Covariant Form of Maxwell's Equations

In 1865 electromagnetism was twenty equations in twenty variables. Heaviside's vector calculus compressed them to the four you have lived with for this whole track. This article performs the final compression, and it is the payoff the last ten articles have been assembling piece by piece: all of electromagnetism fits in two lines,

\partial_\mu F^{\mu\nu}=\mu_0 J^\nu, \qquad\qquad \partial_\mu G^{\mu\nu}=0,

and the second line is bookkeeping — an identity the potentials satisfy automatically. Every object in those lines has already been built and certified: the four-gradient $\partial_\mu$ , the four-current $J^\mu$ , the field tensor $F^{\mu\nu}$ with $\mathbf{E}$ and $\mathbf{B}$ woven into its slots, and its shadow-partner, the dual tensor $G^{\mu\nu}$ . What remains — this article's business — is the system: proving these two lines say exactly what Gauss, Ampère, Maxwell, and Faraday said, slot by slot; watching charge conservation stop being a law and become an algebraic consistency condition; and cashing the cheque relativity has been writing since the postulates article, because an equation between tensors that holds in one inertial frame holds in every one, visibly, with no calculation. That last point is the deep one. Maxwell wrote his equations forty years before special relativity, for a stationary aether, and they turned out to be perfectly Lorentz-invariant — the first relativistic field theory, discovered by accident. The covariant form makes the accident impossible to miss: the equations could not break Lorentz symmetry if they tried. This is also the form in which electromagnetism enters the action principle, quantum electrodynamics, and general relativity — the two lines are the template every later force theory copies.

Key takeaways

All of Maxwell's electrodynamics is two tensor equations: $\partial_\mu F^{\mu\nu}=\mu_0 J^\nu$ (Gauss + Ampère–Maxwell) and $\partial_\mu G^{\mu\nu}=0$ (no monopoles + Faraday), the second equivalent to the Bianchi identity $\partial_\lambda F_{\mu\nu}+\partial_\mu F_{\nu\lambda}+\partial_\nu F_{\lambda\mu}=0$ .
The $\nu$ index files the familiar laws: the time slot of each equation holds a divergence law, the three space slots hold a curl law — eight scalar slots, audited one by one in this article.
Charge conservation $\partial_\mu J^\mu=0$ rides free: a symmetric derivative pair contracted with the antisymmetric $F^{\mu\nu}$ vanishes identically, so the system refuses to accept a non-conserved source.
Covariance is manifest: both sides are four-tensors, so the system boosts as a whole. The individual 3+1 laws do not — a boosted observer's Gauss law is a $\gamma$ -weighted mixture of your Gauss and Ampère laws; only the package is frame-proof.
Written through the potentials, the whole content is $\Box A^\nu-\partial^\nu(\partial_\mu A^\mu)=\mu_0 J^\nu$ ; in Lorenz gauge this collapses to the four wave equations $\Box A^\mu=\mu_0 J^\mu$ already assembled by The Four-Potential.
Key idea: bundle $\mathbf{E}$ and $\mathbf{B}$ into $F^{\mu\nu}$ ; then all of electromagnetism is $\partial_\mu F^{\mu\nu}=\mu_0 J^\nu$ plus an identity, both tensorial, hence valid in every inertial frame.

The core insight

Bundle $\mathbf{E}$ and $\mathbf{B}$ into the antisymmetric tensor $F^{\mu\nu}=\partial^\mu A^\nu-\partial^\nu A^\mu$ ; then all of electromagnetism is $\partial_\mu F^{\mu\nu}=\mu_0 J^\nu$ together with $\partial_{[\lambda}F_{\mu\nu]}=0$ , both manifestly tensorial and hence valid in every inertial frame.

Where this lives today

The covariant system is not a notational luxury; it is the form in which electrodynamics is solved at the frontier. Relativistic particle-in-cell codes that simulate laser–plasma accelerators and pulsar magnetospheres push millions of charges with the covariant Lorentz force while evolving the fields by exactly the slot structure audited above — constraints monitored, evolution slots stepped, charge conservation enforced by the same identity that makes equation (1) consistent. Astrophysicists modeling black-hole jets solve "force-free electrodynamics," which is nothing but $\partial_\mu F^{\mu\nu}=\mu_0 J^\nu$ with the current determined self-consistently by the fields. And spacecraft do the audit in reverse: ESA's four-satellite Cluster mission flies a tetrahedron through Earth's magnetopause and computes the current from the measured curl of $\mathbf{B}$ — a 20 nT variation across 200 km implies $J\approx\Delta B/\mu_0 L\approx8\times10^{-8}$ A/m² — literally reading the $\nu=1,2,3$ slots of equation (1) off the sky. The deeper legacy is structural. The two-line system is the template of modern physics: general relativity's Einstein equations, the Yang–Mills equations of the strong and weak forces, all have the shape "divergence of a field tensor equals a source current, plus a Bianchi identity." When this track reaches the bridge to quantum electrodynamics, the two lines will be re-derived in one stroke from an action principle — Relativistic Lagrangian and Action for Charged Particles plants that seed — and their form will survive quantization untouched. Maxwell's accidental relativistic field theory turned out to be everyone's first draft of everything.

Origins · 1 minOliver Heaviside and how this was discovered

Oliver Heaviside

1850–1925 · self-taught telegrapher, London and Devon

Heaviside left school at sixteen, taught himself Maxwell's Treatise while working as a telegraph operator, and then did to Maxwell what Maxwell had done to Faraday: found the compact structure hiding inside the sprawl. In 1884–85 he recast the twenty-equation system into the four vector equations every student now learns — "Maxwell redressed," in his phrase — inventing much of modern vector calculus in the process and waging a decades-long war against the quaternionists who considered his pragmatic vectors vulgar. He saw the duplex symmetry between the electric and magnetic halves of the system, introduced the magnetic-current term to display it, and grumbled presciently that the equations' beauty was being wasted on component-by-component drudgery. He would have recognized the two-line covariant form instantly as the finish line of his own program: the 1908 compression differs from his 1885 one only in fusing time into the objects as well as space. A recluse who signed letters W.O.R.M., he died poor; the four equations on every physics t-shirt are, by right of authorship, his.

History

1865

— Maxwell's Dynamical Theory of the Electromagnetic Field: twenty equations in twenty variables, radiation at

c

predicted

1884

— Poynting localizes energy flow; Heaviside begins recasting the system in vector form

1885

— Heaviside publishes the four vector equations — the first great compression

1905

— Einstein's electrodynamics paper: the equations' symmetry group, not the aether, is the physics

1908

— Minkowski's Grundgleichungen: electrodynamics written as tensor equations on spacetime — the two-line form

1918

— Noether and Weyl connect the system's structure to conservation and gauge principles, the template for all later field theories

The physical picture

Physics has a habit of measuring understanding in compression. Maxwell's 1865 memoir carried twenty component equations because vectors had not been invented yet; every curl was written out one Cartesian slot at a time. Heaviside's 1884–85 rewriting — "Maxwell redressed," he called it — fused those components into four vector equations by recognizing that $E_x, E_y, E_z$ are one thing, an arrow, whose components mix under rotations. The move you are about to watch is the same move one dimension up. The relativity chapter has spent ten articles demonstrating that $\mathbf{E}$ and $\mathbf{B}$ are themselves components — six slots of one antisymmetric spacetime tensor — that mix under boosts exactly as $E_x$ and $E_y$ mix under rotations. If the fields fuse into one object, the laws governing them should fuse too. They do: four vector equations become two tensor equations, and the second one turns out to be an automatic consequence of the potentials' existence.

Figure

The great compression — twenty component equations to four vector laws to two tensor lines

There is a picture worth holding for the whole article. An antisymmetric tensor's divergence, $\partial_\mu F^{\mu\nu}$ , is the four-dimensional version of "field lines beginning and ending": the inhomogeneous line says the only thing that can start or stop the field object's lines in spacetime is the four-current. Its time slot counts lines of $\mathbf{E}$ starting on charge (Gauss); its space slots count how circulating $\mathbf{B}$ and changing $\mathbf{E}$ trade off against flowing charge (Ampère–Maxwell). The homogeneous line says the dual object's lines never start or stop at all: no magnetic charge (time slot), and Faraday induction as pure geometry (space slots). One object, one source, one identity. What the compression buys is not elegance but inevitability. In the 3+1 language, checking that Maxwell's equations survive a Lorentz boost is a gruesome afternoon: transform the fields with the six mixing rules, transform the coordinates and derivatives, transform $\rho$ and $\mathbf{J}$ , and watch four equations scramble into each other before miraculously reassembling. In the covariant language the same check is one line — both sides are four-tensors, so if they agree in one frame they agree in all. Derivation B below does the afternoon's work in three steps, and then extracts the interesting corollary hiding inside the miracle: the individual laws are not separately covariant. "Gauss's law" is a frame-dependent statement. What is frame-proof is the two-line package.

The two lines, and what sits in their slots

Everything below is written in this track's locked conventions: signature $(+,-,-,-)$ , $A^\mu=(V/c,\mathbf{A})$ , $J^\mu=(c\rho,\mathbf{J})$ , and the component dictionary constructed by The Electromagnetic Field Tensor,

F^{\mu\nu}=\begin{pmatrix}0&-E_x/c&-E_y/c&-E_z/c\\ E_x/c&0&-B_z&B_y\\ E_y/c&B_z&0&-B_x\\ E_z/c&-B_y&B_x&0\end{pmatrix},

where: the first row carries $-\mathbf{E}/c$ , the first column $+\mathbf{E}/c$ , and the spatial block weaves in $\mathbf{B}$ through $F^{ij}=-\epsilon^{ijk}B_k$ . Lowering both indices flips exactly the electric entries ( $F_{0i}=+E^i/c$ ), a sign asymmetry that article prosecutes in detail — including against its own base text. We quote its dictionary and never re-derive it. The dynamical law of electromagnetism — the only dynamical law — is the inhomogeneous equation:

\partial_\mu F^{\mu\nu}=\mu_0 J^\nu. \qquad (1)

where: $\partial_\mu=(\tfrac{1}{c}\partial_t,\nabla)$ is the four-gradient (born covariant, per Four-Vectors and the Metric Tensor), the repeated $\mu$ is summed, and the free index $\nu$ makes this four scalar equations filed in one envelope. Read it physically: the spacetime divergence of the field object equals the source current. It is linear — superposition survives relativity intact — and first-order in the field, which is why six field components can be evolved by it at all. The factor $\mu_0$ on the right is a units choice; with $\mu_0\varepsilon_0 c^2=1$ the time slot will land Gauss's law with $\varepsilon_0$ exactly where it belongs. The second line involves no sources and, once potentials exist, no new physics — the homogeneous equation:

\partial_\mu G^{\mu\nu}=0, \qquad (2)

where: $G^{\mu\nu}=\tfrac12\epsilon^{\mu\nu\alpha\beta}F_{\alpha\beta}$ is the dual tensor, $F$ with the substitution $\mathbf{E}/c\to\mathbf{B}$ , $\mathbf{B}\to-\mathbf{E}/c$ . The Dual Field Tensor owns this object and proved the equivalence chain that matters here: $\partial_\mu G^{\mu\nu}=0$ $\iff$ the cyclic Bianchi identity $\partial_\lambda F_{\mu\nu}+\partial_\mu F_{\nu\lambda}+\partial_\nu F_{\lambda\mu}=0$ $\iff$ automatic whenever $F_{\mu\nu}=\partial_\mu A_\nu-\partial_\nu A_\mu$ for some four-potential. Its time slot is the absence of magnetic charge, its space slots are Faraday's law — that unpacking is that article's theorem, recalled here, not repeated. Which classical law lives where? The filing system deserves its own table — this is the address book for the whole subject:

slot	equation (1), $\partial_\mu F^{\mu\nu}=\mu_0 J^\nu$	equation (2), $\partial_\mu G^{\mu\nu}=0$
$\nu=0$	Gauss: $\nabla\cdot\mathbf{E}=\rho/\varepsilon_0$	no monopoles: $\nabla\cdot\mathbf{B}=0$
$\nu=1$	Ampère–Maxwell, $x$ : $(\nabla\times\mathbf{B})_x-\tfrac{1}{c^2}\partial_t E_x=\mu_0 J_x$	Faraday, $x$ : $(\nabla\times\mathbf{E})_x+\partial_t B_x=0$
$\nu=2$	Ampère–Maxwell, $y$	Faraday, $y$
$\nu=3$	Ampère–Maxwell, $z$	Faraday, $z$

where: each divergence law sits in a time slot and each curl law fills the three space slots — four vector equations were really $1+3+1+3$ scalar equations all along, and the two envelopes hold exactly eight. The pairing across the columns is the duality map: Gauss is to no-monopoles as Ampère–Maxwell is to Faraday, with $\mathbf{E}/c\leftrightarrow\pm\mathbf{B}$ and the source deleted.

Figure

The filing system — eight scalar slots inside two tensor envelopes, each holding one classical law

Derivation AThe audit — four familiar laws extracted from the inhomogeneous line

We open envelope (1) and read out all four scalar equations, confirming that one tensor divergence reproduces Gauss's law and the full Ampère–Maxwell law with every sign and every factor of $c$ in place. The strategy is pure bookkeeping: fix $\nu$ , split the sum over $\mu$ into its time and space parts, and translate tensor entries through the component dictionary quoted above. The two things to watch are index placement — the dictionary is for upper indices, first row $-\mathbf{E}/c$ — and the SI identity $\mu_0\varepsilon_0 c^2=1$ that converts between $\mu_0$ and $\varepsilon_0$ forms. The homogeneous envelope is not re-opened here: its unpacking ( $\nabla\cdot\mathbf{B}=0$ and Faraday) is owned by The Dual Field Tensor, and we import it as a finished theorem at the end.

Step 1. The time slot, $\nu=0$ . In $\partial_\mu F^{\mu 0}$ the $\mu=0$ term dies because $F^{00}=0$ (antisymmetry), leaving the spatial sum over the first column of the dictionary, $F^{i0}=+E^i/c$ :

\partial_\mu F^{\mu 0}=\partial_i F^{i0}=\frac{1}{c}\,\nabla\cdot\mathbf{E}.

The right side of (1) is $\mu_0 J^0=\mu_0 c\rho$ . Equate and multiply by $c$ :

\nabla\cdot\mathbf{E}=\mu_0 c^2\rho=\frac{\rho}{\varepsilon_0}.

Why the plus sign matters: this is the audit that catches convention errors. The first column of $F^{\mu\nu}$ carries $+\mathbf{E}/c$ ; had we misread the first row ( $-\mathbf{E}/c$ ) into this slot, Gauss's law would emerge with negative charge density sourcing outward field — instantly, visibly wrong.

Step 2. A space slot, $\nu=1$ , done in full. Now both the time term and the spatial sum contribute:

\partial_\mu F^{\mu 1}=\partial_0 F^{01}+\partial_2 F^{21}+\partial_3 F^{31} =\frac{1}{c}\,\partial_t\!\left(-\frac{E_x}{c}\right)+\partial_y B_z-\partial_z B_y,

reading $F^{01}=-E_x/c$ , $F^{21}=+B_z$ , $F^{31}=-B_y$ straight from the dictionary ( $F^{11}=0$ by antisymmetry). The last two terms are exactly $(\nabla\times\mathbf{B})_x$ . Setting the slot equal to $\mu_0 J^1=\mu_0 J_x$ :

(\nabla\times\mathbf{B})_x-\frac{1}{c^2}\frac{\partial E_x}{\partial t}=\mu_0 J_x.

Why the displacement current is automatic here: in the 3+1 history Maxwell had to inject the $\partial_t\mathbf{E}$ term by hand to rescue consistency. In the covariant packaging it was never optional — the $\mu=0$ term of the divergence exists because spacetime has a time direction. Relativity retroactively explains Maxwell's repair.

Step 3. The remaining slots, $\nu=2,3$ , run identically with cyclic relabeling ( $x\to y\to z\to x$ ), delivering the $y$ and $z$ components of the same law. Assembling all four slots and using $\mu_0\varepsilon_0c^2=1$ :

\partial_\mu F^{\mu\nu}=\mu_0 J^\nu \;\;\Longleftrightarrow\;\; \nabla\cdot\mathbf{E}=\frac{\rho}{\varepsilon_0} \;\;\text{and}\;\; \nabla\times\mathbf{B}-\mu_0\varepsilon_0\frac{\partial\mathbf{E}}{\partial t}=\mu_0\mathbf{J}.

Both directions of the equivalence hold: we derived the right side from the left, and since the eight scalar slots exhaust both formulations, the classical pair implies the tensor line just as well.

Step 4. The homogeneous half enters as a finished theorem. The Dual Field Tensor proved $\partial_\mu G^{\mu\nu}=0\iff\{\nabla\cdot\mathbf{B}=0,\ \nabla\times\mathbf{E}+\partial_t\mathbf{B}=0\}$ , and that both hold identically once $F_{\mu\nu}=\partial_\mu A_\nu-\partial_\nu A_\mu$ . The system is therefore complete: equations (1) and (2) together are exactly Maxwell's four laws, no more, no less.

Go deeper · Derivation B · 2 minManifest covariance — and how the four laws mix into each other

Derivation BManifest covariance — and how the four laws mix into each other

We prove that the system (1)–(2) is Lorentz covariant in three lines, then extract the sharper statement buried in the proof: the individual 3+1 laws are not frame-proof — a boosted observer's "Gauss's law" is a mixture of the lab's Gauss and Ampère laws. The strategy is to package each equation as the vanishing of a four-vector, then let the transformation law of four-vectors do all the work. Watch how much labor the word "tensor" is doing: every certificate we invoke was earned by a previous article, and none needs to be re-proved here.

Step 1. Define the residual four-vectors — the amounts by which a candidate field configuration fails each equation:

W^\nu\equiv\partial_\mu F^{\mu\nu}-\mu_0 J^\nu, \qquad V^\nu\equiv\partial_\mu G^{\mu\nu}.

$W^\nu$ is a four-vector by construction: $\partial_\mu$ is a covariant four-gradient (Four-Vectors and the Metric Tensor), $F^{\mu\nu}$ a certified tensor (The Electromagnetic Field Tensor), and $J^\nu$ a certified four-vector (The Four-Current Density); a contraction of the first two is a four-vector, and the difference of two four-vectors is one. The same argument certifies $V^\nu$ through the dual's tensor credentials.

Step 2. Maxwell's equations in frame $S$ are the statement $W^\nu=0$ and $V^\nu=0$ at every event. Boost to $\bar S$ : four-vectors transform linearly,

\bar W^\nu=\Lambda^\nu{}_\alpha W^\alpha,

and a linear map sends zero to zero. So $W^\alpha=0$ in one frame forces $\bar W^\nu=0$ in every frame. That is the entire covariance proof — the "gruesome afternoon" of transforming fields, coordinates, and sources component by component is prepaid in the tensor certificates. Why "manifest" is the right word: covariance is a property; manifest covariance is a property you can verify by inspection. Equation (1) equates two objects with the same transformation law, so no computation is even possible in which the two sides transform differently. The 3+1 equations are equally covariant — but invisibly so, which is why nobody noticed for forty years.

Step 3. Now the corollary. Write out the boosted time slot for a frame $\bar S$ moving at $-\beta c$ along $x$ , for which $\Lambda^0{}_1=+\gamma\beta$ and the mixing term carries a plus sign:

\bar W^0=\gamma\left(W^0+\beta W^1\right).

Translate through the audit of Derivation A: $cW^0$ is the lab's Gauss residual $\nabla\cdot\mathbf{E}-\rho/\varepsilon_0$ (up to the overall $1/c$ ), and $W^1$ is the lab's Ampère–Maxwell $x$ -residual. The equation says: the moving observer's Gauss law is a $\gamma$ -weighted blend of the lab's Gauss law and the lab's Ampère law. Suppose a demon violated Gauss's law alone in the lab — $W^0\neq0$ , $W^i=0$ . At $\beta=0.8$ ( $\gamma=5/3$ ) the boosted observer finds $\bar W^0=\tfrac{5}{3}W^0$ and $\bar W^1=\tfrac{5}{3}(0.8)W^0=\tfrac{4}{3}W^0$ : for her, both Gauss and Ampère fail. No single 3+1 law is a frame-independent statement.

Step 4. The same mixing runs inside the homogeneous envelope: $\bar V^0=\gamma(V^0+\beta V^1)$ blends no-monopoles with Faraday. This is the field-law echo of a lesson the chapter has taught twice — Transformation of Electric and Magnetic Fields showed the fields mix; Magnetism as a Relativistic Effect showed the forces mix; now the laws themselves mix. Only the two-line package, never its slots, is physics that all observers share.

Figure

Boosting the system — the package is frame-proof while the individual laws blend

The potential form — one equation, pre-gauge

Substituting the potential representation into the inhomogeneous line produces the master equation of the subject in its most compact dynamical form. Since The Electromagnetic Field Tensor built $F^{\mu\nu}=\partial^\mu A^\nu-\partial^\nu A^\mu$ , equation (1) becomes

\partial_\mu\!\left(\partial^\mu A^\nu-\partial^\nu A^\mu\right) =\Box A^\nu-\partial^\nu(\partial_\mu A^\mu)=\mu_0 J^\nu, \qquad (3)

where: $\Box=\partial_\mu\partial^\mu$ is the d'Alembertian and the derivatives commute, so the second term is the four-gradient of the single scalar $\partial\cdot A$ . This form is gauge-blind — true for every potential representative, before any gauge is chosen — and it carries the whole theory: the homogeneous equation is already automatic in this language, so (3) alone is electromagnetism. Choose the Lorenz condition $\partial_\mu A^\mu=0$ and the second term dies, collapsing (3) to the four uncoupled wave equations $\Box A^\mu=\mu_0 J^\mu$ that The Four-Potential assembled from the E9 machinery; what gauge choices are available, which are covariant, and what freedom survives is the next article's entire subject (Gauge Transformations and Covariant Gauge Conditions). One more piece of structure falls out of (3) by counting. Apply $\partial_\nu$ : the left side gives $\Box(\partial\cdot A)-\Box(\partial\cdot A)=0$ identically, so the right side must obey $\partial_\nu J^\nu=0$ — the system cannot be solved at all unless charge is conserved. This is the potential-language echo of the antisymmetry identity that The Electromagnetic Field Tensor proved directly on (1): $\partial_\nu\partial_\mu F^{\mu\nu}\equiv0$ because a symmetric derivative pair meets an antisymmetric tensor. Either way, the moral is the same and it is one of the deepest in physics: charge conservation is not an extra law of electromagnetism; it is a solvability condition built into the structure of the field equations. Feed the system a non-conserved source and it returns no solution, in any frame. The bookkeeping of equations versus unknowns is worth one honest paragraph, because "two equations" undersells and "eight equations for six unknowns" overfrightens. The eight scalar slots split by role: the two time slots contain no time derivatives of the fields — check Derivation A, Step 1: Gauss's law is a statement about $\mathbf{E}$ now — so they are constraints on initial data, not evolution equations. The six space slots evolve the six field components. And the two identities (antisymmetry on the inhomogeneous side, Bianchi on the homogeneous side) guarantee that the constraints, once satisfied at $t=0$ , are preserved by the evolution: differentiate Gauss's residual in time, use Ampère–Maxwell and charge conservation, and it stays zero. Six evolution equations for six unknowns, two self-maintaining constraints, zero overdetermination — the system is exactly as big as its physics.

question	reach for	why
fields from given sources, any frame	$\partial_\mu F^{\mu\nu}=\mu_0 J^\nu$ + identity (2)	the complete dynamics, manifestly covariant
potentials from sources, gauge not yet chosen	$\Box A^\nu-\partial^\nu(\partial\cdot A)=\mu_0 J^\nu$	gauge-blind master form (3)
potentials from sources, Lorenz gauge	$\Box A^\mu=\mu_0 J^\mu$	The Four-Potential; solved by Covariant Wave Equation and Retarded Solutions
is this source even admissible	$\partial_\mu J^\mu=0$	solvability condition — built into (1) and (3)
how fields respond to the equations' symmetry	$\bar F=\Lambda F\Lambda^{\mathsf T}$	Transformation of Electric and Magnetic Fields

Limiting cases

case	the two lines become	meaning
statics ( $\partial_t=0$ )	$\nabla\cdot\mathbf{E}=\rho/\varepsilon_0$ , $\nabla\times\mathbf{B}=\mu_0\mathbf{J}$ decouple from $\nabla\cdot\mathbf{B}=0$ , $\nabla\times\mathbf{E}=0$	electro- and magnetostatics of E1/E4 — the time slots and space slots stop talking
vacuum ( $J^\nu=0$ )	$\Box A^\nu=0$ in Lorenz gauge	free radiation at $c$ ; the E8 chapter's wave equation, now one covariant line
slow sources ( $v\ll c$ )	$J^0$ dominates $J^i$ by $c/v$	Coulomb physics leads, magnetism enters at order $v/c$ — the hierarchy Magnetism as a Relativistic Effect quantified
magnetic monopoles admitted	$\partial_\mu G^{\mu\nu}=\mu_0 K^\nu$	the symmetrized system catalogued by The Dual Field Tensor; the homogeneous line stops being an identity
one spatial dimension suppressed	slots collapse to $2+1$ electrodynamics	$B$ becomes a scalar; the filing table shrinks with spacetime itself

Common trapHalf of Maxwell is an identity — do not treat the homogeneous pair as independent dynamics

The two lines have profoundly different standing, and conflating them is the classic covariant-electrodynamics error. The inhomogeneous line is dynamics: it can be false for a candidate field, and satisfying it is a genuine constraint tying fields to sources. The homogeneous line, once you write $F_{\mu\nu}=\partial_\mu A_\nu-\partial_\nu A_\mu$ , cannot be false — it is an algebraic identity of second derivatives, as automatic as $\nabla\cdot(\nabla\times\mathbf{A})=0$ . Practical consequences: when counting equations, the homogeneous slots constrain which six-component objects qualify as electromagnetic fields, not how fields respond to matter; when building the theory from an action (next articles), only the inhomogeneous line appears as an equation of motion, because the identity was baked in the moment $A^\mu$ became the fundamental variable; and when someone proposes magnetic monopoles, they are proposing that the identity fails somewhere — which is why monopoles are not a small edit but a change in what the potential is. If your "derivation" of Faraday's law from the covariant formalism used only $F=\partial A$ and no physics, you have not derived a law of nature; you have confirmed your bookkeeping.

Worked examples

Example 1 — auditing the $z$ -slot inside a charging capacitor. Between the plates of a parallel-plate capacitor driven by a fast ramp, the field is uniform in space and rising linearly in time: $\mathbf{E}=k t\,\hat{\mathbf{z}}$ with $k=3.0\times10^{10}$ V m $^{-1}$ s $^{-1}$ , $\rho=0$ , $\mathbf{J}=0$ between the plates. Read the $\nu=3$ slot of equation (1) at radius $r=2.0$ cm from the axis.

Step 1. The slot demands $\partial_0F^{03}+\partial_iF^{i3}=\mu_0J^3=0$ . The time term is $\tfrac{1}{c}\partial_t(-E_z/c)=-k/c^2$ ; it is spatially uniform and nonzero.

Step 2. The spatial terms must therefore supply $+k/c^2$ : from Derivation A these terms are exactly $(\nabla\times\mathbf{B})_z$ , so the slot forces a magnetic field to circulate around the axis even though no charge moves anywhere. This is the displacement-current effect, delivered by index bookkeeping.

Step 3. By symmetry $\mathbf{B}=B_\phi(r)\hat{\boldsymbol{\phi}}$ ; the circulation theorem applied to a circle of radius $r$ gives $2\pi r B_\phi=(k/c^2)\pi r^2$ , so

B_\phi=\frac{k r}{2c^2}=\frac{(3.0\times10^{10})(0.020)}{2(2.998\times10^8)^2} =\boxed{3.3\times10^{-9}\ \text{T}},

about a thousandth of Earth's field — measurable with a fluxgate magnetometer. The equivalent displacement-current density is $\varepsilon_0 k=0.27$ A/m², a genuinely respectable current density flowing through empty space in the ledger of equation (1).

Example 2 — one system, two frames: the boosted line charge. A straight line charge $\lambda=30$ nC/m lies at rest along the $x$ -axis of the lab. In the lab, only the $\nu=0$ slot is active: pure electrostatics, $E_r=\lambda/2\pi\varepsilon_0 r=1.08\times10^4$ V/m at $r=5.0$ cm. Now describe the same physics from a frame $\bar S$ moving at $\beta=-0.8$ along $x$ (so the charges move at $+0.8c$ in $\bar S$ ; $\gamma=5/3$ ).

Step 1. Transform the source with The Four-Current Density's line-charge bookkeeping: $\bar\lambda=\gamma\lambda=50$ nC/m (length contraction packs the charge), and a current appears, $\bar I=\gamma\lambda\beta c=(5/3)(30\times10^{-9})(0.8)(2.998\times10^8)=12.0$ A.

Step 2. In $\bar S$ two slots of equation (1) are active. The time slot gives the new Gauss field $\bar E_r=\bar\lambda/2\pi\varepsilon_0 r=1.80\times10^4$ V/m $=\gamma E_r$ — matching the perpendicular-field rule of Transformation of Electric and Magnetic Fields. The $x$ -slot gives the Ampère field of the current, $\bar B_\phi=\mu_0\bar I/2\pi r=4.8\times10^{-5}$ T.

Step 3. Cross-check the package covariantly. The source invariant, per unit length: in $\bar S$ , $(c\bar\lambda)^2-\bar I^2=(15.0\ \text{A})^2-(12.0\ \text{A})^2=80.9\ \text{A}^2$ ; in the lab, $(c\lambda)^2-0=(8.99\ \text{A})^2=80.9\ \text{A}^2$ — identical, as the four-current's norm must be. And the field ratio: $c\bar B_\phi/\bar E_r=(2.998\times10^8)(4.80\times10^{-5})/(1.80\times10^4)=0.80=\beta$ , precisely the mixing signature the field-transformation rules demand. One frame files this physics under Gauss alone; the other needs Gauss and Ampère together; the two-line system is the same in both.

Going further

Reading. Griffiths, Introduction to Electrodynamics, 4th ed., §12.3 — the 3+1-to-covariant dictionary at this article's register. Jackson, Classical Electrodynamics, 3rd ed., §11.9 — the tensor formulation with the Gaussian-units matrices. Zangwill, Modern Electrodynamics, §22.4 — the covariance argument done as done here. Purcell & Morin, Electricity and Magnetism, 3rd ed., Ch. 5–6 — the physical soul of the enterprise. Component conventions follow the track's $(+,-,-,-)$ dictionary: the corrected set is $F^{0i}=-E^i/c$ , $F_{0i}=+E^i/c$ , $F^{ij}=-\epsilon^{ijk}B_k$ . Key numerical anchors: the covariance mixing coefficients at $\beta=0.8$ are $\gamma=5/3$ and $\gamma\beta=4/3$ ; the worked examples yield $B_\phi=3.3\times10^{-9}$ T at $r=2$ cm for $k=3.0\times10^{10}$ V m $^{-1}$ s $^{-1}$ , and for the boosted line charge: $\bar\lambda=50$ nC/m, $\bar I=12.0$ A, $\bar E_r=1.80\times10^4$ V/m, $\bar B_\phi=4.8\times10^{-5}$ T, $c\bar B/\bar E=0.80$ , with the four-current invariant $80.9$ A² in both frames. The field-tensor construction, its transformation law, and the charge-conservation identity on (1) are treated in The Electromagnetic Field Tensor; the dual object and the homogeneous unpacking in The Dual Field Tensor; $\Box A^\mu=\mu_0J^\mu$ assembly in The Four-Potential; gauge theory in Gauge Transformations and Covariant Gauge Conditions; executed particle dynamics in Relativistic Dynamics of a Charged Particle; the action route to equation (1) in Relativistic Lagrangian and Action for Charged Particles; $T^{\mu\nu}$ and the conservation ledger in The Electromagnetic Stress-Energy Tensor and Conservation Laws and Energy-Momentum Flow; covariant retarded solutions in Covariant Wave Equation and Retarded Solutions. This article owns the system: the slot audit, the counting, and the covariance theorem with its mixing corollary.

Builds on: The Electromagnetic Field Tensor · The Dual Field Tensor · The Four-Current Density · The Four-Potential · Four-Vectors and the Metric Tensor · Lorentz Transformations

Leads to: Gauge Transformations and Covariant Gauge Conditions · Relativistic Dynamics of a Charged Particle · Relativistic Lagrangian and Action for Charged Particles · The Electromagnetic Stress-Energy Tensor

If two equations really contain all four of Maxwell's laws, where did the information go?

Nowhere — it moved into the index. Each covariant equation has a free index $\nu$ running over four values, so the pair holds $4+4=8$ scalar slots, and the audit in Derivation A shows those slots are exactly the $1+3$ scalar equations of the Gauss/Ampère pair plus the $1+3$ of the no-monopole/Faraday pair. The compression is real but honest: it packages, it does not discard. What is genuinely new relative to the 3+1 form is the packaging's covariance guarantee — the eight slots are organized into two four-vector statements, which is precisely the grouping that transforms cleanly under boosts. The 3+1 grouping (four "laws") is the rotationally clean packaging; the covariant grouping is the relativistically clean one.

Why does the inhomogeneous equation carry the source while the homogeneous one carries nothing?

Because nature, as far as every experiment can tell, supplies only electric charge. The two lines are duality partners — swap $\mathbf{E}/c\to\mathbf{B}$ , $\mathbf{B}\to-\mathbf{E}/c$ and they trade places — so a perfectly symmetric world would have $\partial_\mu G^{\mu\nu}=\mu_0 K^\nu$ with a magnetic four-current $K^\nu$ . The Dual Field Tensor catalogues that symmetrized system and what magnetic monopoles would do to it: the key casualty would be the potential representation, since $F=\partial A$ forces the homogeneous line to vanish. So the emptiness of the second line is equivalent to the statement that a global four-potential exists. Finding one monopole would demote the identity to dynamics.

Is charge conservation an input to Maxwell's theory or an output of it?

Structurally, it is a solvability condition — stronger than either word. Take $\partial_\nu$ of equation (1): the left side vanishes identically (antisymmetry against symmetric derivatives), so $\partial_\nu J^\nu=0$ is forced on any source the equations can accommodate at all. You cannot even pose the Maxwell problem for a non-conserved current; the system returns no solution rather than a wrong one. Historically the logic ran forward — Maxwell added the displacement current partly to make the equations consistent with conservation — and the covariant form shows why the repair worked: consistency and conservation are the same algebraic fact. The action-principle articles ahead will deepen this once more: there, conservation follows from the gauge invariance of the coupling, tying the identity to a symmetry.

What exactly does "manifestly" add to "covariant"? The 3+1 equations are covariant too.

Covariance is a fact about equations; manifest covariance is a fact about notation. The 3+1 Maxwell equations are fully Lorentz-covariant — they had better be, since they are the same physics — but verifying it requires the whole apparatus: transform fields by the six mixing rules, sources by the four-current law, derivatives by the chain rule, and confirm the scrambled equations reassemble. Nothing about $\nabla\times\mathbf{E}+\partial_t\mathbf{B}=0$ displays its own covariance. In the tensor form, both sides of each equation carry the same transformation law by construction, so the verification is the one-liner of Derivation B. The practical payoff is error-proofing: in manifestly covariant notation it is impossible to write down a Lorentz-violating equation without visibly mismatching indices — the notation polices the symmetry for you.

If the homogeneous equation is just an identity, why do we call Faraday's law physics?

Because the identity's premise — fields come from a four-potential — is itself physical, and Faraday's law is its experimental fingerprint. The honest chain is: experiment shows induction; induction plus no-monopoles is equivalent to the statement that $F_{\mu\nu}$ can be written as $\partial_\mu A_\nu-\partial_\nu A_\mu$ ; given that representation, the homogeneous line becomes automatic bookkeeping. So the physics lives in the existence of the potential, and "Faraday's law is an identity" really means "we have chosen variables in which Faraday's law cannot be violated." That choice has teeth elsewhere: in quantum mechanics the potential couples directly to phases (the minimal-coupling story of Relativistic Lagrangian and Action for Charged Particles), so the variable choice forced by the homogeneous pair turns out to be the one nature actually uses.

Do the two tensor equations overdetermine the fields near t equals zero?

No, and the resolution is the constraint/evolution split. Slots $\nu=1,2,3$ of each equation contain $\partial_t\mathbf{E}$ or $\partial_t\mathbf{B}$ : six equations that march the six field components forward in time. The two $\nu=0$ slots contain no time derivatives of fields — they are conditions on the initial snapshot ( $\nabla\cdot\mathbf{E}=\rho/\varepsilon_0$ , $\nabla\cdot\mathbf{B}=0$ ). The system stays consistent because the identities guarantee the constraints are transported: if Gauss holds at $t=0$ and you evolve with Ampère–Maxwell while conserving charge, Gauss holds forever. Numerical electrodynamics lives and dies by this structure — solvers monitor "constraint violation" precisely because discretization breaks the exact identities that the continuum system enjoys.

Which of the two lines is responsible for electromagnetic waves?

Both, working together — and the potential form shows the division of labor. Radiation needs Faraday (changing $\mathbf{B}$ makes $\mathbf{E}$ ) and the displacement current (changing $\mathbf{E}$ makes $\mathbf{B}$ ): one lives in the homogeneous line, one in the inhomogeneous. In the potential language the cooperation is already consummated: the homogeneous line is spent building the representation $F=\partial A$ , and the single gauge-blind equation (3) then carries wave propagation entirely — set $J^\nu=0$ , choose Lorenz gauge, and $\Box A^\mu=0$ delivers light at $c=1/\sqrt{\mu_0\varepsilon_0}$ . That constant emerging from a structural equation, with no medium in sight, is the second postulate's engine — the story Postulates of Special Relativity opened this chapter's relativity arc with.

Why is the right-hand side written with mu-zero rather than epsilon-zero?

Pure SI bookkeeping, fixed by what $J^\nu$ carries. With $J^\mu=(c\rho,\mathbf{J})$ , the time slot needs $\nabla\cdot\mathbf{E}$ on the left to meet $\rho$ on the right; the factor connecting them can be written $1/\varepsilon_0$ or $\mu_0c^2$ — identical by $\mu_0\varepsilon_0c^2=1$ — and the space slots need plain $\mu_0$ for Ampère. Writing $\mu_0J^\nu$ makes all four slots uniform with one constant. In Gaussian units the same equation reads $\partial_\mu F^{\mu\nu}=(4\pi/c)J^\nu$ ; in Heaviside–Lorentz natural units, $\partial_\mu F^{\mu\nu}=J^\nu$ . The physics is the slot structure, not the constant; when comparing with Jackson (Gaussian) keep the dictionary conversion at hand.

Covariant Form of Maxwell's Equations

Where this lives today

The physical picture

The two lines, and what sits in their slots

The potential form — one equation, pre-gauge

Limiting cases

Worked examples

Going further

Before this

Where this leads