Renormalization and EFT for working researchers

Use this pathway when a calculation spans separated scales and you need more than the slogan “integrate out the heavy physics.” You will turn a physical question into an effective theory, match its coefficients, evolve them to the scale of the measurement, and report a prediction together with the reasons it should be trusted.

Required background. You should be able to manipulate linear maps and tensors, Fourier transforms and Green functions, complex expansions, and quantum operators. The corresponding preparation checks are available if one of those operations is rusty. You will also use perturbative rules, loop regularization, and renormalization-group reasoning.

Helpful background. Variational reasoning matters when field redefinitions are used. Statistical and numerical reasoning becomes essential when coefficients are fitted, correlated inputs are propagated, or a continuum extrapolation enters the result.

Start with the prediction, not the operator list

Before choosing a basis, write three lines:

1. Observable and kinematics. Name the amplitude, rate, response function, energy shift, or other quantity, including the external states and regime.
2. Scale hierarchy. Order every relevant scale and identify the small ratios: for example $E/M$, $m/M$, a coupling, a velocity, or a distance from a fixed point.
3. Accuracy target. State whether the result needs leading order, next order, leading logarithms, percent precision, or only a parametric estimate.

Those statements determine which fields remain dynamical, which operators can matter, and how far the expansion must be carried. If you cannot write them, the immediate task is to sharpen the research question—not to generate a larger Lagrangian.

Your first three actions are then concrete:

1. Match one amplitude or Green function near a scale $\mu_M$ of order the heavy threshold $M$.
2. Evolve the resulting coefficient vector to the characteristic low scale $\mu_L$.
3. Compute the target quantity at $\mu_L$ and vary unphysical scales and truncation choices to test the claimed accuracy.

Five controls make an EFT predictive

An effective theory is not defined by higher-dimension operators alone. Keep five connected choices visible throughout the calculation.

Degrees of freedom and symmetries

Retain modes that can go on shell, carry long-distance correlations, or are otherwise required in the regime. State the gauge, spacetime, internal, and discrete symmetries, including any controlled breaking. Integrating out a field removes it as a propagating low-energy degree of freedom; its effects remain in Wilson coefficients and nonlocal terms expanded within their domain.

Power counting and domain

Assign an order to fields, derivatives, masses, couplings, loops, and symmetry breaking. For a relativistic theory with heavy scale $\Lambda$, a typical local expansion is

$$\mathcal L_{\mathrm{EFT}} =\mathcal L_{d\le 4} +\sum_{d>4}\sum_i \frac{C_i^{(d)}(\mu)}{\Lambda^{d-4}}\,O_i^{(d)}(\mu).$$

Power counting tells you which terms form one accuracy order, including loop corrections and multiple insertions. The expansion parameter need not be just $E/\Lambda$; threshold velocities, light masses, thermal scales, or critical exponents can reorganize it. The organizing principle and its breakdown criterion must be stated together. See Manohar 2018, Chapters 3–4 for a systematic treatment of power counting with loops.

A nonredundant basis

Operators related by integration by parts, algebraic identities, or allowed local field redefinitions do not represent independent on-shell information. Choose a basis, state flavor and Hermiticity conventions, and preserve the translation to any basis used by data or another calculation. Equations of motion can simplify an EFT basis without changing on-shell observables, but they do not authorize deleting terms indiscriminately inside off-shell Green functions or at inconsistent orders; see Arzt 1995, pp. 189–195.

Matching and running

Matching fixes short-distance information by requiring the full and effective descriptions to agree for chosen low-energy quantities to a declared order. Running then transports that information between scales. These operations solve different problems: matching handles threshold physics, while the renormalization group resums scale logarithms and preserves independence from the arbitrary renormalization scale.

An error model

Track at least the first omitted EFT order, the next perturbative order, parametric inputs, numerical or sampling error, and any model dependence in nonperturbative matrix elements. Correlated contributions cannot automatically be added in quadrature. An error band is useful only when its construction is tied to the expansion and checked against order-by-order behavior.

Running example: remove one heavy scalar

Let $\phi$ be light and $H$ have mass $M$, with

$$\mathcal L =\frac12(\partial\phi)^2-\frac12m^2\phi^2 +\frac12(\partial H)^2-\frac12M^2H^2 -\frac g2 H\phi^2, \qquad E,m\ll M.$$

After integrating the heavy kinetic term by parts, its equation of motion is

$$(\Box+M^2)H=-\frac g2\phi^2.$$

At tree level, eliminating $H$ gives

$$\Delta\mathcal L_{\mathrm{EFT}} =\frac{g^2}{8}\, \phi^2\frac{1}{M^2+\Box}\phi^2 =\frac{g^2}{8M^2}\phi^4 -\frac{g^2}{8M^4}\phi^2\Box\phi^2 +O(M^{-6}).$$

This is an expansion of a propagator, not a claim that the full theory is local at all momenta. In the convention $\Delta\mathcal L=C_4\phi^4/4!$, matching gives $C_4=3g^2/M^2$. The three full-theory exchange channels yield

$$\mathcal M_{\mathrm{full}} =g^2\left(\frac{1}{M^2-s} +\frac{1}{M^2-t} +\frac{1}{M^2-u}\right) =\frac{3g^2}{M^2} +\frac{g^2(s+t+u)}{M^4}+O(E^4/M^6),$$

so the leading contact interaction has the correct sign, factor, and dimension. The derivative operator reproduces the next momentum dependence, up to integrations by parts and use of the light-field equations of motion. This simple calculation illustrates the low-momentum decoupling logic of Appelquist and Carazzone 1975, pp. 2856–2861.

At loop level, match renormalized quantities in the same scheme and with the same infrared regulator on both sides. Infrared contributions then cancel in the difference, leaving the short-distance coefficient. Matching near $\mu_M\sim M$ avoids large threshold logarithms; evolution to $\mu_L\sim E$ avoids large logarithms in the low-energy matrix element.

Operator mixing and scale independence

Write a column of renormalized operators $O(\mu)$ and a column of coefficients $C(\mu)$ so that the interaction is $C^T O$. If

$$\mu\frac{dO}{d\mu}=-\gamma O,$$

then scale independence requires

$$\mu\frac{dC}{d\mu}=\gamma^T C, \qquad \mu\frac{d}{d\mu}\bigl(C^T O\bigr)=0.$$

The transpose is not decoration: it follows from the chosen column-vector convention. At finite order, the derivative is zero only up to omitted terms, and its residual size is one useful diagnostic of truncation.

For one multiplicatively renormalized coefficient with

$$\mu\frac{dC}{d\mu}=\frac{\alpha}{4\pi}\gamma_0 C, \qquad \mu\frac{d\alpha}{d\mu}=-2\beta_0\frac{\alpha^2}{4\pi},$$

the leading-logarithmic solution is

$$C(\mu_L)=C(\mu_M) \left[\frac{\alpha(\mu_L)}{\alpha(\mu_M)}\right]^{-\gamma_0/(2\beta_0)}.$$

With several operators, replace the power by an evolution matrix and treat thresholds in stages. A coefficient by itself is basis- and scheme-dependent; only its consistently combined prediction has physical meaning.

Add the machinery your question actually needs

The five specialist chapters below answer different research problems. They are not a ceremonial sequence.

• Composite operators and mixing is the next stop when insertions renormalize as a coupled system, when contact terms matter, or when nonperturbative step scaling supplies the evolution.
• Fixed points and universality is needed when the prediction concerns a continuum limit, scaling exponent, crossover, or deformation of a critical theory. Separate universal eigenvalues and scaling functions from scheme-dependent coordinates.
• Matching and decoupling deepens threshold matching beyond the tree example, including multiple thresholds, nondecoupling effects, and on-shell versus off-shell strategies.
• Operator bases and field redefinitions is essential when two results use different bases, evanescent operators enter, or a claimed constraint may be a redundancy.
• Multiscale effective theories is required when several dynamical modes share a virtuality, overlap regions must be subtracted, or ordinary virtuality running leaves rapidity logarithms. Mode definitions and factorization become part of the claim.

Choose an application by its failure modes

Standard Model and precision data

Use Standard Model assembly and consistency when gauge representations, symmetry breaking, anomalies, flavor, and input schemes constrain the EFT. Record whether the theory is SMEFT, HEFT, or a more specific low-energy EFT; their degrees of freedom and power countings are not interchangeable.

Functional and nonperturbative methods

Use continuum functional equations when the calculation closes an infinite hierarchy by truncation. The dominant question is then whether symmetries, asymptotic limits, branch selection, and independent observables support that truncation.

Conformal flows

Use Weyl anomalies and conformal perturbation for deformations near a fixed point. Identify the perturbing operator, its dimension, the range over which the flow is controlled, and which anomaly or scaling data remain universal.

Thermal scales

Use thermal EFT, screening, and resummation when temperature generates hard, soft, and ultrasoft scales. Static dimensional reduction and real-time dissipative matching answer different questions; state which one the observable requires.

Quantum critical matter

Use quantum phase transitions and critical metals when patches of a Fermi surface, Landau damping, dangerously irrelevant couplings, or hyperscaling violation alter naive relativistic counting.

Gravity

Use gravity as an effective field theory when curvature and graviton loops are treated below a gravitational cutoff. Keep local counterterm coefficients distinct from universal long-distance nonanalytic effects, and state whether the metric is fixed, semiclassical, or quantized.

Produce a research-ready result sheet

For one prediction, save a compact sheet containing:

• the observable, external states, kinematic cuts, and scale hierarchy;
• retained degrees of freedom, symmetries, gauge choice, and regularization and renormalization schemes;
• the operator basis and the translation from any basis used by inputs;
• matching conditions at every threshold, with an infrared-consistency check;
• anomalous dimensions and evolution kernels, including coefficient/operator sign and transpose conventions;
• matrix elements or response functions at their evaluation scales;
• the central result, first omitted terms, parametric covariance, scale variations, and numerical convergence tests; and
• at least one independent limit, symmetry identity, alternative matching quantity, or benchmark.

A polished coefficient table without this chain is not yet a validated prediction. Conversely, a short leading-order result can be research-useful when its domain and error are explicit.

Exercises

1. Check the heavy-scalar matching

Derive the leading contact coefficient in the running example and show that it matches the low-energy full-theory amplitude.

Solution

Writing the heavy part as $-H(\Box+M^2)H/2-gH\phi^2/2$, its stationary value is $H=-g(M^2+\Box)^{-1}\phi^2/2$. Substitution gives

$$\Delta\mathcal L =\frac{g^2}{8}\phi^2(M^2+\Box)^{-1}\phi^2 =\frac{g^2}{8M^2}\phi^4+O(M^{-4}).$$

Because $4!/8=3$, the coefficient in $C_4\phi^4/4!$ is $C_4=3g^2/M^2$. Each of the $s$, $t$, and $u$ exchange channels contributes $g^2/M^2+O(E^2/M^4)$, so their sum agrees. The coefficient has mass dimension zero in the $C_4\phi^4/4!$ convention because $[g]=1$ and $[M]=1$.

2. Verify coefficient–operator cancellation

Starting from $\mu\,dO/d\mu=-\gamma O$ and $\mu\,dC/d\mu=\gamma^T C$, show that $C^TO$ is scale-independent.

Solution

Differentiate both factors:

$$\mu\frac{d}{d\mu}(C^TO) =(\gamma^TC)^TO-C^T\gamma O =C^T\gamma O-C^T\gamma O=0.$$

For a truncated anomalous dimension and matrix element, the cancellation holds only through the retained order. A residual of the next expected order is consistent; an unsuppressed residual usually signals mismatched conventions, schemes, or perturbative orders.

3. Translate an operator basis

Let a constant invertible matrix define $O'=RO$. Find the coefficient vector and anomalous dimension in the primed basis.

Solution

Invariance of the interaction requires

$$C^TO=C^TR^{-1}O'=C'^TO',$$

so $C'=R^{-T}C$. Differentiating $O'=RO$ gives

$$\mu\frac{dO'}{d\mu}=-R\gamma R^{-1}O',$$

and therefore $\gamma'=R\gamma R^{-1}$. These transformations preserve $C^TO$ and its RG cancellation. If $R$ depends on $\mu$, differentiating $R$ adds an extra term to $\gamma'$; omitting it would create spurious scale dependence.

Decide where to go next

You are ready to leave this pathway when you can reproduce the prediction from the result sheet, translate it between two stated bases or schemes, and show that scale variation and the next EFT order behave consistently with the error claim. If any of those checks fails, return to the first broken link—basis, matching, evolution, matrix element, or uncertainty—instead of restarting the whole subject.

For a process-level continuation, use the scattering phenomenology pathway. For a calculation whose main challenge is numerical evidence, continue to computational field theory.

References

• Appelquist, Thomas, and J. Carazzone. 1975. “Infrared Singularities and Massive Fields.” Physical Review D 11: 2856–2861. DOI.
• Arzt, Christopher. 1995. “Reduced Effective Lagrangians.” Physics Letters B 342: 189–195. DOI. Open PDF.
• Becher, Thomas, Alessandro Broggio, and Andrea Ferroglia. 2015. Introduction to Soft-Collinear Effective Theory. Lecture Notes in Physics 896. Cham: Springer. DOI. Open PDF.
• Manohar, Aneesh V. 2018. “Introduction to Effective Field Theories.” In Les Houches 2017: EFT in Particle Physics and Cosmology. arXiv:1804.05863.
• Weinberg, Steven. 1979. “Phenomenological Lagrangians.” Physica A 96: 327–340. DOI.