Stress Tensor and Conserved Currents

The previous page organized local operators into conformal families. This page focuses on the two most important universal classes of local operators in a CFT:

$$T_{\mu\nu}(x), \qquad J_\mu^a(x).$$

The stress tensor $T_{\mu\nu}$ is the conserved current for spacetime symmetries. A flavor current $J_\mu^a$ is the conserved current for a continuous global symmetry. They are not merely examples of local operators. They are the operators that make symmetry local inside correlation functions.

For AdS/CFT, these operators are unavoidable:

$$T_{\mu\nu} \longleftrightarrow \text{bulk metric}, \qquad J_\mu^a \longleftrightarrow \text{bulk gauge field}.$$

If one wants to understand why gravity and gauge fields appear in the bulk, one must first understand why $T_{\mu\nu}$ and $J_\mu^a$ are universal CFT operators.

The point of this page

A conserved current is a local operator whose divergence vanishes away from contact terms:

$$\partial_\mu J^\mu(x)=0 \qquad \text{for }x\neq x_i$$

inside a correlator with operator insertions at $x_i$. The words “away from contact terms” are not a nuisance. They are the main point. The contact terms say how other operators transform under the symmetry.

For a global symmetry, the schematic Ward identity is

$$\partial_\mu\langle J^\mu_a(x)\mathcal O_1(x_1)\cdots\mathcal O_n(x_n)\rangle = -\sum_{i=1}^n\delta^{(d)}(x-x_i) \langle\mathcal O_1(x_1)\cdots (t_a^{(i)}\mathcal O_i)(x_i)\cdots\mathcal O_n(x_n)\rangle.$$

Here $t_a^{(i)}$ is the symmetry generator acting on $\mathcal O_i$. The overall sign is conventional; it depends on whether one defines $\delta\mathcal O=+\alpha^a t_a\mathcal O$ or $\delta\mathcal O=-\alpha^a t_a\mathcal O$. The invariant lesson is that the divergence of $J^\mu_a$ vanishes at separated points and becomes a delta function when the current hits a charged operator.

For translations, the analogous Ward identity is generated by the stress tensor. For scalar insertions,

$$\partial_\mu\langle T^{\mu\nu}(x)\mathcal O_1(x_1)\cdots\mathcal O_n(x_n)\rangle = -\sum_{i=1}^n\delta^{(d)}(x-x_i) \frac{\partial}{\partial x_i^\nu} \langle\mathcal O_1(x_1)\cdots\mathcal O_n(x_n)\rangle.$$

For spinning operators there are additional contact terms implementing rotations of spin indices. For scale and special conformal transformations there are contact terms involving scaling dimensions and spin representations. Thus the stress tensor is the operator that inserts infinitesimal spacetime transformations into correlators.

This is the operational meaning of a symmetry in QFT:

$$\boxed{ \text{a symmetry is a Ward identity for a conserved current.} }$$

Two complementary definitions

There are two standard ways to define $T_{\mu\nu}$ and $J_\mu^a$.

The first is the Noether definition. If a classical action is invariant under a continuous transformation, there is a conserved current. Translation invariance gives $T^\mu{}_{\nu}$. Internal global symmetry gives $J^\mu_a$.

The second is the source definition. Couple the QFT to nondynamical background fields:

$$g_{\mu\nu}(x), \qquad A_\mu^a(x), \qquad \lambda^i(x).$$

Here $g_{\mu\nu}$ is the background metric, $A_\mu^a$ is a background gauge field for a global symmetry, and $\lambda^i$ are scalar sources for local operators $\mathcal O_i$. Define the Euclidean generating functional

$$W[g,A,\lambda]=- \log Z[g,A,\lambda].$$

Our convention for one-point functions is

$$\delta W = \frac12\int d^d x\sqrt g\,\langle T^{\mu\nu}\rangle\delta g_{\mu\nu} + \int d^d x\sqrt g\,\langle J^\mu_a\rangle\delta A_\mu^a + \int d^d x\sqrt g\,\langle\mathcal O_i\rangle\delta\lambda^i.$$

Equivalently,

$$\langle T^{\mu\nu}(x)\rangle = \frac{2}{\sqrt g}\frac{\delta W}{\delta g_{\mu\nu}(x)},$$ $$\langle J^\mu_a(x)\rangle = \frac{1}{\sqrt g}\frac{\delta W}{\delta A_\mu^a(x)}, \qquad \langle\mathcal O_i(x)\rangle = \frac{1}{\sqrt g}\frac{\delta W}{\delta\lambda^i(x)}.$$

This definition is the one that fits AdS/CFT most naturally. Boundary sources are boundary values of bulk fields, and CFT one-point functions are responses of the on-shell bulk action.

The stress tensor $T_{\mu\nu}$ and global current $J_\mu^a$ are defined as responses to the background sources $g_{\mu\nu}$ and $A_\mu^a$. In a CFT, they are conserved operators with protected dimensions $\Delta_T=d$ and $\Delta_J=d-1$. In AdS/CFT, their sources become boundary values of the bulk metric $G_{MN}$ and bulk gauge field $A_M^a$.

Background gauge invariance and current conservation

Suppose the CFT has a continuous global symmetry group $G$, with Lie algebra generators $t_a$. The word “global” means that the symmetry parameter is constant in the original QFT. But once we introduce a background gauge field $A_\mu^a$, we may write the generating functional in a way that is invariant under local background gauge transformations:

$$\delta_\alpha A_\mu^a = D_\mu\alpha^a,$$

where

$$D_\mu\alpha^a = \partial_\mu\alpha^a+f^a{}_{bc}A_\mu^b\alpha^c.$$

If the symmetry has no anomaly, background gauge invariance gives

$$0=\delta_\alpha W = \int d^d x\sqrt g\,\langle J^\mu_a\rangle D_\mu\alpha^a.$$

Integrating by parts gives

$$D_\mu\langle J^\mu_a\rangle=0$$

in the absence of charged insertions. In flat space and with $A_\mu^a=0$, this becomes

$$\partial_\mu J^\mu_a=0$$

at separated points.

With charged operator insertions, the Ward identity acquires contact terms. These contact terms are precisely what make $J_\mu^a$ the generator of the symmetry.

Charges and radial surfaces

Given a conserved current, the associated charge is an integral over a codimension-one surface $\Sigma$:

$$Q_a(\Sigma)=\int_\Sigma d\Sigma_\mu\,J^\mu_a.$$

If $\partial_\mu J^\mu_a=0$ and no charged operator is crossed while deforming $\Sigma$, then $Q_a(\Sigma)$ is independent of the surface. This is the local-to-global mechanism behind conserved charges.

In Lorentzian quantization with $\Sigma$ a constant-time slice,

$$Q_a=\int d^{d-1}x\,J^0_a.$$

In radial quantization, $\Sigma$ is usually a sphere surrounding the origin:

$$Q_a=\int_{S^{d-1}} dS_\mu\,J^\mu_a.$$

A small sphere surrounding an operator insertion measures its charge. If

$$\mathcal O_i$$

transforms in a representation $R_i$ of $G$, then

$$Q_a\mathcal O_i(0) = (t_a^{(i)}\mathcal O_i)(0)$$

up to the same sign convention used in the Ward identity. This is the cleanest way to think about the current-operator OPE: the singularity of $J_\mu^a(x)\mathcal O_i(0)$ is fixed by the charge of $\mathcal O_i$.

More explicitly, if $S_{d-1}$ is the area of the unit $(d-1)$-sphere, a conventional leading singularity is

$$J^a_\mu(x)\mathcal O_i(0) \sim -\frac{x_\mu}{S_{d-1}|x|^d} (t_a^{(i)}\mathcal O_i)(0) +\cdots.$$

The normalization is chosen so that the flux through a small sphere gives one unit of the symmetry generator:

$$\int_{S_r^{d-1}}dS^\mu\frac{x_\mu}{S_{d-1}|x|^d}=1.$$

This is the higher-dimensional cousin of the familiar two-dimensional OPE

$$J^a(z)\mathcal O_i(w,\bar w) \sim \frac{(t^a\mathcal O_i)(w,\bar w)}{z-w} +\cdots,$$

for a holomorphic current, up to convention-dependent signs.

Stress tensor as the current of spacetime symmetry

The stress tensor is the conserved current for translations:

$$\partial_\mu T^{\mu\nu}=0.$$

The corresponding momentum charges are

$$P^\nu=\int_\Sigma d\Sigma_\mu\,T^{\mu\nu}.$$

If the theory is rotationally invariant, one can choose the stress tensor to be symmetric,

$$T_{\mu\nu}=T_{\nu\mu},$$

possibly after adding an improvement term. The angular-momentum currents are then

$$M^{\mu;\rho\sigma} = x^\rho T^{\mu\sigma}-x^\sigma T^{\mu\rho} + \text{spin terms}.$$

For scalar fields the spin terms are absent. For spinning fields they implement rotations of the operator indices.

In a CFT, after choosing a properly improved stress tensor, one also has

$$T^\mu{}_{\mu}=0$$

at separated points in flat space. Then the dilatation current is

$$D^\mu=x_\nu T^{\mu\nu},$$

and the special conformal currents are

$$K^\mu{}_{\rho} = (2x_\rho x_\nu-x^2\eta_{\rho\nu})T^{\mu\nu}.$$

Their conservation follows from conservation, symmetry, and tracelessness of $T_{\mu\nu}$.

Thus the entire conformal algebra can be generated by integrals of the stress tensor:

$$P^\nu = \int_\Sigma d\Sigma_\mu\,T^{\mu\nu},$$ $$D = \int_\Sigma d\Sigma_\mu\,x_\nu T^{\mu\nu},$$ $$K_\rho = \int_\Sigma d\Sigma_\mu\,(2x_\rho x_\nu-x^2\eta_{\rho\nu})T^{\mu\nu}.$$

This is why the stress tensor is the central universal operator of any CFT.

Ward identity for conformal transformations

Let $\xi^\mu(x)$ be an infinitesimal spacetime transformation. The stress-tensor insertion

$$\int d^d x\,\partial_\mu\left(\xi_\nu\langle T^{\mu\nu}(x)X\rangle\right)$$

acts on a product of local operators

$$X=\mathcal O_1(x_1)\cdots\mathcal O_n(x_n).$$

For a conformal Killing vector,

$$\partial_\mu\xi_\nu+ \partial_\nu\xi_\mu = \frac{2}{d}(\partial\cdot\xi)\eta_{\mu\nu},$$

the induced transformation of a scalar primary is

$$\delta_\xi\mathcal O_i(x_i) = -\xi^\mu(x_i)\partial_\mu\mathcal O_i(x_i) - \frac{\Delta_i}{d}(\partial\cdot\xi)(x_i)\mathcal O_i(x_i).$$

Therefore conformal invariance gives the integrated Ward identity

$$\sum_{i=1}^n \left[ \xi^\mu(x_i)\frac{\partial}{\partial x_i^\mu} + \frac{\Delta_i}{d}(\partial\cdot\xi)(x_i) \right] \langle\mathcal O_1(x_1)\cdots\mathcal O_n(x_n)\rangle =0$$

for scalar primaries. For spinning primaries, one adds the spin-rotation term

$$\frac12\omega_{\mu\nu}^{(i)}(x_i)\Sigma_i^{\mu\nu},$$

where $\Sigma_i^{\mu\nu}$ are spin generators in the representation of $\mathcal O_i$.

This formula is the practical bridge between local stress-tensor conservation and the familiar constraints on CFT correlators.

Protected dimensions of currents

In a unitary CFT, conservation fixes the scaling dimensions of conserved currents:

$$\Delta_J=d-1, \qquad \Delta_T=d.$$

The stress tensor is a spin-two operator. A global current is a spin-one operator. Conservation equations are shortening conditions:

$$\partial_\mu J^\mu_a=0, \qquad \partial_\mu T^{\mu\nu}=0.$$

They remove descendant states from the conformal multiplet. From the representation-theory viewpoint, $J_\mu^a$ and $T_{\mu\nu}$ sit exactly at unitarity bounds. We will derive these bounds later, but it is useful to state the result now:

$$\boxed{ \text{conserved current }J_\mu: \Delta=d-1, }$$ $$\boxed{ \text{stress tensor }T_{\mu\nu}: \Delta=d. }$$

A small but important caveat: in $d>2$, the stress tensor is usually treated as a spin-two primary operator. In $d=2$, the holomorphic stress tensor $T(z)$ transforms with a Schwarzian derivative when the central charge is nonzero, so it is quasi-primary under the global conformal group but not a primary under the full Virasoro symmetry. This exception is not a defect; it is the beginning of the central-charge story.

Two-point functions and normalizations

Conformal symmetry almost completely fixes the two-point functions of conserved currents. For a global symmetry current in flat Euclidean space,

$$\langle J_\mu^a(x)J_\nu^b(0)\rangle = \frac{C_J\delta^{ab}}{(x^2)^{d-1}}I_{\mu\nu}(x),$$

where

$$I_{\mu\nu}(x) = \delta_{\mu\nu}-2\frac{x_\mu x_\nu}{x^2}.$$

The coefficient $C_J$ is physical once the normalization of the symmetry generators is fixed. It is sometimes called a flavor central charge.

The stress-tensor two-point function has the form

$$\langle T_{\mu\nu}(x)T_{\rho\sigma}(0)\rangle = \frac{C_T}{(x^2)^d} \mathcal I_{\mu\nu,\rho\sigma}(x),$$

with

$$\mathcal I_{\mu\nu,\rho\sigma}(x) = \frac12 \left( I_{\mu\rho}(x)I_{\nu\sigma}(x) + I_{\mu\sigma}(x)I_{\nu\rho}(x) \right) - \frac1d\delta_{\mu\nu}\delta_{\rho\sigma}.$$

The coefficient $C_T$ is among the most important pieces of CFT data. In two-dimensional CFT, the central charge $c$ plays an analogous role. In higher dimensions, $C_T$ controls the normalization of stress-tensor fluctuations.

In a holographic CFT with a weakly coupled Einstein gravity dual,

$$C_T\sim \frac{R_{\mathrm{AdS}}^{d-1}}{G_N},$$

up to a known dimension-dependent normalization convention. Thus large $C_T$ is the CFT signal of small bulk Newton constant and semiclassical gravity.

Similarly,

$$C_J\sim \frac{R_{\mathrm{AdS}}^{d-3}}{g_{\mathrm{bulk}}^2}$$

for a bulk gauge field with coupling $g_{\mathrm{bulk}}$, again up to convention-dependent numerical factors.

What is fixed and what is dynamical?

Ward identities fix the singular terms that encode charges. They do not fix all current correlators.

For example, symmetry fixes the leading singularity in

$$J_\mu^a(x)\mathcal O_i(0),$$

because a small sphere around $\mathcal O_i$ must reproduce the representation matrix $t_a^{(i)}$. But the full three-point function

$$\langle J_\mu^a(x_1)\mathcal O_i(x_2)\mathcal O_j(x_3)\rangle$$

contains dynamical coefficients. These are CFT data, constrained by Ward identities but not eliminated by them.

Similarly, the leading contact terms in stress-tensor Ward identities are fixed by the dimensions and spin representations of operators. But coefficients such as

$$C_T, \qquad C_J, \qquad \lambda_{TTT}, \qquad \lambda_{JJT}, \qquad \lambda_{\mathcal O\mathcal O T}$$

are genuine CFT data. Some of them are partially fixed by Ward identities once operator normalizations are chosen. Others encode dynamical information about the theory.

A useful slogan is:

$$\boxed{ \text{symmetry fixes the form; CFT data fixes the numbers.} }$$

Diffeomorphism Ward identity with sources

The source definition makes it easy to write Ward identities on nontrivial backgrounds. If the theory has a metric, a background flavor gauge field, and scalar sources, diffeomorphism invariance implies a relation of the schematic form

$$\nabla_\mu\langle T^\mu{}_{\nu}\rangle = F^a_{\nu\mu}\langle J^\mu_a\rangle + \langle\mathcal O_i\rangle\nabla_\nu\lambda^i + \text{anomaly terms}.$$

When the background sources are turned off and there is no anomaly, this reduces to

$$\partial_\mu T^{\mu}{}_{\nu}=0.$$

This equation has a simple meaning. If a background electric field acts on a charged system, energy-momentum is not conserved by the matter sector alone; the background field can inject energy and momentum. If scalar sources vary in space, translation invariance is explicitly broken and the stress tensor has a corresponding nonzero divergence.

In AdS/CFT, this same identity appears as a boundary constraint following from bulk diffeomorphism invariance.

Current multiplets and supersymmetry

In non-supersymmetric CFTs, $T_{\mu\nu}$ and $J_\mu^a$ may sit in unrelated conformal families. In supersymmetric CFTs, they often belong to larger supermultiplets.

For example, in four-dimensional $\mathcal N=4$ super Yang-Mills, the stress tensor, supercurrents, and $SU(4)_R$ currents all belong to the same protected stress-tensor multiplet. This is one reason the theory is so rigid and so useful in AdS/CFT. The bulk dual does not contain just a graviton in isolation; it contains a whole supergravity multiplet.

We will return to this much later. For now, remember the moral:

$$\text{ordinary symmetry currents become parts of supermultiplets in SCFTs.}$$

Anomalies

A global current may fail to be conserved in the presence of background fields. This is an anomaly. Schematically,

$$D_\mu J^\mu_a=\mathcal A_a[A,g].$$

For an ordinary global symmetry, such an anomaly is not an inconsistency. It is an ‘t Hooft anomaly: a protected piece of CFT data that must be matched along RG flows.

A gauge symmetry of a dynamical theory cannot have an uncanceled anomaly. But a global symmetry can. When we say a CFT has a global symmetry $G$, we often mean its Ward identities hold in flat space with no background fields, while background gauge transformations of $W[A]$ may have a controlled anomalous variation.

In AdS/CFT, boundary anomalies are often encoded by bulk Chern-Simons terms or related topological couplings. This is a beautiful example of how a contact-term-level CFT statement becomes a geometric term in the bulk effective action.

Holographic checkpoint

The source definition of $T_{\mu\nu}$ and $J_\mu^a$ is already the AdS/CFT dictionary in embryonic form.

For the stress tensor,

$$\delta W_{\mathrm{CFT}}[g] = \frac12\int d^d x\sqrt g\,\langle T^{\mu\nu}\rangle\delta g_{\mu\nu}$$

is matched to variation of the renormalized bulk on-shell action with respect to the boundary metric. The bulk field dual to $T_{\mu\nu}$ is the graviton.

For a global current,

$$\delta W_{\mathrm{CFT}}[A] = \int d^d x\sqrt g\,\langle J^\mu_a\rangle\delta A_\mu^a$$

is matched to variation of the bulk on-shell action with respect to the boundary value of a bulk gauge field.

The bulk constraints are the gravitational and gauge-theory versions of the CFT Ward identities:

$$\text{bulk diffeomorphism constraint} \quad\leftrightarrow\quad \nabla_\mu T^{\mu\nu}=0,$$ $$\text{bulk Gauss law} \quad\leftrightarrow\quad D_\mu J^\mu_a=0.$$

This is one of the deepest lessons of holography: conservation laws on the boundary are not optional features of the duality. They are the boundary form of gauge redundancies in the bulk.

Common pitfalls

Pitfall 1: thinking a background gauge field makes the symmetry dynamical

The source $A_\mu^a$ used to define $J^\mu_a$ is nondynamical. It is a book-keeping device for current correlators. In AdS/CFT, the corresponding bulk field is dynamical, but its boundary value is fixed as a source.

Pitfall 2: forgetting contact terms

The conservation equation

$$\partial_\mu J^\mu=0$$

is a separated-point statement. In a correlator with charged operators, the divergence has delta-function contact terms. Those terms are exactly how the current knows the charges of the operators.

Pitfall 3: confusing flavor currents with gauge fields in the CFT

A CFT global current $J_\mu^a$ is gauge-invariant and physical. It is not the same thing as a gauge field inside a Lagrangian description. In holography, global symmetry on the boundary becomes gauge symmetry in the bulk.

Pitfall 4: treating $C_T$ and $C_J$ as conventions only

The numerical values of $C_T$ and $C_J$ depend on normalization conventions. But once conventions are fixed, they are physical CFT data. Their scaling with $N$ is especially important in holographic theories.

Summary

The stress tensor and global currents are the universal conserved operators of a CFT:

$$\partial_\mu T^{\mu\nu}=0, \qquad \partial_\mu J^\mu_a=0$$

at separated points, with contact terms implementing transformations of operator insertions.

They are most cleanly defined by varying the generating functional with respect to background sources:

$$T^{\mu\nu}\sim \frac{\delta W}{\delta g_{\mu\nu}}, \qquad J^\mu_a\sim \frac{\delta W}{\delta A_\mu^a}.$$

In a unitary CFT,

$$\Delta_T=d, \qquad \Delta_J=d-1.$$

Their two-point functions are fixed by conformal symmetry up to constants $C_T$ and $C_J$. In AdS/CFT,

$$T_{\mu\nu}\leftrightarrow G_{MN}, \qquad J_\mu^a\leftrightarrow A_M^a.$$

So the CFT stress tensor and global currents are the boundary origins of bulk gravity and bulk gauge fields.

Exercises

Exercise 1 — Surface independence of a charge

Let $J^\mu$ be a conserved current in Euclidean space away from operator insertions:

$$\partial_\mu J^\mu=0.$$

Let $\Sigma_1$ and $\Sigma_2$ be two closed codimension-one surfaces enclosing the same set of charged insertions. Show that

$$\int_{\Sigma_1}d\Sigma_\mu J^\mu = \int_{\Sigma_2}d\Sigma_\mu J^\mu.$$

Solution

Let $M$ be the region between $\Sigma_1$ and $\Sigma_2$. Since no operator insertion lies in $M$, the current is conserved everywhere in $M$:

$$\partial_\mu J^\mu=0.$$

By Gauss’s theorem,

$$0= \int_M d^d x\,\partial_\mu J^\mu = \int_{\partial M}d\Sigma_\mu J^\mu.$$

The boundary $\partial M$ is $\Sigma_2$ with one orientation and $\Sigma_1$ with the opposite orientation. Therefore

$$\int_{\Sigma_2}d\Sigma_\mu J^\mu - \int_{\Sigma_1}d\Sigma_\mu J^\mu =0.$$

Hence the charge is independent of continuous deformations of the surface, as long as no charged operator crosses the surface.

Exercise 2 — Current conservation of the conformal two-point function

Define

$$I_{\mu\nu}(x)=\delta_{\mu\nu}-2\frac{x_\mu x_\nu}{x^2}.$$

Show that, for $x\neq0$,

$$\partial_\mu\left(\frac{I_{\mu\nu}(x)}{(x^2)^{d-1}}\right)=0.$$

This verifies separated-point conservation of the current two-point function.

Solution

Write

$$\frac{I_{\mu\nu}(x)}{(x^2)^{d-1}} = \delta_{\mu\nu}(x^2)^{1-d} -2x_\mu x_\nu(x^2)^{-d}.$$

The derivative of the first term is

$$\partial_\mu\left[\delta_{\mu\nu}(x^2)^{1-d}\right] = \partial_\nu(x^2)^{1-d} = 2(1-d)x_\nu(x^2)^{-d}.$$

For the second term,

$$\partial_\mu\left[x_\mu x_\nu(x^2)^{-d}\right] = (d+1)x_\nu(x^2)^{-d} -2d x_\nu(x^2)^{-d} = (1-d)x_\nu(x^2)^{-d}.$$

Multiplying by $-2$ gives

$$-2(1-d)x_\nu(x^2)^{-d}.$$

Adding the two contributions,

$$2(1-d)x_\nu(x^2)^{-d} -2(1-d)x_\nu(x^2)^{-d}=0.$$

Thus

$$\partial_\mu\left(\frac{I_{\mu\nu}(x)}{(x^2)^{d-1}}\right)=0$$

for $x\neq0$.

Exercise 3 — Trace of the stress-tensor two-point tensor structure

Using

$$\mathcal I_{\mu\nu,\rho\sigma}(x) = \frac12 \left( I_{\mu\rho}I_{\nu\sigma} + I_{\mu\sigma}I_{\nu\rho} \right) - \frac1d\delta_{\mu\nu}\delta_{\rho\sigma},$$

show that

$$\delta^{\mu\nu}\mathcal I_{\mu\nu,\rho\sigma}(x)=0.$$

Solution

First note that $I_{\mu\nu}(x)$ is an orthogonal matrix:

$$I_{\mu\rho}(x)I_{\mu\sigma}(x)=\delta_{\rho\sigma}.$$

Then

$$\delta^{\mu\nu}\frac12 \left( I_{\mu\rho}I_{\nu\sigma} + I_{\mu\sigma}I_{\nu\rho} \right) = \frac12(\delta_{\rho\sigma}+\delta_{\sigma\rho}) = \delta_{\rho\sigma}.$$

The trace of the subtraction term is

$$\delta^{\mu\nu}\frac1d\delta_{\mu\nu}\delta_{\rho\sigma} = \frac1d d\,\delta_{\rho\sigma} = \delta_{\rho\sigma}.$$

Therefore the two pieces cancel:

$$\delta^{\mu\nu}\mathcal I_{\mu\nu,\rho\sigma}=0.$$

This is the tensor-structure version of stress-tensor tracelessness.

Exercise 4 — Background gauge invariance implies current conservation

Assume the source variation

$$\delta W= \int d^d x\sqrt g\,\langle J^\mu_a\rangle\delta A_\mu^a.$$

Under a background gauge transformation,

$$\delta_\alpha A_\mu^a=D_\mu\alpha^a.$$

Show that gauge invariance of $W$ implies

$$D_\mu\langle J^\mu_a\rangle=0$$

when there is no anomaly.

Solution

Gauge invariance means

$$0=\delta_\alpha W = \int d^d x\sqrt g\,\langle J^\mu_a\rangle D_\mu\alpha^a.$$

Integrating by parts covariantly gives

$$0= -\int d^d x\sqrt g\,\alpha^a D_\mu\langle J^\mu_a\rangle,$$

up to a boundary term. If $\alpha^a(x)$ is arbitrary and the boundary term vanishes, then

$$D_\mu\langle J^\mu_a\rangle=0.$$

If the symmetry is anomalous, the right-hand side is replaced by the anomaly functional.

Exercise 5 — Why large $C_T$ means weak bulk gravity

In a holographic CFT with an Einstein gravity dual, suppose

$$C_T\sim \frac{R_{\mathrm{AdS}}^{d-1}}{G_N}.$$

Explain why a semiclassical bulk gravity limit requires $C_T\gg1$.

Solution

The strength of quantum gravity effects in the bulk is controlled by the dimensionless ratio

$$\frac{G_N}{R_{\mathrm{AdS}}^{d-1}}.$$

If this ratio is small, the bulk Planck scale is much smaller than the AdS curvature radius, and the gravitational path integral is dominated by classical saddle points.

Using

$$C_T\sim \frac{R_{\mathrm{AdS}}^{d-1}}{G_N},$$

small bulk Newton coupling is equivalent to

$$C_T\gg1.$$

Thus large $C_T$ is the boundary signal of weakly coupled semiclassical gravity. In large-$N$ gauge theories, $C_T$ usually scales like the number of degrees of freedom, often $C_T\sim N^2$ for adjoint matrix theories.