The Proca Field

For a real vector field with mass $m>0$, the Proca action propagates exactly three physical modes. Its Euler–Lagrange equation implies both the massive wave equation and the subsidiary condition $\partial\cdot A=0$; in momentum space that condition leaves three polarizations on the massive shell. Independently, the canonical theory has two second-class constraints, so its eight-dimensional phase space per spatial point reduces to six physical dimensions, or three configuration modes. After the nondynamical component $A_0$ is eliminated, the Hamiltonian is nonnegative under the stated boundary assumptions.

The subsidiary condition is an equation of motion, not a gauge choice. This page derives that distinction, checks the mode count and energy from independent descriptions, and explains why setting $m=0$ is not a continuous operation on every state or observable. The Higgs mechanism and interacting massive vectors lie outside the free Proca model.

Required background. The Action Principle and Field Equations supplies the boundary-aware first variation used below. One-Particle States: Mass, Spin, and Relativistic Normalization supplies the massive mass shell, spin-one little-group representation, and invariant momentum measure.

Helpful background. Normal Forms, Spectra, and Projectors supplies the projector language used in the polarization completeness relation.

The mass term makes transversality dynamical

Let $A_\mu$ be a real vector field on four-dimensional Minkowski space and let $m>0$. On a spacetime region $\Omega$, consider

$$\begin{aligned} S_{\mathrm P}[A] &=\int_\Omega \mathrm d^4x\,\mathcal L_{\mathrm P},\\ \mathcal L_{\mathrm P} &=-\frac14F_{\mu\nu}F^{\mu\nu} +\frac12m^2A_\mu A^\mu, \qquad F_{\mu\nu}=\partial_\mu A_\nu-\partial_\nu A_\mu. \end{aligned}$$

The field has engineering dimension one and $m$ has dimension one. The local assumptions that matter here are $m>0$ and an admissible variation class: either $\delta A_\mu$ has compact support inside $\Omega$ or it vanishes on all of $\partial\Omega$. If the region extends to spatial infinity, suitable spatial falloff removes the spatial part of the surface term, while the variations must still vanish on the initial and final time slices.

Varying the lower-index field $A_\nu$ and integrating the kinetic term by parts gives the complete first variation

$$\begin{aligned} \delta S_{\mathrm P} = {}&\int_\Omega \mathrm d^4x\, \left( \partial_\mu F^{\mu\nu}+m^2A^\nu \right)\delta A_\nu\\ &-\int_{\partial\Omega}\mathrm d\Sigma_\mu\, F^{\mu\nu}\delta A_\nu. \end{aligned}$$

For the declared variations, stationarity therefore yields

$$\partial_\mu F^{\mu\nu}+m^2A^\nu=0.$$

Taking a divergence uses no gauge choice. Antisymmetry of $F^{\mu\nu}$ makes $\partial_\nu\partial_\mu F^{\mu\nu}$ vanish identically, so

$$m^2\partial_\nu A^\nu=0.$$

Because $m>0$, every solution obeys

$$\partial\cdot A=0, \qquad (\Box+m^2)A^\nu=0.$$

Thus the Proca mass term does two jobs: it fixes the mass shell and turns transversality into a derived equation. The Maxwell kinetic term alone is invariant under $A_\mu\mapsto A_\mu+\partial_\mu\alpha$, but the mass term changes by

$$\Delta\mathcal L_m =m^2A^\mu\partial_\mu\alpha +\frac12m^2\partial_\mu\alpha\,\partial^\mu\alpha,$$

which is not an off-shell boundary term for an arbitrary local $\alpha$. Proca theory at $m>0$ therefore has no gauge equivalence that removes a field configuration. The action, subsidiary condition, and three-mode conclusion are developed in Schwartz 2014, § 8.2.2, pp. 114–117.

Three polarizations on the massive shell

For a positive-frequency plane wave

$$A^\mu(x)=\varepsilon^\mu(p)e^{-ip\cdot x},$$

the field equation becomes

$$(m^2-p^2)\varepsilon^\nu +p^\nu(p\cdot\varepsilon)=0.$$

The subsidiary condition gives $p\cdot\varepsilon=0$, and a nonzero mode then requires $p^2=m^2$. In the rest frame $p^\mu=(m,\mathbf 0)$, transversality says $\varepsilon^0=0$. The remaining polarization space is the three-dimensional spatial vector representation of the massive little group $SO(3)$, so it carries spin one.

Choose an orthonormal polarization basis with

$$\varepsilon^{(\lambda)*}\cdot \varepsilon^{(\lambda')} =-\delta_{\lambda\lambda'}, \qquad \lambda=1,2,3.$$

The minus sign is a Lorentz norm for spacelike vectors; it is not a negative Hilbert-space probability. A real classical solution can be expanded as

$$\begin{aligned} A_\mu(x)=\sum_{\lambda=1}^{3} \int\frac{\mathrm d^3\mathbf p} {(2\pi)^3\,2E_{\mathbf p}} \big[&a_\lambda(\mathbf p) \varepsilon_\mu^{(\lambda)}(p)e^{-ip\cdot x}\\ &+a_\lambda(\mathbf p)^* \varepsilon_\mu^{(\lambda)}(p)^*e^{ip\cdot x}\big], \end{aligned}$$

where $E_{\mathbf p}=\sqrt{|\mathbf p|^2+m^2}$. The coefficient normalization is conventional; the displayed measure matches the invariant one-particle measure used by the required-background page.

On shell, the basis satisfies the completeness relation

$$\sum_{\lambda=1}^{3} \varepsilon_\mu^{(\lambda)}(p) \varepsilon_\nu^{(\lambda)}(p)^* =-\eta_{\mu\nu}+\frac{p_\mu p_\nu}{m^2}.$$

There are two quick checks. In the rest frame the time–time entry vanishes and the spatial block is the $3\times3$ identity. Contracting with $p^\mu$ gives zero because $p^2=m^2$. After raising one index, the displayed completeness tensor is the negative of the Lorentz-orthogonal projector,

$$C^\mu{}_{\nu}=-T^\mu{}_{\nu}, \qquad T^\mu{}_{\nu} =\delta^\mu{}_{\nu}-\frac{p^\mu p_\nu}{m^2}.$$

Thus $T^2=T$ only on the massive shell; the covariant polarization sum $C_{\mu\nu}$ should not itself be called a projector. Schwartz 2014, § 8.2.2, pp. 115–117 gives the polarization construction and its positive-energy mode interpretation.

Second-class constraints give the same count

The polarization argument is covariant. A canonical calculation exposes the nondynamical component and supplies an independent degree count. In this section only, spatial squares use the Euclidean metric:

$$\mathbf A^2=\delta^{ij}A_iA_j, \qquad \boldsymbol\pi^2=\delta_{ij}\pi^i\pi^j.$$

Writing the Lagrangian in time and space gives

$$\mathcal L_{\mathrm P} =\frac12(\dot A_i-\partial_iA_0)^2 -\frac14F_{ij}F_{ij} +\frac12m^2(A_0^2-\mathbf A^2).$$

The canonical momenta are

$$\pi^0=0, \qquad \pi^i=\dot A_i-\partial_iA_0.$$

The first equality is a primary constraint: no $\dot A_0$ occurs. After one spatial integration by parts, the canonical Hamiltonian density is

$$\begin{aligned} \mathcal H_c={}& \frac12\boldsymbol\pi^2 +\frac14F_{ij}F_{ij} +\frac12m^2\mathbf A^2\\ &-\frac12m^2A_0^2 -A_0\partial_i\pi^i, \end{aligned}$$

up to the boundary term $\partial_i(A_0\pi^i)$. For the canonical reduction, assume that the spatial slice $\Sigma$ is compact without boundary, or that the fields obey spatial boundary conditions for which

$$\int_{\partial\Sigma}\mathrm dS_i\,A_0\pi^i=0.$$

Preserving $\pi^0\approx0$ under Hamiltonian evolution produces the secondary constraint

$$\chi\equiv\partial_i\pi^i+m^2A_0\approx0.$$

With

$$\{A_\mu(\mathbf x),\pi^\nu(\mathbf y)\} =\delta_\mu{}^\nu\delta^{(3)}(\mathbf x-\mathbf y),$$

their bracket is

$$\{\pi^0(\mathbf x),\chi(\mathbf y)\} =-m^2\delta^{(3)}(\mathbf x-\mathbf y).$$

Reversing the bracket reverses the sign; the invariant fact is that the constraint matrix is nonsingular for $m>0$. The pair is therefore second class. Each second-class constraint removes one phase-space dimension, so

$$N_{\mathrm{phase}}=8-2=6, \qquad N_{\mathrm{configuration}}=3.$$

No additional gauge quotient is taken. Weinberg 1995, § 7.6, pp. 326–330 develops this massive-vector constraint pair and its nonsingular bracket matrix. Weinberg uses a mostly-plus metric; translating the action first changes convention-dependent signs but leaves the second-class classification and the three-mode count unchanged. General Dirac–Bergmann theory belongs to Constraints, Dirac Brackets, and Reduction.

Solving the secondary constraint gives

$$A_0=-\frac{\partial_i\pi^i}{m^2}.$$

Substitution into the Hamiltonian, under the spatial boundary condition just stated, yields

$$\begin{aligned} H_{\mathrm{red}} =\int\mathrm d^3x\,\bigg[ &\frac12\boldsymbol\pi^2 +\frac14F_{ij}F_{ij} +\frac12m^2\mathbf A^2\\ &+\frac1{2m^2}(\partial_i\pi^i)^2 \bigg]. \end{aligned}$$

Every term is nonnegative for $m>0$. Positivity is a property of this reduced Hamiltonian; the negative $-m^2A_0^2/2$ term in the unreduced density is not an instability because $A_0$ is constrained rather than an independent oscillator. This elimination, the $d-1$ degree count, and the positive Hamiltonian are checked independently in Zinn-Justin 2021, § 21.1.1, pp. 508–509.

Why the massless limit is conditional

For fixed nonzero spatial momentum and $p^\mu=(E,\mathbf p)$, one normalized longitudinal polarization is

$$\varepsilon_L^\mu(p) =\frac1m\left( |\mathbf p|,E\widehat{\mathbf p} \right), \qquad E=\sqrt{|\mathbf p|^2+m^2}.$$

It obeys $p\cdot\varepsilon_L=0$ and $\varepsilon_L^2=-1$, but along this on-shell family

$$\varepsilon_L^\mu(p) =\frac{p^\mu}{m} +O\!\left(\frac mE\right).$$

The normalized vector and the $p_\mu p_\nu/m^2$ term in the completeness relation therefore have no finite componentwise limit. If a test current is conserved, $p_\mu J^\mu=0$, its contraction with the leading $p^\mu/m$ term vanishes. Conserved-current quantities can consequently have a smooth limit even though the polarization vector itself does not.

The canonical calculation identifies the structural discontinuity. At $m=0$, the constraint bracket above loses its nonzero entry, $A_0=-\partial_i\pi^i/m^2$ ceases to be a valid elimination, and a gauge equivalence appears. One must specify the observable, source conservation, on-shell trajectory, and held-fixed couplings before calling the limit smooth or discontinuous. The singular longitudinal behavior is shown in Schwartz 2014, § 8.2.3, p. 118; the conserved-current qualification is developed in Zinn-Justin 2021, § 21.2, p. 510.

The controlled next comparison is The Free Maxwell Field and Gauge Redundancy, followed by Massive and Massless Spin-One Polarizations for explicit bases, completeness relations, and the developed three-versus-two analysis.

What this free model establishes

The answer to the page’s question is now overdetermined in a useful way. The field equation leaves three transverse directions on a timelike mass shell; the canonical analysis leaves three configuration degrees of freedom; and the reduced Hamiltonian assigns them nonnegative energy. A disagreement among those checks signals a sign, constraint, or normalization error.

This conclusion applies to the free real Proca field with $m>0$ and the stated boundary assumptions. It does not explain how an interacting gauge theory generates a vector mass or whether an arbitrary interacting massive-vector model is consistent at high energy. The gauge-invariant Higgs mechanism belongs to Elitzur’s Theorem and the Gauge-Invariant Higgs Mechanism; electroweak mass generation belongs to Gauge-Boson Masses and Electroweak Mixing; and interacting longitudinal-vector behavior belongs to Longitudinal Vector Bosons and the Equivalence Theorem.

Common pitfalls

Calling $\partial\cdot A=0$ a gauge condition. In Proca theory it follows from the Euler–Lagrange equation because $m>0$. There is no gauge orbit whose representative is being selected.

Setting $A_0=0$. The time component is nondynamical, not generally zero. Its secondary constraint determines $A_0=-\partial_i\pi^i/m^2$.

Diagnosing instability from the unreduced density. The term $-m^2A_0^2/2$ cannot be read independently of the constraint. Eliminating $A_0$ turns the relevant contribution into the positive square $(\partial_i\pi^i)^2/(2m^2)$.

Calling the completeness tensor a projector, or using it off shell. The mixed-index projector is $T^\mu{}_{\nu}=\delta^\mu{}_{\nu}-p^\mu p_\nu/m^2$ on shell; the polarization sum with one index raised is $-T^\mu{}_{\nu}$. Away from $p^2=m^2$, even $T$ is not idempotent or transverse.

Check your understanding

1. Starting from the Proca equation, recover both the subsidiary condition and the massive wave equation. Which step fails at $m=0$?

   Solution

   Apply $\partial_\nu$ to the equation. Antisymmetry gives $\partial_\nu\partial_\mu F^{\mu\nu}=0$, hence $m^2\partial\cdot A=0$. For $m>0$, substitute $\partial\cdot A=0$ into $\partial_\mu F^{\mu\nu}+m^2A^\nu=0$ to obtain $(\Box+m^2)A^\nu=0$. At $m=0$, division by $m^2$ is impossible and the divergence equation becomes an identity.

2. Verify the canonical degree count and explain why it agrees with the rest-frame polarization count.

   Solution

   Four fields and four conjugate momenta give eight phase-space dimensions per spatial point. The primary constraint $\pi^0\approx0$ and secondary constraint $\chi\approx0$ have a nonsingular mutual bracket, so they are second class and remove two dimensions. Six physical phase-space dimensions remain, corresponding to three canonical pairs. In the rest frame, $p\cdot\varepsilon=0$ removes the time component and leaves the same three spatial polarizations.

3. Show that the apparent negative $A_0$ term becomes positive after reduction.

   Solution

   Write $d=\partial_i\pi^i$. The $A_0$-dependent terms are $-m^2A_0^2/2-A_0d$. The constraint gives $A_0=-d/m^2$, so these terms become $-d^2/(2m^2)+d^2/m^2=d^2/(2m^2)$. This is nonnegative for $m>0$.

References

• Schwartz, Matthew D. Quantum Field Theory and the Standard Model. Cambridge: Cambridge University Press, 2014. DOI.
• Weinberg, Steven. The Quantum Theory of Fields. Volume I: Foundations. Cambridge: Cambridge University Press, 1995. DOI.
• Zinn-Justin, Jean. Quantum Field Theory and Critical Phenomena. 5th ed. Oxford: Oxford University Press, 2021. DOI.