Breakdown, Matching, and UV Sensitivity

An EFT breaks down when an omitted contribution is no longer smaller than the retained accuracy—not automatically when a dimensionful quantity looks large. Matching identifies how heavy physics enters local Wilson coefficients, while massless cuts and symmetry relations remain low-energy predictions. At the next EFT order, universality therefore divides sharply: nonanalytic light-field terms are fixed, but new analytic coefficients require measurement or a specified ultraviolet model.

Required background. Flat-Spacetime Quantum Field Theory supplies thresholds, matching, and running. Gravitational Actions and Boundary Terms supplies the local operator basis.

Helpful background. Einstein Equations supplies the leading classical regime, Quantum Information supplies unitarity constraints, and Quantum Gravity supplies the scale and evidence hierarchy.

Control is observable dependent

For an observable computed through order $N$, write schematically

$$O=O^{(0)}\left[1+c_1\epsilon+c_2\epsilon^2+\cdots+c_N\epsilon^N\right] +\Delta_{\mathrm{nonanalytic}}.$$

The parameter $\epsilon$ may stand for $E^2/\Lambda^2$, $GE^2$, curvature divided by a cutoff squared, a velocity, or a threshold ratio. The truncation is controlled only when the first omitted contribution is bounded in the actual kinematics and state. Large Wilson coefficients can make $c_{N+1}\epsilon^{N+1}$ important early. Large logarithms can require renormalization-group improvement even when $\epsilon$ is small.

Breakdown can also be local or observable specific. A heavy total mass does not invalidate an exterior derivative expansion when curvature and transfer are small. A new light particle does invalidate a metric-only description near its threshold even if $GE^2\ll1$. Long-time secular growth, high occupation, forward enhancement, or a rapidly varying background can introduce scales absent from a simple center-of-mass estimate.

Results and limitations as of 2026-08-20

The cutoff is 2026-08-20. The entries compare controlled low-energy statements, not candidate ultraviolet completions as though they had equal empirical support.

Prediction or matching class	Degrees of freedom and observable	Evidence or result	Assumptions and scope	Limitation or UV sensitivity
Massless nonanalytic term	Declared light gravitons and matter; on-shell amplitude	Branch-cut logarithms and square roots are fixed by low-energy unitarity Donoghue 1994, §§II–V, assessed 2026-08-20	Transfer below heavy thresholds; complete light spectrum	Changes if a supposedly heavy field becomes light or the observable changes
Local higher-curvature coefficient	Metric EFT; polynomial momentum term	Renormalized coefficient absorbs UV poles and is measurable or matchable	Complete covariant basis and declared scheme	Finite part is UV-model dependent; Einstein theory alone does not predict it
Heavy-field decoupling	Light external states; low-energy S-matrix	Heavy exchange expands into local operators suppressed by powers of its mass Appelquist and Carazzone 1975, §§II–IV, assessed 2026-08-20	External invariants well below the heavy threshold	Expansion fails at the pole or cut; nondecoupling can occur when couplings scale with the heavy mass
Renormalization-group logarithm	Wilson coefficients and light loops	Running cancels scale dependence and can resum large logarithms	Perturbative beta functions and separated scales	Boundary values still require matching; strong running can end perturbative control
Extra pole from a truncated propagator	Finite set of higher-derivative operators	Algebraic resummation produces additional roots	Treating a finite derivative truncation as exact	Root position changes under omitted same-order operators; not reliable particle evidence near the cutoff

The conservative position says gravitational EFT should stop at the first uncontrolled ratio and remain agnostic about the ultraviolet theory; matching coefficients are inputs, not discoveries of microscopic gravity. A permissive position treats correlated Wilson-coefficient patterns, analyticity bounds, or symmetry-protected zeros as clues that can discriminate families of completions. These competing interpretations depend on the assumed light spectrum, locality, unitarity, regulator, matching scale, and observable. It remains unresolved which microscopic theory supplies nature’s coefficients.

A solvable heavy-field match

Take a light scalar $\phi$ and a heavy scalar $X$ of mass $M$, coupled by $gX\phi^2/2$. Tree-level exchange contributes a momentum kernel proportional to

$$\mathcal A_{\mathrm{full}}(q^2)=\frac{g^2}{M^2+q^2-i0}$$

in the mostly-plus convention, so the heavy timelike pole is at $q^2=-M^2$. For $\lvert q^2\rvert\ll M^2$,

$$\mathcal A_{\mathrm{full}}(q^2)= \frac{g^2}{M^2} \left[1-\frac{q^2}{M^2}+\frac{q^4}{M^4}+O\!\left(\frac{q^6}{M^6}\right)\right].$$

The EFT reproduces these terms with local operators $\phi^4$, $\phi^2\Box\phi^2$, and higher derivatives. Matching at tree level fixes their coefficients. Truncating after $q^2$ leaves a relative error of order $q^4/M^4$ away from accidental zeros.

This model exposes the boundary exactly. The Taylor series has radius set by the nearest singularity, $q^2=-M^2$. Near that pole no finite local expansion is adequate; $X$ must be retained explicitly. Changing the subtraction scheme redistributes finite local coefficients but cannot move the full-theory pole or change the matched low-energy amplitude through the retained order.

Now add a massless light loop. It can produce a term such as $q^2\log[(q^2-i0)/\mu^2]$, with reference mass $\mu$. No local counterterm can cancel its branch point. Its coefficient is determined by the light-field theory and lower-order couplings at the retained order; at higher orders, light loops with matched vertices can inherit UV dependence. If $X$ is made light, its own threshold becomes part of the low-energy nonanalytic structure and the original EFT field content has failed.

Matching and adverse checks

A controlled matching calculation follows six steps:

1. choose on-shell low-energy observables and external-state conventions;
2. compute them in the full model at invariants below every heavy threshold;
3. expand only the hard contribution analytically in those invariants;
4. compute the EFT with the same regulator and infrared content;
5. equate amplitudes to determine coefficients at a matching scale;
6. run coefficients and verify a second observable not used in the match.

Infrared terms must cancel between the two sides before coefficients are extracted. Otherwise a light logarithm can be mistakenly assigned to a Wilson coefficient. Gauge dependence must also cancel in on-shell matching; off-shell matching requires a declared field basis and generates redundant operators.

Adverse checks include varying the matching scale and scheme, changing the operator basis by field redefinitions, adding the next derivative order, and moving toward the nearest threshold. Physical predictions should remain stable within the estimated truncation error. A coefficient that changes under basis choice is not itself an observable, although invariant combinations entering amplitudes are.

When to stop the expansion

Stop or enlarge the EFT when any of the following occurs: an invariant reaches a heavy threshold; curvature or acceleration reaches an operator suppression scale; a Wilson coefficient makes an omitted term order one; perturbative running reaches strong coupling; secular terms grow to the leading size; or a state explores field values outside the expansion domain.

Do not diagnose breakdown by resumming a few higher-curvature terms and finding a pole. Near that pole, every omitted derivative operator can contribute comparably. A pole becomes credible only if protected by a controlled matching relation, stable under adding operators, and located parametrically below the remaining cutoff.

The inverse conclusion is equally important. Null sensitivity to a local coefficient does not test the whole ultraviolet theory, and agreement of one matched coefficient does not validate its high-energy spectrum. The EFT transmits only those UV features that survive in its Wilson coefficients and symmetry pattern.

Exercises

Exercise. In the heavy-scalar model, truncate the amplitude after the $q^2/M^2$ term. At $q^2=M^2/10$, estimate the relative truncation error. What changes if the same kinematics are continued toward the timelike pole $q^2=-M^2$?

Solution

Writing $x=q^2/M^2$, the exact bracket is $1/(1+x)$ and the truncation is $1-x$. Their relative difference is exactly

$$\frac{\left\lvert(1-x)-(1+x)^{-1}\right\rvert} {\left\lvert(1+x)^{-1}\right\rvert}=x^2.$$

Thus at $x=0.1$ the relative error is exactly one percent. As $x\to-1$, it approaches unity while the full amplitude diverges: no fixed finite Taylor truncation describes the pole. The EFT must then retain $X$ explicitly rather than adjusting finitely many local coefficients.

What could change these conclusions

The conclusions could change when an independent matching calculation changes a coefficient relation, a new light threshold alters a claimed universal term, a larger operator basis changes apparent pole stability, a controlled ultraviolet model yields a new discriminating low-energy relation, or evidence after 2026-08-20 changes the comparison.

References

• Appelquist, Thomas, and José Carazzone. “Infrared Singularities and Massive Fields.” Physical Review D 11 (1975): 2856–2861. DOI.
• Donoghue, John F. “General Relativity as an Effective Field Theory: The Leading Quantum Corrections.” Physical Review D 50 (1994): 3874–3888. DOI.