gt; 1/2$ with non-vanishing value of $\int J^0 d^3x$ (i.e. charge) - **Theorem 2**: a theory that allows a Lorentz-covariant, conserved $T^{\mu\nu}$ cannot contain massless particles of spin $j>1$ - **Remarks**: - the theorems apply to both elementary and composite particles - the theorem is compatible with QED - photons: spin 1 but no charge - electrons: spin 1/2 - YM SU(2): - with $A^3_{\mu}, \,A^{\pm}_\mu=\frac{1}{\sqrt{2}}(A_\mu^1\pm iA_\mu^2)$, $A^{\pm}_\mu$ massless spin-1 charged under U(1) subalgebra generated by $\sigma^3/2$ - BUT: there does not exist a conserved Lorentz-covariant current for this U(1) - gauge-invariant and conserved current is not Lorentz-covariant - Lorentz-covariant and gauge invariant current is not conserved - these theorems do not forbid graviton from GR - In GR, there is no *conserved* Lorentz-covariant $T^{\mu\nu}$ - $\nabla_\mu T^{\mu\nu}=0$ <=> *covariantly conserved* (but not conserved) - *conserved* means ordinary derivatives - theorem 2 => none of renormalisable QFTs in Minkowski can have emergent gravity - hidden assumption: the particles live in the same spacetime in the original theory! - **Proof**: Suppose there exist massless particles of spin-j. - one particle state: $|k^\mu,\sigma\rangle$, $\sigma=\pm j$ = helicity - consider rotation $\hat{R}(\theta,\hat k)|k,\sigma\rangle=e^{i\sigma \theta}|k,\sigma\rangle$ around its spatial direction - now conserved current $J^\mu$ => $Q=\int d^3 x J^0$ - $T^{\mu\nu}$=> $\hat P^\mu=\int T^{0\mu}$, $\hat P^\mu|k,\sigma\rangle=k^\mu|k,\sigma\rangle$ - if charged under $\hat Q$: $\hat Q|k,\sigma\rangle=q|k,\sigma\rangle$ - want to show: 1) if $q\ne0$, $j\le1/2$; 2) $j\le1$. - i) Lorentz symmetry: $\left\langle k, \sigma\left|J^{\mu}\right| k^\prime, \sigma\right\rangle$ -> $\frac{qk^\mu}{k^0}\frac{1}{(2\pi)^3}$ and $\left\langle k, \sigma\left|T^{\mu\nu}\right| k^\prime, \sigma\right\rangle$ -> $\frac{k^\mu k^\nu}{k^0}\frac{1}{(2\pi)^3}$ as $k\rightarrow k^\prime$ - intuition: when $\mu=0$, $|k,\sigma\rangle$ is eigenstate of $J^0$ with eigenvalue $q$; when $\mu\ne0$ is must carry a $\mu$ index and must reduce to $q$ when $\mu=0$ - ii) for massless particles, $k^2=0={k^\prime}^2$ => $k.k^\prime<0$, $k+k^\prime$ timelike. -> choose a frame s.t. $\vec k+\vec k^\prime=0$. so $k^\mu=(E,0,0,E),{k^\prime}^\mu=(E,0,0,-E)$ - iii) under rotation of $\theta$ around 3-direction, $\hat R(\theta)|k,j\rangle=e^{\mathrm{i}j\theta}|k,j\rangle$, $\hat R(\theta)|k^\prime j\rangle=e^{-\mathrm{i}j\theta}|k^\prime,j\rangle$. Consider $\langle k^\prime,j|\hat R^{-1}(\theta)J^\mu \hat R(\theta)|k,j\rangle$ => $e^{2 i j{\theta}}\langle k^\prime, j| J^{k}|k, j\rangle ={\Lambda^\mu}_\nu\langle k^\prime, j| J^{k}|k, j\rangle$ ==(4)==. - can do it for $T^{\mu\nu}$ similarly => ==(5)== - now $\Lambda$ can only have eigenvalue $e^{\pm i\theta}, 1$ - => $j\le1/2$ from (4) and $j\le1$ from (5). QED ### 1.2 Black hole thermodynamics - when Schwarzschild radius much larger than Compton, expect quantum effect to be negligible, but this is not true: - BH can have macroscopic quantum effects! - [[0017 Planck length]] is the minimal scale one can probe - different regimes - in particular, semiclassical: $\hbar$ finite, expand in $G_N$ - classical BH - Rindler horizon etc ### 1.3 Holographic principle - in non-gravitational system, number of d.o.f. is proportional to volume, but can easily violate entropy bound - so QG leads to a huge reduction of the number of d.o.f. - i.e. coupling any QFT to gravity leads to formation of BH before too many d.o.f. are stuffed into any given region - **holographic principle**: in QG, a region of boundary area A can be fully described by no more than A/4 d.o.f. (so the number of Hilbert space would be $e^{A/4}$) ### 1.4 Large N matrix theories - QCD: $SU(3)$ gauge theory + quarks - $L = -\frac{1}{g_{YM}}\left[-\frac{1}{4}TrFF- i \Psi(\cancel{D}-m)\Psi\right]$ - 't Hooft (1974): take number of colours as a parameter, i.e. $A_\mu$ are now N by N matrices, and do expansion in 1/N. - **Surprise**: $1/N$ expansion <-> String Theory - key: fields are matrices - **Illustration example**. Consider scalar theory - $\mathcal{L}=-\frac{1}{g^{2}} \operatorname{Tr}\left[\frac{1}{2} \partial_{\mu} \Phi \partial^{\mu} \Phi+\frac{1}{4} \Phi^{4}\right]$ - $\left\langle\Phi_{b}^{a}(x) \Phi_{d}^{c}(y)\right\rangle=g^{2} \delta_{d}^{a} \delta_{b}^{c} G(x-y)$ can represent using **double lines** - 4-vertex: $\frac{1}{g^{2}} \delta_{h}^{a} \delta_{b}^{c} \delta_{d}^{e} \delta_{g}^{f}$ - vacuum diagrams - ![[Rsc0010_ribbons_vac.png|400]] ![[Rsc0010_ribbons_vac_cd.png|400]] - number of closed single-line loops = power of N (because each loop is taking a trace) - **Observations** - (b) and (d) can be drawn on a torus without crossing lines ![[Rsc0010_torus_bd.png|400]] - power of N = number of faces on torus - **Fact**: any orientable 2d surface is classified topologically by an integer h, called genus. Equal to the number of holes. - **Claim 1**. For any non-planar diagram, there exists an integer $h$, such that the diagram can be straightened out on a genus-$h$ surface, but not on a surface with a smaller genus. - **Claim 2**. For any diagram, the power N coming from contracting propagators is given by the number of faces on such a genus-h surface. - => a vacuum diagram has $A\sim(g^2)^{E-V}N^F$ - E: no. of propagators; V: vertices; F: faces - seems no sensible N -> infinity limit - but 't Hooft: possible to have finite limit if $g \to 0$. - define $\lambda\equiv g^2N$ ('t Hooft coupling) - then $A\equiv (g^2N)^{E-V}N^{F+V-E}=\lambda^{L-1}N^{2-2h}$ - $L=E-(V-1)$ = number of loops - number of loops = \# undetermined momentum; each propagator carries a momentum; each vertex fixes a momentum conservation, although the overall momentum conservation is guaranteed (so $V-1$ independent constraints) - $\chi=F+V-E=2-2h$ - this now has a well-defined large-$N$ limit as $h\ge0$. - to leading order in large $N$: planar diagrams (genus 0) has $N^2(c_0+c_1\lambda+...)=N^2f_0(\lambda)$。 - turns out to be convergent: the number of *planar* diagrams grows very slowly with the power of $\lambda$ - in general, **all vacuum diagrams**: $\log Z=\sum_{h=0}^\infty N^{2-2h}f_h(\lambda)=N^2f_0(\lambda)+f_1+\frac{1}{N^2}f_2(\lambda)+\dots$ - **a heuristic way** to understand this expansion - $Z=\int D\Phi e^{iS[\Psi]}$ - $\mathcal{L}=-\frac{N}{\lambda}Tr(\dots)$: this is order $N^2$ (trace is a sum of $N$ things) - so (leap of faith) if we do saddle point approximation, we get powers of $1/N^2$ in the expansion - **Claim**. For *any* Lagrangian of matrix valued fields of the form $\mathcal{L}=\frac{N}{\lambda}Tr(\dots)$, we have $\log Z=\sum_{h=0}^\infty N^{2-2h}f_h(\lambda)=N^2f_0(\lambda)+f_1+\frac{1}{N^2}f_2(\lambda)+\dots$ - **Summary**. In the 't Hooft limit, $1/N$ expansion = expansion in the topology of Feynman diagrams. - In **gauge theories**, $\mathcal{L}=\mathcal{L}(A_\mu,\Phi,\dots)$ - allowed **local operators**: $Tr(F_{\mu\nu}F^{\mu\nu})$, $Tr(\Phi^k)$ <- single trace: denote $O_n$ - $Tr(F_{\mu\nu}F^{\mu\nu})Tr(\Phi^2)$ <- multi-trace" denote $O_1O_2O_3...$ - **general observables**: correlation functions of gauge invariant operators - **Trick to obtain $N$ dependence** of correlation functions - $Z\left[J_{1}, \ldots ,J_{n}\right]=\int D{A_\mu}D\Phi \exp \left[i S_{0}+i N\int J_i(x) O_i(x)\right]$ - => $\left\langle O(x_1) \cdots O_{n}(x_{n})\right\rangle_C=\frac{1}{i^n}\frac{\delta^n\log Z}{\delta J_1(x_1)\dots\delta J_n(x_n)}\frac{1}{N^n}$ - now notice that the whole exponent has $N \text{Tr}(\dots)$ just like before => $\log Z(J_1,\dots, J_n)=\sum_{h=0}^\infty N^{2-2h}f_h(\lambda,J_i)$ - => ==$\left\langle O(x_1) \cdots O_{n}(x_{n})\right\rangle_C=\sum_{h=0}^\infty N^{2-n-2h}f_h(\lambda,J_i)\sim N^{2-n}(1+\mathcal{O}(\frac{1}{N^2})+\dots)$== - **Physical implications**. 1. $O_1\dots O_n(x)|0\rangle$ = $n$-particle state - (a) $\langle O_iO_j\rangle\sim \delta_{ij}$ by choice of basis - (b) $\langle O_i(x) :O_1O_2:(y)\rangle\sim 1/N\rightarrow 0$. i.e. no mixing between single particle and multi-particle states - (c) $\langle :O_1O_2:(x) :O_1O_2:(y)\rangle$ $= \langle O_1(x)O_2(y)\rangle\langle O_1(x)O_2(y)\rangle+\langle O_1O_2O_1O_2\rangle_C$ second terms goes to 0 at large N, so this is like two particles propagating on their own - n.b. may not be physical particles states, just saying they are like them. in QCD they can be short-lived glueball states 2. glueball fluctuations are suppressed - suppose $\langle O\rangle\ne0\sim N$ - variance = $\langle O^2\rangle-\langle O\rangle^2=\langle O^2\rangle_C\sim N^0$ - so $\sqrt{\text{variance}}/\langle O\rangle\sim 1/N \rightarrow 0$ - disconnected correlation always factorise -> like a classical theory - $\langle O_1O_2\rangle=\langle O_1\rangle\langle O_2\rangle+\langle O_1O_2\rangle_C$ first term always dominate 3. If we interpret $\langle O_1(x_1)\dots O_n(x_n)\rangle_C$ as scattering amplitudes of $n$ glueballs (we call each single particle state a glueball, then the scatterings are classical (i.e. tree-level) - (a) think of a 3-pt vertex as a 3-pt. correlator so $\sim 1/N\equiv \tilde{g}$ => can convince oneself that tree-level $n$-particle amplitudes $\sim \tilde{g}^{n-2}\sim N^{2-n}$ - (b) can also include higher order vertices (4, 5, ...) - (c) to leading order in N, in any correlation function, there are only one-particle intermediate states. - try inserting complete basis (which are called intermediate states), then will find that contribution from multi-particle states vanish for large N - contribution from multi-particle states form loops: e.g. $\langle O_1O_2O_3\rangle$ contains possibility of 1 -> i,j (2 particles) -> 2,3, where the splitting and joining would form a loop - **Summary**. at large $N$, we have a *classical* theory of glueballs, with interactions among them given by $\tilde{g}\sim1/N$. - gauge theory at large N but finite $\hbar$ = glueball theory with effective $\tilde{\hbar}\rightarrow0$ - perturbation in $1/N$ <-> semi-classical expansion in $\tilde{\hbar}\sim1/N$ - **Caveat**. We are talking about loops of glueballs (gauge-invariant composite objects) which are suppressed. The gauge-dependent fields such as original *gluons* can have loops in their amplitudes. ### 1.5 Large N expansion as a String theory - QFT: 2nd quantized approach. - String theory: a generalisation of the 1st quantised approach: theory of particles (interaction needs to be added by hand, e.g. a particle splitting into 2 particles; adding interaction = specifying what paths to include in the PI for the particles) - **Vacuum energy**. $Z=\sum_{h}\sum_{\text{all surf. with genus } h}e^{-S_{NG}}$, where we use Euclidean Nambu-Goto action now. - whenever you have a discrete sum, can add a weight by hand: $Z=\sum_{h}e^{\lambda\chi}\sum_{\text{all surf. with genus } h}e^{-S_{NG}}$, where $\lambda$ is like a chemical potential for topology. - turns out that this weighting is what we get if we quantise strings properly, some other powers of $\chi$ will give different theories - write ==$g_S=e^\lambda$== - e.g. genus 0: order $g_S^{-2}$; genus 1: order 1 etc - **Remarkable fact**. Summing over topology automatically includes interactions of the string. In fact this fully specifies string interactions. - effectively giving a factor of $g_S$ for each splitting or joining (adding a genus gives $g_S^2$, and adding a genus adds two vertices) - n.b. can slice the diagrams in different ways and can in principle get a string splitting into 3, but it is consistent to have only 1-to-2 splitting. - **With external strings**. Since now the worldsheets have $n$ boundaries, $\chi=2-2h-n$, the amplitude is given by ==$A_n=\sum_{h=0}^\infty g_S^{n-2+2h}F_n^{(h)}$== - identical structure with large-$N$ theories! - with $g_S\leftrightarrow 1/N$, external strings $\leftrightarrow$ glueballs (single-trace operators), sum over string worldsheets of genus $h$ $\leftrightarrow$ sum over Feynman diagrams with genus $h$ - **BUT**, can you really identify the two ($f_n^{(h)}$ and $F_n^{(h)}$)? - $f_n^{(h)}$ = $\sum_\text{Feynman diagrams of genus h}G=\sum_\text{all triangulation of surface with genus h}G$ = discrete form of summing over all surfaces - so they are really the same if $G=e^{-S_{NG}}$ - what string theory it corresponds to will depend on the nature of Feynman diagrams: given a QFT at large $N$, need to figure out that string theory it maps to. - **Comment**. The identification is difficult: 1. $G$ is expressed as products of field theory propagators integrated over spacetime. they look nothing like $e^{-S_{string}}$ - $S_{string}$ is a map from worldsheet to target space, which needs to be chosen. need to choose (a) the target space, (b) what $S_{String}$ looks like, (c) additional internal d.o.f. on the worldsheet e.g. fermions in Superstring 2. String theory: continuum. Feynman diagrams: at best discrete version. - expect geometric picture for $G$ to emerge only at strong 't Hooft coupling (at very strong coupling, Feynman diagrams with infinite number of vertices dominate: more like continuum) - recall $A\equiv (g^2N)^{E-V}N^{F+V-E}=\lambda^{L-1}N^{2-2h}$ - with fixed number of lines at each vertex, number of loops is proportional to the number of vertices: think about squares on the floor - alternative - $\mathcal{L}=\frac{N}{\lambda}Tr(\frac{1}{2}(\partial\Phi)^2+\frac{1}{4}\Phi^4)$ and $\lambda$ goes to the $\Phi^4$ after rescaling -> i.e. propagator $g^2$, vertex $1/g^2$ becomes propagator $1$, vertex $g^2$ - so no need to count number of propagators; just count vertices 3. For simple systems like a matrix integral ($dM$) or matrix quantum mechanics ($DM(t)$: paths), can guess the corresponding string theory. both are simple. - see [[Klebanov1991]]@[arXiv](https://arxiv.org/abs/hep-th/9108019) sec. 2 - **Generalisations**. - so far matrix valued fields, adjoint rep of the $U(N)$ gauge group. Can also include fields in the fundamental rep. (quarks). These have single index, so are represented by single lines. => Feynman diagrams can be classified by 2d surfaces with boundaries (for vacuum diagrams) - <-> a String theory with both closed and open strings - so far used $U(N)$ as gauge group. Can consider $SO(N)$ or $Sp(N)$ so that fields are $\Phi_{ab}$ -> no up and down indices -> lines are not directed -> non-orientable surfaces - <-> non-orientable string theories - Take large $N$ generalisation of QCD in 3+1 d Mink, can we say anything about its String theory description? - natural guess: a String theory in $M_{3+1}$, $S_{NG}=\frac{1}{2\pi\alpha^\prime}\int dA$. But it does not work: - such a string theory is inconsistent except for $D=26$, or 10 if add fermions - take a string in 10d with $M_{3+1}\times N$ with $N$ compact - would not work, because string theory always contain gravity, and if put in compact space, it always leads to massless spin-2 particle in the $M_{3+1}$ part -> not okay for QCD in 3+1 [[0151 Weinberg-Witten theorem]] - -> what to do now? - (a) more exotic action - (b) other target space - Hints for looking for a string theory in *4+1* non-compact spacetime - holographic principle: String theory contains gravity, and theories with gravity have d.o.f. proportional to area -> an equivalent non-gravitational description should be one dimension lower. (this is hindsight: no one realised it before the discovery) - consistency of string theory itself (sometimes adding d.o.f. in order to have the correct symmetry leads to one higher dimension) - Now suppose a $4+1$ manifold, Y, describes QCD - by Poincare invariance of 3+1, the metric must be of the form $d s^{2}=a^{2}(z)\left(d z^{2}+ \eta_{\mu\nu} d X^{\mu} d X^\nu\right)$ - if the field theory is CFT -> can show that this metric must be AdS (easily) - **History** - late 60's and early 70's, String theory developed to understand strong interactions - 1971 Asymptotic freedom. Eliminated the hope of using string theory to describe strong interactions: QCD very different from string theory - 1974 - 't Hooft: large N like a string theory - Schwarz etc: string theory as a gravity - Hawking radiation - quarks observed (charmonium) - Wilson: lattice QCD. quarks can be confined through the strings. -> revived string theory - 1993, 1994, 't Hooft: holographic principle - 1995 Polchinski: D-branes - 1997 June, Polyakov: realised that it should be 4+1 not 3+1 - 1997 November, Maldacena: construction using D-branes - 1998 February, Witten: made the connection to holographic principle ## 2. Deriving AdS/CFT 1. spectrum of closed and open strings - closed string -> gravity - open string -> gauge theory 2. D-branes - special role of connecting gravity and gauge theory ### 2.1 More about string theory #### 2.1.1 General setup - setup - consider a string with $ds^2=g_{\mu\nu}(X)dX^\mu dX^\nu$ - worldsheet parameterised by $X^\mu(\sigma^a)$ with $\sigma^a=(\tau,\sigma), a=0,1$ - induced metric $ds_{ind}^2=h_{ab}d\sigma^a d\sigma^b$, $h_{ab}=g_{\mu\nu}\partial_a X^\mu \partial_b X^\nu$ - action $S_{NG}=\frac{1}{2\pi\alpha^\prime}\int d^2\sigma \sqrt{-\det h}$ - inconvenient so rewrite as (**Polyakov** form): - $S_P[\gamma.X]=-\frac{1}{4\pi\alpha^\prime}\int d^2\sigma\sqrt{-\gamma}\gamma^{ab}(\sigma^a)\partial_a X^\mu\partial_bX^\nu g_{\mu\nu}(X^\mu)$ - $\gamma^{ab}$ is like Lagrangian multiplier - **sigma model** - Polyakov action has the form of a scalar field on a 2-dim spacetime, but now both $\gamma, X^\mu$ dynamics - -> can be considered as 2d gravity coupled to $D$ scalar fields - path integral over both: $\int D\gamma_{ab}DX^\mu e^{iS_P[\gamma,X]}$ - n.b. one of the scalar fields has the wrong sign for KE term - will be resolved by integration over $\gamma$ - symmetries of the action - (a) Poincare symmetry of $X^\mu$ - (b) 2d coordinate transformation (reparameterisation) - (c) Weyl scaling of $\gamma_{ab}$ - a bit surprising - since NG action has no $\gamma$, this is needed if Polyakov is equivalent to NG - **Euler action** - can use (a,b,c) to (almost) determine the action - another action consistent with 3 symmetries: **Euler action** $S_{Euler}=\frac{\lambda}{4\pi}\int d^2\sigma\sqrt{-\gamma}R$ - but this is a total derivative -> actually topological: depends on genus - in Euclidean path integral $e^{-S_{Euler}}=e^{-\lambda \chi}$ - categorise the symmetries - (a) is global -> conserved charges - (b,c) gauge symmetry #### 2.1.2 Light cone quantisation - why **canonical quantisation** - PI is simpler but does not give string spectrum - basic picture - closed string: each oscillation mode gives rise to a particle - open string: also get gauge fields - infinite number of types of particles - quantization **procedure** - write down EOM - fix gauge - find complete set of classical solutions - promote classical worldsheet fields to quantum operators satisfying canonical commutation relations - then classical solutions become solutions to operator equations - parameters in classical solutions (for SHO: amplitude of oscillations, a and a*) -> creation and annihilation operators - use ladders to find the spectrum - gauge freedom - use 3 gauge freedoms to fix 3 d.o.f. of $\gamma_{ab}$ to something fixed value - this is can be done locally. as spectrum can be obtained using local stuff, no need to worry about global issues - but will worry about global issues when doing PI - lightcone gauge - first solve for true d.o.f. then quantise (rather than quantise and then impose constraints as operator constraints) - $d s^{2}=-d t^{2}+d s^{2}=-2 d \sigma^{+} d \sigma^{-}$ - this metric is preserved by **residual gauge** freedom: $\tilde{\sigma}^{+}=f\left(\sigma^{+}\right)$, $\tilde{\sigma}^{-}=g\left(\sigma^{-}\right)$ followed by a Weyl scaling (this combination is the residual gauge freedom) - lightcone gauge: $\tau=X^+/v^+$ for constant $v^+$, i.e. identify worldsheet time with lightcone time - consequence - $\partial_\tau X^+=v^+$, $\partial_\sigma X^+=0$ - $X^-$ can be fully solved in terms of $X^i