Relativistic Lagrangian and Action for Charged Particles

The last article postulated the covariant Lorentz force and rode it through accelerators. This one asks the deeper question: what principle produces that force — and, while we are at it, produces Maxwell's equations too? The mechanics track has already told half the story: its treatment of the relativistic Lagrangian is built on the maximal-aging idea — a free particle extremizes proper time, $S=-mc^2\int d\tau$ — and derives the Lorentz force by varying the 3+1 Lagrangian $L=-mc^2/\gamma+q\mathbf{A}\cdot\mathbf{v}-qV$ . This edition owns the other half: the manifestly covariant construction, where the interaction is a single spacetime scalar, $-q\int A_\mu\,dx^\mu$ , and the payoffs cascade. The field tensor is born in the variation — antisymmetric because an integration by parts says so, not by decree. Gauge invariance of the action turns out to be exactly charge conservation. And the fields themselves get their own Lagrangian density, $-F_{\mu\nu}F^{\mu\nu}/4\mu_0$ — the invariant that Lorentz Invariants of the Field promised would reappear as dynamics — whose variation delivers the inhomogeneous Maxwell equations in one stroke. The construction runs on a principle with a name: minimal coupling. Matter meets electromagnetism through one term, charge times potential along the worldline, and nothing else — no field gradients, no higher powers, no structure. That austerity is not aesthetic. It hard-codes the split between canonical momentum $\pi_\mu=p_\mu+qA_\mu$ and kinetic momentum (a worked example makes the difference numerically vivid, then spends it), and it is the exact classical statement that survives, symbol for symbol, into the Schrödinger, Klein–Gordon, and Dirac equations. This article is where the track's endgame — the bridge to quantum electrodynamics — is classically prefabricated.

Key takeaways

One action governs everything: $S=-mc\int ds-q\int A_\mu dx^\mu-\tfrac{1}{4\mu_0}\int F_{\mu\nu}F^{\mu\nu}\,d^4x$ — particle term, minimal coupling, field term.
Varying the worldline yields $dp_\mu/d\tau=qF_{\mu\nu}u^\nu$ with $F_{\mu\nu}=\partial_\mu A_\nu-\partial_\nu A_\mu$ appearing automatically antisymmetric — the variational birth certificate of the field tensor.
Varying the potential yields $\partial_\mu F^{\mu\nu}=\mu_0 J^\nu$ — the action route to the covariant Maxwell system; the homogeneous pair is prepaid by using $A^\mu$ as the variable.
Gauge transformation $A_\mu\to A_\mu+\partial_\mu\chi$ shifts the worldline action by endpoint terms only (same dynamics); for a continuous source the interaction shifts by $\int\chi\,\partial_\mu J^\mu\,d^4x$ — gauge invariance of the action ⟺ charge conservation.
The canonical momentum is $\pi_\mu=p_\mu+qA_\mu$ : conservation theorems attach to $\pi$ , not to $\gamma m\mathbf{v}$ — and the difference is measurable (a solenoid quench converts one into the other).
The field Lagrangian $-F^2/4\mu_0$ is essentially unique: quadratic, gauge-invariant, parity-even; the rival invariant $F\cdot G$ is a total derivative, and a photon mass term $A\cdot A$ is forbidden by gauge invariance.
Key idea: extremize a Lorentz scalar built from proper time, the potential along the worldline, and the field invariant — and the whole of classical electrodynamics, forces and fields alike, falls out.

The core insight

All of charged-particle electrodynamics follows from extremizing one scalar action, $S=-mc\int ds-q\int A_\mu dx^\mu-\tfrac{1}{4\mu_0}\int F_{\mu\nu}F^{\mu\nu}d^4x$ : worldline variation gives the covariant Lorentz force with $F=\partial A$ born antisymmetric, potential variation gives Maxwell's equations, and gauge invariance of the coupling is charge conservation.

Where this lives today

Open any modern physics paper at the word "Lagrangian" and you are reading this article's grammar. The Standard Model is specified as an action: gauge-invariant kinetic terms (the $-\tfrac14F^2$ pattern, three times over for three forces), matter coupled by covariant derivatives (minimal coupling, promoted to non-abelian charges), plus the Higgs sector. Predictions flow by exactly the two variations performed here — wiggle matter for forces, wiggle fields for field equations — and by the quantum generalization of the scoreboard, Feynman's sum over histories, where $e^{iS/\hbar}$ weights every path and the classical extremum is merely the stationary-phase survivor. The gauge-invariance-equals-conservation theorem of Derivation B, run in reverse, is how new interactions are proposed: postulate a local symmetry, and the coupling structure plus its conserved current follow. Even gravity plays: general relativity is the action $\int R\sqrt{-g}\,d^4x$ , its "minimal coupling" the replacement of $\eta_{\mu\nu}$ by $g_{\mu\nu}$ in this very article's particle term — the $-mc\int ds$ scoreboard, curved, gives geodesics. The bridge article this track ends with (Bridge from Classical Electrodynamics to QED) will need no new mechanics: only $\hbar$ , and the courage to keep every history.

Origins · 1 minKarl Schwarzschild and how this was discovered

Karl Schwarzschild

1873–1916 · Göttingen and Potsdam

Schwarzschild is remembered for the solution he computed in a wartime field hospital — the black-hole geometry that bears his name — but this article rests on his earlier, less famous first: in 1903, twenty-two years before quantum mechanics and two before relativity, he wrote electrodynamics as a least-action principle, with the interaction between charge and field carried by precisely the worldline integral $\int A_\mu dx^\mu$ this article builds on (he called the key quantity the elektrokinetisches Potential). It was the first time the full Maxwell–Lorentz system — fields and particle motion — flowed from a single variational scoreboard. An astronomer by trade (director of the Göttingen and then Potsdam observatories), he treated theory as an observer treats instruments: something to be engineered until it delivers. His action principle waited quietly while Minkowski supplied spacetime and Noether supplied the symmetry dictionary; by the 1920s it had become the default grammar of fundamental physics, and by 1948 Feynman was summing over the very histories Schwarzschild had taught physics to score. He died at 42, of a disease contracted on the Russian front, months after mailing Einstein the exact solution — and thirteen years after quietly writing down the term that runs this article.

History

1903

— Schwarzschild casts the Maxwell–Lorentz theory as a least-action principle: the

\int A_\mu dx^\mu

coupling appears

1906

— Planck extends least action to relativistic particle dynamics; the

-mc^2\int d\tau

term takes its modern role

1915

— Hilbert derives field equations from an invariant action in parallel with Einstein's general relativity

1918

— Noether's theorems weld the action's symmetries to conservation laws; gauge invariance meets charge conservation

1929

— Weyl's gauge principle: local phase symmetry dictates minimal coupling

1948

— Feynman's path integral: every history weighted by

e^{iS/\hbar}

; the classical scoreboard becomes quantum amplitude

The physical picture

An action principle is a scoreboard: assign every conceivable history a single number, and declare that nature runs the histories where the number is stationary. Relativity adds one commandment — the scoreboard must be a Lorentz scalar, or different observers would disagree about which histories win. For a free particle the mechanics track found the essentially unique scalar a worldline owns: its proper time, giving $S_{\text{free}}=-mc^2\int d\tau$ and the maximal-aging story told there. The question for electrodynamics is: what else can a charged worldline contribute to the scoreboard? Inventory what is locally available. The particle carries a scalar $q$ and its displacement $dx^\mu$ ; spacetime offers the potential $A_\mu(x)$ along the route. The simplest scalar you can build is their contraction, summed along the history:

-q\int A_\mu\,dx^\mu,

charge times "potential flux" through the worldline. No metric needed, no $\gamma$ , no reference to any frame — it is as native to spacetime as a line integral can be. The commitment to this term and nothing more is minimal coupling, and the entire article is the harvest of that commitment. Notice what it refuses: terms sensing $\partial A$ directly (those describe particles with structure — dipoles), higher powers of $A$ (nonlinear couplings nature declines for the photon at accessible field strengths), and any dependence on gauge beyond removable endpoint bookkeeping.

Figure

The relativistic scoreboard — proper time plus potential flux along each candidate history

The deepest habit to acquire here is reading invariance as physics. The scoreboard cannot see gauge: change $A_\mu$ by $\partial_\mu\chi$ and each history's interaction score shifts by $-q[\chi(b)-\chi(a)]$ — the same constant for every competitor, since all share endpoints. A contest whose every entrant gains the same points has the same winner: dynamics untouched. But run the argument for a smeared-out current instead of one worldline and the shift is no longer automatic — it vanishes only if $\partial_\mu J^\mu=0$ . What looked like bookkeeping tolerance becomes a conservation law. This inversion — symmetries of the scoreboard dictating conserved currents — is Noether's theorem working its first shift in electrodynamics, and it is the template for every force theory built since.

The action, term by term

S\;=\;\underbrace{-mc\int ds}_{\text{worldline}}\;\underbrace{-\,q\int A_\mu\,dx^\mu}_{\text{minimal coupling}}\;\underbrace{-\,\frac{1}{4\mu_0}\int F_{\mu\nu}F^{\mu\nu}\,d^4x}_{\text{field}}, \qquad (1)

where: $ds=c\,d\tau$ is the invariant line element (Proper Time and Four-Velocity), so the first term is $-mc^2\int d\tau$ — the mechanics track's maximal-aging scoreboard, quoted, not re-derived. The middle term couples the worldline to the potential; parametrized by proper time it reads $-q\int A_\mu u^\mu\,d\tau$ . The last term gives the field its own dynamics, with the invariant $F_{\mu\nu}F^{\mu\nu}=2(B^2-E^2/c^2)$ computed by Lorentz Invariants of the Field — that article deferred "why this invariant becomes the Lagrangian" to here, and the uniqueness argument below pays the debt. For many particles, add one worldline term and one coupling term per charge; for a continuous distribution the coupling smears into $-\int A_\mu J^\mu\,d^4x$ with $J^\mu$ the four-current of The Four-Current Density. In 3+1 language the middle term unpacks to $\int(q\mathbf{A}\cdot\mathbf{v}-qV)\,dt$ — precisely the interaction of the mechanics track's Lagrangian, confirming the two editions describe one theory in two dialects. The covariant dialect earns its keep in the derivations: each result below is three lines where the 3+1 route needs a page of vector identities.

Derivation AWorldline variation — the Lorentz force, with the field tensor born antisymmetric

We vary the particle terms of (1) over worldlines with fixed endpoints and derive $dp_\mu/d\tau=q(\partial_\mu A_\nu-\partial_\nu A_\mu)u^\nu$ : the covariant Lorentz force of the previous article, now a theorem. The strategy is standard variational bookkeeping done covariantly: parametrize by proper time, integrate by parts, and keep honest track of one subtlety — the potential is sampled along the varied path, so its variation carries a gradient term. Watch the antisymmetric combination assemble itself: we never impose it; the algebra of "total derivative minus chain rule" produces it. This is the variational birth certificate of $F_{\mu\nu}$ — the object the chapter built by hand in The Electromagnetic Field Tensor is what any action of this form must construct.

Step 1. Vary the free term. With $ds=\sqrt{\eta_{\mu\nu}dx^\mu dx^\nu}$ and $u_\mu=dx_\mu/d\tau$ ,

\delta\left(-mc\int ds\right)=-m\int u_\mu\,\frac{d(\delta x^\mu)}{d\tau}\,d\tau =\int m\frac{du_\mu}{d\tau}\,\delta x^\mu\,d\tau,

after integrating by parts and dropping the endpoint terms ( $\delta x^\mu=0$ there). The mechanics track's free-particle result sits inside: alone, this variation gives $du_\mu/d\tau=0$ , straight worldlines, maximal aging.

Step 2. Vary the coupling. Two dependencies on the path: the route element $dx^\mu$ and the sampling point of $A_\mu$ :

\delta\left(-q\int A_\mu dx^\mu\right) =-q\int\left(A_\mu\frac{d(\delta x^\mu)}{d\tau}+\partial_\nu A_\mu\,\delta x^\nu u^\mu\right)d\tau.

Why the second term exists: nudging the path to $x+\delta x$ means the particle feels the potential there, not at $x$ — the same convective bookkeeping that generated the magnetic force in the mechanics track's 3+1 derivation, in covariant dress.

Step 3. Integrate the first piece by parts and rename dummy indices so both pieces multiply $\delta x^\mu$ :

\delta S_{\text{int}} =-q\int\left(-\frac{dA_\mu}{d\tau}+\partial_\mu A_\nu\,u^\nu\right)\delta x^\mu\,d\tau =-q\int\left(\partial_\mu A_\nu-\partial_\nu A_\mu\right)u^\nu\,\delta x^\mu\,d\tau,

using the chain rule $dA_\mu/d\tau=\partial_\nu A_\mu\,u^\nu$ along the worldline. The antisymmetric derivative appears on its own: "gradient of the sampled potential" minus "rate of change along the route" is precisely $\partial_\mu A_\nu-\partial_\nu A_\mu$ . Name it $F_{\mu\nu}$ — the variation has rebuilt the field tensor.

Step 4. Demand stationarity for arbitrary $\delta x^\mu$ :

\frac{dp_\mu}{d\tau}=qF_{\mu\nu}u^\nu,

the covariant Lorentz force — equation (1) of Relativistic Dynamics of a Charged Particle, index-lowered. Everything that article executed (exact 3+1 reduction, mass constancy, the three solved motions) is now downstream of one scalar scoreboard. Note the logic of gauge invariance made mechanical: only the antisymmetric combination survived the variation, and the antisymmetric combination is exactly the gauge-proof content of $A_\mu$ . Verified: variation recomputed symbolically (sympy, arbitrary $A_\mu(x)$ and parametrized path); the 3+1 reduction matches the mechanics track's Lagrangian route (its verification note reciprocates), and the sign conventions land $d\mathbf{p}/dt=q(\mathbf{E}+\mathbf{v}\times\mathbf{B})$ with this track's dictionary.

Go deeper · Derivation B · 2 minGauge invariance of the action — and charge conservation as its price

Derivation BGauge invariance of the action — and charge conservation as its price

We subject each term of (1) to the gauge transformation $A_\mu\to A_\mu+\partial_\mu\chi$ and prove two statements with one computation: for a single worldline, the action shifts by endpoint constants only, so the dynamics is gauge-blind; for a continuous source, the interaction is invariant if and only if charge is conserved. The strategy is integration by parts in one dimension (along the worldline) and in four (over spacetime), watching where the boundary terms land. This derivation is the article's conceptual center: it converts the chapter-long slogan "gauge freedom is redundancy" into a conservation law with teeth.

Step 1. The worldline coupling. Under the transformation,

-q\int_a^b A_\mu dx^\mu\;\to\;-q\int_a^b A_\mu dx^\mu-q\int_a^b \partial_\mu\chi\,dx^\mu =-q\int_a^b A_\mu dx^\mu-q\left[\chi(b)-\chi(a)\right].

The added piece is a pure endpoint term — the same for every candidate history in the variational contest, since all share $a$ and $b$ . Stationary points are unmoved: same physics, shifted scoreboard. (The free term never sees $\chi$ ; the field term is invariant because $F$ is — one line, prepaid by Derivation A's antisymmetry.)

Step 2. The smeared coupling. For a continuous current the interaction reads $-\int A_\mu J^\mu\,d^4x$ , and the same transformation adds

-\int \partial_\mu\chi\,J^\mu\,d^4x =+\int \chi\,\partial_\mu J^\mu\,d^4x\;-\;\oint_{\partial}\chi J^\mu\,dS_\mu,

by the four-dimensional integration by parts. The surface term dies for localized sources (or $\chi$ of compact support); what remains is an integral of $\chi$ against $\partial_\mu J^\mu$ .

Step 3. Read the theorem in both directions. If $\partial_\mu J^\mu=0$ , the shift vanishes for every $\chi$ : the action is gauge-invariant. Conversely, if the action is to be gauge-invariant for arbitrary $\chi$ , the factor multiplying $\chi$ must vanish identically: $\partial_\mu J^\mu=0$ . Gauge invariance of the coupling ⟺ charge conservation. The chapter has now met this law three times — as an experimental fact (The Four-Current Density), as an algebraic identity of the field equations (Covariant Form of Maxwell's Equations) — and here as the deepest version: a symmetry requirement on the scoreboard. The three are one: the field equations have the identity because they come from this action (Derivation C), and the action tolerates only conserved sources.

Step 4. The quantum foreshadow, stated honestly. In quantum mechanics the endpoint shift of Step 1 cannot be discarded — it multiplies the wavefunction by the phase $e^{-iq[\chi(b)-\chi(a)]/\hbar}$ — and consistency forces matter fields to transform along with the potential: $\psi\to e^{iq\chi/\hbar}\psi$ , Fock's gradient invariance (Gauge Transformations and Covariant Gauge Conditions told his story). Demanding local phase freedom then reconstructs the coupling: derivatives must appear as $\partial_\mu+iqA_\mu/\hbar$ , the covariant derivative. Classically we chose minimal coupling for economy; quantum mechanically it is forced. That reversal is the U(1) gauge principle — the Standard Model's first axiom — and it is planted exactly here. Verified: endpoint algebra checked directly; the 4D parts computation re-run with explicit localized test functions (sympy); logic cross-checked against Zangwill §24 and Landau–Lifshitz §16.

Go deeper · Derivation C · 2 minField variation — Maxwell's equations from the invariant, and why the Lagrangian had no choice

Derivation CField variation — Maxwell's equations from the invariant, and why the Lagrangian had no choice

We give the field its dynamics: vary (1) with respect to $A_\mu(x)$ — the potentials, not the worldlines, are now the wiggled variables — and obtain $\partial_\mu F^{\mu\nu}=\mu_0J^\nu$ , the inhomogeneous half of Covariant Form of Maxwell's Equations, by the second route promised there. Then we close the uniqueness case for the Lagrangian density itself: among local, quadratic, gauge-invariant, parity-even scalars, $-F^2/4\mu_0$ is the only candidate. The strategy for the variation is the field-theory analogue of Derivation A — integrate by parts in four dimensions and let antisymmetry organize the terms. Watch that the homogeneous Maxwell pair never appears as an equation of motion: choosing $A_\mu$ as the variable prepays it, exactly as the covariant-Maxwell article's warning band said.

Step 1. Write the field-plus-interaction pieces as a Lagrangian density and vary $A_\nu\to A_\nu+\delta A_\nu$ :

\mathcal{L}=-\frac{1}{4\mu_0}F_{\mu\nu}F^{\mu\nu}-A_\mu J^\mu, \qquad \delta F_{\mu\nu}=\partial_\mu\,\delta A_\nu-\partial_\nu\,\delta A_\mu.

The quadratic term's variation, using the antisymmetry of $F$ to fold the two identical halves together:

\delta\!\left(-\frac{F^2}{4\mu_0}\right)=-\frac{1}{\mu_0}F^{\mu\nu}\partial_\mu\,\delta A_\nu.

Step 2. Integrate by parts (fields die at infinity) and collect:

\delta S=\int\left(\frac{1}{\mu_0}\partial_\mu F^{\mu\nu}-J^\nu\right)\delta A_\nu\,d^4x.

Stationarity for arbitrary $\delta A_\nu$ delivers

\partial_\mu F^{\mu\nu}=\mu_0 J^\nu

— the dynamical line of the two-line system, derived rather than assembled. The homogeneous line costs nothing: with $A_\mu$ fundamental, $F=\partial A$ antisymmetrized satisfies the Bianchi identity automatically (The Dual Field Tensor's theorem). Two routes now meet: the covariant-Maxwell article packaged the 3+1 laws into this equation; the action just generated it from a symmetry-constrained scalar. Physics doubly checked.

Step 3. Uniqueness — why $-F^2/4\mu_0$ and nothing else. Shop the menu of local scalars built from $A$ and one derivative, at most quadratic (linear field equations), gauge-invariant, parity-even. Gauge invariance forces dependence through $F_{\mu\nu}$ alone — any bare $A$ (e.g. a photon mass term $\tfrac12 m_\gamma^2 A_\mu A^\mu$ ) shifts under $\chi$ and is disqualified; this is why gauge invariance means massless light, with the Proca alternative and its finite-range Coulomb law parked as the measurable counterfactual (current bounds: $m_\gamma c^2<10^{-18}$ eV). Two quadratic invariants survive, both computed by Lorentz Invariants of the Field: $F_{\mu\nu}F^{\mu\nu}$ and $F_{\mu\nu}G^{\mu\nu}$ .

Step 4. Eliminate the second by exhibiting it as a total derivative:

F_{\mu\nu}G^{\mu\nu}=\partial_\mu\!\left(2\,\epsilon^{\mu\nu\rho\sigma}A_\nu\partial_\rho A_\sigma\right),

so adding it to $\mathcal{L}$ changes $S$ by boundary terms only — no contribution to any equation of motion. (Its integrand, $\propto\mathbf{E}\cdot\mathbf{B}$ , is also parity-odd: doubly spent.) Classically dead, this " $\theta$ -term" leads a famous afterlife in quantum non-abelian theory, where topology makes boundary terms count — a story for the QFT track. The normalization $1/4\mu_0$ is then fixed by Step 2 landing Maxwell with SI sources, and the sign by energy positivity: the time-slot energy density this Lagrangian generates (next article's $\theta^{00}$ ) must come out positive. The Lagrangian had no choice at all. Verified: the variation recomputed symbolically with arbitrary $\delta A$ ; the total-derivative identity for $F\cdot G$ expanded and confirmed (sympy, all 24 $\epsilon$ terms); the $\mathcal{L}$ sign convention checked against the energy density of E7's ledger and Jackson §12.7.

Canonical versus kinetic momentum — where conservation theorems actually attach

The Lagrangian's momentum is not the particle's $m\gamma\mathbf{v}$ . Differentiating the interaction term along with the free term,

\pi_\mu=\frac{\partial L}{\partial u^\mu}=p_\mu+qA_\mu, \qquad (2)

where: $p_\mu=mu_\mu$ is the kinetic four-momentum that hits detectors, and $qA_\mu$ is the field's bookkeeping share. Noether's machinery — cyclic coordinates, conserved quantities — attaches to $\pi_\mu$ , not to $p_\mu$ : when the potentials are independent of some coordinate, it is the canonical combination that is conserved along the motion. The corresponding Hamiltonian, $H=\sqrt{c^2(\boldsymbol{\pi}-q\mathbf{A})^2+m^2c^4}+qV$ , and its role as quantization's launchpad are owned by the mechanics track (The Relativistic Lagrangian for a Free and Charged Particle); what this edition owns is the covariant statement (2) and its physical teeth — the worked examples below make $qA_\mu$ the dominant part of $\pi$ for a slow electron near a solenoid, then convert it into visible kinetic momentum by switching the field off. Minimal coupling in one prescription: wherever free physics says $p_\mu$ , charged physics says $\pi_\mu-qA_\mu$ — the substitution rule that quantum mechanics will inherit as $\hat p_\mu\to\hat p_\mu-qA_\mu$ and field theory as the covariant derivative.

question	reach for	why
force on a given charge	Derivation A's $dp_\mu/d\tau=qF_{\mu\nu}u^\nu$	worldline variation; executed solutions in the previous article
fields from given sources	Derivation C's $\partial_\mu F^{\mu\nu}=\mu_0J^\nu$	potential variation; matches the covariant-Maxwell package
conserved quantity from a symmetry	canonical $\pi_\mu=p_\mu+qA_\mu$	Noether attaches to the Lagrangian's momentum
is a proposed coupling legitimate	gauge-shift the action	endpoint terms or $\partial\cdot J$ — Derivation B's audit
energy/momentum bookkeeping of the field	next article's $\theta^{\mu\nu}$	this action, Noether-processed

Limiting cases

case	result	meaning
$q\to0$	$S=-mc^2\int d\tau$	free maximal aging — the mechanics track's story
$v\ll c$	$L\to\tfrac12mv^2+q\mathbf{A}\cdot\mathbf{v}-qV$ (+const)	the nonrelativistic charged Lagrangian of classical mechanics
$A_\mu$ pure gauge ( $A=\partial\chi$ )	endpoint scores only	no fields, no forces — flat scoreboard
null field ( $E=cB$ , $\mathbf{E}\perp\mathbf{B}$ )	$\mathcal{L}_{\text{field}}=0$ pointwise	light's Lagrangian density vanishes — worked example 1
add $m_\gamma^2A^2$ term	gauge broken; Yukawa-range Coulomb	Proca counterfactual; bounds $m_\gamma c^2<10^{-18}$ eV
add $F\cdot G$ term	boundary term; classical dynamics blind	the $\theta$ -term, dormant until quantum topology

Common trapThe conserved momentum is canonical — and it is gauge-dependent, so cash it out before comparing

Two linked traps. First: in a magnetic field, symmetry does not conserve $\gamma m\mathbf{v}$ . A charge orbiting in an axially symmetric field conserves the canonical $\pi_\phi=p_\phi+qA_\phi$ ; forgetting the $qA$ share produces "paradoxes" wherever flux changes — the solenoid-quench example below is the classic. Second: $\pi_\mu$ is built from the gauge-dependent $A_\mu$ , so its value depends on your gauge slice. That is not a contradiction — in any fixed gauge the conservation theorem holds exactly, and all physical extractions (kinetic momenta, energies, deflections) come out gauge-proof — but it means canonical momenta from different gauges must never be compared raw. The professional protocol: choose the gauge, exploit the symmetry, and convert $\pi\to p=\pi-qA$ before quoting anything measurable. Quantum mechanics inherits the same discipline verbatim: canonical operators are gauge-covariant, expectation values of kinetic quantities are what experiments see.

Figure

Momentum in escrow — the canonical split for an electron outside a solenoid

Worked examples

Example 1 — the Lagrangian density of light is zero. A 1.0 kW/m² laser beam (bright sunlight's intensity) has peak fields $E_0=\sqrt{2I/c\varepsilon_0}=868$ V/m and $B_0=E_0/c=2.90\ \mu$ T. Evaluate $\mathcal{L}_{\text{field}}=\tfrac{\varepsilon_0}{2}E^2-\tfrac{1}{2\mu_0}B^2$ .

Step 1. Electric share at the peak: $\tfrac{\varepsilon_0}{2}E_0^2=3.34\times10^{-6}$ J/m³. Magnetic share: $B_0^2/2\mu_0=3.34\times10^{-6}$ J/m³. Identical to all digits.

Step 2. So $\mathcal{L}=0$ — at every instant and every point of the wave, not just on average. The Lagrangian density is $-\tfrac{1}{4\mu_0}F^2=-\tfrac{1}{2\mu_0}(B^2-E^2/c^2)$ : a multiple of the invariant, and a plane wave is a null field ( $\mathcal{I}_1=0$ , per Lorentz Invariants of the Field). Every inertial observer agrees light scores zero.

Step 3. Contrast the non-null cases: a 1 MV/m capacitor field scores $\mathcal{L}=+4.4$ J/m³; a 1 T magnet scores $\mathcal{L}=-4.0\times10^5$ J/m³. And note what $\mathcal{L}$ is not: the energy density of the laser field is $u=I/c=3.3\times10^{-6}$ J/m³ $\neq0$ . The Lagrangian weighs $E$ against $B$ ; the energy adds them — the difference between the invariant scoreboard and the frame-dependent ledger, and a preview of why the next article needs a full tensor for the latter.

Example 2 — the solenoid quench: canonical momentum does real work. A long solenoid of radius $r_0=1.0$ cm carries an internal field $B=0.010$ T. An electron circles outside at $r=2.0$ cm, where $\mathbf{B}=0$ but $A_\phi=Br_0^2/2r=2.5\times10^{-5}$ T·m, with canonical angular component $\pi_\phi\equiv\pi=6.0\times10^{-24}$ kg·m/s (per unit: the $\phi$ -components of the vectors at radius $r$ ).

Step 1. Inventory the split (2): the field's share is $qA_\phi=(1.602\times10^{-19})(2.5\times10^{-5})=4.0\times10^{-24}$ kg·m/s — two-thirds of the canonical total. Kinetic share: $p=\pi-qA=2.0\times10^{-24}$ kg·m/s, i.e. $v=p/m=2.2\times10^6$ m/s ( $0.007c$ ; nonrelativistic bookkeeping licensed), kinetic energy $13.6$ eV.

Step 2. Quench the solenoid: $B\to0$ over a few milliseconds. During the quench, the changing flux drives an induced azimuthal electric field at the electron (Faraday, from the same changing $A$ : $\mathbf{E}=-\partial_t\mathbf{A}$ ), delivering an impulse $\int qE_\phi\,dt=-q\Delta A_\phi=+qA_\phi$ .

Step 3. Audit with the symmetry. The configuration stays axially symmetric throughout, so $\pi_\phi$ is conserved: after the quench $A=0$ , hence $p'=\pi=6.0\times10^{-24}$ kg·m/s — all canonical momentum is now kinetic. Check against Step 2's impulse: $p'=p+qA_\phi=2.0+4.0=6.0\times10^{-24}$ ✓. The electron speeds up to $v'=6.6\times10^6$ m/s ( $123$ eV) while the magnet it never touched dies two centimeters away. The canonical momentum was not bookkeeping; it was momentum in escrow, and Faraday's induced field is the escrow agent. (Quantum mechanics sharpens the same geometry into the Aharonov–Bohm phase — the potential outside the solenoid is physically potent even without quenching.)

Going further

Reading. Landau & Lifshitz, Classical Theory of Fields, §§8, 16–17, 23, 27–30 — the covariant action treatment this article's register follows. Jackson, Classical Electrodynamics, 3rd ed., §12.1 and §12.7 — Lagrangians for particles and fields. Zangwill, Modern Electrodynamics, Ch. 24. Griffiths, 4th ed., §12.3.5 for the force-law connection.

Provenance. Built from the AllQFT base article for this topic (action structure, minimal-coupling doctrine, canonical/kinetic split, and gauge-invariance-conservation link retained), conventions locked to this track's $(+,-,-,-)$ dictionary. A verifier should: (1) rerun both variations (worldline and potential) and confirm the endpoint bookkeeping; (2) check the total-derivative identity for $F\cdot G$ ; (3) recompute Example 1 ( $E_0=868$ V/m, $B_0=2.90\ \mu$ T, both Lagrangian shares $3.34\times10^{-6}$ J/m³, $\mathcal{L}=0$ ; capacitor $+4.4$ J/m³ at 1 MV/m, magnet $-4.0\times10^5$ J/m³ at 1 T) and Example 2 ( $qA_\phi=4.0\times10^{-24}$ kg·m/s, $p=2.0\times10^{-24}$ , $13.6$ eV; post-quench $6.0\times10^{-24}$ , $6.6\times10^6$ m/s, $123$ eV; impulse audit closes). Division of labor for this edition: the maximal-aging story, the 3+1 Lagrangian route to the Lorentz force, the Hamiltonian and the einbein formulation are owned by the mechanics track's The Relativistic Lagrangian for a Free and Charged Particle (companion, cross-checked); executed particle dynamics by Relativistic Dynamics of a Charged Particle; the two-line system this action re-derives by Covariant Form of Maxwell's Equations; the invariants menu by Lorentz Invariants of the Field (its Lagrangian deferral delivered here); the stress-energy harvest of this action by The Electromagnetic Stress-Energy Tensor (next); field-theoretic Noether machinery by the mechanics track's field-theory chapter. This edition owns the covariant variations, the gauge-invariance-conservation theorem, the Lagrangian uniqueness case, and the canonical-momentum physics.

Builds on: Proper Time and Four-Velocity · The Four-Potential · The Electromagnetic Field Tensor · Relativistic Energy and Four-Momentum · Four-Vectors and the Metric Tensor

Leads to: The Electromagnetic Field Tensor · The Four-Current Density · Transformation of Electric and Magnetic Fields

Why is the interaction term unique — could nature have chosen something else?

Within the stated rules — locality, Lorentz invariance, gauge invariance up to endpoints, and minimality (no new dimensionful constants) — yes, $-q\int A_\mu dx^\mu$ is the only worldline scalar available, which is the uniqueness half of "minimal coupling." But the rules themselves encode a physical claim: that the particle is structureless. Relax minimality and legitimate extra terms appear, each with its own coupling constant and physics: $-\tfrac12 m_{\mu\nu}\int F^{\mu\nu}d\tau$ -type terms describe intrinsic dipole moments (a neutron couples to fields only this way, having $q=0$ ), and gradient couplings describe polarizable composites. The discovery that the electron's behavior is captured by the minimal term alone — to the parts-per-trillion precision of the $g-2$ industry, where the tiny calculable non-minimal residue comes from QED itself — is among the strongest statements we own about the electron's pointlike nature.

The action uses the gauge-dependent potential. How can physics built on an unphysical quantity be trusted?

Because the theory only ever commits to gauge-proof outputs. Derivation B is the audit: the gauge shift moves the action by endpoint constants (one particle) or by zero (conserved sources), so stationary worldlines, field equations, and everything measurable are unmoved. The potential plays the role of a coordinate system for the interaction — arbitrary in itself, indispensable for local bookkeeping. Could one write the action using only $F_{\mu\nu}$ ? Not locally: no local worldline scalar built from $F$ alone reproduces the Lorentz force (try it — the available candidates like $\int F_{\mu\nu}u^\mu u^\nu d\tau$ vanish identically by antisymmetry). The lesson sharpens in quantum mechanics, where loop integrals $\oint A\cdot dx$ — gauge-proof but not expressible through local fields along the loop — produce the Aharonov–Bohm effect: the potential is more than bookkeeping exactly to the extent topology allows.

What exactly would change if the photon had a small mass?

Add the Proca term $+\tfrac{1}{2\mu_0}(m_\gamma c/\hbar)^2A_\mu A^\mu$ to the Lagrangian and follow the consequences. Gauge invariance dies (Derivation B's audit fails — the action now depends on $\chi$ ), so the potential becomes physical and the Lorenz condition becomes a derived equation rather than a choice. The Coulomb potential acquires a Yukawa cutoff $e^{-r/\lambda}/r$ with $\lambda=\hbar/m_\gamma c$ ; light gains a longitudinal third polarization and travels dispersively ( $v_g$ depends on frequency). Every one of these is hunted experimentally: galactic-scale magnetic fields and torsion-balance tests of Coulomb's law bound $\lambda$ beyond $10^{11}$ m, i.e. $m_\gamma c^2\lesssim10^{-18}$ eV. The theoretical moral runs the other way: because we observe exact gauge invariance, the mass term is forbidden, and light's masslessness is understood structurally rather than accidentally.

How literally does quantum mechanics inherit the minimal-coupling rule?

Symbol for symbol. The classical canonical relation $\pi=p+q A$ becomes the operator statement that the canonical momentum $\hat{\boldsymbol\pi}=-i\hbar\nabla$ generates translations, while the kinetic momentum is $\hat{\boldsymbol\pi}-q\mathbf{A}$ ; every Hamiltonian is built by that substitution (Schrödinger, Pauli, Dirac — all of them). Equivalently, derivatives of charged fields appear only in the combination $D_\mu=\partial_\mu+iqA_\mu/\hbar$ , the covariant derivative, which is Derivation B's endpoint phase made local: gauge-shifting $A$ while phase-rotating $\psi$ leaves $D_\mu\psi$ transforming like $\psi$ itself. The logic even inverts: demand that the phase of $\psi$ be locally unobservable and the existence of $A_\mu$ , its gauge law, and the minimal coupling are forced. That inversion, generalized from U(1) phases to larger symmetry groups, is the construction principle of the entire Standard Model.

Where is the maximal-aging story in all this — did the covariant route abandon it?

It is the first term, quoted whole. The mechanics track's The Relativistic Lagrangian for a Free and Charged Particle owns that narrative: $-mc^2\int d\tau$ scores each worldline by the aging of its traveler, free motion maximizes proper time, and the twin asymmetry is the variational statement seen from inside. This edition deliberately builds on top of that result rather than retelling it: the charged term $-q\int A\cdot dx$ is a second, independent scalar added to the same scoreboard, and the interplay is the physics — a charge's actual worldline is the compromise between aging as much as possible and surfing the potential as profitably as possible. In a static electric field the compromise reads: sag toward low potential energy costs coordinate speed, which buys aging — the electromagnetic version of the "why does a thrown ball arc" story gravity tells in the general-relativity track.

Why does the Lagrangian density of light vanish, and does anything measurable follow from that?

It vanishes because $\mathcal{L}=-\mathcal{I}_1/4\mu_0$ is proportional to the field invariant, and radiation is a null field: $E=cB$ with $\mathbf{E}\perp\mathbf{B}$ makes $B^2-E^2/c^2=0$ pointwise (Example 1's arithmetic). Measurable consequences are indirect but real. First, it certifies frame-independence: no observer can boost light into a "mostly electric" or "mostly magnetic" configuration — the zero is invariant. Second, it explains why free radiation contributes to the action only through interference-style boundary effects while contributing robustly to energy — the scoreboard and the ledger are different projections of $F$ , and the next article's stress-energy tensor exists precisely because energy bookkeeping needs sixteen slots where the action needed one. Third, in nonlinear extensions (Euler–Heisenberg, Born–Infeld) corrections enter as powers of the invariants, so pure plane waves famously fail to self-interact even there — crossed beams, with non-null total fields, are required, which is exactly how light-by-light scattering searches are designed.

What plays this article's role for the other forces?

The same script with a bigger cast. Each Standard-Model force is specified by a gauge-invariant field term (the $-\tfrac14F^2$ pattern generalized: $-\tfrac14F^a_{\mu\nu}F^{a\,\mu\nu}$ with an internal index $a$ and self-interaction terms forced by the non-abelian symmetry) plus matter coupled through covariant derivatives — minimal coupling with matrix-valued charges. Weak isospin and color repeat electromagnetism's grammar; only the symmetry group changes. Gravity fits with one twist: the "potential" is the metric itself, the particle term is this article's $-mc\int ds$ with $g_{\mu\nu}$ replacing $\eta_{\mu\nu}$ (worldline variation now yields geodesics), and the field term is the Einstein–Hilbert action. The uniqueness arguments of Derivation C reappear each time — which is why theoretical physics proposals are argued at the level of which terms the symmetries permit, this article's Step-3 menu audit scaled up to research practice.