Effective Field Theory: Construction and Power Counting

An effective field theory (EFT) is a deliberately limited description: it retains the degrees of freedom that can be resolved in a stated regime and represents shorter-distance effects by local interactions. Its value is not that it is “approximately renormalizable,” but that a power counting turns an infinite allowed Lagrangian into a finite calculation at any declared accuracy.

This chapter builds that logic in dependency order. It first states what makes an EFT controlled, identifies when scale separation supports a local expansion, and uses degrees of freedom and symmetry to determine the allowed operators. It then derives predictive power counting, tests closure under loops, treats the special case in which shallow scales force nonperturbative iteration, and finishes by constructing truncation uncertainties and breakdown diagnostics. Matching the coefficients to a more microscopic theory belongs to the next chapter.

Enter through the calculation you need

If the immediate question is…	Start with…	Leave with…
Why can an unknown ultraviolet theory still yield systematic low-energy predictions?	Effective Field Theory as a Controlled Expansion	A declared observable, small parameter, retained order, and target accuracy
When is a derivative expansion local, and where does it stop working?	Scale Separation, Locality, and the Domain of an EFT	A threshold and analyticity check that distinguishes formal domain from practical breakdown
Which fields and interactions belong in the low-energy action?	Degrees of Freedom, Symmetry, and the Local Operator Expansion	Active degrees of freedom, symmetry realization, spurions, and allowed operator classes
Which trees, loops, and insertions contribute at a requested accuracy?	Power Counting and Predictive Order	A homogeneous counting rule and a complete contribution list at the target order
Why do higher-dimension interactions not destroy predictivity in loops?	Loops, Counterterms, and Closure of an EFT Expansion	The counterterm sector, coefficient running, and a retained-order scale-cancellation check
When must a nominally weak interaction be iterated?	Nonperturbative Iteration, Shallow Scales, and Power-Counting Consistency	A resummed amplitude, promoted interactions, and an order-by-order regulator test
How large are omitted terms, and what would falsify the assumed expansion?	EFT Truncation Errors and Breakdown Diagnostics	A decomposed uncertainty model, residual-scaling test, and breakdown criterion

The natural linear route is the order shown. Readers who already possess a well-defined EFT can enter at power counting, but only if its domain, fields, and symmetry realization are genuinely fixed. Readers approaching a shallow two-body pole should still establish ordinary loop closure before studying promoted iteration.

The control statement

Suppose an observable probes a characteristic low scale $Q$, while the nearest omitted dynamics enters at a hard scale $\Lambda_b$. The basic expansion parameter is

$$q\equiv\frac{Q}{\Lambda_b}<1.$$

The scale $Q$ can stand for external momentum, a light mass, an inverse size, or a combination of several homogeneous low scales. The breakdown scale $\Lambda_b$ need not equal the ultraviolet regulator and need not coincide exactly with the mass of the first omitted particle. It is the scale at which the assumptions supporting the chosen expansion cease to control the calculation.

For fields collectively denoted by $\phi$, write the local action as

$$\mathcal L_{\mathrm{EFT}} = \mathcal L_{\mathrm{LO}} + \sum_{\nu>0}\sum_i \frac{c_i^{(\nu)}(\mu)}{\Lambda_b^{d_i-d}} \mathcal O_i^{(\nu)}(\phi,\partial;m_{\mathrm{light}}).$$

The label $\nu$ is the EFT order assigned by the chosen power counting. It can depend on derivatives, light masses, weak couplings, loops, symmetry-breaking spurions, field multiplicity, or anomalous scaling. It is not automatically the canonical dimension $d_i$. At fixed target order $\nu_{\max}$, only finitely many operator insertions and loop topologies contribute, provided the counting is closed under renormalization.

A useful prediction therefore states at least six things:

1. the observable and kinematic domain;
2. the active degrees of freedom and symmetry realization;
3. the low and hard scales entering each expansion parameter;
4. the coefficient and field normalization;
5. the retained order and every contribution at that order; and
6. the model for omitted terms and the tests that would expose breakdown.

Without those declarations, “higher order” has no operational meaning. Weinberg’s general construction is to write the most general local Lagrangian consistent with the symmetries and then expand observables according to an ordering principle, not to discard interactions merely because their canonical dimension exceeds four Weinberg 1979, pp. 327–331.

The heavy-mediator thread

The chapter uses a transparent scalar model to keep the logic concrete. Let $\phi$ be a light real scalar and $H$ a heavy real scalar of mass $M$, with

$$\mathcal L = \frac12(\partial\phi)^2 +\frac12(\partial H)^2 -\frac12M^2H^2 -\frac g2H\phi^2.$$

At tree level, one exchange channel contributes a factor proportional to

$$\frac{g^2}{M^2-q^2} = \frac{g^2}{M^2} \left( 1+\frac{q^2}{M^2}+\frac{q^4}{M^4}+\cdots \right), \qquad |q^2|<M^2.$$

Each term is reproduced by a local operator with more derivatives. Keeping terms through $1/M^6$ gives an amplitude whose first omitted term scales as $Q^6/M^6$ relative to the leading result. The expansion is useful because a finite set of coefficients determines every low-energy process related by locality and symmetry, not because one has approximated a single number.

The geometric series also exposes the assumptions. Its convergence is limited by the nearest pole in the relevant complex invariant; crossed channels can have different distances to their nearest singularities. If a massless state is removed incorrectly, nonanalytic terms such as $\ln(-q^2)$ or $1/q^2$ cannot be represented by local coefficients. If the external energy crosses the heavy-particle threshold, the active field content and unitarity cuts change. Burgess develops the low-energy action and its relation to a hierarchy of particle states in Burgess 2021, §§ 1.1–1.3, pp. 6–16.

A reproducible calculation uses the fixed kinematics

$$s=E^2, \qquad t=-\frac{E^2}{3}, \qquad u=-\frac{2E^2}{3},$$

with $M=10$ and $g=1$. The static calculation on the chapter pages derives its reference coefficients and the expected slope-six residual. The calculation is a computational companion; the exposition and checks here remain complete without it.

Power counting is a closure claim

An EFT Lagrangian generally contains infinitely many symmetry-allowed terms. Predictivity comes from proving that a finite subset is sufficient through a chosen order. A candidate counting must answer two related questions.

First, enumeration: which tree graphs, loop graphs, operator insertions, and parameter corrections have the same order? Omitting any one of them makes the claimed accuracy inconsistent.

Second, closure: do ultraviolet divergences generated by those contributions lie in the span of operators included at that order or at a systematically higher order? If a loop requires a counterterm earlier than predicted, the counting must be revised. The coefficient running generated by those counterterms then cancels the explicit renormalization-scale dependence of matrix elements through the retained order.

This is why the old distinction between “renormalizable” and “nonrenormalizable” interactions does not separate predictive from unpredictive theories. An EFT with higher-dimension operators is renormalized order by order: new counterterms appear, but only finitely many are needed at each specified accuracy. Burgess derives this power-counting logic in cutoff and dimensional-regularization languages in Burgess 2021, §§ 3.1–3.3, pp. 52–79.

Canonical counting can fail when the dynamics generates a scale much smaller than $\Lambda_b$. A large scattering length, for example, makes repeated contact interactions as important as the first one. The leading interaction must then be summed while derivative corrections are inserted perturbatively according to a reorganized counting. Kaplan, Savage, and Wise exhibit such a controlled expansion around a shallow two-body scale in Kaplan, Savage, and Wise 1998, pp. 390–395. Iteration is therefore a consequence of a diagnosed infrared enhancement, not a generic improvement obtained by summing selected diagrams.

Errors, diagnostics, and validity limits

For an observable with an order-by-order series

$$X(Q)=X_{\mathrm{ref}}(Q) \sum_{n=0}^{\infty}c_n(Q)q^n,$$

the first omitted term suggests

$$\delta X_k(Q) \sim X_{\mathrm{ref}}(Q) q^{k+1} \times \text{a declared coefficient-size estimate}.$$

That is a conditional model, not a theorem and not a complete error budget. Input-parameter, numerical, matching, regulator, and model-discrepancy errors must be tracked separately. Correlations matter when the same omitted coefficient affects several energies or observables. Furnstahl, Phillips, and Wesolowski formulate Bayesian truncation models precisely by adding explicit assumptions about coefficient sizes and priors Furnstahl, Phillips, and Wesolowski 2015, §§ 2–4.

The chapter treats the following as empirical checks of the expansion:

• successive corrections scale with the declared powers of $q$ over a resolved window;
• residual regulator or renormalization-scale dependence moves to higher order;
• fitted coefficients remain stable when the energy and observable set changes;
• uncertainty intervals attain their advertised coverage on withheld data; and
• the inferred breakdown scale is compatible with known thresholds and singularities.

Stop extrapolating when $q$ is no longer small, when a new threshold or degree of freedom enters, when coefficients exhibit unexplained order-by-order growth, when residuals fail the predicted scaling, or when the claimed uncertainty repeatedly misses validation observables. The correct response is to revise the domain, degrees of freedom, or counting—not merely to enlarge an error bar without diagnosis.

Boundaries and handoffs

This chapter develops EFT domains, active content, symmetry-constrained local expansions, power counting, loop closure, promoted iteration, and truncation diagnostics. Independent operator-basis construction is treated in Operator Bases and Field Redefinitions, while Matching, Decoupling, and Threshold Evolution determines ultraviolet coefficients by comparing full and effective descriptions and evolving across thresholds.

The preceding Fixed Points, Universality, and Continuum Limits chapter explains RG eigenoperators. Their relevance can motivate an EFT hierarchy, but EFT order can also include kinematics, weak couplings, symmetry spurions, loop factors, or promoted infrared scales. “Irrelevant operator” and “higher-order EFT correction” are related statements only after a concrete counting has been declared.

Framework-specific applications—including chiral EFT, heavy-particle theories, nonrelativistic theories, SCET, SMEFT, gravitational EFT, and open-system EFT—are introduced and developed later in this volume. The present chapter supplies the construction and validation grammar that those applications must make explicit.

References

• Burgess, C. P. Introduction to Effective Field Theory: Thinking Effectively about Hierarchies of Scale. Cambridge: Cambridge University Press, 2021. DOI
• Furnstahl, R. J., D. R. Phillips, and S. Wesolowski. “A Recipe for EFT Uncertainty Quantification in Nuclear Physics.” Journal of Physics G: Nuclear and Particle Physics 42 (2015): 034028. DOI · Open PDF
• Kaplan, David B., Martin J. Savage, and Mark B. Wise. “A New Expansion for Nucleon–Nucleon Interactions.” Physics Letters B 424 (1998): 390–396. DOI · Open PDF
• Weinberg, Steven. “Phenomenological Lagrangians.” Physica A: Statistical Mechanics and its Applications 96 (1979): 327–340. DOI