Lorentz Covariance of Classical Field Theory

Every article of this chapter has quietly assumed something enormous: that the field theories we build respect special relativity. This article makes the assumption explicit, mechanical, and checkable. The whole discipline reduces to one design rule — make the action a Lorentz scalar — and one accounting fact makes the rule easy to obey: the spacetime volume element $\mathrm{d}^4x$ is itself invariant (boosts contract lengths and dilate times by exactly compensating factors), so a scalar action needs nothing more than a scalar Lagrangian density. From there the machinery of the SR chapter takes over: build $\mathcal{L}$ from complete contractions of tensors — $\partial_\mu\phi\,\partial^\mu\phi$ , $\phi^2$ , $F_{\mu\nu}F^{\mu\nu}$ — and the Euler-Lagrange equations that fall out are tensor identities, true in every inertial frame or none. Counting free indices becomes a physics audit: one uncontracted index anywhere in $\mathcal{L}$ and some observer measures different laws. Two distinctions carry the article's conceptual weight. Covariance of equations is not invariance of solutions: the dynamics is frame-blind, but a particular lump of field happily singles out its rest frame — symmetry of laws, not of states. And manifest covariance is a property of a formalism, not of the physics: the Hamiltonian description sliced spacetime and hid Lorentz symmetry, yet the symmetry survived — as the Poincaré algebra of the ten Noether charges, whose closure under Poisson brackets this article completes and audits. The payoff is the chapter's quality certificate: the field theories about to be quantized are relativistic by construction, and the ten conserved quantities certifying it are already in hand.

Key takeaways

The design rule: demand $S = \int\mathrm{d}^4x\,\mathcal{L}$ be a Lorentz scalar. Since $\mathrm{d}^4x' = |\det\Lambda|\,\mathrm{d}^4x = \mathrm{d}^4x$ , this means $\mathcal{L}$ must be a scalar.
Fields carry Lorentz representations: scalars ( $\phi' (x') = \phi(x)$ ), vectors ( $A'^\mu = \Lambda^\mu{}_\nu A^\nu$ ), spinors ( $\psi' = S(\Lambda)\psi$ ); derivatives transform with $(\Lambda^{-1})^\nu{}_\mu$ .
Legal $\mathcal{L}$ terms are complete contractions — zero free indices, all indices tied by $\eta_{\mu\nu}$ or $\epsilon^{\mu\nu\rho\sigma}$ : $\partial_\mu\phi\,\partial^\mu\phi$ , $m^2\phi^2$ , $F_{\mu\nu}F^{\mu\nu}$ .
A scalar action forces form-invariant Euler-Lagrange equations: the same tensor equation in every inertial frame.
Covariance of equations ≠ invariance of solutions: a static lump picks a rest frame; the boosted lump is another solution of the same equations.
Lorentz + translation invariance = ten Noether currents: $T^{\mu\nu}$ (4) and $M^{\mu\alpha\beta}$ (6) — the Poincaré charges, closing as an algebra under the field Poisson bracket.
Non-covariant terms (free indices, fixed background vectors) produce frame-dependent physics — e.g. modified propagation speeds — and are experimentally constrained to extraordinary precision.
Key idea: scalar action in, covariant dynamics out — relativity is not checked after the fact but installed at the level of the Lagrangian's index bookkeeping.

The core insight

Demand the action be a Lorentz scalar; since $\mathrm{d}^4x$ is invariant, this makes the Lagrangian density a scalar, which forces the Euler-Lagrange equations to be covariant tensor equations valid in all inertial frames.

Where this lives today

Lorentz covariance graduated from design principle to precision-measurement target. The Standard-Model Extension — a systematic catalogue of every index-illegal term that could be appended to known physics, each with its own coefficient — organizes a worldwide hunt: optical and microwave resonators rotating with the Earth, atomic-clock comparisons, antimatter spectroscopy at CERN, ultra-high-energy cosmic rays (whose very arrival constrains dispersion modifications at the Planck-suppressed level), and multimessenger astronomy, with GW170817's photon-graviton speed match the single most dramatic entry. All null so far — making Lorentz invariance among the best-tested structures in physics and disciplining quantum-gravity model-building, since many proposals (spacetime foams, some string-theory backgrounds) would generate exactly the frame-fixing terms the worked example dissects. On the theory side, covariance remains the master constraint from which modern field theory is derived: Wigner's representation theory (mass and spin as the only labels), the CPT and spin-statistics theorems (whose hypotheses are Lorentz invariance plus locality), and the effective-field-theory paradigm, which writes down every covariant operator up to a given dimension and lets experiment prune — the index audit of this article, industrialized. Meanwhile the "broken by the state, not the law" distinction runs whole subfields: condensed-matter systems live in a lattice frame yet engineer emergent Lorentz invariance at critical points (graphene's Dirac cones, with $c$ replaced by the Fermi velocity), and cosmology inhabits a preferred slicing (the CMB frame) atop perfectly covariant dynamics — the pitfall section's lesson, at the scale of the universe. Within the chapter: this article certifies the theories the capstone will quantize; the ten charges whose algebra closed above are the classical skeleton of the quantum Poincaré generators, and Wigner's theorem will read particles off their representations.

The physical picture

Think of the action as a single number summarizing an entire field history — the "cost" the variational principle minimizes. Lorentz covariance is the demand that all inertial observers agree on this number for the same physical history. That agreement decomposes into two factors. The volume element: a boost contracts lengths by $\gamma$ and dilates time intervals by $\gamma$ , so the four-volume $\mathrm{d}^4x = \mathrm{d}t\,\mathrm{d}^3x$ is untouched — formally, $\mathrm{d}^4x' = |\det\Lambda|\,\mathrm{d}^4x$ and the defining condition of the Lorentz group forces $\det\Lambda = \pm 1$ . And the integrand: with the measure already invariant, agreement on $S$ requires agreement on $\mathcal{L}$ point by point — the density must be a scalar, the same number evaluated by every observer at each spacetime event.

Figure

One history, two observers, one action — and the index audit that guarantees it

Once $\mathcal{L}$ is a scalar, covariance of the dynamics is automatic, and the mechanism is worth internalizing as a language fact. Tensor equations are statements in a coordinate-free grammar: "this four-vector equals that four-vector" survives any change of frame, because both sides transform identically. The Euler-Lagrange procedure applied to a scalar density produces exactly such statements — for the Klein-Gordon field, $(\Box + m^2)\phi = 0$ , where $\Box = \partial_\mu\partial^\mu$ is a scalar operator acting on a scalar field: every piece frame-portable by construction. The practical audit is index counting: a scalar has zero free indices, so every legal term of $\mathcal{L}$ must contract all its indices using the two invariant tensors available — the metric $\eta_{\mu\nu}$ and (for parity-odd physics) the Levi-Civita symbol $\epsilon^{\mu\nu\rho\sigma}$ . An uncontracted index, or a contraction against some fixed background vector, is the red flag: it selects a preferred direction or frame, and the worked example below computes precisely what breaks. The subtlety the article guards: the symmetry lives in the laws, not the states. A static soliton solution has a rest frame; the universe's microwave background defines a cosmic frame; a moving wave packet is manifestly not boost-invariant. None of this violates Lorentz covariance — the boosted lump is another solution of the same equations, and the symmetry's content is exactly that solutions map to solutions. Conflating "the equations are covariant" with "the configuration looks the same to all observers" is the standing confusion this article's pitfall section retires.

The design rule, executed

The Lorentz group is defined by preserving the Minkowski interval (the SR chapter's structure):

x'^\mu = \Lambda^\mu{}_\nu\,x^\nu, \qquad \Lambda^\mu{}_\alpha\,\eta_{\mu\nu}\,\Lambda^\nu{}_\beta = \eta_{\alpha\beta} \qquad (1)

where: the defining condition says $\Lambda$ preserves the metric — it is the spacetime analogue of orthogonality. Taking determinants of $\Lambda^{\mathsf{T}}\eta\Lambda = \eta$ gives $(\det\Lambda)^2 = 1$ , hence $\det\Lambda = \pm 1$ : Lorentz transformations preserve four-volume (the $+1$ branch, containing boosts and rotations, is the one connected to the identity; the $-1$ branch holds parity and time reversal — discrete, and outside Noether's jurisdiction). Fields respond according to their representation — the "what kind of field" declaration of the spin-current article:

\phi'(x') = \phi(x), \qquad A'^\mu(x') = \Lambda^\mu{}_\nu\,A^\nu(x), \qquad \psi'(x') = S(\Lambda)\,\psi(x) \qquad (2)

where: the scalar just relabels its argument; the vector's components mix like the coordinates; the spinor mixes under the double-valued matrices $S(\Lambda)$ built from the $\Sigma$ 's. Derivatives transform inversely, $\partial'_\mu = (\Lambda^{-1})^\nu{}_\mu\,\partial_\nu$ , which is exactly what makes contractions like $\partial_\mu\phi\,\partial^\mu\phi$ close on themselves: one $\Lambda$ meets one $\Lambda^{-1}$ and the pair cancels.

Derivation AScalar action implies form-invariant field equations

We prove the central theorem: if $\mathcal{L}$ is a Lorentz scalar (same functional form, same value at corresponding points), the Euler-Lagrange equations take the identical form in every inertial frame, and their solutions map to solutions under boosts. The strategy has three moves — measure invariance, action equality, and re-running the variational argument in the primed frame — and the physical punchline is in the third: stationarity is a property of the number $S$ , which both observers share.

Step 1. Measure invariance. Under (1) the Jacobian is $|\det\Lambda| = 1$ :

\mathrm{d}^4x' = \left|\det\frac{\partial x'}{\partial x}\right|\,\mathrm{d}^4x = |\det\Lambda|\,\mathrm{d}^4x = \mathrm{d}^4x

Why this is the relativistic accident that makes field theory easy: neither $\mathrm{d}t$ nor $\mathrm{d}^3x$ is separately invariant — time dilates, lengths contract — but their product is. This is also why the Lagrangian formalism stays manifestly covariant while the Hamiltonian formalism, which integrates over $\mathrm{d}^3x$ alone, cannot.

Step 2. Action equality. Because $\mathcal{L}$ is a scalar, its value at the event is observer-independent: $\mathcal{L}(\phi'_a(x'), \partial'_\mu\phi'_a(x')) = \mathcal{L}(\phi_a(x), \partial_\mu\phi_a(x))$ . Combined with Step 1:

S' = \int \mathcal{L}\big(\phi'_a, \partial'_\mu\phi'_a\big)\,\mathrm{d}^4x' = \int \mathcal{L}\big(\phi_a, \partial_\mu\phi_a\big)\,\mathrm{d}^4x = S

One history, one number, all frames.

Step 3. Stationarity transports. A variation $\delta\phi$ around a history maps, under the transformation, to a variation $\delta\phi'$ around the transformed history, and $\delta S' = \delta S$ . Hence a history stationary in one frame is stationary in all — and re-running the Euler-Lagrange derivation in the primed frame, where $\mathcal{L}$ has the identical functional form, produces the identical equation with primes everywhere:

\partial'_\mu\!\left(\frac{\partial\mathcal{L}}{\partial(\partial'_\mu\phi'_a)}\right) - \frac{\partial\mathcal{L}}{\partial\phi'_a} = 0

Step 4. Read the two corollaries. Form invariance: the equation of motion is a tensor identity — its left side transforms as a tensor of the same rank as $\phi_a$ 's representation, so "= 0" holds in every frame or none. Solutions map to solutions: if $\phi(x)$ solves the equations, so does the boosted configuration $\phi'(x) = \phi(\Lambda^{-1}x)$ (scalar case) — symmetry acts on the space of solutions, which is exactly the flat-direction picture Noether's theorem then converts into the six conserved currents of the spin-current article. Verified: standard argument per Landau-Lifshitz Vol. 2 §27 and Goldstein ch. 13; $(\det\Lambda)^2 = 1$ recomputed from the defining condition; the boosted-solution statement checked explicitly on the plane wave in the worked example.

The constructive half of the rule is a menu. Any complete contraction of fields, their derivatives, $\eta$ , and $\epsilon$ is admissible; the low-derivative menu for a scalar field is short — $\phi^2$ , $\phi^n$ , $\partial_\mu\phi\,\partial^\mu\phi$ — and the Lagrangian-density article's kinetic-minus-potential structure picks its physical members. For a vector field the gauge-invariant menu begins at $F_{\mu\nu}F^{\mu\nu}$ (the EM action), with $\epsilon^{\mu\nu\rho\sigma}F_{\mu\nu}F_{\rho\sigma}$ as the parity-odd theta-term. What the rule forbids is equally informative, and the audit is mechanical: $(\partial_0\phi)^2$ alone carries two bare time indices — a chosen frame; $n^\mu\partial_\mu\phi$ with a fixed $n^\mu$ smuggles in a background arrow. Such terms do not make the theory inconsistent; they make it non-relativistic, with measurable frame-dependence the worked example quantifies.

The covariance certificate of the canonical formalism

The Hamiltonian half of this chapter chose a time slice and lost manifest covariance; the Poisson-brackets article promised the loss was presentational. Here is the audit. The ten Poincaré charges — $P^\mu$ from translations, $J^{\mu\nu} = L^{\mu\nu} + S^{\mu\nu}$ from Lorentz transformations — are all in hand from the stress-energy and spin-current articles. Covariance of the sliced formalism is the statement that these charges represent the Poincaré algebra under the equal-time bracket: the flows they generate compose exactly as the geometry of (1) demands.

Go deeper · Derivation B · 2 minThe boost and rotation charges transform the momentum as a four-vector

Derivation BThe boost and rotation charges transform the momentum as a four-vector

We verify the key mixed bracket of the Poincaré algebra, $\{J^{\mu\nu}, P^\rho\} = \eta^{\nu\rho}P^\mu - \eta^{\mu\rho}P^\nu$ — the statement that under the flow generated by $J^{\mu\nu}$ , the four charges $P^\rho$ shuffle into one another exactly like the components of a four-vector. This is the algebraic certificate that the equal-time formalism knows special relativity: the bracket (a frame-sliced object) reproduces the transformation law (a spacetime object). Strategy: use the generator property of the charges — each charge's bracket with a field produces its transformation — and apply it to the charges themselves, keeping only the structure needed; the computation runs on the scalar field for concreteness.

Step 1. Recall the two generator facts, both established by the bracket article's machinery: $P^\rho$ generates translations, $\{P^\rho, \phi\} = -\partial^\rho\phi$ ; and $J^{\mu\nu}$ generates Lorentz transformations, $\{J^{\mu\nu}, \phi\} = -(x^\mu\partial^\nu - x^\nu\partial^\mu)\phi$ (scalar field: orbital action only).

Step 2. Write $P^\rho = \int\mathrm{d}^3x\;T^{0\rho}$ as a functional of $(\phi, \pi)$ and act with $J^{\mu\nu}$ 's flow. By the Leibniz rule, the bracket $\{J^{\mu\nu}, P^\rho\}$ replaces each field inside $P^\rho$ by its Lorentz variation from Step 1. But that substitution is precisely the statement that $P^\rho$ is evaluated on the infinitesimally transformed configuration:

\{J^{\mu\nu}, P^\rho\} = -\,\delta_{(\mu\nu)}P^\rho

where $\delta_{(\mu\nu)}$ is the change of the total momentum under the infinitesimal Lorentz transformation with parameter in the $\mu\nu$ plane.

Step 3. Evaluate that change geometrically. The four-momentum of a localized configuration assembles, by the conserved-currents article's slice theorem, into a genuine four-vector $P^\rho$ (its value is slice-independent; its components mix under frame changes). An infinitesimal Lorentz transformation with parameter $\omega_{\mu\nu}$ changes any four-vector by $\delta P^\rho = \omega^\rho{}_\sigma P^\sigma$ ; peeling off the antisymmetric parameter in the $\mu\nu$ plane gives

\{J^{\mu\nu}, P^\rho\} = \eta^{\nu\rho}\,P^\mu - \eta^{\mu\rho}\,P^\nu \qquad (3)

Step 4. Read the certificate, component by component. Take $J^{12}$ (a rotation about $z$ ) and $P^1$ : $\{J^{12}, P^1\} = -P^2$ — the rotation flow turns $x$ -momentum into $y$ -momentum, as rotations must. Take a boost, $J^{01}$ , and the energy $P^0$ : $\{J^{01}, P^0\} = -P^1$ — the boost flow feeds momentum into the energy at exactly the rate the SR chapter's four-momentum transformation prescribes ( $E' = \gamma(E - vp)$ , linearized). Together with $\{P^\mu, P^\nu\} = 0$ (translations commute) and the $\{J, J\}$ brackets closing on $J$ 's (rotations and boosts compose with the Wigner-rotation structure of the SR chapter), the ten charges close on the Poincaré algebra: the sliced formalism carries a complete, exact representation of spacetime symmetry. Nothing was lost — only hidden. Verified: consistent with the algebra stated in the Poisson-brackets article (eq. 7 there); component readings match the four-momentum transformation of the SR chapter's four-momentum article; the generator facts per the bracket article's derivations.

Figure

One wave, two frames, one mass shell — the boost slides the wave along the invariant hyperbola

Limiting cases

Case	Result	Meaning
$v \ll c$ limit of the theory	Galilean-covariant field theory	boosts degenerate; $\mathrm{d}^4x$ -invariance becomes separate $\mathrm{d}t$ , $\mathrm{d}^3x$
scalar field, boosted solution	$\phi(\Lambda^{-1}x)$ solves the same equation	symmetry maps solutions to solutions
plane wave under boost	$k^\mu \to \Lambda^\mu{}_\nu k^\nu$ , $k\cdot k = m^2$ fixed	dispersion relation is the invariant mass shell
rotations only ( $\omega_{0i} = 0$ )	ordinary angular-momentum covariance	the nonrelativistic rotation group sits inside
discrete $\Lambda$ ( $\det\Lambda = -1$ )	parity, time reversal	symmetries without Noether currents; $\epsilon$ -terms test them
fixed background vector in $\mathcal{L}$	frame-dependent propagation	Lorentz violation — computed in the worked example
Hamiltonian formalism	not manifestly covariant; algebra (3) closes	covariance carried by charges, not by the slicing
gravity switched on	Lorentz group becomes local (tangent-space) symmetry	the flat-space audit becomes general covariance — GR's business

Common trapCovariant equations, non-covariant solutions — and invariance of form, not of numbers

Two conflations, both fatal on exams and in research. First: a Lorentz-invariant Lagrangian does not produce Lorentz-invariant field configurations. The action's symmetry constrains the dynamics — the form of the equations, the menu of allowed terms — while any particular solution is free to break the symmetry by its shape: a soliton has a rest frame, a plane wave a propagation direction, the thermal universe a comoving frame. The symmetry's content for solutions is that the boosted configuration is again a solution (Step 4 of the first derivation), not that solutions look the same to everyone. Physicists say the symmetry is "spontaneously broken by the state" when even the ground state does this — but no dynamical law has been violated in either case. Second: covariance preserves tensor relations, not component values. The statement $\partial_\mu T^{\mu\nu} = 0$ holds in all frames; the sixteen numbers $T^{\mu\nu}$ do not — a boosted observer measures different energy density, different fluxes, mixed by two factors of $\Lambda$ . "The equation is the same" means both observers write the same symbolic relation among their own measured components. Claiming " $T^{00}$ is invariant because the theory is covariant" is the classic version of the error; the invariants are the contractions ( $T^\mu{}_\mu$ , $P_\mu P^\mu$ , $k_\mu k^\mu$ ), and everything else transforms. Keep the audit honest: forms are invariant, components covariant, solutions merely mapped.

Worked examples

1 · One plane wave, two observers — the invariant checked to the digit (confidence builder)

A Klein-Gordon plane wave has $k^\mu = (\omega, k) = (1.25, 0.75)$ in frame $S$ (one spatial dimension, $m = 1.00$ ). Observer $S'$ moves at $v = 0.60$ ( $\gamma = 1.25$ ). Compute the wave's frequency and wavenumber in $S'$ , and verify the invariant $k\cdot k$ .

Step 1. Check the mass shell in $S$ : $k\cdot k = \omega^2 - k^2 = 1.5625 - 0.5625 = \mathbf{1.0000} = m^2$ ✓ — the wave is on-shell, as any Klein-Gordon solution must be.

Step 2. Boost the wave four-vector with the SR chapter's Lorentz transformation (it transforms exactly like $(t, x)$ ):

\omega' = \gamma(\omega - v k) = 1.25\,(1.25 - 0.60 \times 0.75) = 1.25 \times 0.80 = \mathbf{1.0000}

k' = \gamma(k - v\omega) = 1.25\,(0.75 - 0.60 \times 1.25) = 1.25 \times 0.0 = \mathbf{0.0000}

Step 3. Interpret: $S'$ is the wave's rest frame — the phase velocity pattern this observer sees does not travel; the field simply oscillates in unison at $\omega' = m = 1.00$ , the mass frequency. (Check: the original wave's group velocity was $v_g = k/\omega = 0.60$ — precisely the boost that catches it.)

Step 4. The invariant: $k'\cdot k' = \omega'^2 - k'^2 = 1.0000 - 0 = \mathbf{1.0000} = m^2$ ✓. Both observers, wildly disagreeing about frequency and wavelength, agree on the contraction — the mass shell is the frame-portable content of the dispersion relation, and the scalar amplitude $A$ is untouched ( $\phi$ is a scalar). Every claim of the first derivation, exercised numerically.

2 · The price of one illegal term (exam level)

A theorist adds a frame-fixed term to the Klein-Gordon density: $\mathcal{L} = \tfrac12\partial_\mu\phi\,\partial^\mu\phi - \tfrac12 m^2\phi^2 + \lambda\,\dot\phi^2$ , with $\lambda = 0.02$ — the last term contracts indices against the fixed vector $n^\mu = (1, \mathbf{0})$ instead of the metric. (i) Find the modified dispersion relation. (ii) Compute the high-frequency propagation speed. (iii) State what an experiment would see.

Step 1. Collect the quadratic form: $\mathcal{L} = \tfrac12(1 + 2\lambda)\dot\phi^2 - \tfrac12(\nabla\phi)^2 - \tfrac12 m^2\phi^2$ . The Euler-Lagrange equation: $(1 + 2\lambda)\ddot\phi - \nabla^2\phi + m^2\phi = 0$ .

Step 2. Plane-wave ansatz $\phi \sim e^{i(kx - \omega t)}$ :

\omega^2 = \frac{k^2 + m^2}{1 + 2\lambda} = \frac{k^2 + m^2}{1.04}

Step 3. High-frequency (mass-negligible) propagation speed: $c_\phi = \omega/k \to (1 + 2\lambda)^{-1/2} = 1.04^{-1/2} = \mathbf{0.9806}$ . The field's causal cone is now inside the light cone by 2%: this theory has two invariant speeds — one for this field, one for everything built covariantly — and the split is frame-dependent physics: only in the preferred frame defined by $n^\mu$ is the dispersion isotropic; any boosted observer finds the field's speed depending on direction.

Step 4. What experiments see, and how hard they look. Direction-dependent propagation is exactly what the modern descendants of Michelson-Morley hunt: optical resonator experiments compare orthogonal light paths as the Earth rotates, constraining photon-sector coefficients like $\lambda$ at the $\mathbf{10^{-18}}$ level; astrophysical time-of-flight comparisons (photons versus neutrinos versus gravitational waves from the same event — GW170817 famously matched light to $\sim 10^{-15}$ ) bound speed splits between different fields. Our $\lambda = 0.02$ is thus experimentally dead by sixteen orders of magnitude — the point of the exercise being the mechanism: one bare index pair against a fixed vector, and the theory acquires a measurable ether. The index audit is not formalism; it is what these experiments test.

Origins · 1 minHermann Minkowski and how this was discovered

Hermann Minkowski

1864–1909 · Göttingen

Minkowski supplied the stage this article performs on. Einstein's 1905 relativity was algebra — transformation rules relating observers — until Minkowski's September 1908 Cologne address recast it as geometry: "Henceforth space by itself, and time by itself, are doomed to fade away into mere shadows, and only a kind of union of the two will preserve an independent reality." The union is Minkowski spacetime: events as points, the interval as an indefinite metric, Lorentz transformations as the rotations of that geometry, worldlines, light cones, proper time as arc length — the entire vocabulary of the SR chapter and of this article's index bookkeeping. He introduced four-vectors and the systematic index calculus, unified the electric and magnetic fields into the tensor $F_{\mu\nu}$ (making the scalar $F_{\mu\nu}F^{\mu\nu}$ of this article's menu even writable), and gave physics its most productive habit: build theories from objects with definite transformation laws. Einstein initially dismissed the reformulation as "superfluous learnedness" — then discovered general relativity was unreachable without it. Minkowski, who as a Königsberg student had won the Paris Grand Prix in mathematics at eighteen and later taught a young Einstein at Zürich (remembering him as a "lazy dog"), died of appendicitis months after the Cologne lecture, at forty-four. The geometry he left is, a century on, the unquestioned arena of every quantum field theory.

History

1887

— Michelson and Morley fail to detect the ether wind: no preferred frame for light — the experimental seed.

1905

— Einstein derives the Lorentz transformation from two postulates; simultaneity becomes frame-dependent.

1908

— Minkowski geometrizes: spacetime, four-vectors, the invariant interval — covariance becomes index bookkeeping.

1918

— Noether converts the ten-parameter Poincaré symmetry into ten conservation laws for any scalar-action field theory.

1928

— Dirac finds the spinor representation: Lorentz covariance plus quantum mechanics forces antimatter into existence.

2017

— GW170817: gravitational waves and light from one neutron-star merger arrive within seconds across 130 million light-years — propagation-speed equality tested to

10^{-15}

Going further

References. Landau & Lifshitz, The Classical Theory of Fields, §27–32 · Goldstein, Poole & Safko, Classical Mechanics, 3rd ed., Ch. 13 · Weinberg, The Quantum Theory of Fields, Vol. 1, Ch. 2 (Poincaré representations) · Jackson, Classical Electrodynamics, Ch. 11–12 (covariant electrodynamics) · Kostelecký & Russell, "Data Tables for Lorentz and CPT Violation" (living review of experimental bounds)

Provenance. Equations and derivations follow standard treatments: invariant measure and scalar-action argument (Landau–Lifshitz Vol. 2 §27–32; Goldstein Ch. 13); Klein–Gordon equation from $\mathcal{L}_{\text{KG}}$ ; canonical stress-energy tensor from translation invariance (Goldstein Ch. 13, L&L §32). A verifier should: (1) confirm $(\det\Lambda)^2=1$ from $\Lambda^\mathrm{T}\eta\Lambda=\eta$ ; (2) re-derive $(\Box+m^2)\phi=0$ from the given $\mathcal{L}_{\text{KG}}$ ; (3) reproduce $\partial_\mu T^\mu{}_\nu=0$ with $T^\mu{}_\nu$ as stated. No numerical constants are asserted beyond $\det\Lambda=\pm1$ . Additionally for this upgrade: worked-example numbers recomputed directly (boost of $k^\mu = (1.25, 0.75)$ at $v = 0.6$ : $\omega' = 1.0000$ , $k' = 0.0000$ , $k\cdot k = 1.0000$ both frames; modified dispersion $\omega^2 = (k^2+m^2)/1.04$ , $c_\phi = 0.9806$ ; resonator bounds $\sim10^{-18}$ and GW170817 $\sim10^{-15}$ cited as orders of magnitude from the standard literature). Scope note: the Lorentz transformation and four-vector machinery is owned by the SR articles of the continuum chapter (recalled, not re-derived); the KG and stress-energy derivations by the field-equations, Noether and stress-energy articles; the Poincaré charges' construction by the stress-energy and spin-current articles; the bracket generator facts by the Poisson-brackets article.

Builds on: The Lagrangian Density and Field Action · Covariant Euler-Lagrange Equations for Fields · From Discrete Systems to Continuous Fields · Four-Vectors and Minkowski Spacetime · Special Relativity Postulates and the Lorentz Transformation Leads to: Noether's Theorem for Continuous Symmetries · The Canonical Stress-Energy Tensor · Angular Momentum and Spin Current in Fields · Symmetrization and the Belinfante-Rosenfeld Tensor

Is "covariant" the same as "invariant"? The words seem to be used interchangeably.

They are distinct, and keeping them apart prevents real errors. Invariant means unchanged: scalars like $k_\mu k^\mu$ , $P_\mu P^\mu$ , the action $S$ , and proper time are invariants — every observer computes the same number. Covariant means transforming-by-the-rules: a four-vector's components change under a boost, but according to a definite law, so that relations between tensors survive frame changes — that is why "the equations are covariant" is the correct claim, while "the equations are invariant" is loose (their symbolic form is invariant; the component values in them are not). A third word completes the kit: manifestly covariant describes a formalism in which covariance is visible term by term (the Lagrangian formulation, everything written with balanced indices), as opposed to formalisms where it is true but hidden (the Hamiltonian slicing, where the certificate is the charge algebra rather than the notation). The working physicist's habit: state what is invariant (contractions, the action, the mass shell), what is covariant (fields, currents, tensors, the equations), and which formalism makes it manifest. Most index-gymnastics confusion dissolves under that triage.

Where do parity and time reversal fit — they satisfy the defining Lorentz condition but seem different?

They occupy the disconnected pieces of the Lorentz group, and the difference matters for everything Noether-related. The defining condition (1) allows four disconnected components, classified by $\det\Lambda = \pm1$ and the sign of $\Lambda^0{}_0$ : the identity-connected piece (boosts and rotations — the "proper orthochronous" group this article works with), plus copies reached by parity $P$ (spatial flip), time reversal $T$ , and their product. Two consequences. First, Noether's theorem needs a continuous parameter to differentiate along; $P$ and $T$ are discrete jumps, so they yield no currents and no charges — they are symmetries you check, not symmetries you harvest. Second, nothing forces a covariant theory to respect them: the Levi-Civita symbol $\epsilon^{\mu\nu\rho\sigma}$ is an honest invariant tensor of the proper group but flips sign under parity, so terms built with it (the $\epsilon FF$ theta-term, chiral couplings) are fully Lorentz-covariant yet parity-violating. Nature exploits exactly this loophole: the weak interaction violates parity maximally (Wu experiment, 1957) and CP subtly, while respecting the connected Lorentz group perfectly. The one discrete statement that is protected is the CPT theorem — the combination of all three flips is forced by proper-Lorentz invariance plus locality — which is why CPT tests (kaon systems, antihydrogen spectroscopy) are really tests of this article's foundations.

If the Hamiltonian formalism hides Lorentz symmetry, why not always work covariantly?

Because the two formalisms sell different theorems, and mature field theory keeps both. The manifestly covariant Lagrangian language is unbeatable for construction (the index audit designs theories), for symmetry accounting (Noether runs on the action), and — in the quantum theory — for path integrals and diagrammatics, where term-by-term covariance of every propagator and vertex is what makes renormalization tractable. The Hamiltonian language is unbeatable for states and evolution: initial-value problems, conserved-energy arguments, stability, the bracket-to-commutator quantization route, and unitarity (probability conservation is a statement about time evolution, which requires choosing a time). History ran the full circle: the first quantum electrodynamics (Heisenberg-Pauli, 1929) was canonical and covariance was agony to check; the postwar covariant formulations (Feynman, Schwinger, Tomonaga — the latter two explicitly reformulating the canonical theory on arbitrary spacelike slices) made calculations feasible; and the modern proof of QED's consistency uses both, with the algebra (3) as the bridge certifying they describe one theory. The practical rule inherited by this chapter: derive dynamics and symmetries covariantly, analyze states and quantize canonically, and let the Poincaré algebra police the seam.

The two-postulate derivation of special relativity never mentioned actions. What does the field-theory formulation add?

A conversion of kinematics into dynamics — and a vast tightening of what "relativistic" means. Einstein's postulates constrain how observers relate measurements: the Lorentz transformation, simultaneity, time dilation — the SR chapter's content. They do not by themselves dictate what the laws of physics are. The scalar-action principle is the bridge: it takes the kinematic group and turns it into a design filter on dynamics, producing the complete menu of allowed theories (and, via Noether, their complete conservation-law inventory — ten charges, no more, no fewer, for any of them). The tightening shows in what becomes derivable: that dispersion relations must be functions of $k_\mu k^\mu$ (mass shells); that a field's spin dictates its equation type (Wigner); that interactions must be built from local contractions (which, plus quantum mechanics, forces the existence of antiparticles and the CPT and spin-statistics theorems). None of that follows from the postulates alone — it follows from implementing the postulates variationally. There is also a practical inversion worth noting: in modern physics the action formulation is taken as primary, and "special relativity" is simply the statement that the world's action is Poincaré-invariant — the two postulates are recovered as properties of that action's solutions. This article, in other words, is where the SR chapter stops being kinematics and becomes physics.

Emergent Lorentz invariance in materials: how can a lattice produce relativistic physics?

By accident of low-energy bookkeeping — and the phenomenon illuminates exactly what this article's covariance is and is not. Graphene's electrons hop on a honeycomb lattice, a maximally frame-fixed environment; yet near two special points of the band structure the energy-momentum relation is $E = v_F|\mathbf{p}|$ — a massless Dirac cone — and the long-wavelength effective theory is the 2+1-dimensional Dirac equation with $v_F \approx c/300$ playing the role of light speed. The "Lorentz invariance" is emergent: an approximate symmetry of the low-energy description, violated by lattice-scale corrections (higher-derivative terms — index-illegal under the emergent symmetry but perfectly legal under the true crystal symmetry — suppressed by powers of energy over bandwidth). The lesson cuts both ways. It shows covariance can be an output of dynamics rather than an axiom, which keeps alive speculation that nature's own Lorentz symmetry might be emergent from something deeper — precisely why the experimental bounds of the worked example matter: emergent symmetries are typically violated at some scale, and none of the predicted violations have shown up. And it provides laboratories: superfluid helium, Weyl semimetals and cold-atom systems realize analogue relativistic field theories — complete with analogue horizons and Hawking-like radiation — where this chapter's structures are engineered and measured rather than postulated.

What happens to Lorentz covariance in general relativity — is it lost, localized, or generalized?

Localized, and the flat-space audit becomes the template for the curved-space one. In GR, spacetime is curved and no global inertial frames exist, so global Lorentz transformations stop being symmetries of anything; what survives is local Lorentz invariance — at every event, the tangent space is Minkowski, freely-falling observers see special relativity hold in their neighborhood (the equivalence principle), and the field-theory implementation attaches a local Lorentz frame at each point with the freedom to rotate/boost it independently everywhere (the tetrad formalism; spinors in curved spacetime require exactly this). Meanwhile the global coordinate freedom enlarges enormously: arbitrary smooth coordinate changes — general covariance — become the gauge group, with Noether's second theorem governing the bookkeeping (which is why gravitational energy resists localization, as the Belinfante article's FAQ discusses). The design rule generalizes verbatim: actions must be scalars under both local Lorentz rotations and general coordinate transformations — $\mathrm{d}^4x \to \sqrt{-g}\,\mathrm{d}^4x$ , $\eta \to g_{\mu\nu}$ , $\partial \to \nabla$ — and the index audit continues to work, with more index types to balance. This chapter's flat-space covariance is thus not discarded by gravity but gauged: special relativity survives as the pointwise structure of general relativity, exactly as the internal-symmetries article's global phase survived gauging as the pointwise structure of electromagnetism.