoyancy force. Consider a box near event horizon, so that its top side is at r2 and the bottom side is at r1 (see Fig. 1). The local temperature T1 (T2 ) on the top (bottom)
1 2
Perhaps this is not a big deal, since individual blackholes can merge. Consequently, this says that the speed of sound is c2 sound = 1/3.
side satises T2 /T1 = gtt (r1 )/ gtt (r2 ). Hence T1 > T2 . Consequently, there is more radiation from the bottom than from the top3 , which produces a buoyancy force displaced radiation. As a result, there is a critical rcrit > rs for which E(displaced radiation) = Ebox (rcrit ). By dropping the box at rcrit , SBH = E/TBH , but Sbox Sdisplaced radiation = E/TBH Now lets return to the question of why TBH takes this particular form. For a generic (possibly nonextremal) blackhole, the event horizon is lo cated at a zero of the emblackening factor f . For example, for Schwarzchild blackhole, ds2 Sch dr 2 = f dt2 + + r 2 d2 f 2 R2 dt2 + dR2 + r(R)2 d2 dT 2 + dZ 2 + dX 2 + dY 2 where in the above we have used:4 f = 1 rH /r, rH = 2GM , R(r) = r(r 2GM ) + . . ., = 1/4GM , = t, T = R sinh and Z = R cosh . Notice that after the transformation, we obtain the Minkowski space R3,1 . In other words, the Rindler space is a Minkowski space in the coordinate frame of a uniformly accelerating observer. Graphically, what we get is Fig. 2. Notice that the two lines dened by {r = 2GM, = } and {r = 2GM, = } divide the space into four regions { I, II, III, IV }. The key observation is that region I is selfcontained. i.e., region II and III cant communicate with it, while information from region IV passes through the line {r = 2GM, = } and hence corresponds to initial data. Therefore, at T = 0, the Z < 0 (marked by L on the gure) degree of freedom (d.o.f.) are totally irrelevant and useless for region I. Thus we should trace over them when computing stu in region I. Now, Claim: R = trL g.s.g.s. = where HR is the riddler Hamiltonian:5 HR = and Z = trR e2HR .
3 4
T2 T1
r2 r1
eve
nt ho riz on
= R2 d 2 + dR2 + . . .
1 2HR e Z Tab a nb
const. T surface
analogous to more water at the bottom of sea is called the Rindler time and plays the role of rapidity 5 Note that [HR ] = 1 since the riddler time [] = 1.
r < 2GM
2G M
I
R
II
to singularity
to asymptotically at space
III
r =
xed R xed
M 2G , =
IV
Notice that the density matrix we obtained is not a pure state but is entangled with the L d.o.f. (hence S = tr R ln R = 0). This is a thermal density matrix with temperature Triddler = 1/2. Proof of Claim [Unruh 1976]. Consider a scalar eld in Rindler space. A complete set of commuting d.o.fs at T = 0 is R (x, y, z) for Z > 0 (x, y, z) = L (x, y, z) for Z < 0 R and L commute because they are at spacelike separated points. Any wavefunctional of the eld can then be written as = [L , R ] = L R . The ground state is just the Minkowski space vacuum. The FeynmanKac formula gives us a path integral representation of its wavefunctional: 1 g.s. [L , R ] = [d]eSEucl [] Z x0 >0,(x,x0 =0)=(L ,R ) This equation is written using constantx0 = iT time slices. We can gain some insight by in stead slicing up the path integral by constant euclidean Rindler time (= i) (see Fig. 3(a) for
II
x0 (= iT )
I
R Z
III
Z
(ii)
(iii)
(i)
(= i) L event horizon
(a)
IV
(b)
Figure 3: (a) Change in slicing sequence. The magenta lines are the old x0 slices while the blue lines are the new slices. (b) Fluctuation in dierent regions of the Rindler space. illustration); euclidean Rindler time is just the angular coordinate in the x0 , Z plane. This yields:6 1 1 g.s. [L , R ] = (1) [d]eSEucl [] = L eHR R 0<< Z (=0)=R Z
(=)=L R (R=0)=0
The last condition R (R = 0) = 0 is simply a requirement that the function is regular at the origin. We identify the RHS of (1) L eHR R as the transition amplitude. Hence, R R  R = [dL ] [L , R ][L , ] R 2 1 = [dL ]R eHR L L eHR  R Z 1 = R e2HR  R Z Some results and comments: 1. a constantly accelerated observer in Rn,1 sees a Unruh radiation with7 Tproper = R a = . 2 2 2. By the equivalance principle, the same happens to a blackhole. The corresponding Schwarzchild time is t = /, which implies that TBH = Triddler = . 2
We are transforming the variable that the function takes, rather than the function itself, and it is the function thats being integrated. Thus there is no Jacobian involved here. 7 2 2 here the proper time is dened by ds2 proper = R d
3. Whence does the radiation comes from? There are three types of uctuations (see Fig. 3(b)): (i) is just ordinary pair production uctuation, with lifetime /E (ii) is something that we need not care in region I (iii) corresponds to stu that enters at = , R = 0 and falls back at = +, R = 0. In Schwarzchild time this particle stays forever. Hence this particle is real. 4. What we have calculated is the density matrix of a quantum eld at r rH . Not all energy will get to . Thus we have to include a greybody factor, so that: (b) Np f = where p is the absolute cuto for mode p. 5. Information paradox. Blackhole can form from dictionary and then evaporate into thermal radiation. The thermal density matrix is determined by one number only. Q1. Is this a unitarity evolution? Q2. Entropy results from coarsegraining. What are the microstate? Q3. How is the information stored in the blackhole? via AdS/CFT it is possible to answer Q1 and Q2. p /TH e 1
Previously in lecture 23 we gave evidence that the Black Hole thermodynamics of the AdS Black Brane (nonextremal) solution was dual to the thermal ensemble of N = 4 SYM on R3 at large N and . Thus the gauge theory provides the microstates that are being coarse grained by the BekensteinHawking entropy of the black hole SBH . A few comments on this observation:
1.1
Hydrodynamics
Perturbing the equilibrium of the boundary theory with a kick will result in thermalization relaxation back to equilibrium. In the bulk the response to such a kick is for the energy of the kick to fall into the black hole. The above two statements are related by the duality. In the long wavelength and small frequency limit both are consistent with the hydrodynamics of a relativistic CFT. Additionally the duality allows one to compute various transport coecients of the gauge theory at large , such as the shear viscosity, Rcharge conductivity, etc. See the review [1].
1.2
Thermal Screening
At nite temperature T correlators die o exponentially at large separation r 1/T , even if in vacuum there are massless elds which mediate long range interactions. This is because such particles develop a thermal mass from continuously interacting with the thermal bath. For example the force between two external charges in a gauge theory should behave at large distances as, Vqq (r) emth r (1)
r1/T
T T
T T
Figure 1: A particle receives multiple kicks from its thermal surroundings. This generates a thermal
contribution to the particles mass.
where mth is the thermal mass.
The qq potential can be calculated using a Wilson loop,
= P exp i exp iVqq (r)T A WCr T
Cr T
(2)
where the contour Cr is a rectangle with spatial separation r and temporal separation T not to be confused with temperature. The expectation value of the Wilson loop is taken in a thermal ensemble. The above formula is dened for T r large. Of course as we learned in a previous lecture we can compute Wilson loops at large using the duality. One must nd the minimal action string in the bulk which approaches Cr at the boundary. Then the expectation of the Wilson loop is WCr = exp(iSmin (Cr )). There are two (stable) saddle points to this problem1 The two string congurations are shown in Fig. 2. For r < r the curved string has the smallest action Smin = VU (r)T while for r > r the two straight strings win Smin = V T which is independent of r. Where the point of cross over is r 1/T = zH .
z0 0.4 0.2 0.2 0.4 r 0.4 0.2 z0 0.2 0.4 r 0.4 0.2 z0 0.2 0.4 r
zH
zH
zH
Figure 2: The two string congurations as a function of r (increasing from left to right) with the minimum action conguration highlighted in red. For large enough r the curved string solution ceases to exist (its action becomes complex!) Both actions diverge at the boundary. To yield a nite result for the qq potential we must renor malize by subtracting V .
1 There is a third (unstable) saddle point, which one can think of as coming from a local maximum of some ducial potential.
(3)
ren For small r 1/T the potential approaches the T = 0 result; Vqq r (1 + c(rT )4 + . . .). See Fig. 3. Since the potential vanishes for r > r there is no trace of the interaction, and the gluons mediating the interaction are screened (more than the expected exponential fall o. 2 )
0.0
0.2
0.4
0.6
0.8
1.0
1.2
0.5
1.0
1.5
Figure 3: The red line follows the actual potential. The other curves represent string congurations with higher action. In particular we have included a third conguration, the highest curve, which corresponds to the unstable solution mentioned in footnote 1.
1.3
PolyakovSusskind Loop
The Polyakov loop is an operator dened in the Euclidean theory by, A U = P exp i
1 S
(4)
The expectation value of this operator gives the free energy in the presence of an external charge in representation R; trR U = eFq (T )/T (5) This gives an order parameter for connement at nite T , connement, trU = 0 Fq = nite deconned trU = 0
2
(6)
From this one may conclude that the screening length is r . However there is another candidate for the screening length which comes from the next order correction to V ren in the 1/ expansion: the two long strings may still interact via exchange of a massive glueball in the highT eectively 3+0 conning gauge theory. This results in a potential emgball r where mgmall is the lightest mass glueball of this 3d theory. See [2].
1 In the gravity dual we can compute trR U using a string which ends on S . The 5 dimensional 2 Euclidean AdSBH has the topology R3 D where D is a disk with boundary S at the boundary of AdS, as in Fig. 4.
z=z H
z=0
D
1 S
1 Figure 4: The Euclidean time circle at the boundary S closes o smoothly in the bulk Euclidean black hole solution. A string is begging to be placed here.
This geometry follows from the form of the metric, L2 ds2 = 2 f (z)d 2 + . . . (7) z Since f (z) vanishes at some z = zH the Euclidean time circle which is periodically identied shrinks 1 to zero size at this point. 3 So S is contractable in the full geometry, and we may wrap a string world sheet on the disk D. It follows that the expectation value of the Polyakov loop is non zero, trU eAD / = 0
(8)
where AD is the area of the disk D. Note actually AD is innite, coming from the usual UV divergence associated with an innite quark mass probe. This can be regulated in the usual way to yield a nite result. We thus conclude that N = 4 SYM at strong coupling is not conning on R3 for any temperature T.
We now want to consider the more interesting case of N = 4 SYM on S 3 which does indeed exhibit a connement to deconnement phase transition. Along the way we will see that the AdS/CFT dual of this phase transition implies a dramatic consequence for Quantum Gravity; we must sum over geometries and topologies (consistent with certain boundary conditions.)
2.1
the at directions are lifted by the conformal coupling term RX 2 there are no zero modes for the fermions because we have taken antiperiodic boundary con ditions for them around the thermal circle (thus breaking SUSY.) there are no harmonic 1forms on S3 so the gauge elds have no zero energy modes A0 can have a zero mode, however it is not propagating: the equation of motion S/A0 = 0 is the Gauss law constraint. Gauss law on a compact surface implies that all physical states must be color singlets, even under the global part of the gauge group SU (N ). This is the (trivial) statement of kinematic conne ment. We can reduce the theory on spherical harmonics on the S3 such that the operators in the theory are Mi {KK modes of:A , , X}. All Mi s are in the adjoint representation. So low lying states are given by tr (M1 . . . MS )  > (9) where the trace ensures only gauge singles contribute. The energy of these states is E = S/RS3 and the density of states with xed S is the number of ways to order and choose the Mi s. This density of states grows exponentially with E, (E) eE/Thag where Thag is called the Hagedorn temperature. For small E, such that E N 2 , the density of states is independent of N , so at least for small T s only these states are excited and we have, F (T 1/RS3 ) O(N 0 ) (10)
the free energy is independent of N . For higher temperatures states with energy of order N 2 become important in the thermodynamic ensemble (in particular this must happen before the Hagedorn temperature where thermodynamic quantities begin to diverge). The number of such states is now no longer independent of N , simply because there are trace relations for (9) when S N 2 and the counting of such states changes. In order to estimate F for high temperatures we can use the fact that for the CFT of N = 4, the physics only depends on the dimensionless parameter T RS3 . So large T s at xed RS3 can instead be achieved by taking RS3 large at xed T . Then the theory limits to a theory on R3 at xed temperature. For the free theory we can estimate the free energy as that of a collection of N 2 free elds, F/T N 2 (RS3 )3 T 3 (11) where (RS3 )3 is the volume factor. Hence the free energy now scales like N 2 , which is a signal of deconnement. At nite N this deconnement phase transition becomes smooth, see g. However, even in QCD (N = 3) it is a very sharp, dramatic feature (QCD does not need to be put on a sphere to see this, because QCD already connes on R3 .) See [3] for more about this.
2.2
We would like to nd the Gravity dual of the situation discussed in the previous section. This was discovered by [5] and applied to the AdS/CFT correspondence by [4]. To nd the gravity dual we look for an asymptotically AdS Solution whos boundary is (conformal) to S 3 S
. That is look for a local extrema of the action I[g] = 1/(16GN ) d5 x g R + 12/L2 .
The answer is the (Euclidean) AdSSchwarzschild Black Hole, which for convenience later we will call the space X1 : ds2 = h(r)d 2 + dr 2 /h(r) + r 2 d2 (12) p where for p = 3, h(r) = 1 + r 2 /L2 /r 2 (13) For large r this metric becomes AdS in stereographic coordinates (with the boundary at r = .) The mass can be computed using the usual formula, 16GN M M= Ttt = 3 2 S3
(14)
The horizon is located where h(r) vanishes. There are two roots, we will look at regions outside the largest of the two r > r+ . 2 r 4 1 1 1+ 2 = (15) L2 2 2 L The temperature is found by demanding no conical singularity at r = r+ : T (r+ ) =
2 2r+ + L2 2L2 r+
(16)
Importantly this T (r+ ) has a minimum as a function of r+, see Fig. 5: that is there is a mini mum temperature T = T1 = 2/L for which there exist black hole solutions. Also above this temperature there are two types of black holes: large and small. 2 L2 T 1 1 2 2 2 r+ (T )large = (17) small 2 L T These have the following thermodynamic properties;
small < L. For r 4,1 Small BH: r+ + L they look like black holes locally on R , that is Schwarzschild black holes. Their mass satises, 2 M small (T ) r+ 1/T 2
small
(18)
So the the specic heat is negative CV = dM < 0. So not only do Schwarzschild BHs dT evaporate, this process cannot be hindered by placing them in equilibrium with their sur roundings because of this thermodynamic instability: if its surroundings have higher T , the BH eats some energy, grows bigger and decreasing its T . 6
Small BH
Large BH T1
r rmin L 2
Figure 5: The temperature T as a function of horizon radius r+ . This is equivalent to Jeans instability in at space; there is no canonical ensemble for gravity in at space.
large Large BH: r+ > L and one nds 4 M large (T ) r+ T 4
(19)
such that CV > 0. So a large BH can be in equilibrium with its surroundings, such as its own Hawking radiation. The heuristic reasoning for this is that AdS acts like a box (an infrared cuto on the canonical ensemble) for the Schwarzschild black hole. In this case if the surroundings have higher T the Schwarzschild BH will eat some energy, grow larger and decrease its T , however now because the surroundings are nite they will also decrease their T . If this is enough to compensate for the decrease in the black holes T an equilibrium can be obtained. From now on the space X2 will refer to the large BHs described here. For temperatures smaller than T < T1 we have not found a solution with the correct assymptotics. Fortunately there is a solution for all T s which is no longer a black hole solution. It is simply the Euclidean AdS space periodically identied, + 1 . This space is called thermal AdS and will 1 be denoted X1 . Note that S is no longer contractable. In order to be able to compare to X2 we must have the same antiperiodic boundary conditions for the fermions (we did not have the choice of periodic fermions for X1 because S was contractable.) X1 and X2 have dierent topologies S 1 B4 and B2 S 3 respectively. Where Bn is an n dimensional ball with boundary Sn1 . This situation is nicely represented by tents in Fig. 6. In order to compare the actions of the two solutions we must make sure we have the same T RS3 for both X1 and X2 at the uv cuto, r = r . This condition denes 1 in X1 in terms of T from
S1 B4 S3 S1 S3
S3
B2 S1
Figure 6: Left: Thermal AdS (X1 ) which is topologically S 1 B4 . Right: AdSBH (X2 ) which is B2 S 3 . X2 . 1+
(1) g 1
2 r 1 = L2
g 2 1+
2 r 2 1/T 2 L r
(2)
(20) (21)
Now recall the central formula of the AdS/CFT correspondence relating the partition function of the CFT to the string partition function, 1 ZCF T (S 3 S ) eIsugra (Xi ) (22)
saddlesi
which for large 1/GN is dominated by the saddle point Xi with the smallest action (free energy.) The actions are innite and require some sort of renormalization. We achieve this by taking the dierent in actions of the two spaces, I = I(X2 ) I(X1 ) = lim 1 (Vr (X2 ) Vr (X1 )) r 2GN L2 (23)
Because R g for these spaces the action is simply proportional to Vr the regulated volume of the space. The dierence in action is nite in the limit where we take away the cuto r . 1/T r=r 3 4 4 1/T (r r+ ) (24) d drr 3 d3 = Vr (X2 ) = 4 r=r+ 0 1 r=r 3 4 Vr (X1 ) = d drr 3 d3 = 1 r (25) 4 0 r=0 (26) Then one nds, I = From which we conclude, r+ < L (lowT): I > 0, such that thermal AdS (X1 ) has the smaller action. The free energy which follows from the CFT partition function is, F 0 N 2 + O(N 0 ) 8 (28)
3 2 3 r+ (L2 r+ ) 2 8GN 2r+ + L2
(27)
This conclusion requires a little more careful analysis of the local renormalization counterterms, used in order to regulate I(X1 ). However because the volume Vr (X1 ) only depends on the upper radius r it should be completely removed by such counter terms. Then ZCF T e0/GN from which (28) follows.
1 The Polyakov order parameter for connement satises: trUX1 = 0 since S is not contractable. Both the above results indicate that the lowT phase is conned.
r+ > L (highT): I < 0, so the large BH solution dominates. In this case the free energy is, F N 2 . The Polyakov order parameter for connement satises: trUX2 = 0 since now 1 is contractable. Indicating a deconned phase at hight T . S
2.3
Using x = trU  as our order parameter we can sketch the free energy as a function of x for various temperatures. (Note this is a sketch because we dont know F or x away from the saddle points that we have computed above.)
F F
small BH
small BH
th. AdS
Figure 7: From left to right, top to bottom: (T < T1 ) Thermal AdS is the only saddle point. (T1 < T < THP ) At T = T1 two new solutions appear corresponding to the small and large BHs. The local maximum is unstable and corresponds to the small BH. One can check that xsmall < xbig and Fsmall > Fbig as suggested by this graph. (THP < T < THag ) At T = THP the large BH wins thermodynamically over thermal AdS. (T > Thag ) String theory on thermal AdS becomes 2 unstable: x F (x = 0) < 0. This is due to a tachyon in the spectrum of a closed string wound around the thermal circle: m2 = 1 + Nosc + 1/(T 2 ) which gives a Hagedorn temperature THag = 1/ 1/4 THP . 9
References
[1] D. T. Son and A. O. Starinets, Ann. Rev. Nucl. Part. Sci. 57, 95 (2007) [arXiv:0704.0240 [hepth]]. [2] D. Bak, A. Karch and L. G. Yae, JHEP 0708, 049 (2007) [arXiv:0705.0994 [hepth]]. [3] O. Aharony, J. Marsano, S. Minwalla, K. Papadodimas and M. Van Raamsdonk, Adv. Theor. Math. Phys. 8, 603 (2004) [arXiv:hepth/0310285]. [4] E. Witten, Adv. Theor. Math. Phys. 2, 505 (1998) [arXiv:hepth/9803131]. [5] S. W. Hawking and D. N. Page, Commun. Math. Phys. 87, 577 (1983).
10
1. holographic RG ows and an atheorem [dHoker Freedman review, hepth/0201253, 910] 2. string duals of gauge theories with matter in the fundamental [KarchKatz, hep th/0205236; SakaiSugimoto, hepth/0412141, hepth/0507073] and thermal aspects [see refs in review by Erdmenger et al 0711.4467] 3. more systematic understanding of the realtime AdS/CFT prescription [Son et al, hepth/0205051, hepth/0212072; Skenderisvan Rees, arXiv:0805.0150; IqbalLiu, 0809.3808.] 4. gravity duals of supersymmetric conning gauge theories with chiral symmetry breaking [KlebanovStrassler, hepth/0007191; MaldacenaNunez, hepth/0008001; Klebanov review, hepth/0205100] 5. AdS/QCD [Erlich, Katz, Son, Stephanov, hepph/0501128; Karch, Katz, Son, Stephanov, hepph/0602229] 6. entanglement entropy from the gravity dual [RyuTakayanagi, hepth/0603001, hepth/0605073 and citations thereof; see in particular Fursaev, hepth/0606184] 7. 2d strongcoupling transport in a magnetic eld [Hartnoll, Kovtun et al, 0704.1160, 0706.3215]; constraints from EM duality in the bulk [Witten, hepth/0307041, Kovtun et al, hepth/0701036] 8. holographic models of superuidity and superconductivity [HHH, 0803.3295,
0810.1563, Kovtun et al 0809.4870, Shieh et al 0809.4494]
9. expanding stronglycoupled plasma [JanikPeschanski, hepth/0512162, hep
th/0606149, 0706.2108]
10. charged particles in stronglycoupled plasma: jet quenching [LiuRajagopalWiedemann, hepph/0605178, hepph/0612168], and/or string dragging [Karch et al hepth/0605158; CasalderreySolana, Teaney, hep ph/0605199; Gubser et al, hepth/0605182, hepth/0605292] 11. attempts to see a fermi surface in the gravity dual [SungSik Lee 0809.3402; Rozali et al 0708.1322; KarchSonStarinets, ParnachevKulaxizi...] 12. niteN eects: the stringy exclusion principle [MaldacenaStrominger, hep th/9804085] and giant gravitons [McGreevySusskindToumbas, hepth/0003075; Myers et al, hepth/0008015; Itzhaki et al, hepth/0008016.] 13. a solvable limit of AdS/CFT [BMN, hepth/0202021;last section of hepth/0309246; GubserKlebanovPolyakov, hepth/0204051] 14. spin chains and the dilatation operator [MinahanZarembo, hepth/0212208 and its vast ocean of citations, especially Kruczenski, hepth/0311203] 15. cusp anomalous dimension from string theory [Kruczenski, hepth/0210115] 16. BPS Wilson loops [EricksonSemenoZarembo, hepth/0003055, DrukkerGross, hepth/0010274] 17. c 1 strings and matrix models, an example of holography [Kiritsis, chapter 15, review by Yu Nakayama, hepth/0402009] 18. AdS3 and 2d CFT [MAGOO 5, e.g. MaldacenaOoguri, hepth/0001053, hep th/0005183, hepth/0111180] 19. holography for largeN vector models? [KlebanovPolyakov, hepth/0210114] 20. behind the horizon by analyticity [Maldacena, eternal BH in AdS, hepth/0106112; Shenker et al, hepth/0306170; FestucciaLiu, hepth/0506202] 21. matrix models of black hole information con sumption [FestucciaLiu, hepth/0611098; IizukaPolchinski et al, 0801.3657, 0808.0530] 22. string theory and the physics of hadrons [PolchinskiSusskind, hepth/0112204, PolchinskiStrassler et al, hepth/0109174, hepth/0209211, hepth/0603115] 23. microscopic accounting of BekensteinHawking entropy for BPS black holes [StromingerVafa, hepth/9601029 et cit.]
24. Birkho/nohair, quasinormal modes, and thermalization [HorowitzHubeny, hepth/9909056] quasinormal modes and hydrodynamic modes [KovtunStarinets, hepth/0506184] 25. computation of hydrodynamic green functions [PolicastroSonStarinets, hep th/0205052, 0210220] 26. anything else we dont get to in lecture thats on the syllabus.
Problem Set 1
Reading: 4 of dHokerFreedman http://arXiv.org/pdf/hepth/0201253
1. Branes ending on branes. The Dpbrane eective action contains a term of the form S F Cp1,
Dp
where Cp1 is the RR p 1 form, which couples minimally to D(p 2)branes. Show that a D(p 2) brane can end on a Dp brane without violating the Gauss law for the RR elds involved. Interpret the boundary of the D(p2)brane in terms of the worldvolume theory of the Dp brane. (If you like, focus on the case p = 3.) 2. Timelike oscillators are evil. Show that the commutation relation [a, a ] = 1 (which we found for the oscillators made from the time coordinates of the string) implies that either a) the energy H = a a + E0 1 is unbounded below (if you treat a as the annihilation operator) or b) there are states with negative norms. 3. Extremal ReissnerNordstrom black hole. As a warmup for the 10d RR soliton, lets remind ourselves how the extremal RN black hole works. a) Consider EinsteinMaxwell theory in four dimensions, with action 1 1 4 SEM = d x g R F F 16GN 4
1
Show that the Einstein equation 0 = SEM implies that g 1 2 . R = aGN 2F. F g F 2 for some constant a. b) Consider the ansatz ds2 = H 2 () dt2 + H 2 () d2 + 2 d2 , 2 1 F = bdt d H()
where b is some constant. Show that the Einstein equation 0 = SEM and g SEM Maxwells equation 0 = A are solved by the ansatz if H is a harmonic function on the IR3 whose metric is ab dxa dxb := d2 + 2 d2 . 2 ab 1 Recall that H is harmonic i 0 = H = a ( b H). c) Find the form of the harmonic function which gives a spherically sym metric solution; x the two integration constants by demanding that i) the spacetime is asymptotically at and ii) the black hole has charge Q, meaning F = Q. S 2 at f ixed d) Take the nearhorizon limit. Show that the geometry is AdS2 S 2 . De termine the relationship between the size of the throat and the charge of the hole. [If you get stuck on this problem, see Appendix F of Kiritsis book.] d) If youre feeling brave, add some magnetic charge to the black hole. You will need to change the form of the gauge eld to F = bdt dH() + G()2 where 2 is the area 2form on the sphere, and G is some function. 4. RR soliton. In this problem were going to check that the RR soliton is a solution of the equations of motion. The action for type IIB supergravity, when only the metric and the RR 5form and possibly the dilaton are nontrivial can be written as 1 1 5 5 ..... 10 2 d x g e + ... SIIB = (R + 4 ) F..... F 16GN 5! 2
(The selfduality constraint F 5 = F 5 must be imposed as a constraint, and means that dF 5 = 0 implies the equations of motion for F 5 .) By the way, this is the action for the string frame metric. a) Show that the equations of motion from this action imply 1 5 2 5 5 .... 2 R = aGN e 5F.... F g (F ) 2 for some constant a. b) Plug the following ansatz into the equations of motion: ds2 = 1 H(r) dx dx + H(r)dy 2
F = b(1 + )dt dx1 dx2 dx3 dH 1 = 0 (b, 0 are constants.) Determine the constant b and the condition on the func tion H for this to solve the equations of motion. To do this, there are two options some kind of symbolic algebra program like Mathematica or Maple, or indexshuing by hand. The latter is much more easily done using tetrad or vielbein methods. I always forget these and have to relearn them every time. For a lightning review of the vielbein method of computing curvatures, I recommend dHokerFreedman http://arXiv.org/pdf/hepth/0201253, pages 100101, or Argurio http://arXiv.org/pdf/hepth/9807171, Appendix C. To help with the former option, Ive posted an example curvature calculation in Math ematica on the pset webpage. Note, by the way, that for values of p other than 3, the dilaton is not constant. With hindsight, this specialness of p = 3 is related to the fact that this is the critical dimension for YM theory, where gY M is dimensionless.
1. How to remember the N = 4 action. Show that tendimensional N = 1 SYM dimensionally reduces to 4d N = 4. The 10d lagrangian density is L10 = 1 tr FM N F M N 2iM DM ; 2 2gY M
M, N are 10d indices; here is a 10d MajoranaWel spinor (16 real com ponents). By dimensionally reduce, we mean consider the 10d theory on a 6torus of volume V , and restrict to eld congurations which have no momen tum along the torus (i.e. are independent of the coordinates on the torus). The 4d N = 4 lagrangian density is1 1 1 2 1 L4 = 2 tr F + D X i D X i + gY M 4 2
i I D I 2 2
1 [X i , X j ]2 + I i [X i , J ] . IJ 2 i<j
Here I are four 4d Weyl spinors (4 real components), are 4d gamma ma trices, and i are gamma matrices for SO(6). IJ 2. Show that the N = 4 lagrangian explicitly breaks the diagonal abelian part U(1) U(N = 4)R
of the Rsymmetry group.
3. Show that the 1loop function for N = 4 SYM vanishes. If you dont like arcane group theory, do it for the case when the gauge group is SU(N).
1
This footnote corrects the expression I wrote in lecture (the Yukawa terms were wrong).
4. N = 4 N = 1. A 4d N = 4 supersymmetric theory is a special case of a 4d N = 1 supersym metric theory. a) Describe the eld content of N = 4 SYM in terms of multiplets of some N = 1 subgroup. b) Write the N = 4 lagrangian in N = 1 superspace. c) Convince yourself that this Lagrangian has SO(6) Rsymmetry (e.g. by examining the Lagrangian written in components). Is N = 1 supersymmetry plus this Rsymmetry enough to convince you that this action is actually N = 4 supersymmetric? 5. Extremal = BPS. In this problem we will consider 4d N = 2 supergravity (which has eight real supercharges). Along the lines of the discussion from lecture 5 one can see that the graviton multiplet for this theory must also contain a spin3/2 gravitino eld , and an abelian vector eld (the graviphoton) A . So this is a simple supersymmetric completion of the system we studied in problem set 1 problem 3. The action is invariant under supersymmetry transformations which include the following variation for the gravitino : 1 = F 4 where 1 F (F ) 2 are the selfdual (SD) and antiselfdual (ASD) parts of the eld strength of the graviphoton eld. The covariant derivative acting on spinors is 1 ab + ab 4
F F , F =
where is the spin connection, dea = ab eb and ab 1 [a , b]. 2 a) Show that the extremal RN black hole from problem set 1 lies in a short (BPS) multiplet of the N = 2 supersymmetry. Do this by examining the supersymmetry variations of the fermionic elds (given above) and showing 2
that they vanish for some choice of spinors . How many supersymmetries (choices of , which are called Killing spinors) are left unbroken? b) Show that the BertottiRobinson solution, i.e. AdS2 S 2 (which arises as the nearhorizon limit of the extremal RN black hole) preserves eight Killing spinors. c) If you are feeling energetic, show that AdS5 S 5 preserves 32 supercharges of 10dimensional type IIB supergravity. The supersymmetry variations of the fermion elds in that case are i = F + ... 1920 5 = ... where
is the dilatino eld and ... means terms that vanish when F5 is the only nontrivial eld other than the metric (i.e. they depend on derivatives of the dilaton, and the other uxes). If you want to know what the full variations are, see Kiritsis eqns (H.2628). 6. Wbosons from adjoint higgsing. Using the N = 4 lagrangian, show that giving an expectation value to the scalars in the N = 4 theory of the form X i=1,...,6 = diag(xi , . . . xi ) 1 N gives a mass to the odiagonal gauge bosons of the form m2 ab
6 i=1
(xi xi )2 xa xb 2 ; a b
this is meant to be the mass of the gauge boson with indices ab. Show that the scalars with the same gauge charges get the same mass.
1. Show that the conformal algebra of IRp,q , with generators P , C , M , D and commutators given in lecture, is isomorphic to SO(p + 1, q + 1). 2. [much more optional than others] Show that the generator of an innitesimal conformal transformation in d > 2 must be atmost quadratic in x. That is, convince yourself that (for d > 2) 2 + = d implies that all third derivatives of must vanish. [I dont actually know if there is a simple way to do this.] 3. [more optional than others] In a generic matrix quantum eld theory, give some kind of proof by induction on the number of vertices and propagators that a planar vacuum diagram always contributes an amplitude of order N 2 . 4. Useful coordinates in AdS Lorentzian AdSp+2 is the locus L = ab X X (X
2 a b p+2 2
) (X ) +
0 2
p+1 i=1
(Xi )2
()
inside Rp+1,2 with metric ds2 = ab dX a dX b . a) Show that in the global coordinates dened by X p+2 = L cosh sin X 0 = L cosh cos 1
X = L sinh i ,
p+1 i=1
2 = 1 i
the induced metric on the locus () becomes ds2 = L2 cosh2 d 2 + d2 + sinh2 d2 p with d2 = di di the metric on the unit psphere, S p . p b) The Poincar patch coordinates x , u are dened via the relations e X p+2 + X p+1 = u X p+2 + X p+1 = v ux . X = L Using the dening equation to eliminate v in terms of the other variables, show that the induced metric takes the form ds2 = L2 du2 u2
+ 2 dx dx .
u2 L
L2 : u
dz 2 + dx dx . z2
The boundary of AdS is at z = 0. Think about what part of the AdS space is not covered by these coordinates. Hint: z runs only over positive values, since at z , the timelike killing vector x0 becomes null. c) Starting from the global coordinates, introduce r L sinh . Show that the metric takes the form ds2 = Hdt2 + H 1 dr 2 + r 2 d2 p with H 1 +
r2 . L2
With the metric in this form, the angular part of the Einstein tensor G is linear in H and is proportional to r (r p r H). Using this information write down the metric for SchwarzschildAdSp+2 , i.e. the spacetime that contains a black hole and is asymptotic to AdS at innity.
5. Stereographic projection coordinates [more optional] Lets think about the euclideansignature hyperbolic space: Hd =
2 {Xd
d i=1
X i X i = L2 } IRd,1
i=1
X i = i
2L2 , r i i. 2 r2 L
a) These coordinates arise by considering the line from a point P on the hy perboloid Hd to the point Q = (L, this line intersects the plane X d+1 = 0 0); at some point P whose position in the i direction is i . Try to parse this last sentence, i.e. draw a gure. b) Show that the induced metric on the hyperbolic space is ds2 = 6. Geodesics in AdS a) Given a stationary observer at xed in global AdS, how long does it take (on his clock) for a radiallydirected lightray (i.e. a massless geodesic) to leave him, hit the boundary of AdS, and come back? (Assume that the observer is living in a circumstance where there are reecting boundary conditions on the electromagnetic eld.) b) Examine the behavior of a massive geodesic in AdS (with zero angular momentum on the spatial sphere). Show that the motion is bounded away from the boundary of AdS. 7. Volumes and areas are not so dierent in AdS In a xed slice of the global coordinates, consider the region . Show that A() lim V () is nite, where A and V are the surface area and volume inside the specied region. 4L2 d id i (1 r 2 )2
1. One more symmetry. Dene xA (z, x ), so that the Poincar patch metric takes the form e ds2 = dz 2 + dx2 dxA dxA = . z2 z2 xA xB xB
z i.e. z z 2 +x2 , x z 2x x2 . This is a conformal transformation which is not + continuously connected to the identity.
[Bonus question: what does this do to the region of the global AdS space covered by the Poincar patch coordinates?] e 2. Schrdinger description of AdS instabilities.1 o a) Consider the scalar wave equation in poincar AdS, in momentum space: e 0 = ( m2 ), with (x, z) = eikx (z) By redening the independent variable, rewrite the scalar wave equation as a Schdinger equation:
o
2 (z + V (z))(z) = 2(z)
with 2 playing the role of energy. (Hint: Rewrite (z) = A(z)(z), and choose the function A to set to zero the term in the wave equation multiplying (z).)
I learned about this trick from Sean Hartnoll. If you get stuck or want to see a recent application of this technique, see e.g. appendix A of 0810.1563.
1
b) A property of a eld conguration which generally determines whether it is allowed to participate in the physics (i.e. whether it is xed by boundary con ditions, or whether it can happen on its own) is whether it is normalizable. In Euclidean space, this generally just means that the euclidean action evaluated on the conguration is nite, since then it contributes to the path integral with a nite Boltzmann factor eS[] . In Minkowski signature, a eld conguration should be considered normalizible if it has nite energy; for a scalar eld, this energy is E[] TAB A nB
where the integral is over some xedtime slice , n is a unit normal vector to , is a timelike killing vector, and TAB is the stress tensor for the bulk eld, 2 S TAB g gAB . Given the redenition you found in part a), what is the relationship between normalizable wave functions in the usual QM sense (i.e. 2 < ) and nor malizable solutions of the AdS wave equation (i.e. those with nite energy)? c) Set k = 0. Show that when m2 passes through the value m2  from above, BF this Schrdinger equation develops a single normalizable negativeenergy state.2 o D2 (Recall that the BF bound is m2 m2 = 4L2 .) This corresponds by the BF map above to a normalizable mode with imaginary frequency i.e. a linear instability. 3. Dimensions of vector operators. By studying the boundary behavior of solutions to the bulk Proca equations, derived from the action m2 1 p+2 Sbulk [A] = F F + A A d x g 4 2 AdS compute the scaling dimension of the current coupled to a bulk vector eld A of arbitary mass: Sbdy = dp+1x A J .
AdS
For purposes of this problem, dene to be the power of z of the subleading powerlaw solution of the bulk equation near the boundary at z = 0.
If you want to learn more about potentials of this form, see H. Hammer and B. Swingle, Annals Phys. 321 (2006) 306317 [arXiv:quantph/0503074v1].
2
4. Saturating the unitarity bound.3 For this problem we will work in Euclidean AdS. Feel free to redo the problem for the Lorenzian case (using the notion of normalizability described in problem two). a) Consider the usual bulk action for a (free) scalar eld in AdSD+1 : 1 dD xdz g ()2 + m2 2 Susual [] = 2 Show that for m2 L2 > D + 1, 4 Susual [z ] = . That is, the integral over the radial direction in this action diverges near z = 0 when evaluated on a solution of the form z , where is the smaller root of ( D) = L2 m2 ; such a solution is nonnormalizable. On the other hand, show that z + gives a nite value, and hence a solution with these asymptotics is normalizable. With this action, show that the smallest conformal dimension = + of a scalar operator in the dual theory is D . 2
2
b) Show that the following modied action 1 D+1 2 d x g ( + m ) = Susual [] n . SKW [] 2 AdS
produces the same bulk equations, but has the following property: in the win dow of mass values (we could call this the KW window) D2 D2 < L2 m2 < + 1, 4 4
both roots give normalizable solutions with respect to the action SKW (i.e. SKW [z ] < ). What is the lowest conformal dimension you can obtain now? (Hint: you can now consider + to be the dimension of the source, and to be the dimension of the operator.)
(where wA (w0 , w)A labels a point in the bulk of AdSD+1 ) is the bulkto boundary propagator for a massless vector eld, i.e. this object solves the bulk Maxwell equations, and D (w ) as w0 0. Here x
M M JA (x) A 2
xA xM . x2
Show that
M so this JA
(J is the conserved current to which the bulk gauge eld AA couples) has the form required by conformal symmetry given in lecture. c) [extra bonus problem] Show that the form of the three point function above is determined by conformal invariance up to a constant. 2. Relation between AdS propagators. Show that the bulktoboundary and boundarytoboundary propagators for a scalar eld in AdS are related by K (z, x; x ) = lim + G (z, x; z , x ). z 1
(n (n )) ,
with = G, = K. The G appearing in this relation is dened to be the normalizable solution to the waveequationwithsource 1 x m2 G(z, x; z , x ) = D (x x )(z z ) g which is regular in the interior; K is dened to be the solution to the homoge neous wave equation which is regular in the interior and approaches lim K (z, x; x ) = D x x .
z
[Note: it is possible to show this using properties of the hypergeometric function appearing in the explicit expression for G; this proof is not so illuminating and requires actually knowing the exact solution] 3. Wilson line with cusp. Use the AdS/CFT duality to compute the strongcoupling behavior of W [v]CF T , i.e. the vev of a Wilson loop (in the fundamental) associated to a curve de scribed by a v of opening angle , in a CFT with a AdS string dual. Show that the renormalized expecation value behaves like lnW [v]renT ln CF L 4 ()
where is some function of the opening angle which vanishes as approaches (this means that the discontinuity in the line is small). and L are UV and IR cutos on the radial coordinate in AdS (i.e. < z < L in the coordinates we have been using). 4. Surface gravity. Compute the periodicity of y for which this metric is regular at z = zm (i.e. has no conical decit): dz 2 2 2 2 ds = (z) f (z)dy + + dsother f (z) 2
where f (z) is a function with a rstorder zero at z = zm (i.e. z f (zm ) = 0 ) and (z) is regular and nonvanishing at z = zm . If we think of y as imaginary time, this periodicity determines the temperature of the black hole, since nite temperature means periodic Euclidean time. a) Specialize your answer to the case of the euclidean Schwarzchild black hole in at space, for which f (z) = 1 2GM 2 , = 1, ds2 x2 other = z d . z
b) Specialize your answer to the case of the euclidean AdS black hole (with planar horizon), for which f (z) 1 z4 1 , = 2 , ds2 x2 other = d . 4 zm z
This geometry is also relevant to the model of connement obtained by com pactifying a supersymmetric gauge theory on a circle with supersymmetry breaking (ScherkSchwarz) boundary conditions. c) Show that you get the same answer by computing the surface gravity of the horizon (the locus z = zm ), which can be dened by 1 2 a b c d gbd g ac z=zm 2 where a is the tangent vector to the shrinking circle, = y .
Note: These problems uctuate wildly in diculty. 1. 3/4. Convince yourself that in the N = 4 SYM, F ( ) 3 = ,
F ( 0) 4
where F is the (Helmholtz, i.e. canonical ensemble) free energy and is the t Hooft coupling. 2. A cubes worth of elds? Compute the BekensteinHawking entropy of the planar black hole in AdS7 S 7 . The metric is L2 dz 2 AdS 2 2 ds = 2 (f dt + d + x ) + L2 4 d2 , 4 S z f
2
with f =1
z6 . 6 zH
1
The parameters are related to those of the theory of M5branes by: L = LAdS = 2LS 4 = 2lp (N ) 3
9 where lp is the elevendimensional Planck length, lp = 16GN . Express the entropy in terms of the BH temperature and the number N of M5branes.
3. Warming your feet on the Hawking re. How big a black hole do you need to boil water (at 1 atm, say) with its Hawking radiation? 1
4. Holographic stress tensor Using the usual AdS/CFT prescription, compute the expectation value of the stressenergy tensor of the N = 4 theory at nite temperature, at large t Hooft coupling. Show that the trace vanishes, as it must by the conformal Ward identity. [If you get stuck, see BalasubramanianKrauss, hepth/9902121.] 5. Screening length. Consider the quarkantiquark potential in N = 4 SYM at large , N . Compute the screening length, r , the separation of quark and antiquark at which the (regulated) action of the straightstring solution equals that of the hanging string. [If you get stuck, see hepth/9803137, hepth/9803135, 0807.4747.] 6. Connement and monopole condensation. There is a beautiful qualitative picture of connement [due to Polyakov, t Hooft, Mandelstam] as a dual Higgs mechanism, meaning that it should in volve the condensation of magnetically charged objects. (This has been ex plicitly demonstrated in examples by SeibergWitten theory.) In 4d YM, this means condensation of magnetic monopoles, which means that magnetic charge should be screened in the conning vacuum: the force between magnetically charged object should be independent of distance beyond some minimum dis tance related to the connement scale. Consider the model of connement obtained from D4branes on a thermal circle. a) What is the string theory description of an external magnetic monopole at strong coupling in this model? (Hint: recall pset 1.) b) Show that magnetic charge is screened by giving a strongcoupling estimate of the force between two such static external monopoles. [If you get stuck, look in MAGOO 6.] 7. Giant gravitons. Consider a D3brane wrapping a (contractible) S 3 whipping around the S 5 of the gravity dual of the N = 4 theory. Using the parametrization given in lecture 19, show that the angular momentum of the brane becomes precisely = N when the brane reaches its maximum size. L 2
8. Area theorem in AdS? The event horizon is a global notion that depends strongly on the asymptotics of the space in which the black hole sits. Prove a version of the Area Theorem for asymptotically AdS spaces. The denition of event horizon for asymptot ically AdS spaces, I believe, is (brace yourself): the boundary of (the past of (the intersection of (future timelike innity) with (the conformal boundary of AdS))). [I am not sure if this has been done. Please tell me if you nd a reference or a proof!] 9. Wald entropy and second law? In Einstein gravity, the entropy of a black hole is given by the area of its horizon in Planck units (over four). The EinsteinHilbert term is the leading irrelevant operator in an eective eld theory; a microscopic theory of gravity such as a string vacuum predicts denite coecients for the higherderivative corrections. In such a more general theory of gravity, where the gravitational action can be written in terms of a lagrangian depending on tensors made from the metric, Igrav = L(gab , Rabcd ) a path integral argument shows that the horizon area is replaced by a more general integral over the horizon, often called the Wald entropy. In the case of gravity in four dimensions, the Wald entropy takes the form: L S = 2 ab cd horizon Rabcd By demanding that the resulting Wald entropy satisfy a second law of thermo dynamics, constrain the coecients of highercurvature terms in the lagrangian of a sensible eective gravity theory. [For help, see JacobsonKangMyers, grqc/9503020. I believe this is the stateoftheart on this question.]
I.
Lets rst remind ourselves why branes typically cant end. A D(p 2) brane action typically contains a term S Cp1
D(p2)
(1)
where Cp1 is a (p 1) form gauge potential whose gauge transformation is given by C(p1) = dp2 , p2 being a (p 2) form parametrizing the gauge freedom (compare with familiar electromagnetism, which has A1 and 0 ). Under this gauge transform the variation of the action is S =
D(p2)
dp2 =
D(p2)
p2
(2)
where we have used Gausss law in fancy dierential forms notation M dF = M F . This gauge variation must be 0 for all p2 ; this means that the boundary D(p 2) must not exist, and thus the brane has no end. On the other hand, consider adding the term mentioned in the problem set Snew = Dp F Cp1 , where Dp is some other pbrane. The gauge variation of this new term is Snew =
Dp
F dp2 =
Dp
dF p2
(3)
(where we have assumed that the Dp brane at least has no boundary to discard a boundary term) and thus the new condition for the gauge invariance of the action is Stotal =
D(p2)
p2
Dp
dF p2 = 0
(4)
This condition can be satised only if the boundary D(p 2) sits on the Dp brane and dF = 0 on D(p 2). In that case the D(p 2) brane can certainly end on the Dp brane. What does this look like from the point of view of the worldvolume theory on the Dp brane? Recall the equations of motion for a 2form eld strength F, d F = je dF = jm (5)
in terms of electric and magnetic currents je and jm . dF = 0 thus implies the existence of something coupling magnetically to the worldvolume gauge eld. Some fun numerology shows that on a (p + 1) dimensional worldvoume, je is always a 1form (the gauge eld couples electrically to a pointparticle sweeping out a onedimensional trajectory) but jm is a (p 2) form; in our case this (p 2) form is related to the volume form of the (p 2) dimensional boundary D(p 2). Finally lets specialize to p = 3; in that case we have D1 branes ending on D3 branes; their endpoints are just points that look like magnetic monopoles to the d = 4 YangMills theory living on the D3 brane.
II. TIMELIKE OSCILLATORS ARE EVIL
We have some oensive oscillators with [a, a ] = 1. Lets do the problems in reverse order. Consider the normal vacuum dened by a0 = 0 and compute the norm of 1 a 0 . We have 0aa 0 = 0 1 + a a0 = 1 00 (6) Thus if the vacuum has positive norm the excited states have negative norm and viceversa. Unfazed, we consider a funky vacuum 0f dened by a 0f = 0. Funky excited states dened by nf = an 0f /n! will have positive norm, and we appear to have a nice Hilbert space. We write the Hamiltonian as H = (E0 1) aa to suit our funky interpretation. Unfortunately if we compute the energy of 1f we get H1f = (E0 1 aa )a0f = (E0 2)a0f (7) and the energy is less than that of the ground state (E0 1)! The energy keeps getting more negative the more we excite the funky vacuum, and we see that it is unbounded below. We conclude that timelike oscillators are evil.
2
III. EXTREMAL REISSNERNORDSTROM BLACK HOLE
and vary with respect to g . The variation of the EinsteinHilbert term R gives the normal Einstein tensor (see e.g. Carrolls GR book chapter 4). Varying the Maxwell part has two pieces; one part from the variation of g, which 1 (leaving out the overall 1/16GN ) is + 8 F F g and one from the variation of the metric tensors tying together 1 the indices, which is 2 F F . The full equations of motion work out to be 1 1 1 R Rg = F 2 g + F F 2 8 2 (9)
which is of the form claimed in the problem set. Before computing the curvature tensors, note that taking the trace gives R = 0 as the righthand side is identically traceless. The fact that the stress tensor of electromagnetism is traceless is a consequence of the classical scale invariance of the Maxwell action and implies that a spacetime supported only by the Maxwell action has vanishing Ricci scalar. We will eventually need the curvature tensors in a tendimensional spacetime; thus for generality we will nd them for a D dimensional spacetime containing a ddimensional at Poincare piece parametrized by xi and a n dimensional at Euclidean piece (the transverse space) parametrized by y a . Our ansatz for the metric is ds2 = 1 dxi dxj 2 (y) ij H + H 2 (y)ab dy a dy b (10)
This is slightly more general than the ansatz in the problem set as that only had H depending on one of the y coordinates whereas we are letting it depend on all of them. Also note I am not using polar coordinates on this transverse space but at ones; the conversion to polar is trivial at the end but for now the at coordinates simplify life quite a bit. We now compute the curvature. This is straightforward but tedious 1 . I am not providing details of the calculation, as a much nicer explanation of a similar (slightly simpler) calculation can be found in Carroll, Appendix J. The answer is R = Rab H (H)2 (n d 3) + 4 2 H H H a Hb H a b H = (2n 4 4d) + (2 n + d) + ab H2 H
a
2 a H; these are all with respect to the transverse space. Ok, for now we
Rtt = Rab
1 (H)2 H + H4 H2 H a Hb H (H)2 H = 2 + ab 2 H H2 H
(13) (14)
which we know is 0 by the Einstein equations. Thus it appears that we need H to be a harmonic function on the transverse space. Next we turn to the Maxwell equation of motion 1 F = ( gF ) = 0 g (16)
By which I mean that it was incredibly unstraightforward for me, but in principle if you make no algebra mistakes it isnt too bad.
3
a Plugging in the ansatz Fta = b HH we nd that all factors of H 2 cancel and this becomes 2
a a H = H = 0
a
(17)
Thus Maxwell equations are also satised. We are not done yet, we also need to verify that all of the components of Einsteins equations (9) are satised. We write the intermediate results F 2 = 2b2 (H)2 H4 Ft Ft = b2 (H)2 H6
Fa Fb = b2
a Hb H H2
(18)
Combining these to form the righthand side of (9) we see that all the Einstein equations can be satised provided b = 2. Hooray! We now write down the most general form of H (subject to the boundary condition that H(y ) = 1. We know what this is from oldfashioned electromagnetism; it is the potential from a series of point charges. H(y a ) = 1 +
i
Ai y yi 
(19)
where {yi } provides a set of 3vectors and Ai a set of corresponding charges, both of which are arbitrary. This function is harmonic everywhere except at the poles y = yi , and it is an exact solution to the EinsteinMaxwell equations. It describes a multicenter black hole solution, where we have a set of black holes with centers located at yi , held somehow in a stable conguration by a delicate interplay of gravitational and electromagnetic forces. Beautiful. What is Ai ? To nd this we nally focus on a one blackhole solution and use polar coordinates. We also let H H depend only on . We obtain Ft = 2 H 2 , and Q
S2
F =
S2
Ft H 2 2 sin()d d =
dd sin()2A = 8A
(20)
Thus the full (oneblack hole) metric is ds2 = 1 2 dt + H 2 (d2 + 2 d2 ) H2 H() = 1 + Q 8 (21)
dt2 +
Q 8
d2 +
Q 8
d2
(22)
Q This is the metric on AdS2 S 2 with AdS length L = 8 . Finally, let us add some magnetic charge. To get the solution, lets use the fact that the Hodge dual of a solution with electric charge is one with magnetic charge. We take the dual of a solution with a pure radial electric eld Fe
Fm = Fe = b H sin()2 d d Now we superpose these two solutions to obtain something dyonic, i.e. with both electric and magnetic charge Fdyonic = H b d dt + a2 sin()d d H2
(23)
(24)
From here the analysis proceeds exactly as above. I will not go into details; suce to say that we now require a2 + b2 = 4 and the nal solution with electric charge Q and magnetic charge P has metric given by (21) except now with H()dyonic = 1 + Q2 + P 2 8 (25)
4
IV. RR SOLITON
Despite its horrifying name and birthplace, this is actually almost exactly the same problem as the one we just did, only with 2.5 times as many indices. We start with the action SIIB = 1 16GN 1 d10 x g e2 (R + 4M M ) (F 5 )2 5! (26)
We now need to vary this with respect to g MN . Note that if the dilaton is not constant this is a bit trickier than you might expect, as the variation of the EinsteinHilbert term will result in extra pieces that are normally a total derivative but now are not. Thus the Einstein equations of motion will contain extra pieces depending on gradients of the dilaton. Luckily we will examine only congurations where the dilaton is constant, so I will not keep track of these terms but you should know that they are there. Anyway, we obtain simply 1 e2 RMN gMN R 2 = 1 1 .... (F 5 )2 + FM.... FN 2.5! 4! (27)
Taking the trace again we obtain as before R = 0 (The RR veform is classically scale invariant in ten dimensions, etc. etc.) A new ingredient is provided by the selfduality of the veform; this means that F = F , and thus F 2 F F = F F = 0. The equation becomes RMN = Recall also that the ansatz is ds2 = 1 dx dx + h(y) h(y)ab dy a dy b (29) h, n = 6, d = 3 in the formulas e2 .... FM.... FN 4! (28)
Luckily we already worked out the curvature, once we make the substitution H 2 = (11) and (12): R = Rab h (h)2 4h h h2 1 a hb h ab (h)2 h = + 2 2 2 h 4 h h
(30) (31)
1 We now proceed in direct analogy to what we did above. Taking the trace we nd R = 2 h = 0, implying that h is h harmonic on the transverse space. We should check that the Maxwell equations for the veform are satised. These equations are just M ( gF MN OP Q ) = 0 (32)
(33) ab ab = h , g = 1/ h ,
4 1 a h h( h) ( h) Ftxyza = ba h2 2 = 0, h
(34)
implying yet again that the Maxwell equation is satised i h is harmonic. By now you probably believe me that this is a solution, but we should conscientiously check that all of the other components of the Einstein equations are satised. If we skipped this step, not only would we not have the pleasure of contracting F 5 with itself in entertaining ways, but we would also not be able to x the value of b. .... So we compute the righthand side of the Einstein equation, FM.... FN . The easy one rst:
.... F.... F = 4!b2
(h)2 h3
(35)
5 To understand this, note that if we x say the two indices t in Ft.... Ft.... and let all the rest vary, it is clear from the form (33) that to contribute the remaining indices must be a permutation of {x, y, z, y a }, where all a can contribute (I apologize for the bad notation; y with no superscript is one of the coordinates along the Poincare piece x . y a with a superscript is a coordinate along the transvers space). There are 4! such permutations for each choice of y a , .... resulting in the structure shown. Now the hard one, the computation of Fa.... Fb . To do this we rst note that (dt dx dy dz dy 1 ) 1 txyzy1 M dxN dxN dxO dxP dxQ = MN OP Q dx 5! = h2 dy 2 dy 3 dy 4 dy 5 dy 6 (36) (37)
Thus for example we see from (33) that F23456 Fy2 y3 y4 y5 y6 = b1 h; if all the indices are y a s, then the value of the component is the derivative of h with respect to the one y that isnt in the string. Now we are ready. There are two cases, start with a = b and take a = 1, b = 2 for concreteness. Then
.... abcd F1.... F2 = F1 F2 + F1abcd F2 = 4! h2 F1txyz F2xyzt +
1 F13456 F23456 h2
(38) (39)
= 2.4!b2
1 h2 h h2
h .... Thus for a = b, Fa.... Fb = 2.4!b2 a h2 b h . Now the case a = b = 1; going through an analogous computation we get 4!b2 .... abcd F1.... F1 = F1 F1 + F1abcd F1 = 2 1 h1 h + a ha h (40) h a=1
4!b2 h2
a hb h +
ab (h)2 2
(41)
Comparing with (30) and the equation of motion (27), we see that the Einstein equations are satised if the value of b is xed to be b = e /2. Hooray! Once again we see that we have a beautiful multicenter solution h(y) = 1 +
i 4 Ri y y i 4
(42)
where we can move the centers y i of the D3 branes wherever we like in the transverse 6dimensional space and they will stay where we leave them due to the delicate yet strangely robust cancellation of the RR veform eld with gravity. As mentioned in lecture, this bizarre property manifests itself in the dual eld theory description of this congurationremember from lecture 6 that the X i in N = 4 SYM parametrize movement in a 6N dimensional space along which the potential of the eld theory is at; there is no energy cost associated with changes in X. Now you know why. Finally we restrict to one D3 brane in a spherically symmetric setup. Doing the integral Q = S 5 F = 2e R4 3 we nd that R4 gs Q, where we have used gs = e . Q is the RR charge, and is quantized in appropriate units. Essentially we have found that
4 R 4 ls g s N
(43)
where ls is the string length scale and N is the integer number of RR ux quanta. This is the reason why AdS/CFT 2 maps a classical gravity setup to a dual conguration with a large tHooft couplingwe need gs N = gY M N to obtain a setup where R (the curvature scale of the solution) is much larger than ls and the classical supergravity solution can be trusted.
This problem set contains a high number density of fermions and their associated minus signs. While everything here appeared to work out, really that only means that I succeeded in making an even number of mistakes, so please treat all formulas here with caution.
HOW TO REMEMBER THAT N = 4 ACTION
I.
The 10d action that we are going to dimensionally reduce is as follows 1 S10d = d10 x 2 tr FM N F M N + 2iM DM 4gYM
(1)
Here M, N are 10d indices, , 4d indices, and i, j run over the six dimensional space we are reducing on torii. We will do this problem in various increasingly detailed stages. Depending on how much you like dimensional reduction and spinor representation theory, you can stop reading whenever you want. 1. No work at all: This 10d theory has 16 supercharges and no elds higher than spin 1. Its dimensional reduction to four dimensions also has 16 supercharges and no elds higher than spin 1. The only such theory in four dimensions is N = 4 SYM, whose Lagrangian is completely determined by its symmetries. Therefore it must work out. 2. Bosons only: Just for peace of mind we might as well work out the bosonic sector. Assuming that nothing depends on xi , we can write down the components of FM N : Fij = i[Ai , Aj ] Fi = Ai i[A , Ai ] F = A A i[A , A ] (2)
From here, the rst kinetic term from (1) becomes FM N F M N = F F + 2( Ai i[A , Ai ])( Ai i[A , Ai ]) [Ai , Aj ][Ai , Aj ] (3)
The components Ai are really just scalars in the remaining 4d space, so from now on we will call them Xi instead. Noting also that the middle term is really the covariant derivative of a scalar in the adjoint of the gauge group, we obtain FM N F M N = F F + 2(D Xi )2 2 [Ai , Aj ]2 (4)
i<j
which is indeed the bosonic part of the N = 4 4d action. Surely the fermions will also work out. 3. Fermions too: First, note that the 10d Majorana Weyl spinor has 16 real degrees of freedom. This is because the 10d Dirac spinor had 32 complex; the Weyl spinor thus has 16 complex, and the Majorana constraint cuts this down to 16 real (recall it is in ten (and two) dimensions that one can have a spinor that is both Majorana and Weyl). The N = 4 action has four Weyl spinors, each with two complex (four real) degrees of freedom; so the counting matches. We now need to work out how they reorganize themselves. This is essentially a nightmare of notation, and I encourage you to stop reading now. We are forced to introduce the bewilderinglooking 10d Dirac spinor aI (5)
where a runs over the four components of a 4d Dirac spinor and I over the 8 components of a 6d Dirac spinor (thus 32 components altogether). The fact that we are dealing with a Weyl spinor means that we need not worry about many of these components; to gure out exactly which ones we should keep track of we note that under the decomposition SO(1, 9) SO(1, 3) SO(6), the 16+ 10d Weyl breaks down 16+ (2+ 4+ ) (2 4 ) (6)
2 where 2 and 4 respectively denote positive and negative chirality 4d and 6d spinors. Note that the total chirality of the 10d representation is always positive (chirality is additive), which is what we want. We now introduce yet more notation: , denote normal 4d Weyl spinor indices in keeping with standard notation, and I denotes positive and negative chirality 6d Weyl spinor indices. Thus I goes from 1 to 4 (whereas I includes I+ and I and so goes from 1 to 8). Thus before we impose any sort of Majorana condition, we have a total of 8 Weyl spinors in 4d: I+ 0 I+ = I = (7) 0 I The Majorana condition should relate these two each other. To see how that works, we write the 10d charge conjugation matrix in terms of its 4d counterpart as 0 I4 (8) (10) CabI J = (4) Cab I4 0 I J where we have picked a basis on the 6d part so that the block diagonal structure is a division into dierent chiralities I (compare with the 4d gamma matrices in a Weyl basis)1 . The Majorana condition reads = C( ). This relates I+ to I above, letting us express both 6d chirality spinors in terms of 4 4d Weyl spinors I . I 0 I+ = I = (9) 0 I We now decompose various 10d matrices. The following decomposition satises the Cliord algebra
(10) abI J i (10) abI J
(4) ab
1I J
i (6) IJ
5 (4) ab
where , i are coordinate indices running from 0 to 3 (noncompact dimensions) and 4 to 10 (compact dimensions) respectively, and the subscripts under the matrices denote their dimensionality. We are nally ready to attack the kinetic term M DM . Treating M = rst, we obtain
0 D = (aI ) ab bc D c I = I ( ) D I + ( ) D I I
(13)
where in the second equality we have decomposed I into a sum over I and used the relation (9). This is precisely the kinetic term for four 4d Weyl spinors, with a manifest SU(4) symmetry. Next we work on the terms where = i; these are more interesting, and we obtain
0 5 0 5 0 5 i i Di = (aI ) ab bc i Di cJ = (aI+ ) ab bc i I+J Di cJ + (aI ) ab bc IJ+ Di cJ+ IJ
(14)
In the second equality we have used the fact that a single gamma matrix always relates spinors of dierent chirality, and thus we can decompose the 6d into only cross terms that are I+J or IJ+ : 0 IJ+ I J = (15) I+J 0 Some normal 4d gamma algebra now results in
J i I i I+J Di I IJ+ Di J
(16)
This looks promising. To make it look slightly nicer, we now relate I+J to IJ+ . This is done using the form for the charge conjugation matrix (8). By using the fact that CC 1 = phases , it is not hard to i i convince yourself that in the basis we are using i I+J IJ = (IJ+ ) . Thus we obtain for this part of the action i I i [Xi , J ] + (i ) [Xi , J ] (17) IJ I IJ Putting these pieces together one obtains the full action for the N = 4 theory, including the fermionic terms. It may look unfamiliar because we have written things in Weyl notation (compare e.g. with (3.1) in [1]).
I believe writing such a form for the 10d charge conjugation matrix is justied given that the 6d charge conjugation matrix changes the chirality of the 6d spinor
3
II. U (1) R SYMMETRY
The question here is: the N = 4 SUSY algebra has a U(4) R symmetry given by the following action on the generators QI :
J QI UI QJ
(18)
J where UI is a U (4) matrix. The I transform as the 4 under this U(4). The question is: is this symmetry of the SUSY algebra also respected by the action? Consider the SO(6) vector X i . This vector arose as the antisymmetric product of two 4s. Thus under the diagonal phase rotation U = ei , we expect something like X e2i . The action is obviously not invariant under such a mutilation of X i . On the other hand, the SU (4) part of this R symmetry is respected by the action, essentially because of careful use of SU (4) gamma matrix machinery to tie together the interactions of X i with those of I (the corresponding machinery does not exist for the full U (4)).
III.
FUNCTION
Since the focus of this class is on the wonderful things that one can do with eld theory without ever touching a Feynman diagram, we will not derive the beta function here but instead just look up the beta function for a gauge theory with Nf Weyl fermions and Ns complex scalars in a book. The answer is (see [2] p 106) where the coecient b is given by b= 11 1 1 T (adj) T (ra ) T (rn ) 6 3 a 6 n (20) dg b 3 = g + O(g 5 ) d 16 2 (19)
a runs over fermions, n over scalars, and T (r) is the index of the representation r. In our case everything is in the adjoint, and we obtain b= T (adj) (11 2Nf Ns ) 6 (21)
putting Nf 4, Ns 3 (three complex scalars, or six real ones) we obtain a vanishing 1loop beta function.
IV. N =4N =1
N = 4 has six real scalars, four Weyl fermions, and one gauge eld. To match these to known N = 1 multiplets, it looks like we need one vector multiplet V and three chiral multiplets M (for this problem M is an SU(3) index, I, J, etc. retain their previous meanings and a is an SU(N) adjoint index); we then end up with 3 complex scalars XM , a total of four fermions (one from the vector, one M from each chiral) and one gauge eld. The action for such a conguration is also very restricted. If we require an SU(3) symmetry along the three chiral multiplets and no mass terms we really have very little choice for the action: when written in supereld notation (see e.g. [3] p47) it becomes 1 2 2 4 M V 2 L= d tr(W W ) + c.c. + d ( ) e M + d M N P fabc a a a + c.c. (22) M N P 16 3 The only real freedom we had here was picking the coecient of the cubic supereld term. For the purposes of this problem I have also set the gauge coupling g to 1. Now: what are the symmetries of this action? There is a manifest SU (3), but we expect the existence of a hidden SU (4), allowing us to rotate all four fermions and M into each other (and having a nontrivial action on the Xs); this is not at all apparent in the supereld formalism. To see it, we need to expand the Lagrangian out into components. It is then rather easy to see that the kinetic terms for the fermions are happy with such a large symmetry. We will thus focus on the Yukawas: since one of these comes from the Kahler term and one from the superpotential term, rotating them into each other appears to be a nontrivial business.
4 The relevant terms from the Lagrangian work out to be ([3], p31 and p50) b a b c LY = 2fabc (X aM ) M c + i 2M N P fabc M N XP + c.c.
(23)
Now, rst we should note that in addition to the manifest SU(3) this also has a U(1) symmetry, under which the elds have charges XM : 2 M : 1 : 3 (24)
On the other hand, imagine decomposing an SU (4) SU (3) U (1); in that case the fundamental 4 of the SU(4) decomposes as 4 31 13 (25)
where the main number is the representation of SU(3) and the subscript denotes the U(1) charge. This is explained using powerful machinery in [4], p439 but can be understood in a lowbrow way if you imagine simply writing an SU(4) matrix and taking the topleft corner to be the SU(3). Any diagonal U (1) must have overall phase 0 in the full S U(4) matrix, which means that it must act three times as strongly (and with opposite sign) on the SU(3) singlet as on the triplet. Note this decomposition is consistent with (, M ) secretly forming an SU (4) fundamental; the SU(3) singlet has (minus) three times the U(1) charge as the SU(3) triplet M . Simultaneously this tells us that XM must me embedded into the full SU (4) in a more complicated way. We will now introduce some notation and gure out what that is. Let us compare (23) with the previously obtained (16). These look rather similar, if we enumerate the 6 Xs with a notation where X M,M , X M = (X M ) and recall that is secretly the 4 component of an SU (4). In that case direct comparison of the terms lets us dene tentative gamma matrices M M M,4+ = 2N N ,P + = i 2M N P (26) N M M M N +,4 = 2N N +,P = i 2M N P (27) And indeed it is not too hard to show (remembering how these form a full 6d gamma matrix from (15)) that these purported gamma matrices really do have an algebra {M , N } = 2 M N
{M , N } = {M , N } = 0
(28)
This is nothing but the fermionic oscillator algebra for three oscillators, which is equivalent up to a change of basis to the 6d Cliord algebra. Thus the random coecients of the Yukawa terms in the action above are actually the coecients of 6d gamma matrices, and the action is guaranteed to be invariant under an SO(6) = SU (4) rotation.
V. EXTREMAL = BPS
2
This is a fairly involved problem. Our mission is to nd solutions to the Killing spinor equation i + F = 0 8 where the covariant deriative acting on spinors is dened to be 1 = + 4
(29)
(30)
We will use notation where hatted indices denote local Lorentz frame indices and unhatted indices represent coordinate indices. The solution we found last time has metric given by ds2 = 1 dt2 + H()2 d2 + d2 2 H() (31)
By hitting this equation with (1 + 5 ), doing some gamma matrix algebra, and assuming that is a spinor of denite chirality it is possible to show that this is equivalent to the expression in the problem set, provided we introduce a factor of i in front of the F .
1 dt H
e = Hd
e = Hd
e = H sin()d
(32)
a The spin connection is found by demanding that dea + deb = 0, and can be worked out to be b
H t = t = 3 dt H H = = 1 + d H H = = 1 + sin()d H
= = cos()d
H H2 .
i =
where the i are the oldfashioned Pauli matrices. The metric signature is thus (, +, +, +). To go back and forth between hatted and unhatted indices we use the tetrad; furthermore we pick an orientation such that (t, , , ) = (0, x, y, z). We are now ready to begin.
A. Warmup: Spinors in at space
We begin in the gentlest possible way by solving (29) in at space in Cartesian coordinates. This is not very dicult; the solution is = const (38)
This simply means that Minkowski space has as many Killing spinors as the dimension of the spinor representation. We will now reproduce this result using spherical coordinates to give us some practice with covariant spinor derivatives. The relevant spin connection is given by putting H 1 in the formulas above. It is fairly easy to see that the and t components of (29) are trivial: = t = 0. Some algebra shows that the nontrivial components work out to be iz I2 = 0 (39) 2 i + (y sin x cos ) I2 = 0 (40) 2 Note that the 4 4 gamma matrices have factored into tensor products of 2 2 matrices; furthermore, none of the terms in these equations mix the two chiralities of the Dirac spinor, and thus we can simply consider each of the two Weyl spinors separately. From now on we also work in the standard basis where z is diagonal, and we denote the two components of the Weyl spinor in this basis by = (+ , ). Examination of (39) then tells us that i i + = exp f+ () = exp f (), (41) 2 2 where f remain to be determined. We now expand (40) to obtain i 0 ei =0 2 ei 0 This equation has two solutions, + = e 2 (+) i e 2 ()
i
(42)
or
e 2 (+) i e 2 ()
(43)
6 Thus we have found two complex solutions for the two complex degrees of freedom in the Weyl spinor, precisely as expected. Note that the equations for the two Weyl spinors in the Dirac representation decoupled completely, and thus either of these two solutions can also be used for the other Weyl spinor; thus we have a total of four complex (or eight real) supersymmetries preserved by at space.
B.
It is time to move on to the real problem. The main new ingredient here is really the electric eld; denoting for convenience a matrix F F , we plug in our solution for Ft to eventually obtain H x 0 F = 4 2 (44) 0 x H Now we use this and compute the t component of (29). A slight amount of algebra gives 1 H x 0 t 1 + i t =0 0 x 2 H3
(45)
Now it seems clear that we do not want the Killing spinors for a static spacetime to depend on time; thus it appears that we will need the spinor to be annihilated by the messylooking matrix. Luckily this is not dicult to arrange, as the factor (1 + i t ) is a projectorthus any spinor dened using the orthogonal projector (, , ) = 1 i t (, , ) (46)
will satisfy the above equation, as can be easily veried by noting that t squares to 1. Here is some Dirac spinor whose form we will now x. Note that this form cuts down by a factor of two the dimension of the solution space; the projector ties together the two Weyl spinors in the Dirac spinor, and we cannot choose them separately as we did in Minkowski space. More explicitly, if we write = (1 , 2 ), then the constraint (46) simply means that 1 = i2 . We now move on to the radial equation; this will allow us to x the radial prole of the solution. Some uninteresting algebra reveals eventually i H 0 1 =0 (47) 1 0 2 H
or
H() i H()
(48)
but of these only the rst one is consistent with the projection (46). Soothingly both of these solutions become constant at innity. We move on now to the angular equations; these become iz H z I2 t I2 i 1 + i =0 (49) H 2 2 i H y I2 + (y sin x cos ) I2 + i sin() 1 + i t = 0 (50) 2 H 2 These both have a very nice structure; the second half of each equation is automatically satised by virtue of the projection (46), and the rst half is precisely the atspace angular equation that we have already solved. Thus we can immediately take over our results from the previous section. This xes the angular dependence of 1,2 (, ); thus we now have a complete solution. The critical dierence between this and the Minkowski space example is that we had to project out half of the degrees of freedom to obtain a solution. We could not x both Weyl spinors in the Dirac spinor independently; having specied one, the projection determines the other. Thus this background is invariant only under half as many supersymmetries as at space, and is annihilated by half of the SUSY generators. This is precisely what one nds for a BPS state; we conclude the that the extremal ReissnerNordstrom black hole is BPS (Hooray!).
7
C. AdS2 S 2 , or The Return of Supersymmetry
We now squint closely at the nearhorizon region of the geometry, i.e. 0. For the rst time we now need the explicit form of H, which is H() = 1 + A . Putting this into (45) we nd something miraculousin the 0 limit, the coecient of the projector vanishes! Thus we no longer need to project away half of the modes. We can now use both of the radial solutions that we found earlier in (48). Note this gives us two dierent solutions, which satisfy respectively 2 = i1 (only the rst of these two solutions was consistent with the projection). Plugging in these two forms and taking the nearhorizon limit in the angular equations leaves us with z i 1 = 0 (51) 2 i 0 ei 1 = 0 (52) ei 0 2 where one should consistently take the top or the bottom of all the depending on which of the two radial solutions is being considered. These equations admit the following solutions i () i () e2 e2 1 = or (53) i i 2 () e e 2 () We now have a total of 2 2 = 4 (picking either of and one of the two solutions above) complex solutions, so 8 real supersymmetries are preserved (Hooray!).
VI. WBOSONS FROM ADJOINT HIGGSING
For the Wbosons, the relevant term from the N = 4 Lagrangian is Linteresting = 1 1 i i i i 2 D X D X = 2g 2 tr([A , X ][A , X ]) + boring stu 2gYM YM (54)
Now imagine that X i = diag(xi ...xi ). Note that unlike the normal Higgs mechanism we have some freedom in the 1 N choice of X; if it is diagonal then the potential term in the action vanishes for any choice of xi . Consider also A to be an N N matrix A = Aab . In that case a quick multiplication shows that [A, X i ]ab = Aab (xi xi ) b a and thus we get a term Linteresting = Aab Aba (xi a 2 2gY M i;a,b 1 xi ) 2 b (56) (55)
Now lets consider the gauge group to be SU(N) for concreteness. In that case Aab = Aba , and this is precisely a mass term for the ab gauge bosons, as claimed (where we are considering each odiagonal element ab of the matrix and its transpose ba together to be a single complex degree of freedom). Next we consider the scalar masses; here the relevant term is 1 Linteresting = 2 tr([X i , X j ]2 ) (57) 4gYM
i=j
Now we expand this out to quadratic order in perturbations X i around the background values X i . Essentially there are two dierent types of terms, which work out after some index shuing to be 1 i Linteresting = 2 tr [X , X j ]2 + [X i , X j ][X i , X j ] (58) 4gYM
i=j
Now we work out these commutators; for notational convenience we dene i xi xi and expand, nding a ab b eventually that 1 1 j j i i Linteresting = 2 Xab (Xab ) i j ij 2 2 Xab (Xab ) mij (59) ab 4gYM 4gYM ab
i,j;a,b i,j;a,b
8 Thus for xed a, b we obtain a mass matrix; to understand more clearly the structure of this matrix it is easiest to imagine doing an SO(6) rotation so that i = (, 0, 0, ..., 0). In that case mij = diag(0, 2 , 2 , ...2 ); thus we obtain for each ab 5 scalars with the same mass as the gauge boson and one massless scalar.
[1] E DHoker and D.Z. Freedman, Supersymmetric Gauge Theories and the AdS/CFT Correspondence [arXiv:hep th/0201253]. [2] P. C. Argyres, An Introduction to Global Supersymmetry, available at http://www.physics.uc.edu/ ar gyres/661/susy2001.pdf [3] J. Wess and J. Bagger, Supersymmetry and Supergravity, Princeton University Press, New Jersey 1992. [4] J. Polchinski, String Theory: Volume II, Cambridge University Press, New York 2001.
I.
Recall the conformal algebra1 for Rp,q is given by translations P , rotations L , special conformal transformations K , and dilatations D, satisfying the algebra [D, P ] = iP [D, K ] = iK [K , P ] = 2i( D L ) [K , L ] = i( K K ) [P , L ] = i( P P ) [L , L ] = i( L + L L L ) (1) (2) (3)
Of this algebra really only the rst line should be unfamiliar, as the second line simply says that P and K are vectors and the last line is just the Lorentz algebra SO(p, q). Now to put this into a more intuitive form we imagine that we are in a bigger space with two extra coordinates which we call 1 and 0 (the remaining coordinates are then 1...p + q). Now consider the algebra dened by the following antisymmetric tensor generators Jab (where a, b include the two new coordinates and , run only over the original p + q) J = L J1,0 = D J1, = J0, 1 (P K ) 2 1 = (P + K ) 2 (4) (5)
I claim these generators secretly obey the algebra of SO(p + 1, q + 1): [Jab , Jcd ] = i(ad Jbc + bc Jad ac Jbd bd Jac )
p+1,q+1
(6)
where ab is the metric tensor for R . To verify this, we compute commutators; it is clear that those of J with itself will work out. The next easiest is [J , J1,0 ]; if this is part of a Lorentz algebra it should vanish from (6) as these two generators have no indices in common, and indeed it does vanish (in the lower dimensional interpretation this is because D is a scalar on the original Rp,q ). The rest require slightly more work: [J0 , J ] = i( J0 J0 ) = i( J0 + 0 J 0 J J0 ), (7)
where I have introduced the objects 0 to make more clear the comparison with (6) ; this is indeed exactly what we expect. A precisely analogous computation can be done for J1, . Slightly more interesting are the following 1 i [J1,0 , J1, ] = D, (P K ) = (P + K ) = iJ0, = i1,1 J0 (8) 2 2 i 1 (9) [J1,0 , J0, ] = D, (P + K ) = (P K ) = iJ1, = i00 J0 2 2 (10) In the last equality I have put in the expected value from (6). Note that this only works out if 1,1 = 1 and 00 = +1; thus the enlarged space has one extra time and one extra spatial dimension, as claimed earlier. There remains only one more [J1, , J0 ] = 1 i [P K , P + K ] = [2 D + L + L ] = i J1,0 , 4 2 (11)
again, as expected from (6). Thus the conformal group on Rp,q is isomorphic to SO(p + 1, q + 1)! Note that this ts nicely into our understanding of AdS/CFT; the conformal group from say a (Euclidean) CF Td is realized as the isometries of AdSd+1 , a space that can be nicely embedded into R1,d+1 as a hyperboloid that is manifestly preserved under the action of SO(1, d + 1), which as we see is nothing but the conformal group of the CF Td .
2
II. THE QUADRATICNESS OF CONFORMAL TRANSFORMATIONS
Here we will prove that the generators of an innitesimal conformal transformation, which by denition satisfy + = 2 d (12)
are at most quadratic in x, for d > 2. To start, we dot the formula with and get 2 ( ) = 0 + 1 d
(13)
Note that for d = 2 we nd only that = 0, so any harmonic function generates a conformal transformation. Now we take a derivative of the above equation and add to it the same equation with and interchanged, resulting in 2 ( + ) + 2 1 ( ) = 0 (14) d Now plugging in (12) we nd that 1 2 = 0 + 1 d d = 0 (15)
Hitting this with we nd soothingly that (2 2/d) = 0. Assuming that d = 1, we nd that (16)
Though this appears to be a scalar equation, it turns out that this is all we need, essentially because equations like (12) mean that the tensor structure of derivatives of is entirely determined by scalar functions like . More concretely, plug (16) into (15) to nd (for d = 2) that in fact ( ) = 0. Finally now take two derivatives of (12); the right hand side vanishes by the relation that we just found, and the left hand side is ( + ) = 0 (17)
We are almost done; we just now need to show that not only the symmetric combination ( + ) but also the antisymmetric combination ( ) has vanishing second derivatives. To that end, subtract from (17) itself with a relabeling of indices: ( + ) ( + ) = 0 (18)
Some rearrangement and use of the commutativity of partials reveals that the rst term in each bracketed expression cancels with its counterpart and we nd ( ) = 0 (19)
Thus both the symmetric and antisymmetric combinations have vanishing second derivatives, and is at most quadratic in x.
III. FUN WITH LARGE N
This problem was basically solved in class, but well review the argument here. Consider a very general matrix quantum eld theory with the Lagrangian L = dd xtr(D D + g 3 + g 2 4 + ...) (20) where is an N N matrix and g is some sort of coupling. If we now redene our elds: = /g, then all the factors of g actually come outside the Lagrangian: 1 L= 2 dd xtr(D D + 3 + 4 + ...) (21) g
3 Now what is the contribution of a given Feynman diagram with V vertices, P propagators, and L loops? Note that with the eld denitions used in (21), the propagator is the inverse of the operator D2 /g 2 and so contributes a factor of g 2 ; similarly each vertex (regardless of the type: cubic, quartic, etc.) contributes 1/g 2 . Finally each loop involves a trace and so contributes a factor of N . Thus the scaling of the diagram is contribution (g 2 )(V P ) N L N (L+V P ) (V P ) (22) where in the second equality I have written things in terms of the tHooft coupling g 2 N . Now howevernote that we learned in lecture how to associate a triangulation of a twodimensional surface to this Feynman diagram; each propagator becomes an edge of the triangulation, each vertex is still a vertex, and each index loop becomes a face. Thus really we have contribution N facesedges+vertices verticesedges (23)
Note that the combination faces  edges + vertices is called the Euler characteristic of the surface ; I say of the surface and not of the triangulation because of the remarkable fact that it is a topological invariant and depends only on the surface being triangulated, not the triangulation itself. Thus the scaling of the diagram with N depends only on the topology of the surface that the diagram represents. Note also that the dependence on the tHooft coupling on the other hand does depend on the triangulation (vertices  edges is not a topological invariant) and so actually depends on the connectedness of the Feynman diagram being considered. Anyway, for a sphere = 2; thus we see that for all diagrams with the topology of a sphere (planar diagrams) we get a contribution that goes like N 2 .
IV. USEFUL COORDINATES IN ADS
We dene Lorentzian AdSp+2 to be the set of points X a inside Rp+1,2 satisfying L2 = ab X a X b (X p+2 )2 (X 0 )2 +
p+1 i=1
(X i )2
(24)
Here Rp+1,2 is a at space with two time directions, so its metric is given by ab = diag(, , +, +..+). For most of the remaining problems we will need to compute induced metrics; for reference, recall that if we dene a subspace in terms of embedding coordinates xi by specifying functions X a (xi ), then the induced metric gij on the surface parametrized by xi gij = Gab X a X b , xi xj (25)
where Gab is the metric on the larger space (i.e. ab in our case).
A. Global Coordinates
We take rst the coordinate choice X p+2 = L cosh() sin( ) X = L cosh() cos( ) X = L sinh()i
i 0
where i are functions on the psphere such that i (i )2 = 1. To compute the induced metric we rst require various partial derivatives, which are very easily found to be X p+2 = L sinh() sin( ) X 0 = L sinh() cos( ) X i = L cosh()i X p+2 = L cosh() cos( ) X 0 = L cosh() sin( ) X i i = L sinh() j j (29) (30) (31) (32)
4 Note for example that to nd the component g we want from (25) g = ab Similarly, we nd g = L2 cosh2 ()(cos2 ( ) + sin2 ( )) = L2 cosh2 () and the mixed term g is easily seen to vanish. For the angles parametrizing the psphere we nd gij = L2
k
X a X b =
X p+2
X 0
p+1 i=1
X i
(33)
(34)
sinh2 ()
k k i j
(35)
However this is by denition simply the metric on a psphere with radius L sinh(). Thus the metric of global AdS is ds2 = L2 ( cosh2 ()dt2 + d2 + sinh2 ()d2 ) p
B. Poincare patch
(36)
+X =v ux X = L
p+1
Now the rst step is to gure out what v is on the AdS surface; to do this we plug the above equations into the dening equation (24) and solve for v to get v= L2 u 2 x x u L (40)
We can now explicitly write X p+2 and X p+1 in terms of the intrinsic coordinates and get 1 L2 u 1 L2 u X p+1 = u X p+2 = u+ 2 x x + 2 x x 2 u L 2 u L
(41)
The remainder of the problem simply involves taking partial derivatives of the X a as before. This is not terribly interesting: I present some intermediate steps 2 u X (p+1),(p+2) 1 L 1 X (p+1),(p+2) 1 2 2 x x (42) = 2 x = x L u 2 u L X x X u = = (43) u L x L Two signs indicate that the top is for p + 1 and the bottom for p + 2. From now putting these into the formula (25) results after some honestly straightforward algebra in ds2 = u2 L2 dx dx + 2 du2 L2 u (44)
Finally, we should worry about what portion of the hyperboloid is covered by these coordinates; noting indeed that the norm of the Killing vector x0 vanishes at u = 0, we see that we cannot actually move to negative u; thus we are conned to the portion of the hyperboloid where X p+2 + X p+1 > 0, implying that we cover only half of the hyperboloid.
5
d
(L,0)
C.
Consider now the new variable r L sinh() where is the global coordinate used in (57). We have dr = L cosh()d and thus the metric (57) becomes ds2 = L2 cosh2 ()d 2 + dr2 + L2 sinh2 ()d2 p cosh2 () (45)
(46)
where I have introduced a dimensional time t L . This is the claimed form. Now let us think about how to put a black hole in this background; to add a black hole without adding any extra matter, we would like to somehow deform the metric without altering the values of the Einstein tensor (which determines what the supporting stress energy must be). Note that if indeed G r (rp r H), then there are two types of terms we can add to H without changing the value of G : a constant term and a term going like 1/rp1 . The constant term is not physically relevant (as it can essentially be absorbed into a coordinate redenition), so we are led to the following form for the metric of SchwarzschildAdSp+2: 2M r2 dr2 ds2 = 1 p1 + 2 dt2 + + r2 d2 (47) SchAdS p r2 2M r L 1 rp1 + L2 Note this becomes the familiar Schwarzschild metric if we set p 2 and smooth out the AdSness by putting L . 2M is a parameter that is indeed related to the mass of the black hole. Keeping L nite, we see that far from the black hole this spacetime asymptotes to AdSp+2 . As we shall soon see in lecture, this metric is thought to represent the boundary CFT on S p R in a thermal state with temperature given by the Hawking temperature of the black hole.
V. STEREOGRAPHIC PROJECTION COORDINATES
We take the Euclidean signature ddimensional hyperbolic space Hd to be dened by the set of points d1 2 X i X i = L2 Rd,1 Hd = Xd +
i=0
(48)
6 d1 2 with metric on Rd,1 given by ds2 = dXd + i=0 dX i dX i (note the labeling of coordinates here is slightly dierent from that in the problem set). There is a cute way to introduce coordinates on this space; consider a point P on the hyperboloid and and shoot a line from this to the point (X d , X i ) = (L, 0). This line will intersect the plane X d = 0 at a point (0, i ); we dene r i i , but for simplicity we now rotate the coordinate system so that i points only in the i = 0 direction, so that r = 0 . Our whole problem is now in the (X d , X 0 ) plane. Now we name this newly constructed line C and note that it is the set of points L L , R Rd,1 (49) C= + r 0 Where does this line hit the hyperboloid? The X 0 coordinate of this point can be seen from (48) and (49) to be the solution to the equation X0 L L = (X 0 )2 + L2 r 2L2 L2 r 2 r ii, (50)
(51)
precisely the coordinate choice made in the problem set. Figure 1 illustrates this construction. The dening equation then implies that the remaining coordinate in the base space X d is given by Xd = L L2 + r 2 L2 r 2 (52)
To compute the induced metric we now take various partial derivatives as before. Those of X i are pretty immediate, X i 2L2 2 i j ij + 2 , (53) = 2 j L r2 L r2 whereas those of X d require slightly more work, and are perhaps more easily found by taking a derivative of the dening equation (48) with respect to j 2Xd X i Xd Xi j =2 j i 2 Xd 1 i 2L2 2 i j ij + 2 = d j X i L2 r 2 L r2 2 j 2L2 = L L2 r 2
(54)
We now use (25) and plug in these results, obtaining 4 2 2L2 2 l i 2 m j ij 2L2 li + 2 mj + 2 d i d j ds2 = 2 + L L2 r 2 L2 r 2 L r2 L r2 Expanding this out and waiting for the dust to settle, we obtain eventually ds = as claimed.
VI. GEODESICS IN ADS
2
(55)
2L2 L2 r 2
d i d i
(56)
For this problem we will use the metric in global coordinates given in (57): ds2 = L2 ( cosh2 ()dt2 + d2 + sinh2 ()d2 ) p (57)
7
2 1 1 2
0.2
0.4
0.6
0.8
1.0
We would rst like to nd the trajectory of a massless geodesic, starting from a point = 0 . Along such a trajectory, ds2 = 0, implying that d 1 = d cosh() 0 2 arctan tanh , () = 2 arctan tanh 2 2 (58)
Note that this is nite regardless of the value of 0 ; no matter where you are, a geodesic takes a nite amount of time to leave you and come back. This is a signature of the badness of the Cauchy problem in AdS, as an observer can communicate with regions that are innitely far away in nite time. Next, we would like to understand the behavior of a massive geodesic. To compute this without any mucking around with Christoel symbols, note that for a massive particle the norm of the p + 2velocity u dx /dT is always 1, where T is the proper time along the worldline. This means that cosh ()
2
where I have adjusted the integration constant so that (0 ) = 0. Now the coordinate time for a geodesic to leave 0 , go to the boundary, and come back is simply 2 (), which means that the proper time elapsed on a stationary observers clock is 0 proper time = 2L cosh(0 ) () = 2L cosh(0 ) 2 arctan tanh (59) 2
d dT
d dT
1 L2
(60)
On the other hand, we also know from oldfashioned GR 2 that the scalar obtained by dotting a Killing vector into a (p + 2)velocity is conserved along a geodesic. For us a Killing vector reecting time translational invariance is given by = , (i.e. = ), implying that u = const along the geodesic. Calling the constant A, we obtain cosh2 () The combination of (60) and (61) results in 1 2 d dT 2 + A2 2 cosh2 () = 1 2L2 (62) d =A dT (61)
This is nothing but the equation for conservation of energy for a particle of unit mass moving in a onedimensional A2 1 potential V () = 2 cosh2 () with negative energy 2L2 . The potential is graphed above in (2); notice that since the particle always has negative energy it is bound to the well and can never reach innity, regardless of the value of A.
8
VII. VOLUMES AND AREAS ARE NOT SO DIFFERENT IN ADS...
...by which we mean that they scale the same way in the large limit. To see this, consider some xed and compute the area A() and volume V () inside it. The area is immediately read o of (57) to be A() = Vol(S p )Lp sinhp () (63)
To compute the volume I assume for simplicity that is big enough that I can take sinh() to be just exp()/2. The volume is then just Lp+1 Vol(S p ) exp(p) (64) V () = dLA() 2p 0 Thus A()/V () p/L, as , which is indeed nite.
[1] P. D. Francesco, P. Mathieu, and D. Senechal, Conformal Field Theory, Springer, 1997. [2] S. Carroll, Spacetime and Geometry: An Introduction to General Relativity, AddisonWesley, San Francisco 2004.
I.
In this problem we will demystify this peculiar inversion that is used to such powerful eect by Witten. Consider AdSd+1 in Poincare coordinates: ds2 = dxA dxA dz 2 + dx2 = 2 z z2 (1)
where xA = (z, x ) and xA xA z 2 + x x . Now I claim that the horribly violent inversion xA xA = xA xB xB (2)
is actually an isometry of AdS, meaning that the metric in the barred coordinates takes precisely the same form as it did in the unbarred coordinates. This is not hard to verify, starting with the fact that 1 xA A A C dx = B dx 2 B xC dx (3) x xB x xB It is now easy to show that (xD xD )2 1 dxA dxA = ds = z2 z2 (xB xB )2
2
xA xA A C C dx 2 B xC dx dxA 2 B xC dx x xB x xB
(4)
(5)
as claimed.
II. SCHRODINGER DESCRIPTION OF ADS INSTABILITIES A. Wave equation in Schrodinger form
In this problem we will derive the celebrated BreitenlohnerFreedman bound, which tells us that a negative mass for a bulk scalar eld in AdSd+1 does not necessarily lead to any instabilities, provided that it isnt too negative. We will work in units where LAdS = 1 and the metric is given by (1) but with Lorentzian signature on the constantz slices. The bulk scalar action is 1 m2 1 d+1 d+1 1d AB 2 2 2 S= (6) d x g () + m = d xz A B + 2 2 2 z The equation of motion can be easily obtained from here to be (1 d) m2 + k2 + 2 = 0 z z (7)
where I have assumed a eld theory spacetime dependence eikx , so k 2 is the momentum 2 + k 2 . We would like to write this equation in a form that makes it look like a Schrodinger equation from nonrelativistic quantum mechanics, i.e. we want to make the pesky term multipling go away. To do this, we write (z) = B(z)(z) and plug in; after the dust settles we get 1 (1 d)B + 2zB + 2 (m2 + k 2 z 2 )B + z(d 1)B + z 2 B = 0 B + (8) z z
(9)
We take the arbitrary constant to be 1 and plug in this form for B into the equation of motion. After the derivatives in the last term have exhausted themselves we obtain nally 1 d2 1 2 2 = 2 (10) + k + 2 m z 4 This is a perfectly normal looking Schrodinger equation, with potential V (z) given by the quantity in square brackets and energy given by 2 . Note that a negativeenergy solution corresponds to imaginary , which means that the solution is growing exponentially in time; this is the instability we are looking for. For the remainder of this problem we will set k = 0. Now this dierential equation will always have solutions for all values of , both real and imaginary, but not all of them will be normalizable. A normalizable solution to the AdS wave equation is one that has nite energy, in a sense that we will now make precise; in particular, normalizable in the bulk AdS wave equation sense will be closely related to normalizability in the usual QM sense (i.e. dz(z)2 < ).
B. Normalizable solutions to the AdS wave equation
We begin by rst noting that if any spacetime has an isometry generated by a Killing vector , the current j = T (where T is the stress tensor is covariantly conserved: j = 0, as can be easily checked using the Killing equation and the conservation of T . Thus given a region R of d + 1 dimensional space 0= dd+1 x g j = (11) dd x hn T
R R
where R is the boundary of R and h is the determinant of the induced metric on this boundary. Let us now specialize to AdSd+1 and let the Killing vector be = t . Let us also take R to be a giant chunk of AdS, extending across all space z [0, ) but bounded in the past and the future by two spacelike slices at ti and tf , i.e. t [ti , tf ]. In that case R has three components; the two spacelike slices at t = ti , tf , and the timelike slice at the AdS conformal boundary at z = 0 (I will assume that all elds decay away exponentially at z ; the boundary term at z = 0 will turn out to be important). This integral then becomes
dz z
1d
The second term is the ux of energymomentum out the AdS boundary; if this is zero then this equation implies that the rst integral has the same value at tf and at ti and should be interpreted as the statement of energy conservation. Now let us explicitly work out what the relevant components of T are in our case. For a scalar eld with action (6) the stressenergy tensor is 1 T = g g m2 2 2 (13)
t=tf Ttt
t=ti
tf
dt z
1d
ti
Ttz
=0
z=0
(12)
Similarly, we will manipulate the second term in (12) with a similar integration by parts (except on t) t=tf tf tf 1 1d 1d 1d = dt z Ttz dt z (t z t z ) + z z 2 ti ti z=0 t=ti
z=0
I would now like to write this in terms of . This is mostly very easy except for the term in (z )2 , which we manipulate with an integration by parts as follows z= 1 d2 2 dzz 1d(z )2 = z 1d z (14) dz z (z 1d z ) = z 1d z dz z + 4z 2 0 0 0 z=0 z=0 (15)
3 Putting the pieces together to assemble (12), we see that the boundary terms from each of the two parts cancel and we obtain the following equation tf 2 t=tf 1 1 dz t t + z + V (z)
=0 (16) dt z 1d (t z t z )
2 2 ti 0 t=ti z=0 where V (z) is the potential from the Schrodinger equation (10) 1 d2 1 2 V (z) = 2 m z 4 (17)
Now let us take a step back and think about what we require of a solution to the wave equation. It seems reasonable to require that the energy (dened by the rst integral above) is nite, and also that it is conserved. We have seen that energy conservation requires that the second term vanish above vanish. Introducing a Fourier expansion in , (t) = d()eit , we see that this implies that d d
tf
dteit(+
ti
Note that the fact that this must hold for all ti , tf essentially means it must hold for all , . To put this into a form that will turn out to be more useful, we swap and in the second term, to obtain d d z 1d [()z ( ) ( )z ()]z=0 = 0 (19) For our nal trick, we write in terms of and notice that the reality of (t) implies that () = () to write this equation as () z ( ) ( )z (() ) = 0 at z = 0 (20)
(18)
This equation embodies the statement that nothing important is leaving through the AdS boundary, and will turn out have rather spectacular consequences in the quantum mechanics problem that we will solve next. Let us return now to the criterion of niteness of the energy. This simply requires that the rst bracketed term in (16) is nite. Evaluating it at arbitrary t, we see that the energy is simply 2 1 E(t) = d ddz ()( ) + () z + V (z) ( ) eit(+ ) (21) 2
We have so far nowhere used the fact that we are evaluating this energy functional onshell, that is, on a solution to the equations of motion (10). Using this fact, noticing that it is exactly the Schrodinger operator that appears in the second term above, and switching , we end up with the following expression for the energy 1 E(t) = d ddz + + 2 () ( )eit( ) (22) 2
This is promising. Now let us think very hard; the Schrodinger operator in (10) is Hermitian (Right? Right. Ignore that prickling feeling at the back of your neck...) Thus its eigenfunctions with dierent eigenvalues are orthogonal. These eigenfunctions are nothing but the (), and so we should have dz() ( ) = ( ). Thus we nd E(t) = ddz 2 ()2 (23)
And so niteness of the energy requires that the wavefunctions () have nite norm in the usual quantum mechanical sense. Wait. Something is wrong; I just convinced you that the energy will be conserved only if some awkward boundary condition (20) was met, yet I just found an explicit formula for the energy that is timeindependent ! Why would I lie to you in this way? Well, I assumed that the Schrodinger operator appearing in (10) is Hermitianis this necessarily true? Consider the operator acting on two eigenfunctions with real eigenvalues 2 2 z + V (z) () = 2 () z + V (z) ( ) = 2 ( ) (24)
4 Now multiply the rst equation by ( ), the second by (), subtract the second from the rst, and integrate them both over z. We obtain 2 2 dz ( )z () ()z ( ) = 2 2 dz ( )() (25)
0 0
If = , the eigenfunctions have dierent eigenvalues and the expression on the righthand side is nonzero and proportional to the inner product of the eigenfunctions. If this operator is Hermitian, these eigenfunctions must be orthogonal, which means that the lefthand side of this expression better vanish. However, if we integrate by parts on this lefthand side, we see that it is not zero but in fact equal to a boundary term 2 2 [( ) z () ()z (( ) )] = dz ( )() (26)
z=0 0
Thus the Schrodinger operator is not Hermitian unless this particular boundary term is zero. This boundary term is precisely the one that we found earlier in (20), and is proportional to the energy ux out the boundary of AdS. Suddenly it all makes sense, and we summarize our ndings below:
1. If the Schrodinger operator corresponding to the AdS wave equation is Hermitian on a particular set of eigen functions, then those eigenfunctions correspond to classical eld congurations with a bulk AdS energy (i.e. the charge under the bulk time translational Killing vector) that is conserved. The condition for Hermiticity is precisely equivalent to the condition that no energy leak out the AdS boundary. 2. Furthermore, if a particular eigenfunction is normalizable in the normal QM sense (i.e. dz(z)2 < ), then it corresponds to a classical eld conguration that has nite bulk AdS energy. We now nally have a wellposed quantum mechanics problem; nd normalizable negativeenergy eigenstates of the Schrodinger operator (10) that satisfy the boundary condition (20).
C. Good oldfashioned Quantum Mechanics
2 where = m2 1d . We now want to examine the behavior of the solutions as a function of ; in particular, we 4 are seeking negativeenergy solutions, which with the convention used in (27) corresponds to real . I should point out that this particular potential has a venerable history, ranging from a paper from 1950 [1] (which I am largely following) to the recent work of our own B. Swingle [2]. In fact David Griths will soon (Dec 4th) be giving an entire MIT Physics colloquium involving this very potential. In any case, the solutions to (27) are writable in terms of Bessel functions: (28) (z) = z [C1 J (iz) + C2 Y (iz)] 1 were = 2 1 + 4. Already we see that something peculiar will happen at the particular value of = 1/4. To nd the spectrum, we would like to impose various conditions on the wavefunctionrst, let us examine the large z behavior. The asymptotic behavior of the Bessel functions with imaginary argument is 2 2 J (iz) cosh(z) Y (iz) i sinh(z) (29) iz iz where I have neglected some factors of /4, etc. in the argument of the exponentials. It is clear that we need the linear combination appearing in to decay exponentially at large z, which means that takes the form (30) (z) = C z [J (iz) + iY (iz)] Now we examine this at small z, where it becomes 1 iz () 2 (z) = C z i ( + 1) 2 iz
(31)
5 Now we nally need to worry about the true nature of = 1 1 + 4. If > 1/4, then is real. Is this wave 2 function then normalizable as we approach z 0? Actually, yes, it is; the relevant terms in (z)2 go like z 1+2 , z, and z 12 respectively; these are actually all integrable at z 0 for suciently small , even if it is real. It appears 1 that even if > 4 we can obtain normalizable negative energy bound states; furthermore we can get them for any value of , i.e. a continuous spectrum of negative energy states. This cant be. Can we do better? Yes we can. Recall our other boundary condition (20); evaluating this on two solutions (31) with dierent energies 1 , 2 we obtain after some irritating algebra and manipulation of gamma functions the condition 2 1 + =0 (32) 1 2 This condition is never satised, regardless of the values of 1 , 2 real. Thus we conclude that for real (i.e. > 1/4) we can never have a negativeenergy bound state that satises the Hermiticity condition. Finally, we move on to the case of interest, where < 1/4 and thus is imaginary; lets write it as = ig, where g is a real number. Our large z analysis goes through untouched, and thus (30) is still the correct form for the wavefunction. On the other hand, in the smallz analysis, z is now raised to an imaginary power and so is oscillatory z ig exp[ig log(z)]. After extensive use of Mathematica and manipulation of gamma functions that is too tedious to recount, you can show that at small z (z) is proportional to z + (33) (z) z cos g log 2 is a gdependent phase that I encourage you to calculate. This oscillates around 0 and so is probably normalizable at the boundary. Now the fun part; imposing the boundary condition (20) results in z1 z2 g cos g log + sin g log + 1 2 = g sin[g(log 1 log 2 )] = 0 (34) 2 2 This boundary condition can be solved, and the answer is that the (negative) energies 1 , 2 must be related by n 2 , nZ (35) 1 = exp g Thus we have a discrete spectrum! Note that = 0 is an accumulation point of the spectrum. And indeed, given some way to x one of the energies in the spectrum, we can x the entire set. However, I will not pursue this any further, and simply write down the punchline: If < 1/4, we have normalizable negativeenergy states on which the Schrodinger operator is Hermitian. Now recall that a negativeenergy state is actually an instability of the classical AdS wave equation, and = 2 m2 1d . Thus the criterion for the existence of a nite(and conserved)energy instability is simply 4 m2 < This is the BreitenlohnerFreedman bound. (Whew!)
III. DIMENSIONS OF VECTOR OPERATORS
d2 4
(36)
x g
m2 1 F F + A A 4 2
(37)
corresponding to a massive bulk vector eld. The equations of motion from variation of this action are simply F = m2 A which when written with normal partials reduces to 1 ( gg g F ) m2 A = 0 g (39) (38)
6 For simplicity let us imagine that we are in Euclidean AdS and thus can put the full momentum in the t direction. We will examine polarizations that are not in the t or z directions, = t, z; in this case the equation of motion becomes after some manipulation
2 2 z 2 z A + (3 d)zz A + z 2 t A m2 A = 0
(40)
Now if we substitute a power law solution z into this equation, and examine the nearboundary behavior, we see that satises the equation ( + 2 d) m2 = 0 whole solutions are 1 = (d 2) (d 2)2 + 4m2 2 (42) (41)
At this point we might remember the scalar eld example and be tempted to simply say that + is the answer and move on with our lives. This would be wrong, and the reason why is that A has a vector index that appears a bit problematic; for example if we had worked instead with A we would have obtained indices that were both shifted upwards by +2. Furthermore, we know that for m = 0 the bulk eld is dual to a conserved current j with dimension d 1; our formula does not reect this. Of course the correct way to really nd the scaling dimension is to do a slightly more extensive calculation and compute e.g. the correlator of the current j . A rather more heuristic way to do this is to work with an orthonormal basis for the A s. This removes the ambiguity (the orthonormal basis components actually have some physical meaning, and A = A ). A = zA , and thus we should shift the indices both upwards by 1, giving us = 1 d (d 2)2 + 4m2 2 (43)
Taking + to be the conformal dimension of the boundary operator now gives us the correct result.
IV. SATURATING THE UNITARITY BOUND
and we would like to evaluate this action on a powerlaw eld prole = z eikx . This is not very dicult; in components the action is 1 1 dd+1 x d+1 z 2 (z )2 + (m2 z 2 k 2 )2 (45) Susual [] = 2 z which on the powerlaw eld becomes Susual [] = 1 2 dd+1 x(2 + m2 z 2 k 2 )z 2(d+1) (46)
We work with the normal massive scalar bulk action 1 dd+1 x g ()2 + m2 2 Susual [] = 2
(44)
Peforming the integral over z, the leading divergence as z 0 is the term of the form Susual [] lim 2 + m2 2d z z0 2(2 d) (47)
This will be nite only if 2 d > 0. Now the conformal dimension of the operator dual to the bulk eld is the exponent of the normalizable solution; thus it is clear that using this action the conformal dimension will always satisfy > d 2 (48)
7 Somewhat more concretely, the two possible indices are = 1 d d2 + 4m2 2 (49)
It is clear from here that only + corresponds to a normalizable solution. However this state of aairs is not completely satisfactory, since the bound (48) on the dimension seems to be a bit too restrictive. To see how to get around it, consider the modied action 1 SKW [] = (50) dd+1 g ( + m2 ) 2 This action has precisely the same equations of motion as the previous one, yet it diers by a (divergent) boundary term and so will have a dierent numerical value. We now work out what this value is. On the AdS metric the action works out to be 1 1 SKW [] = (51) dd+1 x d+1 z d+1z (z 2(d+1) z ) + k 2 z 2 + m2 2 z and when acting on the power law solution a short computation results in 1 dd+1 xz 2(d+1) [(d ) + m2 ] + z 2 k 2 SKW [] = 2
(52)
Now so far we have assumed nothing about ; let us now assume that is one of the indices from (49), in which case the rst and most divergent term in (52) vanishes. Nnote we could actually have guessed this immediately from the form of SKW , as the power law with appropriate is actually an exact solution to the equations of motion in the limit that k 0; thus the only nonzero piece in SKW must be proportional to k 2 . In any case, we see that the leading divergence in SKW is SKW lim k2 z 2d+2 2(2 d + 2) d2 2 (53)
z0
This is nite under the less restrictive condition that > (54)
which is indeed the unitarity bound on the dimension of an operator in a CFT. Thus this lower bound actually makes sense. How can we achieve this? Note that if the mass satises d2 d2 < m2 < + 1 4 4 (55)
both satisfy (54), and so either of them correspond to a normalizable solution and can be used as the dimension 2 of the operator. Indeed, at the critical value m2 = 1 d we have = (d 2)/2 (Recall with the more restrictive 4 bound (48) we could never even use any for the dimension of the operator). More discussion along these lines can be found in [3].
[1] K. M. Case, Singular Potentials, Phys. Rev. 80, 797  806 (1950) [2] H. Hammer, B. Swingle, On the limit cycle for the 1/r 2 potential in momentum space, Annals Phys. 321 (2006) 306317 [arXiv:quantph/0503074] [3] I. Klebanov, E. Witten, AdS/CFT Correspondence and Symmetry Breaking, Nucl.Phys. B556 (1999) 89114. [arXiv:hep th/9905104]
I.
In this problem we will evaluate a certain threepoint function using AdS/CFT, to verify that it has the form required by conformal symmetry. This will require some machinery, and I will follow the approach taken in [1]. We work with AdSd+1 with metric given by d 1 1 i i 2 2 dx dx 2 dxA dxA (1) ds = 2 dx0 + x0 x0 i=1 For this problem we will work in Euclidean signature i runs over all d dimensions of the boundary theory, and xA = xA = (x0 , xi ).
A. Warmup
First, we will study the gauge eld bulkboundary propagator KMN . The equations of motion for a massless gauge eld in AdSd+1 were worked out in Problem Set 4. If we assume there is no momentum in any of the eld theory directions, these equations reduce to (see (40) in Problem Set 4 solutions) x2 0 Ai + (3 d)x0 0 Ai = 0 0 We then have the immediate solution Ai = i xd2 0 (3) (2)
where i is some polarization vector (we work in a gauge where Az = 0). Now let us consider the eect of the infamous inversion on this solution. Recall that the inversion is a relation between coordinate systems xA = xA xA xA (4)
I denote the coordinates of the gauge eld in the new barred coordinate system by AB ; these take the form x x AB () = AC (x()) B x x
C
(5)
Working out the Jacobian of the coordinate transformation (done previously in Problem Set 4, problem 1) we nd that the solution (3) becomes d2 xA xB A i B (C ) = x0 A x B A i (6) x2 x2(d1) This is precisely the bulkboundary propagator from a point in the bulk xC to the point xi = 0 on the boundary. This can be veried in exactly the same way as was done in class for the scalar case. To recap briey: by construction this is a solution to the equations of motion. Furthermore at x0 = 0 the propagator vanishes at all points where xi = 0, and a scaling argument shows that dd xAi (x) is independent of x0 and proportional to i , showing that the boundary value of this propagator is a delta function, as desired. Thus, dropping the bars, the bulktoboundary propagator from an arbitrary point y i at the boundary to a point xA in the bulk is just xd2 (x y)A (x y)i 0 KAi (xA ; yi ) = c 2 Ai (7) (x0 + (x y)i (x y)i )d1 x2 This is (almost) the form claimed in the problem set. As it turns out, we will not actually require this explicit form for the rest of the computation.
2
B. 3point function
Next, we turn to the computation of the following CFT threepoint function O(x1 )O (x2 )Ji (x3 )
CFT
(8)
Here Ji is a symmetry current and O is an operator of dimension that is charged under that symmetry. Let us take Ji to be a U (1) current for simplicity; it is then dual to a bulk U (1) gauge eld AM , and O is dual to a bulk massive scalar eld that is charged under that U (1). The covariant derivative of this eld is then DM = (M igAM ) Thus the bulk scalar action contains a term Sbulk igAM (M M ) (10) (9)
This is the term that will be relevant for us. Indeed, as discussed in lecture it is rather clear that the three point function that we want is proportional to O(x1 )O (x2 )Ji (x3 ) CFT g dd+1 z gg MN KMi (z; x3 ) [K (z; x1 )N K (z; x2 ) K (z; x2 )N K (z; x1 )] , (11) where KMi is the gauge eld bulkboundary propagator found above and K (z; xi ) is the familiar scalar dimension bulkboundary propagator z0 i (12) K (z, x ) = c 2 z0 + (z x)i (z x)i At rst glance, this integral looks completely and utterly impossible. We are going to do it anyway. We make our lives simpler with the following steps: 1. First, set x3 = 0. It should be clear there is no loss of generality here. 2. Next, we stare hard at the righthand side of (11). Clearly we can do this integral in any coordinate system we want; let us thus do it in a coordinate system related to the one where x3 = 0 by an inversion. Why does this help us? After the inversion, the point x3 (which is at the AdS boundary) becomes the point at innity; but this means that for the gauge eld propagator we can use the hilariously simple solution (3) rather than the messy (7). Let us thus work out how everything on the righthand side of (11) transforms under an inversion xA xA = xA xA : (x y)2 = K (z; xi ) = y x 2 x2 y 2 = 1 ( y )2 (2 + y 2 ) x x 1 + 2+ 2 x y x2 y 2 = ( y )2 x x2 y 2 (13) (14)
z0 z 2 x2 2 ( x)2 z z
= K (; xi ) x2 z
Note that in the rst formula x and y are vectors xA , y A labeling a point in AdS and in the second x = (0, xi ) should be viewed as a (d+1) dimensional vector but restricted to the boundary of AdS. Putting these pieces together the integral becomes 1 z0 z0 g(1 x2 )2 dd+1 z x (x1 x2 ) (15) zi z ( x1 )2 z ( x2 )2 z This looks more tractable. 3. Finally, we do an integral: F (x, y) dz0 dd z
0
(16)
Note that it is clear from translational and scale invariance in x, y that the answer can only be a power law in x y, and dimensional analysis then tells us the exponent. To get the overall factor one must work much harder; I believe it can be done with Feynman parameters (eq (22) in [1]), but I am not going to do it here.
3 With this result our work is actually done; if we trade the derivative zi for a derivative xi in (15) and use the 2 integral result above, the answer is simply O(x1 )O (x2 )Ji (0) = const (1 x2 )2 xi x1 x2 d2 (x1 x2 ) x 2 (17)
Here the barred coordinates should be viewed as functions of the unbarred ones. Thus taking the derivative and expressing everything in terms of the unbarred coordinates, we obtain x2 x1 x2 d22 x1 O(x1 )O (x2 )Ji (0) = const 2 (18) 2 x1 x2 i (x1 x2 )d2 It is clear how to extend this to points x3 = 0; translational invariance means that we simply replace x2 with x2 x3 , etc. This expression is exactly the form required by conformal invariance (as discussed in lecture), and it is a rather nontrivial check of the AdS/CFT correspondence that we could obtain this form from this somewhat involved gravity calculation.
II. RELATION BETWEEN ADS PROPAGATORS
Here we will establish a useful relation between bulkboundary and bulktobulk propagators in AdS space. Switch ing notation for simplicity, we will work now with AdS space in the form ds2 = 1 2 dz + (dxi )2 2 z (19)
Let K (z, x; x ) be the bulkboundary propagator dened in (12). This is a solution to the bulk equations of motion that satises
z
lim K (z, x; x ) = d (x x )
(20)
and let G (z, x; z , x ) be the bulkbulk propagator, dened as the normalizable solution to the bulk equations of motion, but with source: 1 (x m2 )G(z, x; z , x ) = d (x x )(z z ) g Finally, we consider Greens second identity, applicable to any two functions and : dd x (n n ) dd+1 x g (x m2 ) (x m2 ) =
U U
(21)
(22)
Now: set (z, x) = G (z, x; z1 , x1 ) and (z, x) = K (z, x; x2 ) in that case we end up with the following: K (z1 , x1 ; x2 ) = lim dd xz 1d [G (z, x; z1 , x1 )z K (z, x; x2 ) K (z, x; x2 )z G (z, x; z1 , x1 )]
z AdS
(23)
Now, if z and z1 are separated, then the fact that G is a normalizable solution to the AdS wave equation means that asymptotically it must behave as G z + . Using this fact and the known behavior of the scalar propagator (20) to evaluate the radial derivatives, we end up with d K(z1 , x1 ; x2 ) = dd x G (, x; z1 , x1 )[ + ] d (x x2 ) (24) Thus using + + = d, we nally end up with K (z1 , x1 ; x2 ) = which is the desired relation. + G (, x2 ; z1 , x1 ), + (25)
4
III. WILSON LINE WITH CUSP
We are now going to compute the value of a Wilson loop with a cusp with opening angle v. This (and several related computations) have been done in [2], which I will mostly follow. We begin (as always) by reducing an impossible strongly coupled QFT problem to a fun classical mechanics exercise via the prescription discussed in lecture, W [v]
CFT
= exp(S[v])
(26)
where S[v] is the classical action of a string worldsheet whose boundary on AdS has the cusp with opening angle v. Note that this action is the normal NambuGoto action, 1 (27) S[v] = d2 (X X )2 X 2 X 2 2 For this problem we will work with the AdS metric in the following form: ds2 = L2 2 dz + dr2 + r2 d2 + r2 (angles that will not matter to us) 2 z (28)
We will use r, as coordinates on the string worldsheet as well; thus our task is to determine the function z(r, ), which completely determines the embedding of the string in the bulk geometry. Now at rst this problem looks very dicult: nding the string solution that asymptotes to a shape as peculiar as a line with a cusp seems like a nontrivial endeavor. Life is made easier by the fact that the boundary conditions are invariant under a scale transformation r r. This means that the bulk worldsheet solution should also be invariant under the bulk dieomorphism r r, z z, leading us to the following scaleinvariant ansatz for the string solution: z(r, ) = In that case the string action (27) becomes L2 S= 2 drd 4 f + f 2 + f 2 r (30) r f () (29)
and we need only solve for f (). This action does not explicitly depend on ; thus it admits a conserved energy E, which works out to be E= f2 + f4 f 4 + f 2 + f 2 (31)
z(r, 0) = z(r, v) = 0
(32)
so we can write
This simply says that at the AdS boundary z = 0 the opening angle is v. Thus we have a setup where f ( = 0) is , it dips down to nite values for (0, v) before shooting o to again at = v. We would like to relate the opening angle v to the parameter E which will be more useful for evaluating the action integral. To do this, note that the minimum value f0 of f () will happen at = v/2, where f (v/2) = 0 and thus we have 2 E = f0 1 + f0 (33) v = 2
v/2
d =
f0
2 1/2 f (1 + f 2 ) df f 1 + f 2 1 2 2 f0 (1 + f0 )
(34)
where in the second integral we have changed integration variables from to f and used (31) and (33) to relate df /d to f0 . In principle this integral can be done as a function of f0 and inverted to give f0 as a function of v. In reality, this turns out to involve elliptic integrals and is rather unenlightening; thus let us instead expand in powers of f0 .
5 It is not immediately clear why you might want to do this, but if we press on we nd that the leading term in the integral is 1 v f0 2 2 + O(f0 ) = arctan df + O(f0 ) (35) = 2 f2 + f4 2 f0 f 1+f f0
Thus we see that the limit f0 0 corresponds to v , i.e. a small opening angle. We will work in this limit for the remainder of the problem (note that there is no problem in principle in extending this to arbitrary opening angle; however the number of elliptic integrals proliferates and so I prefer to avoid this). Next we express the area in terms of f0 ; starting with (30) and expanding in powers of f0 , we obtain eventually the very simple expression dr L2 2 df + O(f0 ) (36) S= r f0 This is divergent at large f ; however if we look back at the denition of f in (29), we see that if we have a UV cuto on z at z = , we should really only integrate up to r/. Thus we nd L2 1 f0 rmax 1 S= = (37) dr rmax f0 log r rmin We are done. The rst term is proportional to the length of the Wilson line and is just the usual linear divergence; the second term is the extra contribution from the discontinuity in the Wilson line. Note it has the expected dependence on the IR and UV cutos rmax and rmin and indeed vanishes as f0 0, i.e. as v from (35).
IV. SURFACE GRAVITY
Here we study the properties of metrics of the form dz 2 ds2 = (z) f (z)dy 2 + + ds2 other f (z)
(38)
where the function f (z) has a rstorder zero at z = zm and (z) is wellbehaved there. We begin by Taylor expanding f (z) near z = zm , dz 2 2 2 ds = (zm ) (z zm )f (zm )dy + + ... (39) (z zm )f (zm ) This peculiarlooking metric is actually something that we are very familiar with. To understand this, imagine picking a coordinate r such that dz 2 f (zm ) 2 (zm ) = dr2 (z zm ) = r (z zm )f (zm ) 4(zm ) (40)
to obtain the second inequality I integrated the rst relation in a small neighborhood of zm . In terms of this new coordinate r, the metric near zm simply becomes ds2 = r2 f (zm )2 2 dy + dr2 + ... 4 (41)
This should look even more familiar if we dene a coordinate = yf (zm )/2, giving us simply ds2 = r2 d2 + dr2 + ... (42)
This is precisely the metric of Euclidean 2d space in polar coordinates! However we know well that this spacetime has a conical singularity at the origin unless has a periodicity of 2. This means that 4 (43) y 0, f (zm )
6 Thus we have found the periodicity of y. Note that metrics such as (38) can arise as the analytic continuations of black holes to Euclidean space, in which case y is imaginary time and its periodicity is nothing less than the inverse temperature of the black hole. We now work out some examples. First we do the analytic continuation of the ordinary 4d Schwarzschild black hole in at space, for which we have f (z) = 1 2GM z = 1 (44)
Taking the derivative and evaluating at z = zm = 2GM , we get ymax = 8GM as the periodicity of y. Note this 1 means the black hole has a temperature of T = 8GM , a rather famous result that we have quickly rederived. Next, we move on to the Euclidean AdS black hole with planar horizon, which has f (z) = 1 z zm 4 = 1 z2 (45)
1 From here we obtain ymax = zm , and thus T = zm . Finally we turn to a slightly dierent way to compute this periodicity using the surface gravity , dened by 1 2 = a b c d gbd g ac (46) 2 z=zm
where xi is the tangent vector to the shrinking circle, = y . Note that the analysis of the earlier section showed us that the nearhorizon metric could be written in the form ds2 = 2 r2 dy 2 + dr2
(47)
where = f (zm )/2. Now to compute the relevant covariant derivatives we need the Christoel symbols; the relevant components work out to be y = ry 1 r r = r2 yy (48)
a As the upper coordinates of a are all constant ( a = y ), only the Christoel symbols contribute to the expression (46), and we obtain
2 =
(49)
(50)
telling us that the surface gravity is simply related to the inverse periodicity (and thus the temperature) of the black hole.
[1] D. Z. Freedman, S. D. Mathur, A. Matusis, and L. Rastelli, Correlation Functions in the CF Td /AdSd+1 correspondence, Nuclear Physics B, Volume 546, Number 1, 26 April 1999 , pp. 96118(23) [arXiv:hepth/9804058]. [2] N. Drukker, D. J. Gross, H. Ooguri, Wilson Loops and Minimal Surfaces, Phys. Rev. D 60, 125006 (1999) [arXiv:hep th/9904191]