Top-Down, Bottom-Up, and UV-Completion Claim Contracts

“Top-down,” “bottom-up,” “consistent truncation,” and “UV complete” are not four rival grades of the same object. The first two describe a model’s provenance, the third describes a classical reduction property, and the last is a much stronger claim about short-distance definition. The central rule is simple: a construction licenses only the conclusions supported by its declared object, domain, controls, and tests. A successful low-energy fit can be excellent physics without identifying a unique microscopic theory.

Required background. Nonperturbative definition and completion criteria supplies the stronger standard for a complete theory. Stringy and quantum corrections supplies the expansion parameters inherited by gravitational models.

Helpful background. Nondecoupling effects and matching validation supplies EFT validation tests. Claim, evidence, replication, and retraction supplies evidence discipline for quantitative fits.

Reading path. First separate the three questions below. Then follow one neutral black-brane calculation from a flux-quantized parent construction to a bottom-up calibration and an adversarial same-fit pair. The final sections turn the result into claim limits, pitfalls, and solved exercises.

Provenance, reduction, and claim strength

Three questions must be answered independently.

1. Where did the model come from? A top-down model descends from a named parent construction after its background, fluxes, global data, and retained fields have been specified. A bottom-up model starts from chosen fields, symmetries, and couplings over a declared phenomenological domain.

2. How is the reduced model related to a parent? For a consistent truncation with uplift map $\mathcal U$, the defining implication is

   $$\mathcal E_{5}[\Phi]=0 \quad\Longrightarrow\quad \mathcal E_{10}[\mathcal U(\Phi)]=0,$$

   where both sides are specified classical equations. The implication is one-way: it does not say that every parent solution is retained, that omitted Kaluza–Klein modes are harmless, or that $\alpha'$ and string-loop corrections vanish. Global regularity, flux quantization, and admissible boundary conditions still require separate checks. The nonlinear type-IIB/$S^5$ construction gives a concrete example of this equation-level statement Cvetič et al. 1999, § 2.1, Eqs. (2.1)–(2.9).

3. How strong is the conclusion? Classical embeddability, controlled EFT accuracy, a calibrated fit, a candidate exact dual, a nonperturbative definition, and identification of a real material are distinct conclusions. None becomes the next merely by changing vocabulary.

The following matrix is a stop rule: read across the row containing the evidence actually in hand, and do not silently promote the conclusion in the last column.

Evidence required by each label and the conclusion that the label does not supply

Label or conclusion	Minimum required information	Licensed inference	Invalid promotion
Top-down provenance	Parent theory; background and vacuum; flux integers and moduli; global form; retained sector; generator normalization; dictionary and boundary data.	Correlated couplings, charges, and fields in that named construction and in its controlled regime.	Not automatically an exact duality, a complete spectrum, a quantum definition, or the theory selected by an experiment.
Bottom-up provenance	Fields, symmetries, action or equations, renormalization prescription, state, fit domain, priors, and omitted-operator assumptions.	Conditional predictions and tests within the declared model family and domain.	Not a string embedding, a unique operator spectrum, or a UV completion.
Consistent classical truncation	An explicit ansatz or theorem proving that every retained classical solution uplifts to the named parent equations, plus global admissibility checks.	One-way classical embeddability of the retained solution sector.	Not surjectivity, stability against omitted modes, quantum equivalence, or string-exact dynamics.
Controlled gravitational EFT	Cutoff and power counting; curvature and field-strength bounds; loop counting; matching and renormalization; truncation-error estimate.	Predictions at stated energies and accuracy, whether or not a full embedding is known.	Not existence or uniqueness of a UV completion.
Phenomenological calibration	Named observables and states; data window and covariance; fitted parameters; numerical error; model discrepancy; held-out tests.	Descriptive or predictive adequacy for those observables in that window.	Not microscopic identification, embeddability, or completion.
Conditional nonperturbative definition	A precise microscopic object; state and observable rules; global data; regulator or cutoff removal; determinacy; and a sufficiently complete exact dictionary.	A definition of the bulk observables covered by the exact dictionary, conditional on the asserted equivalence.	Not an independent proof of the equivalence or identification with a particular material.
Identification of a real system	Independent discriminating measurements across observables and scales, controlled systematics, and serious exclusion of alternatives.	An empirical identification at the stated confidence and resolution.	Not supplied by a theoretical embedding or a two-observable fit alone.

Two consequences deserve emphasis. A consistent truncation need not be a Wilsonian low-energy truncation: on $S^5$, retained and omitted Kaluza–Klein scales are not parametrically separated. Conversely, a well-controlled EFT need not be a known consistent truncation. These are different reasons for trusting different statements.

A neutral black-brane comparison protocol

To compare constructions without moving the goalposts, declare the same tuple for each one:

$$(\text{object and data};\ \text{state};\ \text{observables};\ \text{domain};\ \ \text{controls};\ \text{evidence};\ \text{uncertainty};\ \text{falsifier}).$$

Use the site’s $(+---)$ convention and consider the five-dimensional family

$$\begin{aligned} S_c={}&\frac{1}{2\kappa _5^2}\int d^5x\,\sqrt{|g|} \left[R-\frac12(\partial\phi)^2-V(\phi)\right] \\ &-\frac{1}{4g_5^2}\int d^5x\,\sqrt{|g|}\,Z(\phi)F_{MN}F^{MN} \\ &+\frac{c}{g_5^2\Lambda^4}\int d^5x\,\sqrt{|g|} \left(F_{MN}F^{MN}\right)^2+\cdots . \end{aligned}$$

Here $V(0)=-12/L^2$, $V'(0)=0$, and $Z(0)=1$. The ellipsis is not decorative: it denotes other operators allowed by the symmetries, whose suppression and uncertainty must be stated before this is called an EFT. The neutral solution has $\phi=0$, $F=0$, and

$$ds^2=\frac{r^2}{L^2}\left[f(r)dt^2-d\boldsymbol{x}^{\,2}\right] -\frac{L^2}{r^2f(r)}dr^2, \qquad f(r)=1-\frac{r_h^4}{r^4}, \qquad T=\frac{r_h}{\pi L^2}.$$

The two calibration observables are defined with one fixed current normalization:

$$\begin{aligned} \eta&=-\lim_{\omega\to0}\frac{1}{\omega} \operatorname{Im}G^R_{T_{xy}T_{xy}}(\omega,\boldsymbol{0}), & C_\eta&\equiv\frac{\eta}{T^3},\\ \sigma&=-\lim_{\omega\to0}\frac{1}{\omega} \operatorname{Im}G^R_{J_xJ_x}(\omega,\boldsymbol{0}), & C_\sigma&\equiv\frac{\sigma}{T}. \end{aligned}$$

At classical two-derivative order, horizon regularity gives

$$C_\eta=\frac{\pi^3L^3}{2\kappa _5^2} =\frac{\pi^2L^3}{16G_5}, \qquad C_\sigma=\frac{\pi L}{g_5^2}, \qquad \kappa _5^2=8\pi G_5.$$

The shear and current results, as well as the limitation to the low-frequency linear-response problem, follow from the membrane analysis Iqbal and Liu 2009, Eq. (41) and § IV.3, Eqs. (43)–(49), (58). Away from this limit, the full radial geometry and additional interactions matter.

Top-down anchor: D3-brane flux and the retained sector

Take type-IIB string theory on $AdS_5\times S^5$ with $N$ units of self-dual five-form flux. In the convention $g_{\mathrm{YM}}^2=4\pi g_s$,

$$L^4=4\pi g_sN\alpha'^2=\lambda\alpha'^2, \qquad \lambda=g_{\mathrm{YM}}^2N.$$

This relation comes from the D3-brane near-horizon construction Maldacena 1998, § 2, Eqs. (2.1)–(2.4). The classical supergravity window requires $\lambda\gg1$ so that $\alpha'/L^2\ll1$, together with $g_s=\lambda/(4\pi N)\ll1$ so string loops are suppressed. “Large $N$” and “large $\lambda$” are therefore correlated controls, not magic labels.

The nonlinear $S^5$ ansatz consistently embeds a five-dimensional $U(1)^3$ Einstein–Maxwell–scalar sector into the classical type-IIB equations Cvetič et al. 1999, § 2.1, Eqs. (2.1)–(2.9). Choosing a particular diagonal $U(1)$ generator, the neutral black brane, and the standard R-current normalization gives at leading supergravity order

$$\kappa _5^2=\frac{4\pi^2L^3}{N^2}, \qquad g_5^2=\frac{16\pi^2L}{N^2},$$

and hence

$$\eta=\frac{\pi N^2T^3}{8}, \qquad \sigma=\frac{N^2T}{16\pi}.$$

These normalizations and transport coefficients are computed directly in Policastro, Son, and Starinets 2002, § 5, Eqs. (18), (34a), and § 6, Eq. (53). The result is an analytic, leading-order prediction for a named state and named operators, with corrections from $\alpha'$ effects and string loops. The construction correlates couplings and identifies a candidate boundary theory; it does not, by uplift alone, prove exact finite-$N$ duality, provide the full string spectrum, settle every boundary-condition choice, or identify a laboratory material.

Bottom-up calibration: two numbers and two parameters

Now forget the D3-brane origin and treat the same five-dimensional action as a bottom-up model. To make the inference test reproducible, choose $L=1$ and synthetic targets equal to the leading top-down values at $N=10$:

$$\begin{aligned} C_\eta^\star&=\frac{100\pi}{8} =39.26990817\pm0.39269908,\\ C_\sigma^\star&=\frac{100}{16\pi} =1.98943679\pm0.01989437. \end{aligned}$$

The displayed one-standard-deviation errors are independent illustrative $1\%$ measurement errors. They are not estimates of finite-$N$ or finite-$\lambda$ corrections; $N=10$ is used only to make the arithmetic transparent. A real calibration must include correlated experimental or numerical covariance and a separate theory-discrepancy budget.

Solving the two horizon equations gives

$$\kappa _5^2=\frac{\pi^3}{2C_\eta^\star} =0.39478418\pm0.00394784, \qquad g_5^2=\frac{\pi}{C_\sigma^\star} =1.57913670\pm0.01579137,$$

where the uncertainties use linear propagation. Back-substitution returns both central values exactly. If the covariance is diagonal, then

$$\chi^2=(\boldsymbol C-\boldsymbol C^\star)^{\mathsf T} \Sigma^{-1}(\boldsymbol C-\boldsymbol C^\star)=0.$$

This is not a goodness-of-fit victory: two fitted parameters for two calibration numbers leave zero degrees of freedom. Validation begins with observables that were not used in the fit. Bottom-up scalar potentials can be calibrated very effectively to thermodynamic data, but the chosen functions remain model hypotheses rather than a microscopic identification Gubser and Nellore 2008, §§ 2–3.

The same neutral transport calculation with different provenance and different evidence ceilings

Test field	Flux-quantized type-IIB construction	Calibrated bottom-up model
Object and fixed data	Type IIB on AdS5 × S5; flux integer N; chosen U(1) generator; neutral black brane; boundary dictionary and global data still declared separately.	Five-dimensional fields, potential V, gauge function Z, omitted-operator basis, cutoff, boundary conditions, and fitted couplings.
Observables and domain	Cη and Cσ for the maximally supersymmetric Yang–Mills stress tensor and a normalized R-current, at zero density and in the hydrodynamic limit.	The same two dimensionless coefficients, here only at zero density, infinitesimal source, and classical linear-response order.
Controls	Small α′/L² and string-loop coupling; regular uplift; stated generator and boundary data; higher-derivative and finite-N corrections retained as uncertainties.	Curvature and frequency below the assumed cutoff; small field strength; stable background; numerical error; omitted operators and model discrepancy.
Evidence class	Primary analytic supergravity calculation plus an explicit consistent classical uplift; exact duality remains an additional assertion.	Exact algebraic calibration within the stated classical model, using synthetic data; no parent construction has been supplied.
Uncertainty and decisive next test	α′ and loop corrections, dictionary and global choices; compare corrected or finite-N observables beyond the leading sector.	Statistical covariance plus theory discrepancy; predict an unused correlator, nonlinear response, finite-density response, or operator dimension.
Strongest warranted statement	A controlled leading prediction in a classically embeddable sector of a named parent construction.	Two neutral transport coefficients are reproduced in the calibration domain.
Statement not established	Exact quantum/string completion from the truncation alone, or empirical selection by nature.	A string embedding, UV completion, unique operator spectrum, or identification of a material.

The adversarial test, carried out

Keep $V$, $Z$, $\kappa _5$, $g_5$, the cutoff $\Lambda$, and the neutral state fixed. Compare

$$\mathcal M_A:\ c_A=0, \qquad \mathcal M_B:\ c_B=1,$$

in the explicitly normalized action above. Model $B$ is a local EFT hypothesis with the additional $(F^2)^2/\Lambda^4$ interaction; the value $c_B=1$ is a bookkeeping choice, not evidence that a UV completion exists.

Write the gauge fluctuation as $F=\bar F+f$, with $f=O(a)$ in the source amplitude. At zero density, $\bar F=0$, so

$$\left(F_{MN}F^{MN}\right)^2 =\left(f_{MN}f^{MN}\right)^2=O(a^4).$$

The added term has no background, metric-quadratic, or gauge-quadratic contribution. Therefore $\mathcal M_A$ and $\mathcal M_B$ have exactly the same neutral geometry and exactly the same classical two-point functions $C_\eta$ and $C_\sigma$. Both have $\chi^2=0$ for the two synthetic calibration values.

They are nevertheless inequivalent theories. The quartic term contributes a contact interaction to the connected four-current correlator and changes nonlinear conductivity. It also enters linear response once a finite-density background is present. Indeed, through quadratic order in $f$,

$$\left[(\bar F+f)^2\right]^2 =\cdots+2\bar F^2 f^2+4(\bar F\!\cdot\! f)^2+O(f^3),$$

which is nonzero when $\bar F_{rt}\ne0$. At finite density the background itself must also be recomputed; the neutral equality must not be extrapolated there.

Two exact calibration matches and the hidden interaction that the fit cannot determine

Test	Model A: c = 0	Model B: c = 1	Inference
Neutral background	AdS5 Schwarzschild black brane.	The identical black brane because F = 0.	The geometry does not determine the hidden quartic coupling.
Calibrated Cη	39.26990817.	39.26990817.	The stress-tensor fit cannot select either action.
Calibrated Cσ	1.98943679.	1.98943679.	The linear current fit cannot select either action.
Held-out response	No quartic gauge contact from this operator.	A nonzero quartic contact suppressed by Λ−4; finite-density response also changes after recomputation.	A four-current or finite-density measurement can break the degeneracy.
Embedding evidence	None supplied by the fit.	None supplied by the fit.	Agreement with two numbers proves no string embedding.
Strongest surviving claim	Both models reproduce the two named coefficients at zero density and classical linear-response order.		Downgrade reason: an unconstrained operator is invisible to every calibrated observable but changes a held-out observable.

The counterexample is enough to refute both proposed promotions:

$$\text{two-coefficient fit}\not\Rightarrow\text{unique microscopic theory}, \qquad \text{two-coefficient fit}\not\Rightarrow\text{string embedding}.$$

It does not show that either local action has a UV completion. That question requires independent consistency and completion evidence.

Consistency tests and their own assumptions

Analyticity and causality can constrain EFT coefficients, but the assumptions are part of the conclusion. The standard forward-limit positivity argument assumes a nongravitational local Lorentz-invariant EFT with a unitary, analytic, crossing-compatible, sufficiently bounded S-matrix and the required gap or infrared control Adams et al. 2006, § 1 and § 4, Eqs. (19), (21)–(24). Under those hypotheses, a forbidden coefficient can rule out a proposed standard completion. Passing the bound does not construct a completion and does not identify one uniquely.

With a massless graviton, the forward $t$-channel pole obstructs the naive proof. Regge behavior can permit controlled, parametrically suppressed negativity Tokuda, Aoki, and Hirano 2020, §§ 3.2–3.4, Eqs. (3.12)–(3.37), and loop analyses exhibit cancellations between low- and high-energy contributions Caron-Huot and Tokuda 2024, § 1 and § 2.1. A gravitational positivity statement must therefore name its subtraction, infrared, Regge, and scale-separation assumptions.

Evidence cutoff: 30 August 2026. A current top-down example reinforces the same discipline: in the five-dimensional STU model, charged black-brane branches develop dynamical modes at the thermodynamic-instability threshold Gladden et al. 2025, abstract and § 3. A known parent construction and a consistent classical truncation do not make every state stable or every approximation reliable.

For more specialized tests, continue to calibration uncertainty and cross-model inference for parameter degeneracy, Kubo formulae and horizon response for transport, quantum-gravity consistency claims for conjectural constraints, and existence, uniqueness, and equivalence claims for the distinct mathematical standard. The next chapter compares nonperturbative definition proposals.

The chapter overview contains the structure diagram and validity and failure diagram. They show the shared construction and failure routes; the calculations and semantic tables on this page carry the page-specific comparison.

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

Common pitfalls

Treating “top-down” as “exact.” A top-down derivation may itself use classical supergravity, a large-$N$ limit, weak string coupling, or a restricted vacuum. State those controls before quoting the prediction.

Treating consistent truncation as decoupling. Consistency proves an uplift implication for the retained classical equations. It does not require a mass gap to omitted modes and does not show that those modes are irrelevant to stability or quantum corrections.

Counting a calibration residual as validation. With two parameters and two data, a zero central residual is automatic when the inverse map exists. Test a third observable and carry covariance, numerical error, and model discrepancy separately.

Calling numerical robustness “universality.” Insensitivity over a sampled parameter range is valuable robustness evidence. Universality additionally requires a reason—such as a symmetry, conservation law, fixed point, or theorem—that identifies the whole relevant class.

Using flat-space positivity without the gravity caveat. First declare whether a massless graviton, forward pole, Regge contribution, or insufficient scale separation invalidates the strict sign argument.

Exercises

1. Classify the evidence

A five-dimensional solution uplifts through a published ansatz to classical type-IIB supergravity. Its shear viscosity agrees with one experimental ratio. Which conclusions follow: consistent classical embeddability, controlled string accuracy, exact duality, or identification of the material?

Solution

The uplift establishes consistent classical embeddability, provided the global and boundary data satisfy the ansatz. Controlled string accuracy additionally requires small curvature in string units, suppressed loops, and an estimate of omitted operators. Exact duality requires an independent exact dictionary and equivalence claim. One matching ratio does not identify the material because alternative EFTs can share that ratio.

2. Recover the fitted couplings

Starting from $C_\eta=\pi^3L^3/(2\kappa _5^2)$ and $C_\sigma=\pi L/g_5^2$, derive the fitted couplings and verify the $L=1$, $N=10$ numbers.

Solution

Invert each equation:

$$\kappa _5^2=\frac{\pi^3L^3}{2C_\eta}, \qquad g_5^2=\frac{\pi L}{C_\sigma}.$$

With $C_\eta=100\pi/8$ and $C_\sigma=100/(16\pi)$,

$$\kappa _5^2=\frac{4\pi^2}{100}=0.39478418, \qquad g_5^2=\frac{16\pi^2}{100}=1.57913670.$$

Substitution recovers both target coefficients. The calculation checks arithmetic, not the size of string or model errors.

3. Locate the first observable changed by the quartic operator

Show by source counting why $(F^2)^2$ changes neither the neutral background nor the classical current two-point function, but can change a four-current correlator.

Solution

At zero background field, $F=f=O(a)$, where $a$ is the boundary gauge-source amplitude. Therefore $(F^2)^2=O(a^4)$. The zeroth functional derivative vanishes, so the background is unchanged; the second derivative at $a=0$ also vanishes, so the two-point function and linear conductivity are unchanged. The fourth derivative is nonzero and supplies a contact contribution to the connected four-current response.

4. Explain the finite-density failure of the same-fit claim

Let $F=\bar F+f$ with $\bar F_{rt}\ne0$. Identify the terms quadratic in $f$ in $[(\bar F+f)^2]^2$ and state the consequence.

Solution

Writing $(\bar F+f)^2=\bar F^2+2\bar F\!\cdot\! f+f^2$ and squaring gives the quadratic terms

$$2\bar F^2f^2+4(\bar F\!\cdot\! f)^2.$$

They modify the gauge fluctuation equations whenever the background field is nonzero. The quartic operator also alters the charged background equations, so both the solution and its linear response must be recomputed. The neutral same-fit result has no automatic finite-density extension.

5. Design a discriminating test

The two bottom-up models match $C_\eta$ and $C_\sigma$. Give one next measurement, its expected logical role, and a possible outcome that would falsify one model without proving the survivor is UV complete.

Solution

Measure a connected four-current response in the same neutral state, with source amplitude and frequency kept below the declared cutoff. Model $B$ predicts a local contribution proportional to $c_B/\Lambda^4$ that model $A$ lacks. A result incompatible with that term can reject $B$ in the tested domain. Agreement with $A$ would only improve its phenomenological support; it would not construct a parent string background or a nonperturbative definition.

References

• Adams, A., Arkani-Hamed, N., Dubovsky, S., Nicolis, A., and Rattazzi, R. (2006), “Causality, Analyticity and an IR Obstruction to UV Completion,” Journal of High Energy Physics 2006(10), 014. DOI. arXiv.
• Caron-Huot, S., and Tokuda, J. (2024), “String Loops and Gravitational Positivity Bounds: Imprint of Light Particles at High Energies,” arXiv:2406.07606 [hep-th]. arXiv.
• Cvetič, M., Duff, M. J., Hoxha, P., Liu, J. T., Lü, H., Lu, J. X., Martinez-Acosta, R., Pope, C. N., Sati, H., and Tran, T. A. (1999), “Embedding AdS Black Holes in Ten and Eleven Dimensions,” Nuclear Physics B 558, 96–126. DOI. arXiv.
• Gladden, L., Ivo, V., Kovtun, P. K., and Starinets, A. O. (2025), “Hydrodynamics with Multiple Charges and Holography,” arXiv:2507.21346 [hep-th]. arXiv.
• Gubser, S. S., and Nellore, A. (2008), “Mimicking the QCD Equation of State with a Dual Black Hole,” Physical Review D 78, 086007. DOI. arXiv.
• Iqbal, N., and Liu, H. (2009), “Universality of the Hydrodynamic Limit in AdS/CFT and the Membrane Paradigm,” Physical Review D 79, 025023. DOI. arXiv.
• Maldacena, J. M. (1998), “The Large $N$ Limit of Superconformal Field Theories and Supergravity,” Advances in Theoretical and Mathematical Physics 2, 231–252. DOI. arXiv.
• Policastro, G., Son, D. T., and Starinets, A. O. (2002), “From AdS/CFT Correspondence to Hydrodynamics,” Journal of High Energy Physics 2002(09), 043. DOI. arXiv.
• Tokuda, J., Aoki, K., and Hirano, S. (2020), “Gravitational Positivity Bounds,” Journal of High Energy Physics 2020(11), 054. DOI. arXiv.