Holographic Anomalies and the Central Charges of N=4 SYM

A conformal field theory can have a nonzero stress-tensor trace on a curved background even when all beta functions vanish. This Weyl anomaly measures a quantum response to geometry. In holography, it comes from a logarithmic divergence of the gravitational action near the AdS boundary.

For five-dimensional Einstein gravity, the result is

$$a=c=\frac{\pi R^3}{8G_5}.$$

On $AdS_5\times S^5$, flux quantization then gives $a=c=N^2/4$ at leading large $N$. We will derive the sign and coefficient through an exact spherical example and a general near-boundary calculation, then compare with the protected field-theory answer. The gravitational calculation uses the leading two-derivative action; it does not include quantum bulk corrections.

The boundary metric and the stress tensor

The shared metric-source definition, together with the Euclidean source prescription, gives

$$\delta W_E[g] =\frac12\int d^4x\,\sqrt g\, \langle T^{ij}\rangle\,\delta g_{ij}, \qquad W_E=\log Z_E=-S_{E,\mathrm{ren}}.$$

Thus the stress tensor extracted from the Euclidean on-shell action is

$$\langle T^{ij}\rangle =-\frac{2}{\sqrt g} \frac{\delta S_{E,\mathrm{ren}}}{\delta g_{ij}}.$$

Here $T^{ij}$ is the response to the four-dimensional boundary metric. Its definition is distinct from the worldsheet stress tensor used in the earlier string quantization pages.

A Weyl variation changes lengths locally:

$$\delta_\sigma g_{ij}=2\sigma(x)g_{ij}.$$

Consequently,

$$\delta_\sigma W_E =\int d^4x\,\sqrt g\,\sigma(x)\,\mathcal A(x), \qquad \mathcal A\equiv\langle T^i{}_i\rangle.$$

This fixes the sign of the anomaly. The variation of $S_{E,\mathrm{ren}}$ has the opposite sign.

A diffeomorphism instead has $\delta g_{ij}=\nabla_i\xi_j+\nabla_j\xi_i$. Integrating by parts gives $\nabla_i\langle T^{ij}\rangle=0$ when there are no other sources, no diffeomorphism anomaly, and no boundary contribution. Conservation and a nonzero Weyl anomaly are therefore compatible: they concern different variations of the metric.

The three curvature terms in four dimensions

Write the anomaly as

$$\mathcal A =\frac{1}{16\pi^2} \left(c\,C_{ijkl}C^{ijkl}-a\,E_4+b\,\Box\mathcal R\right),$$

where $\mathcal R_{ij}$ and $\mathcal R$ are the Ricci tensor and scalar of the boundary metric. The two curvature invariants are

$$\begin{aligned} C_{ijkl}C^{ijkl} &=\mathcal R_{ijkl}\mathcal R^{ijkl} -2\mathcal R_{ij}\mathcal R^{ij} +\frac13\mathcal R^2,\\ E_4 &=\mathcal R_{ijkl}\mathcal R^{ijkl} -4\mathcal R_{ij}\mathcal R^{ij} +\mathcal R^2. \end{aligned}$$

The coefficient $a$ multiplies the Euler density; on a smooth closed four-manifold,

$$\int d^4x\,\sqrt g\,E_4=32\pi^2\chi.$$

The coefficient $c$ multiplies the Weyl tensor squared and also fixes the separated-point stress-tensor two-point function. They are independent data in a general four-dimensional CFT.

The coefficient $b$ depends on the subtraction scheme: a finite local $\int\sqrt g\,\mathcal R^2$ counterterm changes it. Such a term cannot change $a$ or $c$. We use the scheme with $b=0$ for the Einstein calculation below; Exercise 2 derives the effect of changing this scheme.

A sphere measures the Euler anomaly

On a round four-sphere of radius $r$,

$$\mathcal R_{ij}=\frac3{r^2}g_{ij}, \qquad \mathcal R=\frac{12}{r^2}, \qquad C_{ijkl}=0.$$

Its volume is $8\pi^2r^4/3$, while $E_4=24/r^4$. Therefore

$$\int_{S^4}d^4x\,\sqrt g\,\mathcal A=-4a.$$

A constant Weyl rescaling is a change of $r$, so

$$r\frac{\partial W_E}{\partial r}=-4a, \qquad r\frac{\partial S_{E,\mathrm{ren}}}{\partial r}=+4a.$$

The sphere partition function thus determines $a$ from its logarithmic radius dependence. It does not determine $c$, because the Weyl tensor vanishes on the sphere.

The Euclidean AdS filling

Euclidean $AdS_5$ with a spherical conformal boundary has metric

$$ds^2=R^2\left(d\rho^2+\sinh^2\rho\,d\Omega_4^2\right).$$

The center is $\rho=0$, and the boundary is approached as $\rho\to\infty$. Cut off the geometry at $\rho=\rho_0$. With the shared curvature and outward-normal definitions, the required action is

$$\begin{aligned} S_{\rm bulk} &=-\frac1{16\pi G_5}\int_M d^5x\,\sqrt G \left(\mathcal R_5+\frac{12}{R^2}\right),\\ S_{\rm GH} &=-\frac1{8\pi G_5}\int_{\partial M}d^4x\,\sqrt\gamma\,K,\\ S_{\rm ct,power} &=\frac1{8\pi G_5}\int_{\partial M}d^4x\,\sqrt\gamma \left(\frac3R+\frac R4\mathcal R[\gamma]\right). \end{aligned}$$

The Gibbons–Hawking term makes the metric Dirichlet variation well defined. The two displayed counterterms remove the leading power divergences. A logarithmic counterterm will be added after its coefficient has been found.

The Einstein equation and its trace give

$$\mathcal R_{MN}=-\frac4{R^2}G_{MN}, \qquad \mathcal R_5=-\frac{20}{R^2}.$$

Hence the bulk integrand is negative, and the Euclidean action becomes positive:

$$S_{\rm bulk} =\frac{1}{2\pi G_5R^2}\int_Md^5x\,\sqrt G.$$

At the outer surface,

$$n=\frac1R\partial_\rho, \qquad K=\frac4R\coth\rho_0, \qquad \mathcal R[\gamma]=\frac{12}{R^2\sinh^2\rho_0}.$$

Cancel the power divergences explicitly

Introduce the dimensionless prefactor

$$\mathcal P\equiv \frac{\operatorname{Vol}(S^4_{\rm unit})R^3}{8\pi G_5} =\frac{\pi R^3}{3G_5}.$$

The three action terms are

$$\begin{aligned} \frac{S_{\rm bulk}}{\mathcal P} &=4\int_0^{\rho_0}d\rho\,\sinh^4\rho,\\ \frac{S_{\rm GH}}{\mathcal P} &=-4\sinh^3\rho_0\cosh\rho_0,\\ \frac{S_{\rm ct,power}}{\mathcal P} &=3\sinh^4\rho_0+3\sinh^2\rho_0. \end{aligned}$$

The primitive needed in the first line is

$$\int_0^{\rho_0}d\rho\,\sinh^4\rho =\frac{\sinh4\rho_0}{32} -\frac{\sinh2\rho_0}{4} +\frac{3\rho_0}{8}.$$

Combining all three terms cancels their exponentially growing pieces and gives the exact finite-cutoff expression

$$S_{\rm sub}\equiv S_{\rm bulk}+S_{\rm GH}+S_{\rm ct,power} =\mathcal P\left( \frac32\rho_0-\frac38+\frac38e^{-4\rho_0} \right).$$

Only the bulk integral contributes the term linear in $\rho_0$, but the boundary terms are essential to obtain a finite subtraction with the stated constants.

To identify the boundary radius and the ultraviolet regulator, set

$$z=2r\,e^{-\rho}.$$

The metric becomes

$$ds^2=\frac{R^2}{z^2} \left[ dz^2+r^2\left(1-\frac{z^2}{4r^2}\right)^2d\Omega_4^2 \right].$$

Its boundary representative is $r^2d\Omega_4^2$. A cutoff $z=\epsilon$ corresponds to $\rho_0=\log(2r/\epsilon)$.

The coefficient of this logarithm is $3\mathcal P/2$. Comparison with the sphere response $4a$ gives

$$a=\frac{\pi R^3}{8G_5}, \qquad S_{\rm sub}=4a\rho_0-a+ae^{-4\rho_0}.$$

The logarithmic counterterm on this geometry is $4a\log(\epsilon\mu)$, where $\mu$ is a subtraction scale with inverse-length units. Taking $\epsilon\to0$ gives

$$\begin{aligned} S_{E,\mathrm{ren}}(S^4_r) &=4a\log(2r\mu)-a,\\ W_E(S^4_r) &=-4a\log(2r\mu)+a. \end{aligned}$$

The displayed constants belong to the chosen minimal subtraction scheme. A finite local curvature counterterm can change them. The universal result is

$$r\frac{\partial W_E}{\partial r}=-4a,$$

with exactly the sign predicted by the boundary anomaly.

For the exact spherical Einstein filling, $D(\rho_0)=(4a)^{-1}\partial_{\rho_0}S_{\rm sub}=1-e^{-4\rho_0}$. Both axes are dimensionless. The action includes the bulk, Gibbons–Hawking, and power counterterms; the remaining logarithm is subtracted afterward. The limiting value measures its coefficient, while the finite-cutoff deviation is explicitly $e^{-4\rho_0}$.

The local anomaly from the near-boundary geometry

The spherical calculation supplies one integrated answer. To determine the anomaly for an arbitrary smooth boundary metric, use Fefferman–Graham coordinates:

$$ds^2=\frac{R^2}{z^2} \left(dz^2+g_{ij}(z,x)\,dx^i dx^j\right).$$

The local expansion in four boundary dimensions is

$$g(z,x) =g_{(0)}+z^2g_{(2)} +z^4\left[g_{(4)}+\log(z^2\mu^2)\,h_{(4)}\right]+\cdots.$$

The leading coefficient $g_{(0)}$ is the boundary metric source. The equations determine $g_{(2)}$ and the relevant traces locally from it, while part of $g_{(4)}$ contains state-dependent response data. This distinction is the metric counterpart of the source/response separation on page 32.

We need only enough of the Einstein equations to determine the logarithmic volume term. All traces and curvatures in the following calculation use $g_{(0)}$.

The first Einstein coefficient

At the first nontrivial order, the tangential Einstein equation is

$$2g_{(2)ij} +\operatorname{tr}(g_{(2)})\,g_{(0)ij} =-\mathcal R_{ij}.$$

Taking its trace in four dimensions gives $6\operatorname{tr}(g_{(2)})=-\mathcal R$. Substitution then yields

$$g_{(2)ij} =-\frac12\left(\mathcal R_{ij}-\frac16\mathcal R\,g_{(0)ij}\right).$$

For a sphere, this is $g_{(2)}=-g_{(0)}/(2r^2)$, agreeing with the exact spherical metric.

The radial Einstein equation can be written directly in terms of $z$ derivatives as

$$\operatorname{tr}(g^{-1}g'') -\frac1z\operatorname{tr}(g^{-1}g') -\frac12\operatorname{tr}(g^{-1}g'g^{-1}g')=0.$$

Let $A=g_{(0)}^{-1}g_{(2)}$, $B=g_{(0)}^{-1}g_{(4)}$, and $H=g_{(0)}^{-1}h_{(4)}$. Inserting the expansion, the coefficient at order $z^2$ is

$$8\operatorname{tr}B-2\operatorname{tr}(A^2) +\left[8\log(z^2\mu^2)+12\right]\operatorname{tr}H=0.$$

The logarithmic and nonlogarithmic terms must vanish separately. Thus

$$\operatorname{tr}H=0, \qquad \operatorname{tr}B=\frac14\operatorname{tr}(A^2).$$

This supplies the trace needed for the action without determining the full state-dependent tensor $g_{(4)}$. The general expansion and its response interpretation are developed by de Haro, Solodukhin, and Skenderis.

The determinant term that produces a logarithm

Use $\sqrt{\det(1+M)}=\exp[\tfrac12\operatorname{tr}\log(1+M)]$. Since $\operatorname{tr}H=0$,

$$\frac{\sqrt{\det g(z)}}{\sqrt{\det g_{(0)}}} =1+v_2z^2+v_4z^4+\cdots,$$

with

$$\begin{aligned} v_2&=\frac12\operatorname{tr}A,\\ v_4 &=\frac12\operatorname{tr}B +\frac18(\operatorname{tr}A)^2 -\frac14\operatorname{tr}(A^2)\\ &=\frac18\left[(\operatorname{tr}A)^2-\operatorname{tr}(A^2)\right]. \end{aligned}$$

The first Einstein coefficient gives

$$\operatorname{tr}A=-\frac{\mathcal R}{6}, \qquad \operatorname{tr}(A^2) =\frac14\mathcal R_{ij}\mathcal R^{ij}-\frac1{18}\mathcal R^2.$$

Therefore

$$v_4=-\frac1{32} \left(\mathcal R_{ij}\mathcal R^{ij}-\frac13\mathcal R^2\right).$$

The on-shell bulk action contains

$$S_{\rm bulk} =\frac{R^3}{2\pi G_5} \int d^4x\,\sqrt{g_{(0)}} \int_\epsilon dz\,z^{-5} \left(1+v_2z^2+v_4z^4+\cdots\right).$$

The first two terms give powers of $\epsilon^{-1}$. The third gives $\int_\epsilon dz/z=-\log\epsilon+\text{finite}$, so its local logarithmic coefficient is

$$\mathcal L_{\log} =-\frac{R^3}{2\pi G_5}v_4 =\frac{R^3}{64\pi G_5} \left(\mathcal R_{ij}\mathcal R^{ij}-\frac13\mathcal R^2\right).$$

Thus $S_{\rm reg}$ contains $+\log\epsilon\int\sqrt{g_{(0)}}\,\mathcal L_{\log}$. The logarithm here is understood together with a reference length; the counterterm combines it into $\log(\epsilon\mu)$.

The remaining terms preserve this coefficient. A finite logarithm in $\sqrt\gamma K$ or the volume counterterm would be proportional to $\operatorname{tr}H$, which vanishes. In the intrinsic-curvature counterterm, the logarithmic correction first appears at a positive power of $\epsilon$ and vanishes at the boundary. These terms remove power divergences and affect finite local terms, but supply no additional finite logarithm at this order.

A radial cutoff change gives a Weyl response

For a constant passive change of defining coordinate,

$$z'=e^\sigma z, \qquad g_{(0)}'=e^{2\sigma}g_{(0)}.$$

The same regulated surface is described by $\epsilon'=e^\sigma\epsilon$. A space-dependent $\sigma(x)$ also requires a tangential coordinate shift to maintain Fefferman–Graham gauge.

The logarithmic counterterm has the form

$$S_{\rm ct,\log} =-\log(\epsilon\mu) \int d^4x\,\sqrt{g_{(0)}}\,\mathcal L_{\log},$$

or its equivalent local expression in the induced cutoff metric. Changing the defining function shifts the renormalized action by $-\int\sqrt{g_{(0)}}\,\sigma\mathcal L_{\log}$. Since $W_E=-S_{E,\mathrm{ren}}$, the anomaly is

$$\mathcal A=\mathcal L_{\log} =\frac{R^3}{64\pi G_5} \left(\mathcal R_{ij}\mathcal R^{ij}-\frac13\mathcal R^2\right)$$

in the chosen scheme. This is the four-dimensional holographic anomaly derived by Henningson and Skenderis.

Finally,

$$C_{ijkl}C^{ijkl}-E_4 =2\left(\mathcal R_{ij}\mathcal R^{ij}-\frac13\mathcal R^2\right).$$

Comparison with the anomaly definition establishes

$$a=c=\frac{\pi R^3}{8G_5}.$$

The equality follows from the curvature combination produced by the Einstein equations. It is not inferred merely from there being one gravitational coupling. Additional higher-derivative terms can change that combination and allow $a\ne c$.

Two complementary uses of the anomaly. The sphere has zero Weyl tensor, so its logarithmic radius response determines $a$ alone. General curved boundary data distinguish the Euler and Weyl-squared terms; the two-derivative Einstein calculation yields their equal coefficients. This is a schematic comparison of measurements, with no geometric scale implied.

Flux quantization fixes the N-squared coefficient

The preceding result concerns five-dimensional gravity. To compare it with the D3-brane gauge theory, reduce on the physical sphere of radius $R$. The shared Einstein-frame normalization and D3-brane parameters give

$$G_{10}=8\pi^6g_s^2\alpha'^4, \qquad R^4=4\pi g_sN\alpha'^2.$$

The Einstein term integrated over $S^5_R$ gives

$$G_5=\frac{G_{10}}{\pi^3R^5}.$$

It follows that

$$\frac{R^3}{G_5} =\frac{\pi^3R^8}{G_{10}} =\frac{2N^2}{\pi}, \qquad a=c=\frac{N^2}{4}.$$

Every factor has a role: $G_{10}$ has units of length to the eighth power, the physical internal volume has units of length to the fifth power, and $R^3/G_5$ is dimensionless. Using the unit-sphere volume without its $R^5$ would spoil both the dimensions and the coefficient.

For the unwarped $AdS_5\times S^5$ solution in the shared Einstein-frame normalization, radius and flux data combine with the physical internal volume to give $R^3/G_5=2N^2/\pi$. In the two-derivative approximation this fixes $a=c=N^2/4$. The arrows represent algebraic substitutions, not a flow between theories.

Free fields and the protected answer

The free-field anomaly contributions in four dimensions are

Field	$a$	$c$
One real conformal scalar	$1/360$	$1/120$
One Weyl fermion	$11/720$	$1/40$
One gauge vector	$31/180$	$1/10$

These weights, including the gauge-field contributions, are summarized in Gubser’s discussion of Einstein spaces and CFTs. An $\mathcal N=4$ vector multiplet contains six real scalars, four Weyl fermions, and one vector, all in the adjoint representation. Consequently,

$$\begin{aligned} a_{\rm multiplet} &=\frac6{360}+\frac{44}{720}+\frac{31}{180} =\frac14,\\ c_{\rm multiplet} &=\frac6{120}+\frac4{40}+\frac1{10} =\frac14. \end{aligned}$$

For gauge group $SU(N)$, there are $N^2-1$ generators:

$$a_{\rm SYM}=c_{\rm SYM}=\frac{N^2-1}{4}.$$

The equality with the interacting answer uses supersymmetry and the protected anomalies of the superconformal R current. It is not established by free-field arithmetic alone. Exercise 5 derives the result from the R-anomaly relations of Anselmi, Freedman, Grisaru, and Johansen.

For $U(N)$, the additional free center-of-mass vector multiplet contributes $1/4$ to both coefficients. The interacting $SU(N)$ sector and that decoupled multiplet must be distinguished. Classical supergravity reproduces the leading $N^2/4$ term; resolving the order-one difference on the gravity side requires a quantum calculation beyond the one performed here.

What this comparison does and does not protect

The anomaly coefficients remain fixed along the $\mathcal N=4$ conformal family. The finite-temperature entropy does not enjoy the same protection. In the classical black-brane regime,

$$s=\frac{\pi^2}{2}N^2T^3 =2\pi^2a\,T^3$$

at leading large $N$, as derived on page 28. The weakly coupled gas has a different entropy coefficient while retaining the same $a$ and $c$.

Thus $a$ and $c$ measure specific stress-tensor responses, rather than a universal coefficient for every observable with an $N^2$ scaling. A large value of $a$ also does not by itself guarantee a classical Einstein dual. For the simplest D3 supergravity calculation, small string curvature requires $\lambda\gg1$, and weak string coupling requires $g_s=\lambda/(4\pi N)\ll1$. The planar limit followed by large $\lambda$, or the sufficient window $1\ll\lambda\ll N$, makes those approximations explicit.

The stress-tensor two-point coefficient

In flat Euclidean space, conformal invariance fixes the separated-point form to

$$\langle T_{ij}(x)T_{kl}(0)\rangle =\frac{C_T}{|x|^8}\,\mathcal I_{ij,kl}(x), \qquad x\ne0,$$

where

$$\begin{aligned} I_{ij}(x)&=\delta_{ij}-2\frac{x_ix_j}{x^2},\\ \mathcal I_{ij,kl}(x) &=\frac12\left(I_{ik}I_{jl}+I_{il}I_{jk}\right) -\frac14\delta_{ij}\delta_{kl}. \end{aligned}$$

The tensor is symmetric in each index pair and trace-free. Its radial dependence and inversion matrices together enforce conservation away from coincidence; Exercise 1 verifies these statements directly.

With this definition,

$$C_T=\frac{40}{\pi^4}\,c.$$

One way to see why the normalization is fixed is the distributional argument of Osborn and Petkou, section 8. Define the transverse differential operators

$$\begin{aligned} P_{ij}&=\partial_i\partial_j-\delta_{ij}\Box,\\ \mathcal D_{ij,kl} &=\frac12(P_{ik}P_{jl}+P_{il}P_{jk})-\frac13P_{ij}P_{kl}. \end{aligned}$$

The same separated-point correlator can be written as

$$\langle T_{ij}(x)T_{kl}(0)\rangle =\frac{C_T}{320}\,\mathcal D_{ij,kl}\frac1{|x|^4}.$$

The subtraction coefficient here is $1/3$, because $P_{ij}$ projects transverse to one direction. It differs from the $1/4$ subtraction in the inversion tensor.

To extend $|x|^{-4}$ to coincidence, differential regularization gives

$$\left[\frac1{|x|^4}\right]_{\rm ren} =-\frac14\Box\frac{\log(\mu^2x^2)}{x^2}, \qquad \mu\partial_\mu\left[\frac1{|x|^4}\right]_{\rm ren} =2\pi^2\delta^{(4)}(x).$$

Hence the scale derivative of the correlator has coefficient $\pi^2C_T/160$ multiplying $\mathcal D_{ij,kl}\delta^{(4)}(x)$. To compare with the anomaly, perturb the flat metric by a compactly supported $h_{ij}$. The quadratic term is

$$\left(\int d^4x\,\sqrt g\,C_{ijkl}C^{ijkl}\right)_{h^2} =\frac12\int d^4x\,h_{ij}\mathcal D_{ij,kl}h_{kl}.$$

Its second variation therefore contributes $c\mathcal D/(16\pi^2)$. Each metric-source derivative carries a factor of $1/2$, so extracting the stress correlator requires an additional factor of four, as derived in Exercise 1:

$$\mu\partial_\mu\langle T_{ij}(x)T_{kl}(0)\rangle =\left.4\frac{\delta^2\!\int\sqrt g\,\mathcal A} {\delta h_{ij}(x)\delta h_{kl}(0)}\right|_{h=0} =\frac{c}{4\pi^2}\mathcal D_{ij,kl}\delta^{(4)}(x).$$

The Euler integral is topological and the $b$ term has been set to zero. Equating the two scale-derivative coefficients gives the stated relation.

For $\mathcal N=4$ $SU(N)$ theory this means

$$C_T=\frac{10(N^2-1)}{\pi^4}.$$

Classical gravity gives its leading $10N^2/\pi^4$ term. Local changes of scheme alter contact terms, not this separated-point coefficient.

Common pitfalls

Using an action derivative as a generating-functional derivative. The shared Euclidean relation is $W_E=-S_{E,\mathrm{ren}}$. It fixes the sign of the metric response and gives a negative integrated trace on $S^4$ for positive $a$.

Inferring both coefficients from a sphere. A round sphere has $C_{ijkl}=0$. The equality $a=c$ requires the general curvature calculation or another independent observable.

Treating a cutoff expression as already renormalized. The finite-cutoff sphere action has a correction proportional to $e^{-4\rho_0}$. Counterterms and the boundary limit are part of extracting the anomaly.

Extending protected data to thermal physics. The exact field-theory anomaly coefficients and the coupling-dependent entropy coefficient answer different questions. Their common leading $N^2$ scaling is not a nonrenormalization theorem for the entropy.

Exercises

1. Conservation, traces, and metric derivatives

Verify that the separated-point stress-tensor two-point function above is trace-free and conserved. Then determine the factor relating two covariant metric derivatives of $W_E$ to the connected correlator at distinct points.

Solution

First use $I_{ik}I_{il}=\delta_{kl}$. Tracing the inversion tensor gives

$$\delta_{ij}\mathcal I_{ij,kl} =\delta_{kl}-\frac44\delta_{kl}=0.$$

Conservation involves differentiating both the power of $|x|$ and the inversion matrices. Let $n_i=x_i/|x|$. The useful identities in four dimensions are

$$\partial_i|x|^{-8}=-8n_i|x|^{-9}, \qquad \partial_i n_j=\frac{\delta_{ij}-n_in_j}{|x|}.$$

They imply

$$n_i I_{ik}=-n_k, \qquad \partial_i I_{ik}=-\frac{6n_k}{|x|}.$$

For the symmetrized product, direct differentiation gives

$$\begin{aligned} \partial_i\!\left[ \frac{I_{ik}I_{jl}+I_{il}I_{jk}}{2|x|^8} \right] =-\frac{2}{|x|^9}\,\delta_{kl}n_j. \end{aligned}$$

The derivative of $-\delta_{ij}\delta_{kl}/(4|x|^8)$ is $+2\delta_{kl}n_j/|x|^9$, so the two contributions cancel. The identity holds for $x\ne0$; it does not discard the contact terms required by Ward identities at coincident insertions.

To identify the source factor, perturb the metric by $h_{ij}$. The Euclidean action varies as

$$\delta S_{\rm CFT} =-\frac12\int d^4x\,\sqrt g\,T^{ij}h_{ij}.$$

Thus the Boltzmann weight has a linear insertion $+\tfrac12\int\sqrt g\,h_{ij}T^{ij}$. The quadratic connected term in its logarithm is

$$W_E^{(2)} =\frac18\int d^4x\,d^4y\, \sqrt{g(x)}\sqrt{g(y)}\, h_{ij}(x)h_{kl}(y) \langle T^{ij}(x)T^{kl}(y)\rangle_c +\text{local terms}.$$

At distinct points,

$$\frac{\delta^2W_E}{\delta g_{ij}(x)\delta g_{kl}(y)} =\frac14\sqrt{g(x)}\sqrt{g(y)} \langle T^{ij}(x)T^{kl}(y)\rangle_c.$$

The two source factors supply $1/4$. The metric dependence of the measure and of the stress tensor itself supplies additional terms supported at coincidence.

2. Change the subtraction scheme

Add the finite term

$$\Delta S_{E,\mathrm{ren}} =\frac{\kappa}{16\pi^2}\int d^4x\,\sqrt g\,\mathcal R^2$$

on a smooth closed boundary. Determine the resulting change of $b$ and show why it cannot change the logarithmic radius response on $S^4$.

Solution

In four dimensions,

$$\delta_\sigma\sqrt g=4\sigma\sqrt g, \qquad \delta_\sigma\mathcal R=-2\sigma\mathcal R-6\Box\sigma.$$

The terms proportional to $\sigma\mathcal R^2$ cancel, leaving

$$\delta_\sigma\Delta S_{E,\mathrm{ren}} =-\frac{12\kappa}{16\pi^2} \int d^4x\,\sqrt g\,\mathcal R\,\Box\sigma.$$

On a closed manifold, integration by parts twice moves the Laplacian onto $\mathcal R$. Because $\Delta W_E=-\Delta S_{E,\mathrm{ren}}$,

$$\delta_\sigma\Delta W_E =\frac{12\kappa}{16\pi^2} \int d^4x\,\sqrt g\,\sigma\,\Box\mathcal R.$$

Therefore $b\mapsto b+12\kappa$. The sign would reverse if the same finite term were added to $W_E$ rather than to the action.

On the round sphere,

$$\int_{S^4_r}d^4x\,\sqrt g\,\mathcal R^2 =\frac{144}{r^4}\frac{8\pi^2r^4}{3} =384\pi^2.$$

It shifts the action by $24\kappa$, independent of $r$. It changes the finite constant but leaves $r\partial_rW_E=-4a$ unchanged.

3. Distinguish a and c on a product of spheres

Take the boundary metric to be a product $S^2_{r_1}\times S^2_{r_2}$ and set $k_i=r_i^{-2}$. Calculate the integrated anomaly. Explain how it complements the round-$S^4$ measurement.

Solution

For each two-sphere, $\mathcal R=2k$, $\mathcal R_{ij}\mathcal R^{ij}=2k^2$, and $\mathcal R_{ijkl}\mathcal R^{ijkl}=4k^2$. Since the product has no mixed curvature components,

$$\begin{aligned} \mathcal R&=2(k_1+k_2),\\ \mathcal R_{ij}\mathcal R^{ij}&=2(k_1^2+k_2^2),\\ \mathcal R_{ijkl}\mathcal R^{ijkl}&=4(k_1^2+k_2^2). \end{aligned}$$

Substitution into the invariant definitions gives

$$E_4=8k_1k_2, \qquad C_{ijkl}C^{ijkl}=\frac43(k_1+k_2)^2.$$

The volume is $16\pi^2/(k_1k_2)$. The Euler integral is therefore $128\pi^2$, consistent with $\chi(S^2\times S^2)=4$. The scalar curvature is constant, so the $b$ term vanishes. The integrated anomaly is

$$\int\sqrt g\,\mathcal A =\frac{4c}{3} \left(\frac{k_1}{k_2}+\frac{k_2}{k_1}+2\right)-8a.$$

For equal radii it is $16c/3-8a$, which becomes $-8a/3$ in the Einstein case $c=a$.

Unlike the round four-sphere, this product has a nonzero Weyl tensor even when its two radii agree. Once the round sphere has fixed $a$, the product response can determine $c$. This calculation uses the local CFT anomaly; it does not assume a unique or explicitly known bulk filling of this boundary geometry.

4. Normalizable metric data change the response

For a flat boundary, consider the asymptotic solution

$$g_{ij}(z,x)=\delta_{ij}+z^4t_{ij}(x)+\cdots, \qquad \delta^{ij}t_{ij}=0, \qquad \partial^it_{ij}=0.$$

Find its stress-tensor response and explain why it does not change the anomaly. Work to the displayed order and keep the outward normal directed toward decreasing $z$.

Solution

The induced metric and normal are

$$\gamma_{ij}=\frac{R^2}{z^2}\left(\delta_{ij}+z^4t_{ij}\right), \qquad n=-\frac zR\partial_z.$$

Thus

$$K_{ij}=\frac12\mathcal L_n\gamma_{ij} =\frac R{z^2}\delta_{ij}-Rz^2t_{ij}+\cdots, \qquad K=\frac4R+\cdots.$$

The trace has no order-$z^4$ correction because $t$ is trace-free.

For the displayed Euclidean action, varying the bulk plus Gibbons–Hawking term with respect to $\gamma_{ij}$ gives

$$\delta(S_{\rm bulk}+S_{\rm GH}) =\frac1{16\pi G_5}\int\sqrt\gamma\, (K^{ij}-K\gamma^{ij})\,\delta\gamma_{ij}.$$

The volume counterterm supplies $+3\gamma^{ij}/R$ inside the same bracket. Multiplying by $-2/\sqrt\gamma$, as required by $W_E=-S_E$, gives the cutoff response

$$\tau_{ij} =\frac1{8\pi G_5} \left(-K_{ij}+K\gamma_{ij}-\frac3R\gamma_{ij}\right) +\text{curvature-counterterm contributions}.$$

The intrinsic-curvature terms vanish in the boundary response for this flat-source perturbation: derivatives of $t$ carry additional positive powers of $z$. Inserting $K_{ij}$ and $\gamma_{ij}$ into the displayed terms gives

$$\tau_{ij}=\frac{Rz^2}{4\pi G_5}t_{ij}+\cdots.$$

The response to the boundary representative is obtained by undoing its conformal scale:

$$\langle T_{ij}\rangle =\lim_{z\to0}\left(\frac Rz\right)^2\tau_{ij} =\frac{R^3}{4\pi G_5}t_{ij}.$$

It is conserved and trace-free. Meanwhile $g_{(2)}=0$ and $\operatorname{tr}g_{(4)}=0$, so $v_4=0$: the anomaly of the flat source remains zero.

The calculation separates response data from the local anomaly determined by the source metric. It establishes the formal near-boundary response; it does not prove that every transverse trace-free choice of $t_{ij}$ extends to a regular global bulk state.

5. Recover the protected result from R anomalies

For a four-dimensional $\mathcal N=1$ SCFT, use

$$a=\frac3{32}\left(3\operatorname{Tr}R_{\!f}^3-\operatorname{Tr}R_{\!f}\right), \qquad c=\frac1{32}\left(9\operatorname{Tr}R_{\!f}^3-5\operatorname{Tr}R_{\!f}\right),$$

where the traces sum over Weyl fermions with their superconformal R charges. Regard $\mathcal N=4$ SYM as one vector multiplet and three adjoint chiral multiplets. Derive $a$ and $c$ and distinguish $SU(N)$ from $U(N)$.

Solution

The gaugino has R charge $+1$. Each chiral scalar has superconformal R charge $2/3$, so its Weyl fermion has charge $-1/3$. Writing $D=\dim G$,

$$\begin{aligned} \operatorname{Tr}R_{\!f} &=D\left[1+3\left(-\frac13\right)\right]=0,\\ \operatorname{Tr}R_{\!f}^3 &=D\left[1+3\left(-\frac13\right)^3\right] =\frac89D. \end{aligned}$$

The supplied anomaly relations give

$$a=c=\frac D4.$$

Hence $SU(N)$ gives $(N^2-1)/4$, whereas $U(N)$ gives $N^2/4$. The difference is precisely the anomaly of one free $\mathcal N=4$ vector multiplet.

The protection argument uses the superconformal R current and its fermionic anomaly coefficients. In this theory the charges are fixed by the supersymmetry and superconformal structure; their protected anomalies determine $a$ and $c$ throughout the conformal family. This does not license replacing an unknown interacting superconformal R symmetry by arbitrary ultraviolet free-field charges in another $\mathcal N=1$ theory.

6. Replace the internal sphere

Consider an unwarped $AdS_5\times X_5$ solution with constant dilaton and only self-dual five-form flux. Normalize the dimensionless internal metric by $\mathcal R_{ab}[X_5]=4g_{ab}[X_5]$ and let $v=\operatorname{Vol}(X_5)$. The physical internal metric is $R^2ds_{X_5}^2$.

With the shared orientation and positive D3 charge, the calibrated internal flux is $F_5|_{X_5}=-4R^4\operatorname{vol}_{X_5}/g_s$. Use the corresponding oriented flux quantization to derive the leading central charges. Here $B_2=C_2=0$, so $F_5$ equals the shared modified field strength $\widetilde F_5$. Then insert $v(T^{1,1})=16\pi^3/27$.

Solution

Flux quantization gives

$$N=-\frac1{(2\pi)^4\alpha'^2}\int_{X_5}F_5 =\frac{4R^4v}{g_s(2\pi)^4\alpha'^2}.$$

Therefore

$$R^4=\frac{4\pi^4g_sN\alpha'^2}{v}.$$

The physical internal volume is $R^5v$, so $G_5=G_{10}/(R^5v)$. Substituting into the Einstein result gives

$$\begin{aligned} a=c &=\frac{\pi R^8v}{8G_{10}}\\ &=\frac{\pi R^8v}{64\pi^6g_s^2\alpha'^4} =\frac{\pi^3N^2}{4v}. \end{aligned}$$

At fixed flux, the radius also changes when the dimensionless volume changes. Accounting for both effects is why the final answer is inversely proportional to $v$.

For $S^5$, $v=\pi^3$ recovers $N^2/4$. For the supplied $T^{1,1}$ volume,

$$a=c=\frac{27N^2}{64}, \qquad \frac{a_{T^{1,1}}}{a_{S^5}}=\frac{27}{16}.$$

This inverse-volume law was analyzed by Gubser. The geometry and associated field theory are developed on page 35.

The calculation assumes the stated product solution and leading two-derivative action. It does not establish that every positive Einstein five-manifold defines a stable string background or has an identified CFT dual.

References

• Anselmi, Damiano, Daniel Z. Freedman, Marcus T. Grisaru, and Andreas A. Johansen. “Nonperturbative Formulas for Central Functions of Supersymmetric Gauge Theories.” Nuclear Physics B 526 (1998), 543–571. arXiv:hep-th/9708042. doi:10.1016/S0550-3213(98)00278-8.
• de Haro, Sebastian, Sergey N. Solodukhin, and Kostas Skenderis. “Holographic Reconstruction of Spacetime and Renormalization in the AdS/CFT Correspondence.” Communications in Mathematical Physics 217 (2001), 595–622. arXiv:hep-th/0002230. doi:10.1007/s002200100381.
• Gubser, Steven S. “Einstein Manifolds and Conformal Field Theories.” Physical Review D 59 (1999), 025006. arXiv:hep-th/9807164. doi:10.1103/PhysRevD.59.025006.
• Henningson, Måns, and Kostas Skenderis. “The Holographic Weyl Anomaly.” Journal of High Energy Physics 07 (1998), 023. arXiv:hep-th/9806087. doi:10.1088/1126-6708/1998/07/023.
• Osborn, Hugh, and Anastasios C. Petkou. “Implications of Conformal Invariance in Field Theories for General Dimensions.” Annals of Physics 231 (1994), 311–362. arXiv:hep-th/9307010. doi:10.1006/aphy.1994.1045.