Skip to content
QFT .org

Holographic Duality: Claims, Dictionaries, and Regimes

A holographic duality is neither a resemblance between two Lagrangians nor a formula evaluated in one convenient limit. The strongest claim is an equivalence between specified physical sectors of two quantum theories: their states, observable algebras, dynamics, symmetries, global data, and parameters must correspond. Most calculations establish a smaller result—a dictionary entry, a protected equality, a term in an expansion, or a saddle approximation—and should be described at that strength.

Helpful background. Duality Claims, Dictionaries, Regimes, and Evidence supplies the general QFT vocabulary for duality. Claim–Evidence Records, Replication, and Retraction Handling explains why several consequences of one assumed dictionary are not automatically independent tests.

A holographic claim has several maps

Section titled “A holographic claim has several maps”

Let TB\mathcal T_{\mathrm B} be the boundary theory and Tbulk\mathcal T_{\mathrm{bulk}} the bulk theory. Even before asking whether they are equivalent, specify their spacetime and asymptotic structures, gauge-group global forms, boundary conditions, coupling constants, superselection sectors, and allowed defects. Let AB\mathfrak A_{\mathrm B} and Abulk\mathfrak A_{\mathrm{bulk}} denote the physical observable algebras in the sectors being compared, and let SB\mathfrak S_{\mathrm B} and Sbulk\mathfrak S_{\mathrm{bulk}} denote the corresponding state families.

A dictionary contains at least an observable map and a state map,

α:Abulk⟶AB,σ:SB⟶Sbulk,\alpha:\mathfrak A_{\mathrm{bulk}}\longrightarrow \mathfrak A_{\mathrm B}, \qquad \sigma:\mathfrak S_{\mathrm B}\longrightarrow \mathfrak S_{\mathrm{bulk}},

such that matched expectation values obey

ωB ⁣(α(A))=σ(ωB)(A)\omega_{\mathrm B}\!\left(\alpha(A)\right) =\sigma(\omega_{\mathrm B})(A)

for every declared state ωB\omega_{\mathrm B} and observable AA in the domain. Products, adjoints, symmetry actions, and time evolution must also be preserved. A full exact equivalence of the declared sectors requires α\alpha to be an invertible ∗*-isomorphism that intertwines dynamics and symmetries, with σ(ωB)=ωB∘α\sigma(\omega_{\mathrm B})=\omega_{\mathrm B}\circ\alpha an induced bijection of states. An injective map, an exact equivalence of a protected subalgebra, or a collection of matched expectation values is a restricted dictionary—not by itself a duality of the complete specified theories.

When both sides admit the relevant Hilbert-space description, the same equivalence can be represented by a unitary isomorphism intertwining the represented algebras, dynamics, symmetries, and declared state sectors Aharony et al. 2000, §3.4. The algebraic wording is more general and avoids pretending that a fixed density matrix is itself a theory map.

Four logically separate questions then arise:

  • Well-definedness: does every input in the declared domain have an image?
  • Injectivity: can two distinct bulk objects map to the same boundary object?
  • Surjectivity: is every boundary object in the claimed sector represented in the bulk description?
  • Reconstruction: can the inverse map be constructed, exactly or with a controlled error?

A successful check of one entry does not answer all four. Subregion reconstruction, finite-NN completeness, and the global spectrum can therefore carry different claim strengths even when their notation looks uniform.

Claim classes and what they license

Section titled “Claim classes and what they license”
Claim classes, required data, and the conclusions they license
Claim class Required content Licensed conclusion
Dictionary entry Specified objects, normalization, domain, and direction of the map Those objects are proposed or shown to correspond
Protected match The protecting symmetry or index and the parameters it excludes Equality in that protected sector, not completeness of the theory
Perturbative equality Expansion parameter, order, scheme, and remainder estimate Agreement through the stated order
Large-$N$ saddle relation Saddle, contour, boundary data, and $1/N$ and string-scale corrections A semiclassical result in that saddle’s domain
Numerical test Observable, discretization, extrapolation, and statistical and systematic errors Agreement within the reported uncertainty
Conjectural equivalence Complete theory pair and claimed state, operator, and sector coverage A precise conjecture that can organize tests
Definition proposal Which otherwise-unconstructed quantities are defined by the other description A proposed nonperturbative definition, conditional on consistency and completeness

These classes are not a ladder that every result climbs. A protected anomaly can be exact while probing a narrow subsector; a numerical result can probe unprotected dynamics while remaining finite-precision. “Exact,” “broad,” and “independently supported” are different attributes.

GKPW: a conjectured exact identity and a saddle calculation

Section titled “GKPW: a conjectured exact identity and a saddle calculation”

In Euclidean signature, the compact GKPW statement is

ZCFT[M,J;P]=conjectureZbulk[∂confXd+1=M, ϕ∼J;P],ϕ(z,x)=zd−ΔJ(x)+⋯(standard scalar quantization).\begin{aligned} Z_{\mathrm{CFT}}[M,J;\mathcal P] &\overset{\text{conjecture}}{=} Z_{\mathrm{bulk}}[\partial_{\mathrm{conf}}X_{d+1}=M, \,\phi\sim J;\mathcal P],\\ \phi(z,x)&=z^{d-\Delta}J(x)+\cdots \quad\text{(standard scalar quantization)}. \end{aligned}

Here JJ couples to a normalized operator in a dd-dimensional boundary theory, MM fixes the conformal boundary of the (d+1)(d+1)-dimensional bulk geometry, and P\mathcal P collects the state preparation, other boundary data, integration contour, and renormalization prescription. The shorthand ϕ∼J\phi\sim J means the specified asymptotic falloff, illustrated by the second line; logarithmic or resonant terms and alternate quantization require their own boundary data. Local counterterms and source normalizations are part of the equality because they control contact terms. Witten states the generating-functional prescription and its classical evaluation in Witten 1998, §2.3, eqs. (2.10)–(2.13), printed pp. 9–10; the parallel Gubser–Klebanov–Polyakov construction appears in Gubser, Klebanov, and Polyakov 1998, §2, eq. (12), printed p. 4, and eq. (23), printed pp. 6–7.

At finite NN and finite coupling, if the exact bulk partition function on the right has not been independently constructed, this equality is both a strong-form conjecture and a possible definition proposal. It is not an already-established identity merely because its semiclassical expansion is computable.

The bulk path integral is schematically a sum over admissible saddles,

Zbulk[J]∼∑se−SE,ren[Φs;J]Z1-loop(s)[J] (1+⋯ ).Z_{\mathrm{bulk}}[J] \sim \sum_s e^{-S_{E,\mathrm{ren}}[\Phi_s;J]} Z_{\text{1-loop}}^{(s)}[J]\,(1+\cdots).

The sum over admissible geometries and topologies is itself part of the specification; its role is explicit in Witten 1998, §3.1, eqs. (3.1)–(3.3), printed pp. 28–30.

Only after choosing a dominant saddle and a controlled low-energy truncation may one write

WE[J]≡log⁡ZCFT[J]≃−SE,ren[Φcl;J].W_E[J]\equiv\log Z_{\mathrm{CFT}}[J] \simeq -S_{E,\mathrm{ren}}[\Phi_{\mathrm{cl}};J].

Thus the first equation is a conjectured exact identity or definition proposal for complete generating functionals; the last is a saddle approximation. The approximation can fail because another saddle competes, a string or loop correction is not small, the boundary-value problem changes, or the chosen state and contour do not match. Mixed scalar boundary conditions, for example, can implement multi-trace deformations rather than a different state of the same undeformed CFT Witten 2001, §3, eqs. (3.6)–(3.9), printed pp. 5–6, and §4, printed pp. 9–10. The dedicated exact-statements and saddle-expansions page develops this distinction further.

The AdS₅/CFT₄ parameter map

Section titled “The AdS₅/CFT₄ parameter map”

Consider the canonical local-sector correspondence: four-dimensional SU(N)SU(N) N=4\mathcal N=4 super-Yang–Mills theory and type-IIB string theory on AdS5×S5AdS_5\times S^5 with NN units of five-form flux Maldacena 1998, §§2–3. Fix the Euclidean gauge-action and generator conventions by

SYM(E)⊃12gYM2∫d4x tr⁡FμνFμν,tr⁡(TaTb)=12δab.S_{\mathrm{YM}}^{(E)}\supset \frac{1}{2g_{\mathrm{YM}}^2} \int d^4x\,\operatorname{tr} F_{\mu\nu}F^{\mu\nu}, \qquad \operatorname{tr}(T^aT^b)=\frac12\delta^{ab}.

Then the parameter map is

gYM2=4πgs,λ≡gYM2N=L4α′2,gs=λ4πN.g_{\mathrm{YM}}^2=4\pi g_s, \qquad \lambda\equiv g_{\mathrm{YM}}^2N =\frac{L^4}{\alpha'^2}, \qquad g_s=\frac{\lambda}{4\pi N}.

The complex couplings obey

τYM≡θYM2π+4πigYM2=C0+igs.\tau_{\mathrm{YM}} \equiv\frac{\theta_{\mathrm{YM}}}{2\pi} +\frac{4\pi i}{g_{\mathrm{YM}}^2} =C_0+\frac{i}{g_s}.

With the five-dimensional Einstein term normalized as S5=(16πG5)−1∫d5xg R+⋯S_5=(16\pi G_5)^{-1}\int d^5x\sqrt g\,R+\cdots, use G10=8π6gs2α′4G_{10}=8\pi^6g_s^2\alpha'^4 and Vol⁡(S5)=π3L5\operatorname{Vol}(S^5)=\pi^3L^5. Classical reduction then gives

G5=G10π3L5,G5L3=π2N2.G_5=\frac{G_{10}}{\pi^3L^5}, \qquad \frac{G_5}{L^3}=\frac{\pi}{2N^2}.

Thus Einstein gravity predicts a=c=πL3/(8G5)=N2/4a=c=\pi L^3/(8G_5)=N^2/4 at leading large NN. The exact interacting SU(N)SU(N) theory has a=c=(N2−1)/4a=c=(N^2-1)/4, whose −1-1 is beyond that classical reduction; see the holographic anomaly calculation Henningson and Skenderis 1998 and the field-theory comparison Aharony et al. 2000, §3.2.2, eqs. (3.31)–(3.32), printed pp. 79–80. The flux, axiodilaton, and regime relations are summarized in Aharony et al. 2000, §3.1, eqs. (3.7)–(3.10), printed pp. 59–60.

Two small parameters govern the familiar ten-dimensional supergravity corner:

ϵα′≡α′L2=λ−1/2,ϵstring loop∼gs2=(λ4πN)2.\epsilon_{\alpha'}\equiv\frac{\alpha'}{L^2}=\lambda^{-1/2}, \qquad \epsilon_{\mathrm{string\ loop}}\sim g_s^2 =\left(\frac{\lambda}{4\pi N}\right)^2.

Consequently, in this weakly coupled type-IIB frame a parametric classical-supergravity window is

1≪λ≪N,1\ll\lambda\ll N,

with numerical factors and observable-dependent cancellations stated separately. Large NN at fixed λ\lambda suppresses string loops but does not suppress α′\alpha' corrections. Large λ\lambda at fixed NN makes the geometry smooth in string units but eventually makes this string frame strongly coupled. The exact-duality conjecture is broader than either approximation; the approximation window is not evidence that the full claim has been proved.

A row-by-row AdS₅/CFT₄ dictionary

Section titled “A row-by-row AdS₅/CFT₄ dictionary”
Row-by-row AdS₅/CFT₄ dictionary with status, control, and uncertainty
Boundary datum Bulk datum Logical and evidence status Domain and control Leading caveat or uncertainty
$N$ and $\tau_{\mathrm{YM}}$ Five-form flux and axiodilaton $C_0+i/g_s$; $L/\sqrt{\alpha'}$ follows from $\lambda$ Conjectural exact dictionary. D3-brane construction and flux quantization are primary evidence Specified theory, global form, theta periodicity, and boundary data; supergravity further needs $1\ll\lambda\ll N$ A local parameter map does not fix the genuine-line spectrum or every discrete datum
Half-BPS or other short-multiplet single-trace operator Protected Kaluza–Klein mode or string state with matching representation Protected match. Superconformal representation theory and matched protected spectra or correlators provide evidence The stated short multiplet and protected quantity, at the parameters where it is defined Protection of one datum does not establish completeness; finite-$N$ trace relations still matter
Generic single-trace primary $\mathcal O$ Single-string state, or a bulk field when the state is light Asymptotic or perturbative dictionary. Tested through declared large-$N$, strong-coupling, or string expansions Single-particle organization at leading order in $1/N$; a supergravity field also needs a suitable dimension gap and small $\alpha'/L^2$ Stringy states, anomalous dimensions, finite-$N$ mixing, and nonperturbative completeness remain
Multi-trace operator Multiparticle bulk state Asymptotic large-$N$ organization. Factorization and perturbative bulk calculations provide evidence Fixed low excitation number in a declared code sector Finite-$N$ trace identities, mixing, and high-energy backreaction spoil the simple particle-number mnemonic
Stress tensor $T_{\mu\nu}$ or conserved current $J_\mu$ Boundary metric or gauge potential as source; normalizable metric or gauge data as response Symmetry-fixed dictionary entry. Ward identities, anomalies, and correlator normalizations test it Specified symmetry, source convention, boundary condition, and linear or nonlinear response regime The current algebra alone does not fix the global charge lattice; protection of a coefficient is a separate claim
Source-dependent generating functional Renormalized string path integral with asymptotic fields fixed Conjectured exact identity or definition proposal; controlled saddle calculations are perturbative evidence Complete state, contour, source, boundary-condition, counterterm, and normalization data The exact bulk functional may be unconstructed, and a single saddle has a smaller domain
Fundamental or BPS Wilson line; ’t Hooft or dyonic line; higher-representation line F1 worldsheet; D1 worldsheet; $(p,q)$ string; in suitable regimes D3- or D5-brane Conjectural extended-operator dictionary. Protected and semiclassical tests provide evidence Fixed representation, genuine-line lattice, global form, discrete theta data, and approximation regime Which line is genuine and which bulk object dominates can change with those data
CFT state or thermal density matrix Bulk state or weighted set of geometries Conjectural state map. Protected quantities, correlators, and thermodynamics provide sector-specific evidence Declared preparation, ensemble, contour, charges, and semiclassical or exact regime A single classical geometry need not represent a full ensemble or an exact state

Here single-trace and protected are independent attributes. A single-trace operator is a gauge-invariant local operator written as one color trace; at large NN it organizes a single-string or single-particle sector at leading order. Protection instead means that symmetry, shortening, topology, or an index excludes specified corrections. A generic single-trace operator need not be protected, while a protected statement can concern a more complicated object.

The line-operator row is concrete only after its representation and global data are fixed. The fundamental-string prescription is given in Maldacena 1998, “Wilson Loops,” §3, eqs. (3.1)–(3.2), printed p. 3, and §4, printed p. 6. A D3-brane with electric flux describes suitable multiply wound or high-representation loops Drukker and Fiol 2005, §§2–3, while a D5-brane realizes the antisymmetric representation Yamaguchi 2006, §1. Each is a regime-specific semiclassical entry, not a representation-independent replacement for the F1 prescription.

The global form is information beyond the Lie algebra. For example, PSU(N)=SU(N)/ZNPSU(N)=SU(N)/\mathbb Z_N, but naming this quotient does not yet select a unique theory: one must also choose a maximal mutually local lattice of genuine Wilson–’t Hooft lines, equivalently a discrete variant (SU(N)/ZN)n(SU(N)/\mathbb Z_N)_n. Distinct variants can have identical local correlators on R4\mathbb R^4 but different line and topological sectors Aharony, Seiberg, and Tachikawa 2013, §2.1, eq. (2.1), printed pp. 12–13, and §2.3, eqs. (2.8)–(2.9), printed pp. 15–16. This is why a dictionary of local single-trace operators cannot by itself establish a complete duality.

Three adversarial changes

Section titled “Three adversarial changes”

Change the global form. Replacing SU(N)SU(N) by a specified (SU(N)/ZN)n(SU(N)/\mathbb Z_N)_n variant preserves the local Lie algebra but changes genuine line operators and topological sectors. Local protected correlators may survive unchanged; the complete theory-pair claim does not.

Change the AdS boundary condition. In a mass window admitting more than one quantization, or with an allowed mixed condition, the same bulk equation can define a different operator assignment or a deformed boundary theory. The boundary condition belongs in the theory specification, not in an afterthought.

Change the order of limits. Taking N→∞N\to\infty at fixed λ\lambda yields a tree-level string regime with generally unsuppressed string-scale corrections. Taking λ→∞\lambda\to\infty inside that large-NN regime yields classical supergravity. Reversing the operations at fixed finite NN can leave the weakly coupled string frame. A result must retain the limit order under which its remainder was controlled.

The strongest surviving statement after any change is the intersection of the original and modified domains. Often that is a protected local match, not an unchanged claim of full equivalence.

Common pitfalls

Section titled “Common pitfalls”

Treating one saddle as the bulk theory. A classical solution is one term in a path integral and may cease to dominate. State the contour, competing saddles, and omitted loop and string corrections.

Calling every exact number evidence for exact duality. A nonrenormalized anomaly coefficient can agree exactly because both calculations reduce to protected data. It is strong evidence for that entry, but its breadth and its independence from the proposed dictionary must be assessed separately.

Suppressing global data. A Lie algebra and a list of local operators do not determine genuine line operators, discrete theta angles, boundary conditions, or superselection sectors. Those omissions can change the theory while preserving many familiar checks.

Exercises

Section titled “Exercises”

1. Derive the supergravity window. Starting from λ=L4/α′2\lambda=L^4/\alpha'^2 and gs=λ/(4πN)g_s=\lambda/(4\pi N), find conditions that suppress string-scale corrections and string loops in the displayed type-IIB frame. What does large NN at fixed moderate λ\lambda fail to accomplish? Then set λ=Np\lambda=N^p and determine the allowed range of pp as N→∞N\to\infty.

Solution

String-scale corrections are organized by α′/L2=λ−1/2\alpha'/L^2=\lambda^{-1/2}, so they are small when λ≫1\lambda\gg1. String loops require gs=λ/(4πN)≪1g_s=\lambda/(4\pi N)\ll1, hence parametrically λ≪N\lambda\ll N. Together these give 1≪λ≪N1\ll\lambda\ll N. Sending N→∞N\to\infty at fixed moderate λ\lambda makes gsg_s small but leaves α′/L2\alpha'/L^2 finite, so tree-level string theory may be appropriate while a two-derivative supergravity truncation is not.

For λ=Np\lambda=N^p, the conditions α′/L2=N−p/2→0\alpha'/L^2=N^{-p/2}\to0 and gs=Np−1/(4π)→0g_s=N^{p-1}/(4\pi)\to0 require

0<p<1.0<p<1.

This exhibits a shared approximation window. It does not show that exact observables have noncommuting limits; that stronger statement would require following a specified observable and its remainders.

2. Classify the evidence. Suppose a protected anomaly matches exactly, a nonprotected four-point function matches a tree-level Witten diagram at leading order in the genus and α′/L2=λ−1/2\alpha'/L^2=\lambda^{-1/2} expansions, and no nonperturbative bulk construction is known. Classify the two results and state the strongest combined conclusion.

Solution

The anomaly is a protected match in its symmetry-controlled sector. The four-point result is a large-NN, strong-coupling saddle relation with omitted loop and string-scale corrections. Together they support two distinct parts of a conditional dictionary, one protected and one dynamical, but they do not prove surjectivity of the map, equality of all sectors, or a nonperturbative definition of the bulk theory. The strongest conclusion is agreement of the named observables in their stated domains.

The observable and regime matrix now asks which of these mapped quantities remains defined and controlled outside this special supergravity corner. Dictionary Completeness and Global Data develops the extended-operator and global-sector tests.

Status boundary. This page gives durable definitions and source-supported examples, not a current comparative verdict on holographic programs. Any time-sensitive assessment requires dated claim-level sources, contrary evidence, and specialist review.

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

Section titled “References”