Background-Field Yang–Mills Effective Action

The background-field method splits the gauge potential into a prescribed background and an integrated quantum fluctuation, then fixes only the fluctuation with a background-covariant condition. Ordinary gauge redundancy is still fixed, but a separate background gauge invariance remains manifest. In perturbative Yang–Mills theory with an anomaly-free matter sector and a symmetry-preserving renormalization prescription, its Ward identity constrains the background effective action and yields $Z_gZ_{\bar A}^{1/2}=1$ . This makes the coupling counterterm accessible from a background two-point calculation; it does not make a generic off-shell effective action an observable.

Required background. Gauge-fixed Yang–Mills action and ghost sector supplies the Faddeev–Popov and BRST construction. Coupling to background gauge fields and bundles supplies the general background Ward-identity viewpoint.

Helpful background. The 1PI effective action and mean-field equations supplies the Legendre transform and loop expansion used below.

Background and quantum transformations

Write the full integration variable as

A_\mu=\bar A_\mu+a_\mu.

There are two useful transformation laws. A quantum gauge transformation leaves $\bar A$ fixed and transforms the full field in the usual way:

\delta_{\rm q}\bar A_\mu=0, \qquad \delta_{\rm q}a_\mu=D_\mu[\bar A+a]\alpha.

A background transformation acts on the two pieces so that the background is a connection and the fluctuation is covariant:

\bar A_\mu' =U\bar A_\mu U^{-1} -\frac{i}{g}(\partial_\mu U)U^{-1}, \qquad a_\mu'=Ua_\mu U^{-1}.

The total $A=\bar A+a$ therefore transforms as a connection in either description, but the roles differ: the quantum transformation is the redundancy to be fixed, while the background transformation will remain as a manifest symmetry.

Choose

G^a[a;\bar A]=(\bar D^\mu a_\mu)^a, \qquad \mathcal L_{\rm gf} =-\frac{1}{2\xi}(\bar D^\mu a_\mu)^a(\bar D^\nu a_\nu)^a.

The orbit variation is $\delta_{\rm q}G=\bar D^\mu D_\mu[\bar A+a]\alpha$ . With the same Faddeev–Popov sign convention as in ordinary covariant gauge, define

\mathcal M^{ab}[a;\bar A] =-\bigl(\bar D^\mu D_\mu[\bar A+a]\bigr)^{ab},

so the ghost term is

\mathcal L_{\rm gh} =\bar c^a\mathcal M^{ab}c^b.

For compactly supported fields, or boundary conditions that make the integration-by-parts surface term vanish, the same ghost action can be written as

S_{\rm gh} =\int\mathrm d^4x\, (\bar D^\mu\bar c)^a(D_\mu[\bar A+a]c)^a.

These are two representations of one term. The covariant product rule follows from the invariant adjoint contraction; it preserves the order of $\bar c$ and $c$ . Expanding $D[\bar A+a]$ gives one background-covariant ghost kinetic term and its quantum-gauge-field interaction, as in Schwartz 2014, § 34.3, Eq. (34.83), p. 753.

Under a background transformation, $G$ , $c$ , and $\bar c$ all transform in the adjoint. Contracting their adjoint indices makes $\mathcal L_{\rm gf}+\mathcal L_{\rm gh}$ background invariant. This is a simultaneous transformation of the background, fluctuation and ghosts, not invariance under changing the background alone. The transformation and effective-action argument are developed in Srednicki 2006, §78, pp.465–467, Eqs.(78.5)–(78.21), author-manuscript PDF. His ghost normalization has the opposite overall sign to the convention used here; a constant antighost sign change translates it without changing the field-dependent determinant.

Curvature expansion and quadratic operators

The full curvature separates as

F_{\mu\nu}[\bar A+a] =\bar F_{\mu\nu} +\bar D_\mu a_\nu-\bar D_\nu a_\mu -ig[a_\mu,a_\nu].

Terms quadratic in $a$ from the Yang–Mills action are

\begin{aligned} \mathcal L_{\rm YM}^{(2)}={}& -\frac12(\bar D_\mu a_\nu)^a(\bar D^\mu a^\nu)^a +\frac12(\bar D_\mu a_\nu)^a(\bar D^\nu a^\mu)^a\\ &-\frac g2f^{abc}\bar F^a_{\mu\nu}a^{b\mu}a^{c\nu}. \end{aligned}

To see the second curvature coupling, integrate the crossed derivative term by parts and use $[\bar D_\mu,\bar D_\nu]=-ig\bar F^a_{\mu\nu}T^a_{\mathrm{adj}}$ . With the same vanishing surface terms as above,

\begin{aligned} \int d^4x\,(\bar D_\mu a_\nu)^a(\bar D^\nu a^\mu)^a =\int d^4x\,\Bigl[ (\bar D_\mu a^\mu)^a(\bar D_\nu a^\nu)^a -g f^{abc}\bar F^a_{\mu\nu}a^{b\mu}a^{c\nu} \Bigr]. \end{aligned}

The divergence-square term combines with gauge fixing, while the commutator doubles the original curvature coupling. After adding the gauge-fixing term, the quadratic kernel is

\mathcal L_a^{(2)} =\frac12a_\mu^b\left[ (\bar D^2)^{bc}\eta^{\mu\nu} +\left(\frac1\xi-1\right)(\bar D^\mu\bar D^\nu)^{bc} -2g f^{abc}\bar F^{a\mu\nu} \right]a_\nu^c.

In background Feynman gauge, $\xi=1$ , this reduces to a minimal covariant Laplacian plus the spin-one curvature term:

\mathcal L_{a,\xi=1}^{(2)} =-\frac12(\bar D_\mu a_\nu)^a(\bar D^\mu a^\nu)^a -g f^{abc}\bar F^a_{\mu\nu}a^{b\mu}a^{c\nu}.

The ghost quadratic operator at $a=0$ is the adjoint scalar Laplacian $-\bar D^2$ . Thus the one-loop contribution has the schematic determinant form

\Gamma^{(1)}[\bar A] =\frac i2\operatorname{Tr}\ln\Delta_1 -i\operatorname{Tr}\ln\Delta_0 +\Gamma_{\rm matter}^{(1)},

where $\Delta_1$ is the vector kernel and $\Delta_0=-\bar D^2$ is the ghost kernel, with Feynman boundary conditions and field-independent normalizations suppressed. The factor $1/2$ belongs to a real vector Gaussian; the independent Grassmann ghost–antighost pair gives the minus sign and no additional half. The quadratic expansion and diagrammatic evaluation are given in Srednicki 2006, §78, pp.468–471, Eqs.(78.24)–(78.36), author-manuscript PDF. Evaluating the local ultraviolet pole is sufficient for the coupling counterterm; the finite nonlocal functional and infrared behavior are separate questions.

Background Ward identity

Keep a source $J$ for the fluctuation while forming the connected functional, then Legendre transform with respect to that source. Let $\varphi=\langle a\rangle_J$ denote the mean fluctuation and set the mean ghost fields to zero. The background functional is

\Gamma_{\rm bg}[\bar A] =\Gamma[\varphi=0;\bar A], \qquad \Gamma[\varphi;\bar A]=W[J;\bar A]-\int d^4x\,J\cdot\varphi.

At a generic off-shell background, the source required to impose $\varphi=0$ need not vanish. Thus this is a restriction of the 1PI effective action, not simply the logarithm of the source-free partition function. The Legendre transform removes the linear fluctuation term in the one-loop expansion; it must not be discarded by pretending that every background is stationary.

If the source also transforms covariantly, a background change of variables preserves the gauge-fixed functional integral and hence its Legendre transform. At $\varphi=0$ the homogeneous variation of the mean fluctuation vanishes. Varying the background as $\delta\bar A_\mu=\bar D_\mu\alpha$ and integrating by parts therefore gives

\bar D_\mu^{ab} \frac{\delta\Gamma_{\rm bg}[\bar A]}{\delta\bar A_\mu^b}=0.

Consequently the divergent background-only effective action must be a local background-gauge invariant. In four-dimensional pure Yang–Mills theory, the CP-even dimension-four term is proportional to

-\frac14\int\mathrm d^4x\, \bar F^a_{\mu\nu}\bar F^{a\mu\nu},

apart from total-derivative and field-independent terms; BRST-exact terms organize additional counterterms in the full gauge-fixed functional. The two-point divergence therefore determines the same coefficient that renormalizes all background three- and four-point vertices.

Use dimensional regularization with $d=4-2\epsilon$ and let

\bar A_{0\mu}=Z_{\bar A}^{1/2}\bar A_\mu, \qquad g_0=\mu^\epsilon Z_g g.

Because the background covariant derivative contains the single combination $g\bar A$ , the Ward identity requires

Z_g Z_{\bar A}^{1/2}=1.

This is the background-field analogue of the QED charge identity. It means that the coupling beta function can be extracted from the background two-point function alone. In particular, the minimal-subtraction pole and its coupling consequence have the form

\begin{aligned} Z_{\bar A}&=1+\frac{b_0g^2}{16\pi^2\epsilon}+\cdots, &Z_g&=1-\frac{b_0g^2}{32\pi^2\epsilon}+\cdots,\\ \beta_d(g)&=-\epsilon g-\frac{b_0g^3}{16\pi^2}+\cdots. \end{aligned}

The dimensional term in $\beta_d$ is needed when differentiating the pole at fixed $g_0$ . The local ultraviolet calculation continues the minimal operators to Euclidean signature and computes their curvature traces. Its weighted vector-minus-ghost coefficient is $11g^2C_A\bar F^2/12$ ; matching the kinetic normalization supplies the factor four. With Dirac fermions in the $f$ sum and complex scalars in the $s$ sum, the result is

b_0 =\frac{11}{3}C_A -\frac43\sum_fT(R_f) -\frac13\sum_sT(R_s),

with flavor and spectator-representation multiplicities included. Weyl fermions and real scalars carry half these weights when their gauge representations are allowed. This coefficient agrees with an ordinary covariant-gauge calculation. The method simplifies the organization; it does not change the beta function. Higher-loop terms in theories with other couplings belong to the full coupled flow.

Dependence that remains

Background gauge invariance does not make $\Gamma_{\rm bg}$ an observable. At generic off-shell $\bar A$ , the effective action can depend on the quantum gauge parameter $\xi$ . The split $A=\bar A+a$ is also an auxiliary organization, and the gauge-fixing term distinguishes the two pieces. Physical on-shell quantities agree with ordinary gauges after the appropriate BRST identities are imposed.

Nor is $\bar A$ necessarily a solution of the quantum equations. If the expansion is performed away from a stationary background, terms linear in $a$ occur and the interpretation of the loop expansion must retain them or the associated sources. Background covariance alone cannot replace an on-shell or gauge-independence check.

Independent checks and limits

Transformation check: $\bar D^\mu a_\mu$ must transform homogeneously under the combined background transformation; otherwise its square is not invariant.
Operator check: setting $\bar F=0$ and $\bar A=0$ reduces the vector kernel to the ordinary covariant-gauge inverse propagator and the ghost kernel to $-\partial^2$ .
Ward check: every local background counterterm must be expressible through $\bar F$ , background-covariant matter fields, and covariant derivatives; a standalone $\bar A^2$ term fails.
Renormalization check: the same $Z_g$ inferred from $Z_{\bar A}$ must reproduce the one-loop $b_0$ obtained in an ordinary covariant gauge.

Common pitfalls

Transforming only the background. The fluctuation and ghosts must rotate covariantly for $G^aG^a$ and the ghost action to be background invariant.

Confusing background and quantum gauge invariance. Quantum redundancy is fixed and governed by BRST; background invariance is a manifest Ward symmetry of the organized effective action.

Treating off-shell background invariance as gauge-parameter independence. A functional can be background gauge invariant and still depend on $\xi$ and on the split away from physical observables.

Exercise: covariance does not determine the coefficient

Suppose a pure Yang–Mills calculation omits the ghost determinant but keeps the complete background-covariant vector operator. Does the background Ward identity necessarily fail? Use the local coefficients $h_4(P_1)=5\mathcal Q/3$ and $h_4(P_0)=-\mathcal Q/12$ , with $\mathcal Q=g^2C_A\bar F^2$ , to compare the resulting coupling pole with the correct one.

Solution

Each operator transforms by conjugation under a background gauge transformation. A symmetry-preserving trace of its logarithm is therefore background invariant separately. Omitting the ghost does not necessarily spoil that Ward identity, even though it is not the gauge-fixed Yang–Mills path integral.

The real-vector Gaussian alone has coefficient $5\mathcal Q/6$ . Multiplying by four to match the kinetic term gives the wrong value

b_0^{\mathrm{without\ ghost}}=\frac{10}{3}C_A, \qquad b_0^{\mathrm{full}} =4C_A\left(\frac56+\frac1{12}\right) =\frac{11}{3}C_A.

Covariance restricts the allowed counterterm to the curvature invariant but does not fix its numerical coefficient. Correct field content, statistics and Gaussian weights remain essential. In particular, transversality of a background two-point result is a necessary structural check, not a substitute for the determinant calculation.

References

Matthew D. Schwartz, Quantum Field Theory and the Standard Model, Cambridge University Press (2014), § 34.3, DOI.
Mark Srednicki, Quantum Field Theory, prepublication author manuscript (2006), §78. Author’s draft and published-edition errata. Open PDF. Page locators above refer to this manuscript.