D3-Branes and AdS5/CFT4: Parameter-Controlled Regimes and Evidence

The D3-brane construction proposes an equivalence between two complete quantum theories: the interacting four-dimensional $\mathcal N=4$ super-Yang–Mills sector and type-IIB string theory on AdS$_5\times S^5$ with $N$ units of five-form flux. Classical supergravity is only one approximation to the string side. At fixed $\lambda$, large $N$ suppresses higher string genera, while large ’t Hooft coupling $\lambda$ suppresses string-scale curvature corrections; neither condition implies the other. This page makes those controls explicit and then asks, observable by observable, what the evidence actually establishes Maldacena 1998, § 2, pp. 2–7, PDF, Aharony et al. 2000, §§ 3.1–3.3, pp. 55–90, PDF.

Required background. The D3 low-energy limit supplies the decoupling argument. Five-form flux on the five-sphere supplies the relation among $N$, the AdS radius, and the compact Kaluza–Klein spectrum.

Helpful background. $\mathcal N=4$ superconformal data fixes the boundary normalization. Local protected operators explains why some quantities can be transported across coupling. Independent evidence and its ceiling supplies the comparison standard used below.

For a quick route through the article, first separate the two control questions, then compare the evidence classes, stress-test them in the intermediate-coupling limit, and finish with the solved exercises.

The D3-brane pair and its defining data

A stack of $N$ coincident D3-branes first produces a $U(N)$ gauge theory. In the near-horizon decoupling limit, its center-of-mass $U(1)$ multiplet is free and decouples; the nontrivial local dynamics has Lie algebra $\mathfrak{su}(N)$. We therefore use $SU(N)$ as the conventional representative for the local observables calculated below, while remembering that local correlators alone do not choose among global forms such as $SU(N)$ and $PSU(N)$. That choice changes the spectrum of line operators and the admissible bulk boundary conditions; see Aharony, Seiberg, and Tachikawa 2013, § 1, pp. 1–3, PDF.

Defining data of the D3-brane dual pair

Item	Boundary description	Bulk description
Dynamical theory	Interacting four-dimensional maximally supersymmetric Yang–Mills theory with local algebra su(N)	Type-IIB string theory on AdS5 × S5
Discrete datum	Rank parameter N	N units of self-dual five-form flux through S5
Continuous datum	Complex exactly marginal coupling τYM	Axiodilaton C0 + ie−φ
Symmetry	The superconformal group PSU(2,2|4), including SO(4,2) conformal symmetry and SO(6) R-symmetry
Representative observables	Local correlators, anomalies, thermal observables, Wilson and ’t Hooft lines, and partition functions	Boundary amplitudes, on-shell actions, string worldsheets and branes, black-brane thermodynamics, and quantum bulk corrections

The precise conjectural source relation is the GKPW generating-functional relation: an allowed boundary value of a bulk field is a source for a boundary operator. Saying “a bulk field is a single-trace operator” is useful large-$N$ shorthand for single-particle states in the bulk expansion, not an exact finite-$N$ identity. Multi-trace mixing, finite-$N$ trace relations, gauge constraints, and the choice of boundary condition all qualify that slogan Gubser, Klebanov, and Polyakov 1998, § 2.2, pp. 6–9, eqs. (19)–(32), PDF.

Two independent control questions

We use the site’s (+---) metric convention and normalize the fundamental generators by

$$\operatorname{Tr}_{\mathrm{fund}}(T^aT^b)=\frac12\delta^{ab}, \qquad S_{\mathrm{YM}}\supset -\frac{1}{2g_{\mathrm{YM}}^2} \int d^4x\,\operatorname{Tr}_{\mathrm{fund}}F_{\mu\nu}F^{\mu\nu}.$$

In this convention,

$$\tau_{\mathrm{YM}} =\frac{\theta}{2\pi}+\frac{4\pi i}{g_{\mathrm{YM}}^2} =C_0+\frac{i}{g_s}, \qquad g_{\mathrm{YM}}^2=4\pi g_s, \qquad \lambda=g_{\mathrm{YM}}^2N.$$

Some foundational papers instead write $g_{\mathrm{YM,old}}^2=2\pi g_s$; their combination $2g_{\mathrm{YM,old}}^2N$ is the $\lambda$ used here. Stating this conversion prevents a factor-of-two error in string corrections.

Flux quantization gives

$$L^4=4\pi g_sN\alpha'^2=\lambda\alpha'^2, \qquad \alpha'=\ell_s^2.$$

Moreover, $G_{10}=8\pi^6g_s^2\alpha'^4$ and $\operatorname{Vol}(S_L^5)=\pi^3L^5$, so dimensional reduction yields

$$G_5=\frac{G_{10}}{\operatorname{Vol}(S_L^5)} =\frac{\pi L^3}{2N^2}.$$

These equations separate the relevant questions Maldacena 1998, § 2, pp. 2–4, eqs. (2.1)–(2.5), PDF:

$$\underbrace{\frac{\alpha'}{L^2}=\lambda^{-1/2}}_{\text{string-scale curvature}}, \qquad \underbrace{g_s=\frac{\lambda}{4\pi N}}_{\text{ten-dimensional string genus}}, \qquad \underbrace{\frac{G_5}{L^3}=\frac{\pi}{2N^2}}_{\text{five-dimensional bulk loops}}.$$

Thus $N\gg1$ suppresses five-dimensional quantum loops, whereas $\lambda\gg1$ makes the background weakly curved in string units. Keeping the displayed type-IIB frame weakly coupled also requires $\lambda\ll4\pi N$. A simple sufficient parametric window for tree-level, two-derivative type-IIB supergravity is therefore

$$1\ll\lambda\ll N,$$

with constants suppressed and with additional observable-specific requirements on energy, state, and truncation. The first maximally supersymmetric local higher-derivative interaction is schematically $\alpha'^3R^4$, whose relative size on this background is $\lambda^{-3/2}$; that does not mean every observable has a nonzero correction at that order.

One further scale matters. A string excitation has

$$m_sL\sim\frac{L}{\sqrt{\alpha'}}=\lambda^{1/4},$$

but the $S^5$ Kaluza–Klein tower has $m_{\mathrm{KK}}L=O(1)$. Even at large $\lambda$, there is no parametric gap that removes all Kaluza–Klein modes while retaining the full protected spectrum. The more complete derivation and its caveats live on the radius–coupling parameter map.

The expansion vocabulary is topological as well as numerical. Planar means leading order in the boundary $1/N$ expansion; large-$N$ factorization means that connected products of suitably normalized gauge-invariant operators are suppressed relative to disconnected products. On the string side, the genus counts worldsheet handles, so genus zero is the leading closed-string topology. Suppressing higher genera does not make the genus-zero worldsheet theory on an RR-supported AdS$_5\times S^5$ background generally solvable.

Regimes are parametric and observable-dependent, not sharp phase boundaries

Regime	What remains parametrically controlled	What is not licensed
Small λ	Boundary perturbation theory; with large N, its planar reorganization	A weakly curved bulk derivative expansion
Large N, λ of order one	Large-N factorization and suppression of higher string genera, conditional on the dictionary	Two-derivative supergravity, decoupling of the string tower, or a generally tractable full-string calculation
1 ≪ λ ≪ N	Weakly coupled, weakly curved type-IIB supergravity for suitable low-energy observables	A universal five-dimensional truncation or exact finite-N result
λ comparable to or larger than N	Curvature may remain small; a duality-related frame may reorganize some questions	Weak coupling in the displayed type-IIB frame merely from large λ

Protected benchmark: the Weyl anomaly

For a four-dimensional theory on a curved background, define the anomaly convention by

$$\langle T^\mu{}_{\mu}\rangle =\frac{c}{16\pi^2}W_{\mu\nu\rho\sigma}W^{\mu\nu\rho\sigma} -\frac{a}{16\pi^2}E_4+b\,\Box R.$$

The coefficient $b$ can be shifted by a local counterterm, whereas $a$ and $c$ are physical. In the interacting $\mathfrak{su}(N)$ theory, a protected field-theory calculation gives, for every value of the exactly marginal coupling Anselmi et al. 1998, §§ 2–4, PDF,

$$a_{\mathrm{exact}}=c_{\mathrm{exact}}=\frac{N^2-1}{4}.$$

The tree-level five-dimensional gravitational action instead gives

$$a_{\mathrm{tree}}=c_{\mathrm{tree}} =\frac{\pi L^3}{8G_5} =\frac{N^2}{4}.$$

This is an exact-in-$\lambda$ boundary datum compared with the leading large-$N$, large-$\lambda$ bulk result Henningson and Skenderis 1998, § 3.2, p. 7, eqs. (20)–(23), PDF. The absolute difference is $1/4$ and the fractional difference, measured relative to the exact answer, is $1/(N^2-1)$. A one-loop sum over the complete type-IIB Kaluza–Klein spectrum supplies the missing $-1/4$ Mansfield, Nolland, and Ueno 2003, pp. 1–4, eqs. (1)–(3) and (9)–(10), PDF. The massless five-dimensional truncation alone cannot see that correction.

Logical status. This is a stringent normalization and spectrum check, strengthened by the finite-$N$ correction. Because the quantity is protected, it does not by itself test generic strongly coupled dynamics.

Normalization benchmark: the stress-tensor two-point function

At separated Euclidean points, use the Osborn–Petkou convention

$$I_{\mu\nu}(x)=\delta_{\mu\nu}-2\frac{x_\mu x_\nu}{x^2},$$ $$\mathcal I_{\mu\nu,\rho\sigma}(x) =\frac12\left(I_{\mu\rho}I_{\nu\sigma}+I_{\mu\sigma}I_{\nu\rho}\right) -\frac14\delta_{\mu\nu}\delta_{\rho\sigma},$$ $$\langle T_{\mu\nu}(x)T_{\rho\sigma}(0)\rangle =\frac{C_T}{(x^2)^4}\,\mathcal I_{\mu\nu,\rho\sigma}(x), \qquad C_T=\frac{40c}{\pi^4}.$$

Local contact terms are scheme-dependent; the separated-point coefficient $C_T$ is not. These conventions and the relation to $c$ follow Osborn and Petkou 1994, § 2, p. 7, eqs. (2.22)–(2.24), and § 8, pp. 31–32, eq. (8.12), PDF. Substituting the exact and tree-level values of $c$ gives

$$C_T^{SU(N)}=\frac{10(N^2-1)}{\pi^4}, \qquad C_T^{\mathrm{tree}}=\frac{5L^3}{\pi^3G_5} =\frac{10N^2}{\pi^4}.$$

The bulk equality checks that differentiating the gravitational generating functional with respect to the boundary metric has the correct source and kinetic normalization Gubser, Klebanov, and Polyakov 1998, § 2.2, pp. 6–9, eqs. (19)–(32), PDF.

Logical status. The check is important but not independent of the anomaly match: supersymmetry relates both to the same protected $c$, and the tree-level bulk calculations use the same $L^3/G_5$ normalization. Counting them as two independent votes would double-count shared premises. The bulk metric and boundary stress tensor page develops this dictionary in detail.

Unprotected benchmark: thermal entropy

Thermodynamics provides a simple quantity that is not protected. For free $\mathcal N=4$ $SU(N)$ super-Yang–Mills on $\mathbb R^3$ at temperature $T$, the eight bosonic and eight fermionic on-shell degrees of freedom per generator give

$$s_{\mathrm{free}} =\frac{2\pi^2}{3}(N^2-1)T^3, \qquad s_{\mathrm{SB}}^{\mathrm{planar}} =\frac{2\pi^2}{3}N^2T^3.$$

The near-extremal D3 black brane predicts at planar strong coupling

$$s_{\mathrm{strong}} =\frac{\pi^2}{2}N^2T^3 \left[ 1+\frac{15}{8}\zeta(3)\lambda^{-3/2} +O\!\left(\lambda^{-5/2}\right) \right] +O(N^0),$$

or, normalized to the planar free value,

$$\frac{s_{\mathrm{strong}}}{s_{\mathrm{SB}}^{\mathrm{planar}}} =\frac34 +\frac{45}{32}\zeta(3)\lambda^{-3/2} +O\!\left(\lambda^{-5/2}\right) +O(N^{-2}).$$

The $\lambda^{-3/2}$ coefficient comes from the type-IIB $\alpha'^3R^4$ interaction Gubser, Klebanov, and Tseytlin 1998, § 1, pp. 2–3, eqs. (4)–(8), § 2, p. 7, eqs. (24)–(28), and § 3, pp. 11–12, eqs. (45)–(46), PDF. The calculation assumes the planar limit first, $\lambda\gg1$, a weakly coupled displayed IIB frame, and the flat-space thermodynamic limit. Its uncertainty is asymptotic: higher planar string corrections and $1/N^2$ quantum corrections have been omitted, and no numerical remainder bound follows from writing the first neglected order.

Logical status. This is a controlled holographic prediction for an unprotected strong-coupling observable and a test of the string effective-action dictionary. It is not, by itself, an independent gauge-theory verification at the same large value of $\lambda$. In particular, $3/4$ compares the two coupling endpoints; it is not a discrepancy between two calculations performed at one coupling. The black-brane thermodynamics page derives the gravitational side.

Unprotected dynamical benchmark: a four-point coefficient

Consider the connected four-point function of the dimension-two scalar $\mathcal O_2(x,Y)$ in the stress-tensor multiplet, normalized by

$$\langle\mathcal O_2(x_1,Y_1)\mathcal O_2(x_2,Y_2)\rangle =\frac{(Y_1\mathbin{\cdot}Y_2)^2}{x_{12}^4}.$$

The null polarization $Y$, with $Y\mathbin{\cdot}Y=0$, packages the $SO(6)$ indices. The external operator is protected, but its full four-point function and the long-multiplet OPE data are not. After stripping the known free and kinematic pieces, let $\mathcal M_2$ denote the Mellin transform of the reduced dynamical function $\mathcal T_2$. In the convention of Binder, Chester, Pufu, and Wang, with $c=(N^2-1)/4$ and $u=4-s-t$,

$$\mathcal M_2(s,t) =\frac{8}{c} \left[ \frac{1}{(s-2)(t-2)(u-2)} +\frac{15\zeta(3)}{\lambda^{3/2}} +O\!\left(\lambda^{-5/2}\right) \right] +O(c^{-2}).$$

The first term is tree-level supergravity; the constant term is the first stringy contact interaction. An integrated $S^4$ localization constraint fixes the latter coefficient and agrees with the coefficient obtained from the flat-space type-IIB string amplitude Binder et al. 2019, § 4, especially eq. (4.2), pp. 23–27, PDF. These are meaningfully different calculation routes, although they still share supersymmetry, the operator normalization, and the holographic dictionary. The result is a nontrivial planar strong-coupling asymptotic check, not a proof at all $N$ and $\lambda$. The string and quantum correction hierarchy explains the omitted orders.

Comparing the evidence without double-counting

The useful unit of evidence is not a celebrated equation but a comparison record: observable, convention, domain, controls, shared premises, and omitted error. The comparison records above therefore have different strengths.

Evidence classes, control domains, and claim ceilings

Class and observable	Domain and controls	What it checks	Uncertainty and logical ceiling
Protected anomaly: a and c	Exact in boundary coupling; tree bulk result at large N and large λ, plus a full Kaluza–Klein one-loop correction	Flux/radius normalization, five-dimensional Newton constant, protected spectrum, and the finite-N shift	Tree-level absolute remainder 1/4 and relative remainder 1/(N2−1); protection limits its reach into generic dynamics
Protected two-point normalization: CT	Separated Euclidean points; exact boundary value versus tree supergravity in its controlled regime	GKPW source differentiation and the graviton kinetic normalization	Relative remainder of order 1/N2; strongly correlated with the anomaly row rather than independent evidence
Unprotected thermodynamics: entropy density	Planar limit first, then large λ; thermal state on flat space; higher derivatives ordered asymptotically	Black-brane dictionary and the first string effective-action correction	Higher powers of 1/√λ and 1/N2 omitted; prediction rather than same-coupling gauge-theory confirmation
Unprotected dynamics: reduced four-point Mellin coefficient	Large N, large λ, stated Mellin normalization, and protected external operators	Agreement of a localization-integrated constraint with the flat-space string coefficient	Shared supersymmetric and dictionary premises remain; no all-coupling or finite-N conclusion

The evidence is cumulative but not interchangeable. Protected results can be exact while dynamically narrow. Unprotected results probe more of the interaction but usually only in an asymptotic corner. For a general framework for separating shared inputs from independent tests, see evidence, circularity, and double-counting.

Adversarial limit: large N at intermediate coupling

Now hold $\lambda=\lambda_0=O(1)$ while taking $N\to\infty$. Each control parameter has a definite fate:

$$g_s=\frac{\lambda_0}{4\pi N}\longrightarrow0, \qquad \frac{G_5}{L^3}=\frac{\pi}{2N^2}\longrightarrow0,$$

but

$$\frac{\alpha'}{L^2}=\lambda_0^{-1/2}=O(1), \qquad m_sL=\lambda_0^{1/4}=O(1).$$

String topology is controlled: nonplanar boundary corrections and successive closed-string genera are suppressed. The low-energy curvature or $\alpha'$ expansion is not controlled: the entire tower of string excitations remains at the AdS scale, so infinitely many higher-derivative terms can compete.

The exact downgrade is therefore from two-derivative supergravity to a genus-zero full-string problem, conditional on the duality. Large-$N$ factorization, protected data, and suppression of higher genera survive; locality at scales parametrically larger than the string length and an Einstein-gravity truncation do not. This identifies the conjecturally appropriate layer, not a generally tractable worldsheet calculation on AdS$_5\times S^5$. It would be too strong to say that “tree-level string theory fails”: what fails is its low-energy supergravity expansion. Conversely, $\lambda\to\infty$ at fixed small $N$ weakens curvature but does not control quantum gravity.

This adversarial limit is why a single shaded “classical” region would be misleading: the relevant boundaries depend on the observable, the energy scale, and the duality frame. The semantic regime table records those conditions more faithfully than a sharp phase diagram.

What the evidence establishes

Within its controlled domain, the D3 system supplies:

• matching local superconformal symmetry and a convention-complete parameter identification;
• protected normalization checks, including a nontrivial one-loop finite-$N$ correction;
• controlled predictions for unprotected thermal and correlation observables;
• mutually constraining calculations from supergravity, string amplitudes, localization, and field-theory structure.

It does not supply a theorem that every finite-$N$, finite-$\lambda$ observable agrees. Nor does a successful supergravity calculation license a claim about the full parameter space. The strongest statement remains a broad quantum equivalence conjecture supported by many compatible, partly independent tests; every concrete result inherits the ceiling of its own approximation.

Higher-dimensional M2- and M5-brane examples show which parts of this reasoning survive when the boundary theory has no conventional weakly coupled Lagrangian.

Common pitfalls

Equating the duality with supergravity. AdS/CFT relates full quantum theories. Supergravity is the low-curvature, weak-loop corner of one side.

Treating large $N$ as large $\lambda$. The two parameters control different expansions. The adversarial limit above is large $N$ and still maximally stringy at the AdS scale.

Calling a four-point function protected because its external operators are protected. Protected dimensions and normalizations do not freeze the full connected correlator or long-multiplet OPE data.

Counting anomaly and $C_T$ as independent votes. They are valuable cross-checks of different bulk variations, but both use the same protected coefficient and the same gravitational normalization.

Forgetting the decoupled $U(1)$ or the global form. The D3 stack begins with $U(N)$; the interacting local sector removes the free center. Local $\mathfrak{su}(N)$ correlators do not determine the line-operator spectrum.

Universalizing the $\lambda^{-3/2}$ correction. It is the first nonzero correction for the specified thermal and four-point observables. Symmetry can remove it from other quantities.

Exercises

1. Locate a correlated supergravity limit

Let $\lambda=N^p$ as $N\to\infty$. For which $p$ are both string-scale curvature and the displayed-frame string coupling parametrically small?

Solution

Curvature control requires

$$\frac{\alpha'}{L^2}=N^{-p/2}\longrightarrow0,$$

so $p>0$. Weak string coupling requires

$$g_s=\frac{N^{p-1}}{4\pi}\longrightarrow0,$$

so $p<1$. Therefore $0<p<1$. This is a sufficient displayed-frame window; duality-related frames can reorganize other correlated limits.

2. State the anomaly error budget

Compare $c_{\mathrm{tree}}=N^2/4$ with $c_{\mathrm{exact}}=(N^2-1)/4$. Find the absolute and fractional differences and state the conclusion licensed by tree-level gravity.

Solution

The absolute difference is

$$c_{\mathrm{tree}}-c_{\mathrm{exact}}=\frac14.$$

Relative to the exact answer, it is

$$\frac{c_{\mathrm{tree}}-c_{\mathrm{exact}}}{c_{\mathrm{exact}}} =\frac{1}{N^2-1}=O(N^{-2}).$$

Tree-level gravity reproduces the leading large-$N$ coefficient, not the exact finite-$N$ answer. The full Kaluza–Klein one-loop calculation is needed for the displayed subleading shift.

3. Normalize the stress-tensor correlator

Use $C_T=40c/\pi^4$ to compute the exact $SU(N)$ and tree-level bulk coefficients. Why are this result and the anomaly match correlated evidence?

Solution

Substitution gives

$$C_T^{SU(N)}=\frac{40}{\pi^4}\frac{N^2-1}{4} =\frac{10(N^2-1)}{\pi^4},$$

whereas

$$C_T^{\mathrm{tree}}=\frac{40}{\pi^4}\frac{N^2}{4} =\frac{10N^2}{\pi^4}.$$

Both checks use the same protected $c$ and the same bulk factor $L^3/G_5$. The two-point calculation still verifies the source and graviton normalization, but it is not logically independent of the anomaly comparison.

4. Read an asymptotic error estimate honestly

Using only the first thermal correction, estimate when it is less than ten percent of the leading strong-coupling entropy. Does that estimate prove ten-percent accuracy?

Solution

Relative to the leading $3/4$ term, the displayed correction is

$$\frac{15}{8}\zeta(3)\lambda^{-3/2}.$$

Requiring it to be below $0.1$ gives

$$\lambda> \left(\frac{15\zeta(3)}{0.8}\right)^{2/3} \approx7.98.$$

This is only a scale estimate. An asymptotic series with no bound on the next terms does not prove that the full error is below ten percent, and finite-$N$ corrections must also be controlled.

5. Execute the adversarial limit

At $N\to\infty$ and $\lambda=1$, evaluate $g_s$, $G_5/L^3$, and $m_sL$. What is the strongest bulk approximation that can survive?

Solution

One finds

$$g_s=\frac{1}{4\pi N}\to0, \qquad \frac{G_5}{L^3}=\frac{\pi}{2N^2}\to0, \qquad m_sL=1.$$

String genus and bulk-loop effects are suppressed, but the string tower does not decouple. The surviving candidate is a genus-zero full-string problem on the curved background, conditional on the duality—not two-derivative supergravity and not, in general, an explicitly tractable worldsheet calculation.

The chapter overview contains the structure diagram and validity and failure diagram. They are embedded there once so that their shared chapter-level context is not repeated on every article.

For the chapter-wide comparison of assumptions, counterevidence, falsifiers, and claim ceilings, see the claim-domain table.

References

• Aharony, O., Gubser, S. S., Maldacena, J. M., Ooguri, H., and Oz, Y. (2000), “Large $N$ Field Theories, String Theory and Gravity,” Physics Reports 323, 183–386. DOI. Open PDF.
• Aharony, O., Seiberg, N., and Tachikawa, Y. (2013), “Reading between the Lines of Four-Dimensional Gauge Theories,” Journal of High Energy Physics 2013(8), 115. DOI. Open PDF.
• Anselmi, D., Freedman, D. Z., Grisaru, M. T., and Johansen, A. A. (1998), “Nonperturbative Formulas for Central Functions of Supersymmetric Gauge Theories,” Nuclear Physics B 526, 543–571. DOI. Open PDF.
• Binder, D. J., Chester, S. M., Pufu, S. S., and Wang, Y. (2019), “N = 4 Super-Yang–Mills Correlators at Strong Coupling from String Theory and Localization,” Journal of High Energy Physics 2019(12), 119. DOI. Open PDF.
• Gubser, S. S., Klebanov, I. R., and Polyakov, A. M. (1998), “Gauge Theory Correlators from Non-Critical String Theory,” Physics Letters B 428, 105–114. DOI. Open PDF.
• Gubser, S. S., Klebanov, I. R., and Tseytlin, A. A. (1998), “Coupling Constant Dependence in the Thermodynamics of $\mathcal N=4$ Supersymmetric Yang–Mills Theory,” Nuclear Physics B 534, 202–222. DOI. Open PDF.
• Henningson, M., and Skenderis, K. (1998), “The Holographic Weyl Anomaly,” Journal of High Energy Physics 1998(7), 023. DOI. Open PDF.
• Maldacena, J. M. (1998), “The Large $N$ Limit of Superconformal Field Theories and Supergravity,” Advances in Theoretical and Mathematical Physics 2, 231–252. DOI. Open PDF.
• Mansfield, P., Nolland, D., and Ueno, T. (2003), “Order $1/N^2$ Test of the Maldacena Conjecture II: The Full Bulk One-Loop Contribution to the Boundary Weyl Anomaly,” Physics Letters B 565, 207–210. DOI. Open PDF.
• Osborn, H., and Petkou, A. C. (1994), “Implications of Conformal Invariance in Field Theories for General Dimensions,” Annals of Physics 231, 311–362. DOI. Open PDF.