Slavnov–Taylor and Zinn-Justin Identities

BRST invariance becomes useful for quantum Green functions only after the nonlinear variations are treated as composite operators with their own sources. A BRST change of variables then gives one identity for the connected functional and, after a Legendre transform, the quadratic Zinn–Justin equation

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

Differentiating it produces the Slavnov–Taylor hierarchy among 1PI vertices. Linearizing it produces a nilpotent consistency differential that tests local counterterms at ghost number zero and possible anomalies at ghost number one. The discussion below is local and perturbative: it assumes a BRST-stable field domain and makes every regulator, measure, boundary, and cohomological qualification explicit.

Required background. The BRST Differential and Gauge-Fixed Complex supplies the off-shell nilpotent differential, the retained $b$ field, and the gauge-fixed action. Current Sources and Generating Functionals supplies source differentiation, the connected functional, and the ordinary field Legendre transform.

Helpful background. BRST Cohomology and Physical Observables distinguishes global functional cohomology from local cohomology modulo total derivatives and explains why ghost numbers zero and one play different roles.

Nonlinear BRST variations need composite sources

Use the off-shell four-field complex

$$\begin{aligned} sA_\mu^a&=(D_\mu c)^a, & sc^a&=-\frac g2f^{abc}c^bc^c, \\ s\bar c^a&=b^a, & sb^a&=0. \end{aligned}$$

The first two variations are composite operators. Their renormalization cannot be recovered by coupling sources only to $A,c,\bar c,b$. Introduce instead the inert Zinn sources $K^{a\mu}$ and $L^a$:

object	parity	ghost number	source term
$K^{a\mu}$	odd	$-1$	$K^{a\mu}(sA_\mu^a)$
$L^a$	even	$-2$	$L^a(sc^a)$

The source terms are even and have ghost number zero. They are external: no integral over $K$ or $L$ is introduced.

For a linear covariant gauge, take

$$F^a[A]=\partial^\mu A_\mu^a, \qquad (M_Ac)^a=\partial^\mu(D_\mu c)^a,$$

and retain the auxiliary field. The extended classical functional is

$$\begin{aligned} \Sigma={}&S_{\mathrm{YM}} +\int\mathrm d^d x\, \left[ b^aF^a+\frac\xi2b^ab^a -\bar c^a(M_Ac)^a \right] \\ &+\int\mathrm d^d x\, \left[ K^{a\mu}(D_\mu c)^a -\frac g2L^af^{abc}c^bc^c \right]. \end{aligned}$$

The last line is simply $\int(KsA+Lsc)$. Since $K$ is odd and $L$ is even, the left Leibniz rule and $sK=sL=0$ give

$$s(KsA)=0, \qquad s(Lsc)=0.$$

Consequently $s\Sigma=0$ off shell. Sources for nonlinear BRST variations and the resulting quadratic identity are developed in Zinn-Justin 2021, § 26.8, pp. 639–641, eqs. (26.100)–(26.114) and in the current review Bélusca-Maïto et al. 2023, § 2.3, pp. 12–15, eqs. (46)–(63), published Open PDF.

One derivative convention fixes every sign

For every field or source $z$, use a left functional derivative defined by

$$\delta X =\int\mathrm d^d x\, \delta z\, \frac{\delta_LX}{\delta z}.$$

Variations are moved to the left before the derivative is read off. Source factors also precede the composite operators in $\Sigma$. These two declarations are enough to fix the signs below. To keep displays readable, write

$$X_z\equiv\frac{\delta_LX}{\delta z}.$$

For comparison, a right derivative is defined by $\delta X=\int(\delta_RX/\delta z)\,\delta z$. When $X$ is even and $z$ is odd, $\delta_RX/\delta z=-\delta_LX/\delta z$; importing a right-derivative formula without this conversion is a sign error.

For an even functional $X$ of ghost number zero, define the Slavnov functional

$$\mathcal S(X) =\int\mathrm d^d x\, \left[ X_{K^{a\mu}}X_{A_\mu^a} +X_{L^a}X_{c^a} +b^aX_{\bar c^a} \right].$$

At tree level,

$$\Sigma_{K^{a\mu}}=sA_\mu^a, \qquad \Sigma_{L^a}=sc^a,$$

so the ordering in $\mathcal S$ gives

$$\mathcal S(\Sigma) =(sA)\Sigma_A+(sc)\Sigma_c+b\Sigma_{\bar c} =s\Sigma=0.$$

Changing to right derivatives or placing a source after its composite operator changes intermediate signs. It does not change the identity once one convention is used consistently.

A BRST change of variables becomes a 1PI equation

Add ordinary sources in source-before-field order,

$$S_J =\int\mathrm d^d x\, \left( J_AA+\bar\eta c+\eta\bar c+J_bb \right),$$

where $J_A,J_b$ are even and $\bar\eta,\eta$ are odd. With the Lorentzian convention defined on the source-functional page,

$$\begin{aligned} Z[J;K,L] &=\int\mathcal D\Phi\, e^{i(\Sigma+S_J)}, \\ W[J;K,L] &=-i\log\frac{Z[J;K,L]}{Z[0;0,0]}. \end{aligned}$$

The following hypotheses are part of this equation, not consequences of it:

• the regulator and renormalized composite insertions respect the declared BRST differential, or any breaking is retained explicitly;
• the measure has unit BRST Jacobian and no anomaly at the order considered;
• the integration cycle and field domain are BRST stable, with no unaccounted boundary in field space;
• spacetime boundary conditions remove the relevant BRST-current flux or add the required boundary degrees of freedom.

Under those hypotheses, the change of variables $\Phi\mapsto\Phi+\varepsilon s\Phi$ has vanishing integral. The ordinary source term varies as

$$sS_J =\int\mathrm d^d x\, \left[ J_A(sA)-\bar\eta(sc)-\eta b \right].$$

The two minus signs come from moving the left odd differential through an odd source. Because $K$ and $L$ generate the composite insertions, the connected identity is

$$\int\mathrm d^d x\, \left[ J_AW_K -\bar\eta W_L -\eta W_{J_b} \right]=0.$$

Now Legendre transform only the ordinary sources. With

$$A=W_{J_A}, \quad c=W_{\bar\eta}, \quad \bar c=W_\eta, \quad b=W_{J_b},$$

set

$$\Gamma[\Phi;K,L] =W-\int\mathrm d^d x\, \left( J_AA+\bar\eta c+\eta\bar c+J_bb \right).$$

At fixed $K,L$, the declared left-derivative convention gives

$$\Gamma_A=-J_A, \qquad \Gamma_c=\bar\eta, \qquad \Gamma_{\bar c}=\eta, \qquad \Gamma_b=-J_b.$$

The plus signs for the odd mean fields follow by moving $\delta c$ or $\delta\bar c$ to the left of its odd source. Substitution into the connected identity yields

$$\boxed{\mathcal S(\Gamma)=0.}$$

$K$ and $L$ remain external arguments of $\Gamma$; Legendre transforming them would define a different object. The formal change-of-variables and Legendre steps are derived in Bélusca-Maïto et al. 2023, § 2.5, pp. 19–21, eqs. (82)–(93), published Open PDF.

This compact equation is usually called the Zinn–Justin equation for the 1PI functional. Its field derivatives generate the Slavnov–Taylor identities among individual 1PI vertices. Historically, Taylor and Slavnov derived the generalized Yang–Mills Ward identities independently Taylor 1971, pp. 436–444; Slavnov 1972, pp. 99–104. Zinn-Justin’s source formulation packages the nonlinear hierarchy into one quadratic functional equation Zinn-Justin 1975, pp. 1–39.

Classical invariance is not yet the quantum identity

Three statements should remain separate:

level	equation	additional content
classical action	$\mathcal S(\Sigma)=0$	off-shell BRST algebra and an invariant classical domain
regulated integral	connected source identity	measure, regulator, contour, and boundary control
renormalized 1PI functional	$\mathcal S(\Gamma)=0$	compatible subtraction and restoration of composite identities

Writing the last line does not prove it. A regularization may break BRST even when the classical differential is nilpotent, and a nontrivial anomaly may prevent restoration.

Linearization supplies the consistency differential

Let $X$ be even of ghost number zero. For an even ghost-number-zero variation $Y$, define the linearization by the directional derivative

$$\mathcal B_XY \equiv \left. \frac{\mathrm d}{\mathrm d\tau} \mathcal S(X+\tau Y) \right|_{\tau=0},$$

with $\tau$ even. Expanding without reordering odd factors gives

$$\begin{aligned} \mathcal B_XY =\int\mathrm d^d x\Bigl[ &X_KY_A+Y_KX_A \\ &+X_LY_c+Y_LX_c +bY_{\bar c} \Bigr]. \end{aligned}$$

Indices and spacetime arguments are suppressed in these displays. As a functional differential operator,

$$\begin{aligned} \mathcal B_X =\int\mathrm d^d x\Bigl[ &X_K\frac{\delta_L}{\delta A} +X_A\frac{\delta_L}{\delta K} \\ &+X_L\frac{\delta_L}{\delta c} +X_c\frac{\delta_L}{\delta L} +b\frac{\delta_L}{\delta\bar c} \Bigr]. \end{aligned}$$

This operator display extends $\mathcal B_X$ to an arbitrary homogeneous functional. The even directional derivative above is only its ghost-number-zero linearization.

At the classical solution,

$$\begin{aligned} \mathcal B_\Sigma A_\mu^a&=sA_\mu^a, &\mathcal B_\Sigma c^a&=sc^a, \\ \mathcal B_\Sigma\bar c^a&=b^a, &\mathcal B_\Sigma b^a&=0. \end{aligned}$$

Thus $\mathcal B_\Sigma$ reduces to $s$ on functionals that do not depend on $K,L$. The graded Jacobi relation behind the quadratic functional implies

$$\mathcal S(X)=0 \quad\Longrightarrow\quad \mathcal B_X^2=0.$$

This is why the linearization is the correct differential for perturbations of a solution. The construction, its nilpotency, and its counterterm use are given in Bélusca-Maïto et al. 2023, § 6.1, pp. 74–76, eqs. (289)–(302), published Open PDF and summarized cohomologically in Barnich, Brandt, and Henneaux 2000, § 2.6, p. 16, eqs. (2.32)–(2.38), arXiv v3 Open PDF.

Gauge fixing contributes independent functional equations

The Slavnov equation is not the only condition on $\Gamma$. In a linear gauge, the retained $b$ field gives the gauge-fixing equation

$$\frac{\delta_L\Gamma}{\delta b^a} =F^a[A]+\xi b^a.$$

The dependence on $\bar c$ and $K$ also obeys the local antighost equation

$$\frac{\delta_L\Gamma}{\delta\bar c^a} +\partial^\mu \frac{\delta_L\Gamma}{\delta K^{a\mu}} =0.$$

At tree level its sign is immediate:

$$\Sigma_{\bar c^a}=-(M_Ac)^a, \qquad \partial^\mu\Sigma_{K^{a\mu}}=(M_Ac)^a.$$

In a subtraction scheme that preserves these linear identities, their right-hand sides are not replaced by arbitrary loop corrections. They remove counterterms that would satisfy the Slavnov equation alone. Zinn-Justin derives the linear-gauge auxiliary and ghost-field equations at Zinn-Justin 2021, § 26.10.5, pp. 646–647, eqs. (26.140)–(26.147).

There is also a stronger integrated ghost equation in Landau gauge. With the all-left convention used here, define

$$\mathcal G^a =\int\mathrm d^d x\, \left[ \frac{\delta_L}{\delta c^a} -g f^{abc}\bar c^b \frac{\delta_L}{\delta b^c} \right].$$

For $\xi=0$,

$$\mathcal G^a\Sigma=\Delta_{\mathrm{cl}}^a.$$

On a shift-admissible domain and in a subtraction scheme preserving this linearly broken Ward identity, the quantum equation is likewise

$$\mathcal G^a\Gamma=\Delta_{\mathrm{cl}}^a.$$

Here the source-linear breaking is

$$\Delta_{\mathrm{cl}}^a =-g f^{abc}\int\mathrm d^d x\, \left( K^{b\mu}A_\mu^c-L^bc^c \right).$$

The signs reverse in the familiar right-derivative convention. For $\xi\ne0$, applying $\mathcal G^a$ to $\xi b^2/2$ produces $-\xi g f^{abc}\bar c^bb^c$, so this simple linearly broken equation is special to Landau gauge. “Ghost equation” and “antighost equation” are not interchangeable labels unless the fields and derivative convention have been declared. The integrated equation also assumes that a constant ghost shift is an admissible variation and that its boundary term vanishes. It therefore does not act on the based bounded-region complex below, where a nonzero constant violates $c|_{\partial\Sigma_{\mathrm{sp}}}=0$. The Landau-gauge ghost Ward identity is developed, in a different derivative and source convention, in Blasi, Piguet, and Sorella 1991, pp. 154–162.

Counterterms and anomalies occupy different ghost numbers

Suppose the renormalized identity is valid through order $\hbar^{n-1}$. A candidate local counterterm $\mathcal C^{(n)}$ at order $n$ must satisfy

$$\mathcal B_\Sigma\mathcal C^{(n)}=0, \qquad \operatorname{gh}\mathcal C^{(n)}=0,$$

together with the gauge-fixing, antighost, ghost, power-counting, global symmetry, and boundary restrictions of the theory. Counterterms differing by

$$\mathcal C^{(n)} \longmapsto \mathcal C^{(n)}+\mathcal B_\Sigma\Xi^{(n)}$$

lie in the same linearized cohomology class. Exact terms commonly encode allowed field, source, or gauge-fixing redefinitions; nontrivial ghost-number-zero classes encode invariant couplings or deformations. Full stability still requires showing that the available parameters span every allowed class.

Now suppose instead that the identity first breaks at order $n$:

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

Under the hypotheses of the Quantum Action Principle, $\Delta^{(n)}$ is an integrated local functional of ghost number one. Linearized consistency gives

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

If

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

then adding the finite local counterterm $-\hbar^n\widehat\Delta^{(n)}$ restores the identity at that order. A nontrivial ghost-number-one class is a candidate anomaly. Cohomology alone does not determine its coefficient, prove that it is generated, or detect a global anomaly outside the local functional complex.

For the source-dependent local complex used on this page, the relevant sectors are $H^{0,d}(\mathcal B_\Sigma\mid\mathrm d)$ and $H^{1,d}(\mathcal B_\Sigma\mid\mathrm d)$. Their reduction to ordinary $s$-cohomology uses the standard source and doublet hypotheses. On a boundary, a $\mathrm d$-exact term is trivial only when its surface integral vanishes or an allowed boundary counterterm removes it. The corresponding local classes are analyzed in Barnich, Brandt, and Henneaux 2000, §§ 12.2–12.3, pp. 117–121, arXiv v3 Open PDF. The loopwise locality, consistency, and restoration logic is reviewed in Bélusca-Maïto et al. 2023, §§ 6.2.1–6.2.3, pp. 77–81, eqs. (303)–(317), published Open PDF.

That structural review treats local perturbative algebraic restoration; it does not claim that a suitable regulator always exists, that every gauge theory is anomaly-free, or that the construction defines a nonperturbative quotient. A 2026 functional-renormalization-group calculation instead uses a modified Ward–Takahashi/Slavnov–Taylor identity at nonzero cutoff and recovers the ordinary identity only as that cutoff is removed Echigo et al. 2026, § 1, pp. 1–3. Its QED result is truncation- and approximation-specific, not a replacement for the local algebraic theorem.

Bounded Maxwell theory checks the identity without hiding a boundary term

Return to a smooth bounded connected spatial region $\Sigma_{\mathrm{sp}}$. Use the based identity component, $c|_{\partial\Sigma_{\mathrm{sp}}}=0$, zero tangential pullback of $A$, the smooth BRST-stable core, and the direct $L^2$ codomain pairing for $(\bar c,b)$ from the BRST differential page. Assume the remaining temporal and electric boundary data make the action and BRST change of variables well defined. Concrete BRST-compatible electromagnetic boundary sets with the auxiliary field retained are exhibited in Moss and Silva 1997, § III, open-manuscript pp. 7–8, eqs. (30), (31), (33), and (37)–(38), arXiv v1 Open PDF, though their Euclidean one-loop setting is not itself a proof of the quantum identity used here.

For Coulomb gauge,

$$F[A]=\partial_iA_i, \qquad M_0=\Delta_D=\partial_i\partial_i$$

on the based Dirichlet ghost domain. Maxwell theory has $sc=0$, so no $L$ source is needed. Its extended functional is

$$\begin{aligned} \Sigma_{\mathrm M} =S_{\mathrm M}+\int\mathrm dt\Bigl[ &\langle b,\partial_iA_i\rangle +\frac\xi2\langle b,b\rangle \\ &-\langle\bar c,\Delta_Dc\rangle +\langle K^\mu,\partial_\mu c\rangle \Bigr]. \end{aligned}$$

No spatial integration by parts is needed. Direct application of $s$ gives

$$\begin{aligned} s\langle b,\partial_iA_i\rangle &=\langle b,\Delta_Dc\rangle, \\ s\bigl[-\langle\bar c,\Delta_Dc\rangle\bigr] &=-\langle b,\Delta_Dc\rangle, \end{aligned}$$

while the other terms are closed. Hence

$$\mathcal S(\Sigma_{\mathrm M})=s\Sigma_{\mathrm M}=0.$$

The subsidiary equations are equally transparent:

$$\begin{aligned} (\Sigma_{\mathrm M})_b &=\partial_iA_i+\xi b, \\ (\Sigma_{\mathrm M})_{\bar c} &=-\Delta_Dc, \\ (\Sigma_{\mathrm M})_{K^\mu} &=\partial_\mu c. \end{aligned}$$

Thus the antighost equation is simply $-\Delta_Dc+\partial_i(\Sigma_{\mathrm M})_{K^i}=0$. The admissible local ghost variation is instead the weak equation

$$\delta_c\Sigma_{\mathrm M}[\delta c] =-\langle\bar c,\Delta_D\delta c\rangle +\langle K^\mu,\partial_\mu\delta c\rangle,$$

for Dirichlet $\delta c$; no constant shift or spatial integration by parts has been used. At a finite mode regulator chosen to preserve this linear differential and the boundary domain, the same Gaussian identity holds for $\Gamma$; a field-independent determinant normalization does not create new 1PI vertices.

There is nevertheless a nontrivial 1PI check. Differentiate the Maxwell identity once with respect to $c$ and once with respect to $A_\nu$, then set fields and sources to zero. For an admissible Dirichlet ghost test $u$ and gauge-field test $v$, the direct weak form is

$$\left\langle \partial_\mu u, \Gamma^{(2)}_{A_\mu A_\nu}v_\nu \right\rangle=0.$$

Thus the $AA$ Hessian with $b$ retained annihilates based gauge directions in its first slot. The full coupled $(A,b)$ block supplies the gauge-fixing mixing; its invertibility must be checked after the temporal and boundary domains are specified. Eliminating $b$ instead adds a longitudinal term and changes this off-shell form, so the reduced $AA$ inverse should not be inferred from the displayed identity.

The three descriptions now say different things:

reading	bounded Maxwell statement
orbit	$sA_i=\partial_ic$ is the odd tangent to the based identity-component orbit; $K^\mu$ records the full $sA_\mu=\partial_\mu c$ insertion.
charge	boundary-nonzero transformations are absent from $c$ and from this Slavnov equation; they may instead carry surface charge.
gauge-fixed	$b$ enforces $\partial_iA_i+\xi b=0$ and imposes $\partial_iA_i=0$ sharply at $\xi=0$; $\Delta_D$ controls the ghost sector, and the functional identity relates the resulting vertices.

The charge distinction is a statement about the chosen redundancy group, not a consequence of the Zinn–Justin equation. Boundary-supported gauge charges in the classical covariant phase-space setting are separated from based degeneracies in Assanioussi et al. 2024, §§ 3.1–3.3, arXiv v2 pp. 13–16, eqs. (3.24)–(3.26), Open PDF.

For compact Yang–Mills theory, $D_\mu c$ and $sc$ are nonlinear, so both $K$ and $L$ are essential. The same algebraic equation holds on a BRST-stable local domain. Invertibility of $M_A$ is separately required to define the local ghost propagator and Faddeev–Popov slice. A Gribov horizon or a nonperturbative restriction of the integration region can make the domain fail to be BRST stable without making the algebraic identity $s^2=0$ false. The zero-mode and horizon limits of the local Faddeev–Popov construction are reviewed in Vandersickel and Zwanziger 2012, §§ 2.2–2.2.1 and 3.6.2–3.6.4, arXiv v2 pp. 20–25 and 87, Open PDF.

What the functional identity does—and does not—imply

premise	valid conclusion	conclusion that does not follow
$\mathcal S(\Sigma)=0$	the extended classical action is BRST invariant	the regulated measure is invariant
$\mathcal S(\Gamma)=0$	differentiation yields relations among complete 1PI vertices	each diagram or each longitudinal vertex vanishes separately
$\mathcal B_\Gamma^2=0$	consistent deformations and breakings have a cohomological grading	the cohomology at ghost number one is empty
$\mathcal B_\Sigma\mathcal C=0$	$\mathcal C$ is a candidate symmetry-compatible counterterm	$\mathcal C$ obeys power counting and every subsidiary equation
$\mathcal B_\Sigma\Delta=0$	$\Delta$ satisfies the local consistency condition	$\Delta$ is a realized anomaly
no local anomaly	perturbative restoration may be possible with compatible counterterms	absence of global anomalies or a nonperturbative gauge-fixed measure

The identity constrains a complete Green-function or vertex hierarchy. It is not a diagram-by-diagram cancellation rule, a proof of gauge-parameter independence for arbitrary insertions, or a replacement for the physical BRST cohomology.

Common pitfalls

Omitting the Zinn sources. $sA$ and $sc$ are composite operators. Ordinary field sources alone do not close their renormalized insertion identities.

Equating $s\Sigma=0$ with $\mathcal S(\Gamma)=0$. The first is classical. The second also needs a controlled measure, regulator, composite-operator renormalization, boundary domain, and subtraction prescription.

Using only the Slavnov equation to list counterterms. The gauge-fixing, antighost, Landau ghost, power-counting, and global-symmetry equations impose additional restrictions.

Calling every ghost-number-one class an anomaly. Such a class is a candidate local obstruction. A realized anomaly additionally requires a nonzero quantum coefficient after removable breakings have been subtracted.

Extending a local identity through a Gribov restriction. Algebraic nilpotency survives, but a restricted integration domain may have a horizon boundary that invalidates the change-of-variables proof.

Check your understanding

Check 1: ghost number of the Slavnov functional

Show that every term in $\mathcal S(X)$ has ghost number $+1$ when $\operatorname{gh}X=0$.

Check

Since $\operatorname{gh}K=-1$ and $\operatorname{gh}L=-2$,

$$\operatorname{gh}X_K=1, \qquad \operatorname{gh}X_L=2.$$

Also $\operatorname{gh}X_A=0$, $\operatorname{gh}X_c=-1$, and $\operatorname{gh}X_{\bar c}=1$. Therefore $X_KX_A$, $X_LX_c$, and $bX_{\bar c}$ all have ghost number $+1$.

Check 2: why the simple ghost equation is Landau-specific

Apply $\mathcal G^a$ to the $\xi b^2/2$ term.

Check

Only the $b$ derivative contributes:

$$\mathcal G^a \int\mathrm d^d x\,\frac\xi2b^db^d =-\xi g f^{abc} \int\mathrm d^d x\,\bar c^bb^c.$$

This is not the source-linear breaking $\Delta_{\mathrm{cl}}^a$. It vanishes when $\xi=0$, which is why the displayed integrated ghost equation has its simple form in Landau gauge.

Check 3: the bounded Maxwell cancellation

Verify $\mathcal S(\Sigma_{\mathrm M})=0$ directly from the unintegrated pairings.

Check

$sS_{\mathrm M}=0$, $sb=sc=0$, and $s\langle K^\mu,\partial_\mu c\rangle=0$. The remaining terms give

$$\langle b,\Delta_Dc\rangle -\langle b,\Delta_Dc\rangle=0.$$

No adjoint operator or boundary integration by parts was used.

Where the derivation continues

Symmetry and Counterterms develops the full loopwise stability and restoration proof. BV Fields, Antifields, and the Odd Symplectic Structure geometrizes the Zinn sources and extends the construction to open or reducible gauge algebras. Model-specific non-Abelian vertex identities and amplitude checks belong to Gauge Theories and the Standard Model, while Boundary Symmetry, Surface Charges, and Edge Modes determines which boundary transformations were physical rather than ghost directions.

References

• Assanioussi, Mehdi, Jerzy Kowalski-Glikman, Ilkka Mäkinen, and Ludovic Varrin. “On the Covariant Formulation of Gauge Theories with Boundaries.” Classical and Quantum Gravity 41, no. 11 (2024): 115007. DOI. Open PDF (arXiv v2)
• Barnich, Glenn, Friedemann Brandt, and Marc Henneaux. “Local BRST Cohomology in Gauge Theories.” Physics Reports 338, no. 5 (2000): 439–569. DOI. Open PDF (arXiv v3)
• Bélusca-Maïto, Hermès, Amon Ilakovac, Paul Kühler, Marija Mađor-Božinović, Dominik Stöckinger, and Matthias Weißwange. “Introduction to Renormalization Theory and Chiral Gauge Theories in Dimensional Regularization with Non-Anticommuting $\gamma_5$.” Symmetry 15, no. 3 (2023): article 622. DOI. Published Open PDF
• Blasi, Alberto, Olivier Piguet, and Silvio P. Sorella. “Landau Gauge and Finiteness.” Nuclear Physics B 356, no. 1 (1991): 154–162. DOI
• Echigo, Yoshio, Yuji Igarashi, Katsumi Itoh, Jan M. Pawlowski, and Yu Takahashi. “Functional Renormalization Group Flows and Gauge Consistency in Quantum Electrodynamics.” Progress of Theoretical and Experimental Physics 2026, no. 1 (2026): 013B01. DOI. Open PDF
• Moss, Ian G., and Pedro J. Silva. “BRST-Invariant Boundary Conditions for Gauge Theories.” Physical Review D 55, no. 2 (1997): 1072–1078. DOI. Open PDF (arXiv v1)
• Slavnov, A. A. “Ward Identities in Gauge Theories.” Theoretical and Mathematical Physics 10, no. 2 (1972): 99–104. DOI
• Taylor, J. C. “Ward Identities and Charge Renormalization of the Yang–Mills Field.” Nuclear Physics B 33, no. 2 (1971): 436–444. DOI
• Vandersickel, Nele, and Daniel Zwanziger. “The Gribov Problem and QCD Dynamics.” Physics Reports 520, no. 4 (2012): 175–251. DOI. Open PDF (arXiv v2)
• Zinn-Justin, Jean. “Renormalization of Gauge Theories.” In Trends in Elementary Particle Theory, edited by Hans Rollnik and Kurt Dietz, 1–39. Lecture Notes in Physics 37. Berlin: Springer, 1975. DOI
• Zinn-Justin, Jean. Quantum Field Theory and Critical Phenomena. 5th ed. Oxford: Oxford University Press, 2021. DOI