Bulk Interaction Scaling and Effective Cutoffs

A holographic effective field theory is controlled by several independent small parameters, and the first one to fail sets the useful range of the chosen truncation. Heavy single-trace states control local higher-derivative terms, bulk loops depend on the gravitational coupling and weighted species sums, and compactification or string towers can appear before Planckian energies. This page builds an observable-specific estimate for a weakly curved $AdS_{d+1}$ bulk; it does not assign a universal numerical cutoff to every holographic theory.

Required background. Central Charge, Newton Coupling, and the Planck Scale fixes gravitational normalization; Weakly Coupled Bulk Fields from Connected Correlators fixes vertex scaling; Effective Field Theory as a Controlled Expansion supplies generic EFT logic.

Helpful background. Applying EFT Power Counting to Gravity develops the gravity expansion; Effective Field Theory of Gravity: Architecture and Power Counting provides the broader gravitational-EFT architecture; and Loops, Counterterms, and Closure of an EFT Expansion explains loop closure.

Power counting in a weakly curved bulk

Write $D=d+1$ for the bulk spacetime dimension, $L$ for the AdS radius, and $G_L$ for the renormalized Newton coupling fixed by the AdS-scale stress-tensor normalization. Let $G_{\mathrm{eff}}(Q)$ denote the effective gravitational coupling entering the declared local process after threshold matching and running, and define $r_G(Q):=G_{\mathrm{eff}}(Q)/G_L$. This split is bookkeeping; the complete observable is scheme independent. Let $Q$ denote the largest physical scale sampled by the observable: local momentum or mass, together with the curvature scale $L^{-1}$. Let $\Lambda_{\mathrm{gap}}$ be the lowest physical heavy-state threshold integrated out in the chosen observable and field-content truncation; it is not automatically equal to a boundary dimension gap. For a localized sub-AdS collision, $Q$ is approximately the local center-of-mass energy. A boundary frequency is not automatically this local energy; redshift and the bulk region sampled by the state must be specified.

Because

$$[G_L]=[G_{\mathrm{eff}}]=\mathrm{mass}^{2-D}=\mathrm{mass}^{-(d-1)},$$

the gravitational loop-counting combination $G_{\mathrm{eff}}(Q)Q^{d-1}$ is dimensionless. For a gravitationally normalized process whose nonzero leading term comes from the two-derivative bulk action, a useful schematic organization is

$$\frac{\mathcal A(Q)}{\mathcal A_{\mathrm{tree}}(Q)} =1+\epsilon_{\mathrm{hd}}(Q)+\epsilon_{\mathrm{loop}}(Q)+\cdots,$$

with

$$\begin{aligned} \epsilon_{\mathrm{hd}}(Q) &=a_p\left(\frac{Q}{\Lambda_{\mathrm{gap}}}\right)^p,\\ \epsilon_{\mathrm{loop}}(Q) &=b_D(Q)N_{\mathrm{eff}}(Q)G_{\mathrm{eff}}(Q)Q^{d-1}. \end{aligned}$$

Here $p>0$ is the number of extra derivatives relative to the leading interaction, $a_p$ is the matched dimensionless Wilson coefficient, and $N_{\mathrm{eff}}(Q)$ is a threshold- and spin-weighted species count for the particular loop. The dimensionless $b_D(Q)$ contains the loop measure, masses, spins, kinematics, tensor contractions, and the ratio to the chosen tree amplitude. The factorization $b_DN_{\mathrm{eff}}$ is bookkeeping for contributions with comparable loop functions; otherwise the explicit weighted sum must be retained. For a conservative estimate, it bounds the sum of magnitudes rather than relying on cancellations. There is therefore no universal replacement $b_D=(4\pi)^{-D/2}$. A flavor contraction can remove a species enhancement, while a diagram with several independent internal sums can enhance it further.

Choose one renormalization organization. Contributions absorbed into $G_{\mathrm{eff}}(Q)$ must be excluded from $b_D(Q)N_{\mathrm{eff}}(Q)$, which then denotes the residual renormalized loop coefficient in the same scheme and at the same scale. Equivalently, retain $G_L$ in the leading action and put those contributions in the explicit loop term—but never in both.

The ratio form applies to a fixed tensor or helicity component, away from kinematic zeros, for which $\mathcal A_{\mathrm{tree}}$ is nonzero and of its nominal order. If that tree component vanishes or is parametrically suppressed, power-count the absolute amplitude relative to its first nonzero contribution; a divergent ratio to zero is not a cutoff.

The gravitational estimate is only one part of bulk power counting. For a loop built from two nongravitational cubic vertices, use the actual canonically normalized couplings instead; after extracting the powers of $L$ or $Q$ needed to make $\lambda_a$ dimensionless, a representative one-sum contribution has the form

$$\epsilon_{\mathrm{matter\ loop}} \sim \sum_a \lambda_a^2 F_a(m_aL,QL,\ldots),$$

where $F_a$ is the regulated and renormalized loop function in the declared kinematic domain. The connected-correlator analysis explains when $\lambda_a\sim N^{-1}$ and when a growing sum can defeat that suppression. A process is controlled only if every coupling expansion relevant to it is small.

The derivative and loop expansions of gravitational EFT are developed in Donoghue 1994, §§ 3–4 and Burgess 2004, §§ 2.5 and 3.2. The formula above is a dimensional benchmark, not a substitute for evaluating the relevant diagrams and counterterms.

For boundary dimension $d>2$, at two-derivative Einstein order and in the stress-tensor convention fixed on the prerequisite page,

$$\begin{aligned} C_T&=\kappa_d\frac{L^{d-1}}{G_L},\\ \kappa_d&= \frac{d+1}{d-1} \frac{\Gamma(d+1)}{8\pi^{d/2+1}\Gamma(d/2)}. \end{aligned}$$

In $d=2$, use the Virasoro central charge and the Brown–Henneaux relation with an explicit convention conversion; do not simply continue the displayed $d>2$ tensor-normalization formula. For $d>2$, the loop estimate becomes

$$\begin{aligned} \epsilon_{\mathrm{loop}}(Q) &=b_D(Q)N_{\mathrm{eff}}(Q)G_{\mathrm{eff}}(Q)Q^{d-1}\\ &=b_D(Q)\,\kappa_d\, \frac{N_{\mathrm{eff}}(Q)}{C_T} r_G(Q)(QL)^{d-1}. \end{aligned}$$

At fixed $QL$, fixed light spectrum, $r_G$ with no relevant $N$ dependence, and $C_T\sim N^2$, a one-loop correction is suppressed by one additional power of $N^{-2}$ relative to the tree connected correlator. This conclusion is about normalized observables. Rescaling the boundary operator or bulk field changes intermediate couplings, but not the complete correlator.

Corrections and thresholds are different

A small correction estimates an error below a threshold; crossing a threshold changes the degrees of freedom that must be retained. The distinction prevents a low-dimensional truncation from being mistaken for a statement about the full string theory.

Small corrections estimate errors below a threshold; crossing a physical threshold changes the required field content or perturbative organization.

Control datum	Small quantity below the threshold	What fails first
Heavy single-trace scale Λgap	|ap| (Q/Λgap)p	The local derivative truncation, or the omission of the heavy field itself
Gravitational loop scale Λloop	|bD| Neff Geff(Q) Qd − 1	Fixed-loop perturbation theory
Kaluza–Klein mass MKK	Model-dependent powers of Q/MKK after matching	The lower-dimensional field-content truncation
String mass Ms = 1/√α′	α′Q2 = (Q/Ms)2, with interaction-dependent powers	The finite-field local action; massive string states must be resolved
String coupling gs	Usually powers of gs2 per additional closed-string handle	The genus expansion; this is not by itself an energy threshold

A boundary dimension gap becomes a bulk mass scale only through the appropriate field dictionary. For an omitted scalar, $m^2L^2=\Delta(\Delta-d)$, so $mL=\Delta-d/2+O(\Delta^{-1})$ at large $\Delta$. Thus $\Lambda_{\mathrm{gap}}L$ often tracks a large single-trace dimension gap, but spin, compactification, and which fields were retained must still be specified; the sparse-spectrum analysis keeps those notions separate.

On a background of radius $L$, $\alpha'/L^2$ tests string-scale curvature, whereas $\alpha'Q^2$ tests the derivatives in a local process. They coincide only when $Q\sim L^{-1}$. Similarly, reaching $M_{\mathrm{KK}}$ invalidates an EFT that omitted the KK tower, not the higher-dimensional theory that contains it. In the perturbative scalar four-point setting, Heemskerk et al. 2009, § 2.4, § 4.1, and § 7, especially pp. 38–39, PDF formulate the large-$N$, large higher-spin single-trace-gap locality conjecture, match bounded-spin crossing solutions to local quartic interactions, and discuss inverse-cutoff suppression of higher-derivative coefficients as an EFT expectation—not as a general gap-to-coefficient theorem.

First application: a four-point error budget

Consider a normalized connected four-point process. Assume that its leading contribution is tree level, the first allowed higher-derivative contact has four extra derivatives, and the one-loop term has one independently summed species label. Then

$$\begin{aligned} \frac{\mathcal A_4}{\mathcal A_{4,\mathrm{tree}}} &=1+a_4\left(\frac{Q}{\Lambda_{\mathrm{gap}}}\right)^4\\ &\quad+b_D\kappa_d\frac{N_{\mathrm{eff}}}{C_T} r_G(Q)(QL)^{d-1} +\cdots. \end{aligned}$$

If $a_4$, $b_D$, $N_{\mathrm{eff}}$, and $r_G$ are approximately constant throughout the proposed window, with $\lvert b_D\rvert N_{\mathrm{eff}}r_G>0$, the contact term and loop term become order one at

$$\Lambda_{\mathrm{hd}} =\Lambda_{\mathrm{gap}} \min\left(1,\lvert a_4\rvert^{-1/4}\right),$$

and

$$\Lambda_{\mathrm{loop}}L =\left( \frac{C_T}{\lvert b_D\rvert\kappa_dN_{\mathrm{eff}}r_G} \right)^{1/(d-1)}.$$

The minimum in $\Lambda_{\mathrm{hd}}$ matters. If $\lvert a_4\rvert<1$, the contact remains perturbative up to the gap, but the omitted heavy state still appears at $\Lambda_{\mathrm{gap}}$. If $\lvert a_4\rvert>1$, the higher-derivative term becomes comparable to the chosen leading term earlier, so the displayed derivative truncation fails; this alone does not establish strong coupling. If a symmetry sets $a_4=0$, the next nonzero operator—not this formula—sets the derivative estimate. Likewise, if the displayed loop coefficient vanishes, omit that root and inspect the next nonzero loop contribution.

For the stated field content and fixed-order truncation, assuming no independently lower strong-coupling scale, define the first prediction stop by

$$\Lambda_{\mathrm{stop}} \lesssim \min\left( \Lambda_{\mathrm{hd}}, \Lambda_{\mathrm{loop}}, M_{\mathrm{KK}}, M_s \right).$$

Only scales belonging to the specified model and field-content truncation should appear in this minimum. If $N_{\mathrm{eff}}(Q)$, $b_D(Q)$, or $r_G(Q)$ varies appreciably, define $\Lambda_{\mathrm{loop}}$ as the first $Q\ge L^{-1}$, approached from a controlled region, at which $\lvert\epsilon_{\mathrm{loop}}(Q)\rvert=1$. If the correction is already order one at $Q=L^{-1}$, no loop-controlled AdS window exists; if it never reaches one in the declared domain, this contribution supplies no loop stop. The prediction stop is a physical EFT cutoff only when it corresponds to an omitted threshold or genuine UV strong coupling. Resummable logarithms, threshold enhancements, or failure of the chosen perturbative organization can invalidate this calculation without invalidating the EFT.

As a reproducible bookkeeping example, take an $AdS_5/CFT_4$ process, so $d=4$ and $\kappa_4=5/\pi^3\simeq0.1613$. Choose the deliberately synthetic inputs

$$\begin{gathered} C_T=10^4, \qquad N_{\mathrm{eff}}=10, \qquad b_5=1,\\ r_G=1, \qquad \Lambda_{\mathrm{gap}}L=24, \qquad a_4=16,\\ M_{\mathrm{KK}}L=30, \qquad M_sL=40. \end{gathered}$$

They give

$$\begin{aligned} \Lambda_{\mathrm{hd}}L&=12,\\ \Lambda_{\mathrm{loop}}L &=\left(\frac{10^4}{10(5/\pi^3)}\right)^{1/3}\\ &\simeq18.37. \end{aligned}$$

so the contact interaction sets the first stated stop, $\Lambda_{\mathrm{stop}}L\lesssim12$. At $QL=6$,

$$\begin{aligned} \lvert\epsilon_{\mathrm{hd}}\rvert &=16\left(\frac6{24}\right)^4=0.0625,\\ \lvert\epsilon_{\mathrm{loop}}\rvert &=\frac{5}{\pi^3}\frac{10}{10^4}6^3 \simeq0.0348. \end{aligned}$$

For a tree-level prediction, the two displayed omissions are therefore about $6.25\%$ and $3.48\%$. If their signs and correlations are unknown, the sum of magnitudes, about $9.73\%$, is a conservative bound for these two terms only. If both corrections are actually computed, the residual EFT uncertainty begins with the next allowed derivative order, two-loop terms, mixed terms, and any separately unbounded threshold effect.

What the calculation licenses

This is a deterministic power-counting fixture, not data from a specific dual pair. The observable is the normalized four-point ratio in an AdS$_5$ process at $QL=6$; the retained approximation is tree-level two-derivative exchange; the declared controls are one four-extra-derivative contact, one gravitational loop with one weighted species sum, $r_G=1$, and the heavy, KK, and string thresholds. The conclusion is conditional: for the synthetic coefficients above, the contact term sets the earliest stated prediction stop and the two displayed omissions are bounded in magnitude by about $9.73\%$. That number is neither a statistical confidence interval nor a bound on omitted sectors that were not specified; the stop is not automatically a physical cutoff.

Stress tests and nonuniform limits

Hold $G_L$ fixed and increase the weighted species count. If $b_D$, $r_G$, and $N_{\mathrm{eff}}$ vary slowly with energy across the tested window,

$$\Lambda_{\mathrm{loop}} \propto \bigl(N_{\mathrm{eff}}r_G\bigr)^{-1/(d-1)},$$

so the loop window shrinks even though the AdS-scale Planck length is unchanged. At fixed $C_T$ and $L$, the same conclusion follows from the CFT form of the estimate. If $C_T\sim N^2$, $N_{\mathrm{eff}}\sim N^\alpha$, and $r_G\sim N^\gamma$, then

$$\Lambda_{\mathrm{loop}}L \sim N^{(2-\alpha-\gamma)/(d-1)}.$$

It grows only when $\alpha+\gamma<2$. For $\alpha+\gamma=2$, the loop scale remains $O(L^{-1})$. For $\alpha+\gamma>2$, the formal root lies below the minimum $Q=L^{-1}$ used in this power counting: the AdS-scale loop expansion is already uncontrolled, not meaningfully cut off below $L^{-1}$.

Alternatively, hold $L$ and $C_T$—hence $G_L$ through the normalization above—fixed while lowering $\Lambda_{\mathrm{gap}}L$. The derivative window shrinks without changing the AdS-scale graviton kinetic normalization. Either deformation disproves a Planck-only error estimate; neither disproves the underlying holographic theory. It only shows that the selected low-energy field content or perturbative order has a smaller domain.

The large-$N$ limit must therefore keep the tested operator set, $QL$, gap ratios, and relevant species sums under control. Taking $N\to\infty$ does not suppress a loop if the number of comparably coupled fields grows like $C_T$, and taking $Q$ toward a heavy threshold can defeat a fixed-$Q$ expansion.

What to report with a bulk prediction

A reproducible holographic EFT estimate should state the observable and its normalization; the bulk dimension and curvature scale; the local kinematic domain; the retained fields, vertices, derivative order, and loop order; the matching scheme and the ratio $r_G(Q)$; every heavy, KK, and string threshold; the weighted species sums and their contraction pattern; and the first omitted contributions. Stop using the stated truncation or perturbative order when any declared expansion parameter becomes order one or when an omitted state can be produced or resolved, then decide whether the remedy is resummation, reorganization, adding fields, or abandoning the EFT.

Common pitfalls

Calling a regulator the cutoff. A momentum regulator is a calculational device. A physical EFT cutoff is set by omitted degrees of freedom or genuine UV strong coupling. Failure of a fixed-order expansion can instead define only a prediction stop.

Calling every prediction stop a physical cutoff. A large logarithm or an enhanced known interaction may require resummation or a different leading action while the same EFT remains valid. A new threshold or genuine UV strong coupling is a stronger failure.

Using one universal species factor. $N_{\mathrm{eff}}$ is a weighted sum for a specified diagram and energy range. Mass thresholds, spin, coupling normalization, and flavor contractions determine whether a species contributes and how many independent sums occur.

Trusting a cancellation without improving the error. A small sum of contact and loop corrections does not imply that each expansion is controlled. Quote their magnitudes separately unless a symmetry or a calculation protects the cancellation.

Exercises

1. General contact and loop scales

Let the first nonzero contact correction be $a_p(Q/\Lambda_{\mathrm{gap}})^p$, and let $b_D$, $N_{\mathrm{eff}}$, and $r_G>0$ be constant with $\lvert b_D\rvert N_{\mathrm{eff}}r_G>0$. Derive the contact-control and loop-control scales, including the heavy-state stop.

Solution — contact and loop control scales

Setting the contact magnitude to one gives $Q=\Lambda_{\mathrm{gap}}\lvert a_p\rvert^{-1/p}$. The chosen field-content truncation cannot pass the state that was integrated out, so

$$\Lambda_{\mathrm{hd}} =\Lambda_{\mathrm{gap}} \min\left(1,\lvert a_p\rvert^{-1/p}\right).$$

Setting $\lvert b_D\rvert N_{\mathrm{eff}}r_GG_LQ^{d-1}=1$ gives

$$\Lambda_{\mathrm{loop}} =\left( \frac{1}{\lvert b_D\rvert N_{\mathrm{eff}}r_GG_L} \right)^{1/(d-1)}.$$

Using $C_T=\kappa_dL^{d-1}/G_L$ reproduces

$$\Lambda_{\mathrm{loop}}L =\left( \frac{C_T}{\lvert b_D\rvert\kappa_dN_{\mathrm{eff}}r_G} \right)^{1/(d-1)}.$$

The fixed-order prediction stops at the minimum of these scales and every lower KK, string, or strong-coupling threshold. This minimum is a physical EFT cutoff only when the limiting scale is an omitted threshold or genuine UV strong coupling.

2. Boundary large-N order

Suppose a normalized connected four-point function has a tree contribution of order $C_T^{-1}$, while $N_{\mathrm{eff}}$ and $r_G$ remain fixed. What is the absolute order of the one-loop contribution at fixed $QL$? What changes if $N_{\mathrm{eff}}\sim C_T$ or if $r_G$ itself grows with $C_T$?

Solution — species scaling of loop order

At fixed $QL$, the relative loop factor is $\epsilon_{\mathrm{loop}}=O(N_{\mathrm{eff}}r_G/C_T)$. For fixed $N_{\mathrm{eff}}$ and $r_G$, multiplying by the tree order gives

$$\mathcal A_{4,\mathrm{1-loop}} =O(C_T^{-2}).$$

When $C_T\sim N^2$, this is $O(N^{-4})$, one power of $N^{-2}$ below the tree connected term. If $N_{\mathrm{eff}}\sim C_T$ with $r_G=\Theta(1)$, or more generally if $N_{\mathrm{eff}}r_G\sim C_T$, the relative loop factor can be order one, so the loop need not be suppressed at all. The raw number of fields is only a proxy; the relevant quantity is the weighted diagrammatic sum together with the matched gravitational coupling.

The next article, Approximate Bulk Locality from Spectral and Mellin Data, applies this cutoff discipline to the Mellin derivative expansion. Loop renormalization and the separation of nonlocal data from local counterterms continue in Loop Diagrams, Bulk EFT, and Renormalization. The distinct $\alpha'$, genus, KK, and nonperturbative corrections of a top-down construction continue in Stringy and Quantum Corrections to Supergravity.

Evidence cutoff: 28 August 2026. The relations above are controlled power-counting statements once the observable, normalization, spectrum, couplings, and kinematic domain are supplied. They do not determine a Wilson coefficient from the gap alone, replace an explicit loop calculation, or identify a universal quantum-gravity cutoff. A new species-enhanced loop claim, large-gap theorem, or top-down threshold claim should be compared against the separate contact, loop, KK, string, and strong-coupling stops used here.

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

• Burgess, C. P. 2004. “Quantum Gravity in Everyday Life: General Relativity as an Effective Field Theory.” Living Reviews in Relativity 7, 5. doi:10.12942/lrr-2004-5. Open PDF.
• Donoghue, John F. 1994. “General Relativity as an Effective Field Theory: The Leading Quantum Corrections.” Physical Review D 50, 3874–3888. doi:10.1103/PhysRevD.50.3874. Open PDF.
• Heemskerk, Idse, João Penedones, Joseph Polchinski, and James Sully. 2009. “Holography from Conformal Field Theory.” Journal of High Energy Physics 2009, 079. doi:10.1088/1126-6708/2009/10/079. Open PDF.