Composite Operators and Mixing

A local operator is not renormalized merely because the fields and couplings in the action have been renormalized. Bringing fields to the same point creates new short-distance singularities; bringing two already-renormalized insertions together creates further contact singularities. The resulting operators generally form a mixing system, and the scale dependence of that system must be compensated by the opposite evolution of any Wilson coefficients multiplying it.

This chapter develops that structure in a fixed order: define one insertion with an external source, renormalize coincident products, close an operator sector, derive its anomalous-dimension matrix, and then evolve coefficients in the dual space. Protected currents and nonperturbative step scaling are later branches. Throughout, a convention is accepted only when the coefficient–operator pairing and separated-point matrix elements pass explicit invariance checks.

Renormalized local insertions

This chapter develops the renormalization of local composite insertions and their finite-dimensional mixing sectors. Its basic objects are renormalized time-ordered Green functions

$$G_{a_1\cdots a_n}(x_1,\ldots,x_n) = \left\langle T\,[O_{a_1}](x_1)\cdots[O_{a_n}](x_n)\,\Phi_1\cdots\Phi_m \right\rangle_{\rm R},$$

where the square brackets emphasize that the local operators carry renormalization data in addition to the ordinary field, mass, and coupling counterterms. At mutually separated $x_i$, a linear operator-renormalization matrix is enough. On diagonals such as $x_i=x_j$, new local distributions and source counterterms are generally required.

The starting definitions of local operators, point splitting, equation-of-motion relations, and the free-field operator-product expansion belong to Local Operators and Short-Distance Structure. This chapter turns those objects into interacting renormalized insertions. It uses the local-counterterm machinery of Ultraviolet Renormalization and Locality and the current identities of Symmetry and Gauge Structure. It stops before process-specific form factors in Scattering Amplitudes and Observables, complete EFT basis construction later in this volume, and production lattice determinations in Lattice and Hamiltonian Field Theory.

The treatment is perturbative through the coefficient-evolution page. The final step-scaling page explains how a finite renormalization condition can instead be measured and continuum-extrapolated nonperturbatively; it does not supply ensembles or current numerical constants. Collins develops the source method, additional insertion counterterms, operator mixing, and composite-operator RG equations in a unified perturbative construction Collins 1984/2023, ch. 6, pp. 138–167, and § 7.12, pp. 219–222.

Check your preparation

This table routes study and is not a scored assessment. A ready answer should identify the object and its domain, not just recall a formula.

Can you perform this task?	Ready: enter here	Unsure: repair
Distinguish a local insertion from an ordinary interaction vertex integrated over spacetime	Renormalized insertions	Review Local and Composite Operator Insertions and ask whether the insertion momentum is independently fixed.
Explain why a product at $x\ne y$ need not determine its extension to $x=y$	Contact terms	Review Coincident Products and Contact Terms and separate diagonal-supported distributions from separated-point data.
Reduce a proposed operator list by quantum numbers, total derivatives, and equations of motion without confusing local and integrated equivalence	Mixing matrices	Review Local versus Integrated Operator Redundancies.
Differentiate a matrix inverse and track whether a matrix acts on a column from the left	Anomalous dimensions	Write indices on $O_{0a}=Z_{ab}O_b$ and derive the sign from $\mu\,dO_{0a}/d\mu=0$.
Preserve a scalar pairing under a basis change	Coefficient evolution	Starting from $C^{\mathsf T}O$, verify $O'=BO$ and $C'=B^{-\mathsf T}C$.
State the Ward identity that fixes the normalization of a current	Protected operators	Review Quantum Currents, Improvements, and Conservation.
Distinguish the continuum limit at fixed physical volume from a change of renormalization scale	Step scaling	Keep $L$ fixed while $a/L\to0$, then compare the renormalization conditions at $L$ and $sL$.

The first five pages form the core dependency chain. The protected-operator and step-scaling pages add depth but are not prerequisites for entering Renormalization-Group Equations and Running.

Choose a route

In the route column, ⇒ marks a required dependency and → a useful continuation.

Goal	Route	Result and stopping point
Renormalize one local observable	Insertions ⇒ contact terms	Define source derivatives, insertion counterterms, and the diagonal-supported freedom. Stop before introducing a matrix if the sector is genuinely one-dimensional and closed.
Construct a closed mixing sector	Insertions ⇒ mixing matrices	State quantum numbers, redundant sectors, $Z$ direction, and normalization conditions; verify every divergent insertion projects back into the declared sector.
Derive scaling operators	Mixing matrices ⇒ anomalous dimensions	Obtain $\gamma$, transform it under finite basis changes, and diagonalize only where that operation is justified.
Run Wilson coefficients	Core chain through anomalous dimensions ⇒ coefficient evolution	Derive the transpose and path ordering from scale invariance of $C^{\mathsf T}O$; stop after an invariant round trip.
Test a claimed protected current	Mixing matrices + current identities ⇒ protected operators	Distinguish exact protection, normalization by a Ward identity, improvement freedom, and anomaly or regulator-restoration qualifications.
Connect a low scale to perturbation theory nonperturbatively	Anomalous dimensions → step scaling	Define a finite-volume scheme, take each continuum limit, test composition, and convert only in an overlap window.
Explore matrix ordering computationally	Core chain through coefficient evolution → ordered-matrix benchmark	Compare operator evolution with its contragredient coefficient evolution and expose the failure caused by a reversed product or missing transpose.

The seven pages in order

1. Renormalized Composite-Operator Insertions couples a local operator to a source and differentiates the renormalized generating functional. It separates external-leg renormalization, operator normalization, vacuum subtraction, mixing, and contact terms in a one-loop scalar $\phi^2$ example.
2. Contact Terms and Renormalized Operator Products explains why separated-point products do not fix coincident products. It classifies the allowed delta-supported extensions and shows where they enter multiple source derivatives and Ward identities.
3. Operator Mixing and Renormalization Matrices constructs a closed sector using dimension, Lorentz representation, internal quantum numbers, BRST class, equations of motion, and total derivatives. Its two-operator benchmark includes a physical and a redundant direction.
4. Operator Anomalous-Dimension Matrices derives $\gamma$ from $Z$ with sign, index direction, and basis covariance explicit. It identifies fixed-point scaling operators without assuming that a scale-dependent matrix is everywhere diagonalizable.
5. Dual Evolution of Operators and Wilson Coefficients derives contragredient coefficient flow and the required path ordering. It verifies that a noncommuting two-step evolution preserves the coefficient–operator pairing.
6. Symmetry-Protected Operators, Currents, and Improvement uses Ward identities to distinguish exact protection from a convenient scheme choice. It treats conserved currents, stress-tensor improvement, redundant shifts, trace identities, and anomaly caveats.
7. Nonperturbative Renormalization Schemes and Step Scaling defines finite-volume or momentum-subtraction normalization without relying on weak coupling at the low scale. It separates continuum extrapolation, scale stepping, scheme conversion, the window problem, and systematic errors.

One convention from sources to coefficients

Let $O$ be a column vector of renormalized local operators and $O_0$ the corresponding bare vector. This chapter fixes

$$O_0=Z\,O, \qquad \gamma\equiv Z^{-1}\mu\frac{dZ}{d\mu}, \qquad \mu\frac{dO}{d\mu}=-\gamma O.$$

The first equality fixes the direction of $Z$; the second fixes the sign of $\gamma$; the third follows because $O_0$ is independent of $\mu$ at fixed bare data. A source term may be written as

$$\int d^dx\,J^{\mathsf T}O,$$

so the bare and renormalized sources obey the inverse-transpose relation needed to preserve that pairing. Likewise, if an effective interaction contains

$$\mathcal L_{\rm eff}\supset C^{\mathsf T}O,$$

then scale independence gives

$$\mu\frac{dC}{d\mu}=\gamma^{\mathsf T}C.$$

No transpose should be memorized in isolation: it is forced by the declared use of column vectors and the scalar pairing. A source that uses row operators, $O=Z O_0$, or the opposite definition of $\gamma$ will display different signs or transposes while making the same predictions if translated consistently.

Under a finite, possibly scale-dependent basis change

$$O'=B\,O, \qquad C'=B^{-\mathsf T}C,$$

the anomalous-dimension matrix becomes

$$\gamma' =B\gamma B^{-1} -\left(\mu\frac{dB}{d\mu}\right)B^{-1}.$$

For constant $B$, this is an ordinary similarity transformation. For coupling-dependent $B$, omitting the derivative term changes the RG equation. The invariants are the pairing $C^{\mathsf T}O$, complete matrix elements of the effective interaction, and—at a fixed point under an admissible finite basis change—the spectrum of scaling dimensions.

Closure includes more than physical operators

An operator sector is closed only if every ultraviolet divergence generated by its insertions can be cancelled within the declared set. Dimension and exact quantum numbers provide the first filter, but several qualifications matter:

• total derivatives can contribute to nonforward matrix elements even when their spacetime integrals vanish;
• equation-of-motion operators vanish in appropriate on-shell matrix elements but are needed in off-shell Green functions and field redefinitions;
• gauge-fixed calculations can require BRST-exact and equation-of-motion sectors before projection to physical cohomology;
• the identity operator absorbs vacuum or lower-point pieces when quantum numbers permit;
• coincident insertions require source-local contact terms not represented by the one-insertion matrix $Z$;
• dimensional regularization can require evanescent operators that vanish in four dimensions but feed finite physical coefficients after pole subtraction.

Consequently, “these are the operators I care about” is not a closure argument. A useful declaration separates the physical quotient, the larger renormalization sector, and the matrix elements or observables on which redundant directions disappear.

Triangular mixing also requires a qualifier. In a mass-independent scheme with no positive powers of a cutoff, canonical dimension and quantum numbers often give a block-triangular organization. Dimensionful masses, power-divergent regulators, or explicit symmetry breaking can permit mixing into lower-dimensional operators with compensating powers. The regulator, scheme, and parameter dimensions must therefore accompany any triangular matrix.

A scalar insertion thread

For a scalar theory, promote the mass parameter to a local source:

$$S\longrightarrow S+\frac12\int d^dx\,J(x)\phi^2(x).$$

One derivative with respect to $J(x)$ inserts $\phi^2(x)/2$. Ordinary field and coupling counterterms remove the subdivergences already present without the insertion, but the insertion vertex has its own ultraviolet singularities. Renormalizing the source-dependent action introduces a normalized operator $[\phi^2]$, possible mixing with all operators of allowed quantum numbers and dimension, and a source-linear vacuum counterterm.

Two derivatives with respect to $J$ expose a new fact. Even if each $[\phi^2](x)$ is finite by itself, the product as $x\to y$ can require local terms such as

$$\delta^{(d)}(x-y)[O_c](y), \qquad \partial^2\delta^{(d)}(x-y)[O_d](y),$$

with degree bounded by power counting and coefficients restricted by symmetry. These terms vanish at separated points but are essential in integrated identities, susceptibilities, and source derivatives. They cannot be reconstructed by multiplying the one-insertion $Z$ factors.

After a closed basis is chosen, the same matrix controls three views of the physics:

1. $Z$ cancels short-distance poles in operator insertions.
2. $\gamma$ describes how the renormalized basis changes with $\mu$.
3. $\gamma^{\mathsf T}$ evolves coefficient coordinates so that $C^{\mathsf T}O$ is unchanged.

This is the bridge from local renormalization to the running and matching chapters. The operator matrix belongs here; the extraction of short-distance coefficients from a full theory belongs to Matching, Decoupling, and Threshold Evolution.

Convention and validation card

Every matrix calculation in the chapter uses the following fields. A result with any field omitted is not convention-complete.

Field	Chapter convention or required declaration	Check
Operator orientation	$O$ is a column	Write one indexed equation before using matrix notation.
Renormalization direction	$O_0=ZO$	Re-expand the renormalized insertion and verify cancellation of its poles.
Anomalous dimension	$\gamma=Z^{-1}\mu\,dZ/d\mu$	Differentiate $O_0$ and recover $\mu\,dO/d\mu=-\gamma O$.
Coefficient pairing	$\mathcal L_{\rm eff}\supset C^{\mathsf T}O$	Verify $\mu\,d(C^{\mathsf T}O)/d\mu=0$ through the retained order.
Coefficient flow	$\mu\,dC/d\mu=\gamma^{\mathsf T}C$	A missing transpose must fail for a nonsymmetric test matrix.
Basis change	$O'=BO$, $C'=B^{-\mathsf T}C$	Perform an exact round trip and recover the original pairing.
Sector declaration	physical, EOM, total-derivative, BRST-exact, evanescent, identity, and contact sectors named as applicable	Insert every basis element and show that divergent projections remain in the declared enlarged sector.
Contact products	diagonal-supported terms recorded separately from one-insertion $Z$	Compare source derivatives at separated and coincident points.
Protection	identity, normalization condition, regulator, anomaly assumption, and improvement freedom stated	Test the complete Ward identity, not only one matrix entry.
Nonperturbative running	finite-volume condition, continuum limit, scale factor, composition, conversion window, and uncertainties stated	Take the continuum limit at every step and verify composition within errors.

A reproducible calculation uses the deterministic piecewise matrices

$$\gamma_A= \begin{pmatrix} 1&2\\ 0&-1 \end{pmatrix}, \qquad \gamma_B= \begin{pmatrix} 0&1\\ -1&0 \end{pmatrix},$$

on successive RG-time intervals. Because they do not commute, reversing the evolution factors changes the answer. Evolving $O$ with $-\gamma$ and $C$ with $\gamma^{\mathsf T}$ must nevertheless preserve $C^{\mathsf T}O$. The chapter derives that benchmark analytically before offering it as an executable check.

Limits of the chapter

The operator formalism does not make all local products observables. Gauge-variant insertions, off-shell Green functions, and individual Wilson coefficients can be scheme and basis dependent. Physical meaning attaches to symmetry-qualified matrix elements and complete coefficient–operator combinations.

Nor does diagonalizing a perturbative anomalous-dimension matrix prove that the corresponding operators exist as exact scaling operators away from a fixed point. Matrices at different scales may not commute, degeneracies can produce Jordan structure, and a running basis change contributes its own connection term.

Finally, a nonperturbative renormalization condition does not by itself remove lattice artifacts or guarantee perturbative matching. The continuum limit, finite-volume effects, symmetry restoration, scale-setting covariance, and a genuine overlap window must all be demonstrated.

Review the chapter

Use these prompts to test whether the operator conventions and closure conditions are secure.

Capability	Prompt	Successful response and repair
Explanation — extra renormalization	Explain why renormalized elementary fields do not automatically make $\phi^2(x)$ finite.	Identify the new coincident contraction and source-local counterterms. Repair in renormalized insertions.
Classification — contacts	List the contact terms allowed when two scalar insertions approach coincidence.	Use support, dimension, Lorentz symmetry, and internal quantum numbers; distinguish them from separated-point OPE coefficients. Repair in contact terms.
Construction — closure	Given two physical operators and one EOM operator, decide whether a displayed $2\times2$ matrix is sufficient.	State which Green functions and quotient are intended, then include the redundant direction if off-shell closure requires it. Repair in mixing matrices.
Convention translation	A reference defines $O=Z_{\rm ref}O_0$ and $\gamma_{\rm ref}=-\mu(dZ_{\rm ref}/d\mu)Z_{\rm ref}^{-1}$. Translate it to the chapter convention.	Invert $Z$, track index orientation, and recover the same operator evolution. Repair in anomalous dimensions.
Ordering test	Solve two constant noncommuting intervals and decide which exponential acts first.	The earlier operator evolution appears on the right; the coefficient product is its inverse transpose. Repair in coefficient evolution.
Symmetry check	A current has a zero diagonal entry in one computed matrix. Is it protected?	Demand the exact Ward identity, closure, mixing with total derivatives or BRST-exact operators, and anomaly assumptions. Repair in protected operators.
Evidence design	Design a step-scaling determination between $\mu$ and $4\mu$.	Define the finite-volume observable, take two separate continuum limits, test the two-step composition, and state the conversion window and covariance. Repair in step scaling.

Continue from a closed operator sector

• Continue to Renormalization-Group Equations and Running for beta functions, Callan–Symanzik equations, coupled flows, and RG improvement.
• Continue to Effective Field Theory: Construction and Power Counting when the task is to organize an infinite symmetry-allowed expansion by a small parameter.
• Continue to Matching, Decoupling, and Threshold Evolution when coefficient boundary conditions must be extracted from a more microscopic theory.
• Continue to Operator Bases and Field Redefinitions for complete basis certification, Hilbert-series counting, evanescent closure, and automated translation.
• Continue to Lattice and Hamiltonian Field Theory for ensemble generation, continuum extrapolation, and production nonperturbative matrix elements.
• Return to the Volume 5 overview to choose another route.

References

• Collins, John C. 1984; open-access reissue 2023. Renormalization: An Introduction to Renormalization, the Renormalization Group and the Operator-Product Expansion. Cambridge Monographs on Mathematical Physics. Cambridge University Press. DOI and Open PDF.