The Ward–Takahashi Identity and Charge Renormalization

The Ward–Takahashi identity says that contracting the exact QED vertex with the photon momentum equals a difference of exact inverse electron propagators. It is the off-shell, all-orders expression of electric-current conservation. When the regulator and counterterms preserve the Abelian gauge symmetry, it forces the vertex and electron-field renormalizations to agree, $Z_1=Z_2$; charge running is then controlled by the photon factor $Z_3$, not eliminated.

Required background. Electron and photon renormalization defines $Z_1$, $Z_2$, and $Z_3$. Contact terms in Ward identities explains why differentiating a time-ordered product produces delta functions.

Helpful background. Currents, sources, and generating functionals gives the general functional derivation.

The current identity with contact terms

For clarity, factor the signed electric charge out of the current and write

$$j^\mu=\bar\psi\gamma^\mu\psi.$$

Current conservation does not imply that the divergence of a time-ordered product vanishes. Time ordering differentiates step functions, producing equal-time commutators at operator insertions. In the convention used below,

$$\begin{aligned} \partial_\mu^x \langle 0|T\{j^\mu(x)\psi(y)\bar\psi(z)\}|0\rangle ={}&-\delta^{(4)}(x-y) \langle 0|T\{\psi(y)\bar\psi(z)\}|0\rangle \\ &+\delta^{(4)}(x-z) \langle 0|T\{\psi(y)\bar\psi(z)\}|0\rangle. \end{aligned}$$

The two terms have opposite signs because $\psi$ and $\bar\psi$ carry opposite current-generator action. Take incoming electron momentum $p$, incoming photon momentum $k$, and outgoing electron momentum $p+k$. After the common momentum-conserving delta function and the propagator factors of $i$ have been removed, the contracted, electron-leg-attached identity reads

$$S(p+k)\,[k_\mu\Gamma^\mu(p+k,p)]\,S(p) =S(p)-S(p+k).$$

Multiplication on the left by $S^{-1}(p+k)$ and on the right by $S^{-1}(p)$ amputates the legs. The two contact terms therefore become

$$\boxed{ k_\mu\Gamma^\mu(p+k,p) =S^{-1}(p+k)-S^{-1}(p)}.$$

Here $S$ is the reduced propagator, with $S_F=iS$ and $S^{-1}(p)=p\!\!\!/-m-\Sigma(p)$. The proper photon-vertex rule is $+iq\Gamma^\mu$, so $\Gamma^\mu$ has both the signed charge and the factor $i$ removed. For an electron $q=-e$ this rule is $-ie\Gamma^\mu$. At tree level,

$$\Gamma^\mu_{(0)}=\gamma^\mu, \qquad k_\mu\gamma^\mu =(p+k)\!\!\!/-p\!\!\!/ =S_{(0)}^{-1}(p+k)-S_{(0)}^{-1}(p).$$

The original short Ward identity and Takahashi’s off-shell generalization are Ward 1950, p. 182 and Takahashi 1957, pp. 371–375. The local contact identity and its Fourier transform are developed in Schwartz 2014, §§ 14.8.1–14.8.2, pp. 278–280; the explicit amputation appears in Schwartz 2014, § 19.5.1, pp. 352–353, Eqs. (19.77)–(19.82). His inverse propagator contains $+\Sigma$ rather than the $-\Sigma$ used here; the identity between the full inverse propagators is unchanged.

Equal-momentum limit and form factors

If the vertex and inverse propagator are regular as $k\to0$, expand the right-hand side:

$$S^{-1}(p+k)-S^{-1}(p) =k_\mu\frac{\partial S^{-1}(p)}{\partial p_\mu}+O(k^2).$$

The identity then gives

$$\Gamma^\mu(p,p)=\frac{\partial S^{-1}(p)}{\partial p_\mu}.$$

For the on-shell application, keep a nonzero electron mass and an explicit infrared prescription, as in electron and photon renormalization. This is a regulated perturbative pole calculation, not an assertion of an isolated exact charged-particle pole. Let $R$ be the pole residue of the reduced electron propagator:

$$S(p)=\frac{R(p\!\!\!/+m)}{p^2-m^2+i0} +\text{terms less singular at the pole}.$$

Here $m$ is the perturbative pole mass, and $R$ is not the bare-to-renormalized field constant $Z_2$. To see its relation to the vertex, write $S^{-1}(p)=A(s)p\!\!\!/-mB(s)$ with $s=p^2$ and $A(m^2)=B(m^2)$. Expanding the denominator $A(s)^2s-m^2B(s)^2$ at its simple pole gives the first equality below. Differentiating the inverse propagator and using $\bar u(p)\gamma^\mu u(p)=p^\mu\bar u(p)u(p)/m$ gives the second:

$$\begin{aligned} R^{-1}&=A(m^2)+2m^2[A'(m^2)-B'(m^2)],\\ \bar u(p)\Gamma^\mu(p,p)u(p) &=R^{-1}\bar u(p)\gamma^\mu u(p). \end{aligned}$$

Primes denote derivatives with respect to $s$. Thus the proper vertex’s Dirac coefficient is $F_1^{\rm proper}(0)=R^{-1}$ in this regular limit. The two external electron legs supply a product of LSZ residue factors equal to $R$. With the photon field normalized to unit pole residue and the charge defined in the Thomson limit, the residue-included form factor is therefore

$$F_1^{\rm physical}(0)=R F_1^{\rm proper}(0)=1.$$

Thus charge normalization does not by itself set the proper form factor to one in an arbitrary electron-field scheme. The cancellation between vertex and external-line factors is explicit in Weinberg 1995, § 10.4, pp. 447–448, especially Eq. (10.4.28).

For the following decomposition, choose the on-shell electron-field normalization $R=1$, so that the proper and residue-included form factors coincide. Between on-shell spinors the parity-even vertex has the form

$$\bar u(p')\Gamma^\mu u(p) =\bar u(p')\left[ F_1(k^2)\gamma^\mu +\frac{i\sigma^{\mu\nu}k_\nu}{2m}F_2(k^2) \right]u(p),$$

Under these charge and field conditions, $F_1(0)=1$. The transverse Pauli term is unconstrained by the contraction because $k_\mu\sigma^{\mu\nu}k_\nu=0$; its value $F_2(0)$ is the anomalous magnetic moment developed on the lepton magnetic-moments page. General nonzero-transfer form factors can still contain infrared singularities; the static normalization does not make the whole vertex an infrared-finite observable.

Why Z₁ = Z₂

With $e>0$ the electron charge magnitude, the bare and renormalized Lagrangians use

$$\psi_0=Z_2^{1/2}\psi, \qquad A_0=Z_3^{1/2}A, \qquad e_0=\mu^\epsilon Z_e e, \qquad Z_1=Z_eZ_2Z_3^{1/2}.$$

The divergent local part of the inverse propagator proportional to $p\!\!\!/$ is cancelled by $\delta Z_2$. The divergent local part of the vertex proportional to $\gamma^\mu$ is cancelled by $\delta Z_1$. The mass counterterm cancels out of the difference of inverse propagators, leaving the same coefficient of $k\!\!\!/$ on both sides. The finite local terms must obey the same identity when the renormalized charge and fields use this common gauge-covariant normalization. Symmetry-preserving subtraction therefore requires

$$\boxed{Z_1=Z_2}.$$

Consequently,

$$e_0=\mu^\epsilon Z_3^{-1/2}e.$$

This is the precise sense in which charge renormalization is determined by photon vacuum polarization. It does not say $Z_3=1$, and it does not make the beta function vanish. The charge-universality argument and the relation to the full Ward–Takahashi identity are given in Schwartz 2014, § 19.5, pp. 351–353.

For massless QED at nonexceptional Euclidean external momenta in a covariant gauge,

$$\delta Z_1^{\rm UV} =\delta Z_2^{\rm UV} =-\frac{\alpha\xi}{4\pi\bar\epsilon}, \qquad \frac1{\bar\epsilon}=\frac1\epsilon-\gamma_E+\ln4\pi.$$

The equality holds at $\xi=0,1,3$ even though each constant changes with $\xi$. At a fixed perturbative order, a common finite shift

$$\delta Z_1\to\delta Z_1+c, \qquad \delta Z_2\to\delta Z_2+c$$

preserves this counterterm equality, with the remaining parameter and field definitions converted consistently. Shifting only one of them would leave an uncancelled local difference between the contracted vertex and the inverse propagator.

Charge universality—and its limits

Suppose several matter fields carry integer or rational representation labels $Q_i$ and couple through

$$D_\mu\psi_i=(\partial_\mu-ieQ_iA_\mu)\psi_i.$$

Their vector Ward identities preserve one common renormalized gauge coupling $e$ multiplying the fixed labels $Q_i$. Radiative corrections do not generate a different electromagnetic coupling for each process. The statement assumes that the fields really transform under the same unbroken $U(1)$ gauge symmetry and that matching across thresholds is performed consistently.

The common photon-normalization factor for all charges is derived in Weinberg 1995, § 10.4, p. 445, Eqs. (10.4.17)–(10.4.18). When a charged species is integrated out, the resulting threshold matching is an additional calculation.

The identity does not determine the allowed charge lattice or the global form of the gauge group, does not prove anomaly cancellation in a chiral theory, and does not replace matching when heavy charged particles are removed. Those are separate structural questions.

When the inference can fail

The bare identity is exact for vector QED, but a calculation may violate it if the setup is inconsistent:

• a hard momentum cutoff can spoil shift invariance of linearly divergent integrals;
• different momentum routings can leave spurious surface terms;
• vertex and self-energy graphs can be truncated at different orders;
• counterterms can be chosen in mismatched schemes;
• a regulator can break gauge symmetry without the required restoring counterterms; or
• exceptional on-shell momenta can mix infrared singularities into a purported ultraviolet test.

Dimensional regularization preserves the vector identity in ordinary QED. A regulator that breaks it is not automatically forbidden, but the breaking must be local, diagnosed, and removed by symmetry-restoring counterterms before physical conclusions are drawn.

Independent checks

1. Tree check: contract $\gamma^\mu$ with $k_\mu$ and recover the difference of free inverse propagators.
2. On-shell current check: sandwich the identity between spinors satisfying the left and right pole equations, with the infrared prescription retained. The inverse propagators then annihilate the spinors, so the contracted current matrix element vanishes; the photon momentum can be off shell. For a scattering amplitude, test the polarization-to-momentum replacement on the complete sum of photon attachments and required contact contributions, as in Compton scattering. This is not an isolated real-photon emission or absorption process for one free massive electron: nonzero real-photon momentum is incompatible with two future-directed, equal-mass on-shell electron momenta.
3. Pole check: at one loop, verify $\delta Z_1^{\rm UV}-\delta Z_2^{\rm UV}=0$ for more than one gauge parameter.
4. Scheme check: apply the same finite shift to both constants and verify the equality survives.

These checks distinguish the local Ward identity, external-state normalization, and the gauge cancellation of a complete process.

Common pitfalls

$Z_1=Z_2$ is conditional, not regulator-free magic. It follows when the regulated and renormalized theory satisfies the vector Ward identity.

The identity does not forbid charge running. It moves charge renormalization into $Z_3$, whose scale dependence is nonzero.

Dropping contact terms destroys the off-shell identity. Ordinary current conservation applies away from insertions; the delta functions are what become the inverse propagators.

A longitudinal vertex test is not the whole amplitude. Transverse structures such as $F_2$ contain physical information that the contraction with $k_\mu$ cannot determine.

Exercises

A finite change of electron-field normalization. Keep the photon field, Thomson charge, pole mass, and infrared prescription fixed. Replace the renormalized electron field by $\chi=\lambda\psi$, with a constant real $\lambda>0$. Find the transformed reduced propagator, proper vertex, and pole residue. Does the physical zero-transfer charge change if the original electron field had $R\ne1$?

Solution

The two-point function acquires two factors of $\lambda$, whereas amputation of the transformed three-point function leaves two inverse factors:

$$\begin{aligned} S_\chi&=\lambda^2 S, &S_\chi^{-1}&=\lambda^{-2}S^{-1},\\ \Gamma_\chi^\mu&=\lambda^{-2}\Gamma^\mu, &R_\chi&=\lambda^2 R. \end{aligned}$$

Both sides of the proper Ward identity acquire the same factor $\lambda^{-2}$. In particular $F_{1,\chi}^{\rm proper}(0)=\lambda^{-2}/R$, so

$$R_\chi F_{1,\chi}^{\rm proper}(0) =(\lambda^2R)\frac{\lambda^{-2}}{R}=1.$$

The physical charge is unchanged. A proper vertex evaluated without its external-leg normalization would instead appear to change with an arbitrary choice of field variable.

References

• Matthew D. Schwartz, Quantum Field Theory and the Standard Model, Cambridge University Press (2014), §§ 14.8 and 19.4–19.5, doi:10.1017/9781139540940.
• Yasushi Takahashi, “On the Generalized Ward Identity,” Il Nuovo Cimento 6 (1957) 371–375, doi:10.1007/BF02832514.
• John C. Ward, “An Identity in Quantum Electrodynamics,” Physical Review 78 (1950) 182, doi:10.1103/PhysRev.78.182.
• Steven Weinberg, The Quantum Theory of Fields, Volume I: Foundations, Cambridge University Press (1995), § 10.4, doi:10.1017/CBO9781139644167.