Symmetry Constraints and the Space of Counterterms

Locality gives an upper bound on the counterterm space; symmetry usually makes that space much smaller. Ward identities relate Green functions in Abelian theories, while Slavnov–Taylor or Zinn-Justin identities encode BRST invariance after gauge fixing. They can forbid a local term, tie several renormalization constants together, or require a finite symmetry-restoring counterterm.

A failed identity is not automatically an anomaly. After subdivergences are removed, a regulator-induced breaking is first tested for locality and consistency. If it is the variation of another local functional, a finite local counterterm removes it. Only a nonzero breaking in a nontrivial cohomology class is a candidate genuine anomaly; establishing its coefficient and physical interpretation belongs to the anomaly analysis, not to a failed diagram check.

Required background. Renormalized Perturbation Theory and Counterterm Rules supplies the local action and order-by-order counterterm expansion. Slavnov–Taylor and Zinn-Justin Identities defines the gauge-fixed functional identity and its sources.

Helpful background. BRST Cohomology and Physical Observables explains closed and exact BRST classes. What Is an Anomaly? develops the physical classification of quantum obstructions and anomaly matching.

Symmetry selects a subspace of local counterterms

Suppose power counting and locality allow integrated local monomials $\mathcal O_i$ at the working order. Before imposing a quantum identity, the most general counterterm is

$$\Sigma_{\rm ct} = \sum_i c_i\int d^dx\,\mathcal O_i(x).$$

Discrete symmetries, Lorentz covariance, internal representations, ghost number, and mass dimension remove some coefficients immediately. A Ward or Slavnov–Taylor identity then imposes linear relations among those that remain. The correct object is the complete counterterm functional at a fixed loop order, not the counterterm assigned to one graph.

This distinction matters in gauge theory. Individual diagrams contributing to a photon or gluon self-energy need not be transverse, and individual subdivergence counterterms need not obey every final relation. The identity constrains the sum of graphs plus all lower-order counterterm insertions after subdivergences have been removed. Collins makes this order-by-order distinction explicit in the gauge-theory induction Collins 1984/2023, §§ 12.1–12.2, pp. 296–306.

For a linear Ward operator $\mathcal W$, invariant counterterms satisfy

$$\mathcal W\Sigma_{\rm ct}=0.$$

For gauge-fixed non-Abelian theories, the Slavnov functional $\mathcal S$ is nonlinear. Let $\Sigma$ be the classical gauge-fixed action, including sources for nonlinear BRST variations, with

$$\mathcal S(\Sigma)=0.$$

Linearizing around $\Sigma$ defines an operator $\mathcal B_\Sigma$. When the classical identity holds,

$$\mathcal B_\Sigma^2=0.$$

An invariant counterterm $\Sigma_{\rm inv}$ of ghost number zero must obey

$$\mathcal B_\Sigma\Sigma_{\rm inv}=0.$$

Solutions fall into two structural classes. Nontrivial ghost-number-zero cohomology classes change physical couplings or other invariant parameters. Exact terms,

$$\Sigma_{\rm exact} = \mathcal B_\Sigma\widehat\Sigma,$$

describe allowed field, source, or gauge-fixing redefinitions. Total derivatives and equations-of-motion terms are quotiented or retained according to the declared off-shell problem. This classification prevents “one coefficient per superficially divergent vertex” from overcounting the independent theory parameters.

The QED Ward identity gives a bounded test

Write the QED fields and coupling as

$$A_0^\mu=Z_3^{1/2}A^\mu, \qquad \psi_0=Z_2^{1/2}\psi, \qquad e_0=\mu^\epsilon Z_e e$$

in $d=4-2\epsilon$. Define $Z_1$ as the coefficient multiplying the renormalized interaction $e\bar\psi\gamma^\mu\psi A_\mu$ after substituting the fields:

$$Z_1=Z_eZ_2Z_3^{1/2}.$$

Ignoring the vacuum term and displaying the physical local structures, the counterterm Lagrangian can be parameterized as

$$\begin{aligned} \Delta\mathcal L ={}& -\frac14\delta_3F_{\mu\nu}F^{\mu\nu} +\delta_2\bar\psi\,i\gamma^\mu\partial_\mu\psi -\delta_m m\bar\psi\psi \\ &- \mu^\epsilon e\,\delta_1 \bar\psi\gamma^\mu\psi A_\mu +\Delta\mathcal L_{\rm gf}, \end{aligned}$$

where $\delta_i=Z_i-1$ for $i=1,2,3$, while the mass convention is stated separately. Gauge-fixing and BRST-exact terms are collected in $\Delta\mathcal L_{\rm gf}$.

Let $S_{\rm R}^{-1}(p)$ be the renormalized inverse electron propagator and let $\Gamma_{\rm R}^\mu(p+q,p)$ be the renormalized 1PI photon–electron vertex with the overall factor $e$ removed. The Ward–Takahashi identity is

$$q_\mu\Gamma_{\rm R}^\mu(p+q,p) = S_{\rm R}^{-1}(p+q)-S_{\rm R}^{-1}(p).$$

At zero momentum transfer,

$$\Gamma_{\rm R}^\mu(p,p) = \frac{\partial S_{\rm R}^{-1}(p)}{\partial p_\mu}.$$

The local divergent parts on the two sides have the same coefficient, so

$$Z_1=Z_2.$$

Using the definition of $Z_1$ then gives

$$Z_eZ_3^{1/2}=1.$$

Thus electron wave-function and vertex renormalization are not independent, and charge renormalization is fixed by the photon field renormalization in this convention. Ward’s original identity is the historical source of this QED constraint Ward 1950, p. 182.

The same vector identity requires the loop correction to the photon two-point function to be transverse:

$$q_\mu\Pi^{\mu\nu}(q)=0.$$

Its local ultraviolet part therefore has the form

$$\Pi_{\rm div}^{\mu\nu}(q) = a \left( q^2\eta^{\mu\nu}-q^\mu q^\nu \right),$$

which is generated by $F_{\mu\nu}F^{\mu\nu}$. A photon mass term $m_\gamma^2\eta^{\mu\nu}$ fails the test and is forbidden in the final symmetric counterterm action. The longitudinal gauge-fixing sector is handled separately and must not be confused with the transverse vacuum polarization.

These relations are stronger than power counting. Lorentz covariance and dimension alone would allow independent kinetic and vertex coefficients and would not by themselves exclude every gauge-breaking photon term. The Ward identity combines the electron kinetic and interaction terms into the covariant structure $\bar\psi i\gamma^\mu D_\mu\psi$ and restricts the photon term to $F^2$.

Regulator breaking versus a genuine obstruction

Assume all identities have been restored through loop order $n-1$. At order $n$, a regulated or subtracted calculation may give

$$\mathcal S(\Gamma) = \hbar^n\Delta^{(n)} +\mathcal O(\hbar^{n+1}).$$

Under the hypotheses of the quantum action principle, $\Delta^{(n)}$ is an integrated local polynomial with bounded dimension and ghost number one. Applying the linearized identity gives the consistency condition

$$\mathcal B_\Sigma\Delta^{(n)}=0.$$

There are then two cases.

Removable breaking. If a local ghost-number-zero functional $\Xi^{(n)}$ exists such that

$$\Delta^{(n)} = \mathcal B_\Sigma\Xi^{(n)},$$

then the finite redefinition

$$\Gamma^{(n)} \longrightarrow \Gamma^{(n)}-\Xi^{(n)}$$

restores the identity at that order. The restoring counterterm need not itself look invariant; its purpose is to cancel the regulator’s local breaking so that the final functional is invariant.

Candidate anomaly. If $\Delta^{(n)}$ is closed but not exact, it represents a class in

$$H^1(\mathcal B_\Sigma).$$

That is a necessary structural condition for an anomaly, not proof that the class appears with a nonzero coefficient. One must compute or otherwise determine the coefficient and impose any matter-content cancellation conditions. Piguet and Sorella develop the quantum action principle, stability problem, and BRST cohomology in precisely this order Piguet and Sorella 1995, chs. 2 and 4, pp. 21–36 and 59–79.

Vector QED is the bounded nonanomalous example. A momentum cutoff that obstructs loop-momentum shifts can produce a local violation of transversality or a local mismatch between the vertex and electron self-energy. If the complete breaking satisfies the consistency condition, finite local restoring terms can enforce the vector Ward identity. Dimensional regularization with a consistent vector-current treatment normally preserves it directly. Neither situation should be confused with the axial-current anomaly, and an anomalous chiral gauge symmetry cannot be made consistent by an arbitrary finite local patch.

A symmetry-restoration workflow

At each loop order:

1. Enumerate all local monomials allowed by dimension, field content, statistics, discrete symmetries, and ghost number.
2. Insert every lower-order counterterm required by subdivergences.
3. Compute the complete breaking of the Ward or Slavnov–Taylor identity, not a selected diagram.
4. Verify that the breaking is local. A nonlocal residue usually signals an omitted subdivergence, an unresolved infrared term, or inconsistent conventions.
5. Apply the consistency condition.
6. Decompose the local breaking into exact and nontrivial cohomology sectors.
7. Cancel the exact part with a finite local restoring counterterm.
8. Determine the coefficient of every nontrivial candidate and test anomaly cancellation.
9. Solve the homogeneous identity for the remaining invariant counterterms and fix their finite coefficients by renormalization conditions.
10. Re-evaluate the identity and the physical prediction through the declared order.

For QED, the corresponding validation record is:

Identity or sector	Local consequence	Diagnostic	Accepted resolution
Electron Ward identity	$Z_1=Z_2$	Compare vertex contraction with the inverse-propagator difference	Match the coefficients, including finite restoring terms
Photon two-point identity	$q_\mu\Pi^{\mu\nu}=0$	Contract the complete renormalized two-point function	Retain only the transverse $F^2$ structure in the loop correction
Photon mass	Forbidden in the final vector-gauge-invariant action	Test the constant $\eta^{\mu\nu}$ term	Cancel a regulator artifact locally; do not fit it as a new physical input
Gauge-fixing sector	BRST-exact and convention dependent	Track longitudinal terms separately	Renormalize consistently without treating them as observables
Breaking functional	Local, ghost number one, consistency closed	Check dimension, ghost number, and $\mathcal B_\Sigma\Delta=0$	Remove the exact part; classify any nontrivial remainder
Non-input prediction	Independent of restoring convention through the retained order	Repeat after an allowed finite symmetric redefinition	Agreement after translating renormalized inputs

This is the symmetry specialization of the chapter’s prediction-validation record. “Identity restored” must mean that the complete renormalized functional passes the displayed contractions, not merely that the pole terms vanished.

Assumptions and boundaries

The algebraic argument presumes perturbative locality, a valid quantum action principle, complete lower-order subtraction, and a well-defined BRST complex. Boundaries, nonlocal regulators, curved backgrounds, reducible gauge symmetries, open gauge algebras, and effective theories with infinitely many operators can enlarge the functional space, although the same loopwise logic often survives after the space is defined correctly.

An invariant counterterm can still be redundant by a field redefinition, equation of motion, or total derivative. Conversely, an off-shell identity may require source and BRST-exact counterterms that drop out of on-shell observables. Do not prune those terms with on-shell reasoning before the off-shell functional closes.

This page does not derive anomaly descent equations, classify gauge representations, or compute model-specific Standard Model restoring terms. Those tasks are developed in the symmetry, mathematical-QFT, and gauge-theory pages linked above.

Common pitfalls

Checking one diagram for transversality. Ward and Slavnov–Taylor identities constrain the complete order, including counterterm insertions. A single graph may fail while the sum is correct.

Equating local with symmetry allowed. A photon mass is local and power-counting allowed, yet the vector Ward identity forbids it in the final QED action.

Calling every regulator breaking an anomaly. First test whether the breaking is a local exact variation. Exact breakings are removed by finite counterterms.

Calling every cohomology class an anomaly. Nontrivial cohomology identifies possible structures. A genuine anomaly also requires a nonzero coefficient.

Discarding BRST-exact terms too early. They may be needed for off-shell closure, field renormalization, and gauge-parameter control even though they do not label physical observables.

Using $Z_1=Z_2$ without defining the $Z$ factors. Different authors distribute field factors between vertex and coupling constants differently. State $Z_1=Z_eZ_2Z_3^{1/2}$ before drawing the charge-renormalization conclusion.

Exercises

1. QED charge factor. Given $Z_1=Z_eZ_2Z_3^{1/2}$ and the Ward identity $Z_1=Z_2$, derive the charge relation.

Solution

Cancel $Z_2$ on the two sides to obtain $Z_eZ_3^{1/2}=1$, hence $Z_e=Z_3^{-1/2}$. The relation is convention dependent in form but not in content once the definitions are fixed.

2. Transverse polynomial. Show that

$$P^{\mu\nu}(q) = a(q^2\eta^{\mu\nu}-q^\mu q^\nu)$$

passes the photon Ward identity, while $b\,\eta^{\mu\nu}$ does not.

Solution

Contracting gives $q_\mu P^{\mu\nu}=a(q^2q^\nu-q^2q^\nu)=0$. The constant tensor gives $q_\mu b\eta^{\mu\nu}=bq^\nu$, which is nonzero unless $b=0$.

3. Exact breaking. If $\Delta^{(n)}=\mathcal B_\Sigma\Xi^{(n)}$, why does subtracting $\Xi^{(n)}$ not prove that all anomalies vanish?

Solution

It removes only the exact component of this breaking. A closed non-exact component may still remain in $H^1(\mathcal B_\Sigma)$, and its coefficient must be evaluated. Other theories or matter contents can have a nonzero nontrivial component.

Symmetric counterterm result

At fixed order, the admissible counterterm space is

$$\left\{ \text{local terms allowed by power counting} \right\} \cap \left\{ \text{solutions of the quantum identity} \right\}.$$

Homogeneous ghost-number-zero solutions parameterize invariant renormalizations; exact ghost-number-one breakings are removable; nontrivial ghost-number-one breakings are anomaly candidates. This hierarchy is what turns symmetry from a diagram-level hope into a reproducible counterterm constraint.

Continue to Renormalization Conditions, Schemes, and Finite Parts to see how finite symmetric counterterms implement different schemes. Continue to Regulator Removal and Renormalized Predictions to include the identity in the final removal test. Model-specific gauge-theory restoration belongs to the later gauge-theory volume.

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.

• Piguet, Olivier, and Silvio P. Sorella. 1995. Algebraic Renormalization: Perturbative Renormalization, Symmetries and Anomalies. Lecture Notes in Physics Monographs 28. Springer. DOI.

• Ward, J. C. 1950. “An Identity in Quantum Electrodynamics.” Physical Review 78: 182. DOI.