Relativistic Electrodynamics and the Field Tensor

Maxwell wrote the laws of electromagnetism in the 1860s, and for four decades nobody noticed the strangest thing about them: they already obeyed relativity. Cast in three-vector language, an electric field $\mathbf{E}$ and a magnetic field $\mathbf{B}$ look like two independent physical entities — one pushes charges, the other deflects them — yet a mere change of inertial frame scrambles the two into each other by rules that, before 1905, seemed like coincidences. A charge sitting still makes a pure electric field; walk past it and you measure a magnetic field too. Which is "the real" field? Special relativity's answer, delivered by Hermann Minkowski in 1908, is that the question is malformed. $\mathbf{E}$ and $\mathbf{B}$ are not two fields but six components of one geometric object living on spacetime: the antisymmetric Faraday field-strength tensor $F_{\mu\nu}$ . What an observer calls electric versus magnetic depends on their worldline, exactly as what they call space versus time does. This synthesis borrows two results from its prerequisites — the tensor transformation law $F'_{\mu\nu}=\Lambda^\alpha{}_\mu\Lambda^\beta{}_\nu F_{\alpha\beta}$ from aq-gr-lorentz-transformations, and the packaging of the scalar and vector potentials $(\phi,\mathbf{A})$ into a four-vector $A^\mu$ from aq-gr-four-vectors — and asks what falls out. What falls out is everything. All four of Maxwell's equations collapse into two covariant tensor statements; the Lorentz force becomes a single four-vector law; two frame-independent scalars classify every possible field; and the whole structure carries over to curved spacetime by a one-line substitution, making $F_{\mu\nu}$ the natural bridge from flat-space electromagnetism to Einstein's gravity.

Key takeaways

Electric and magnetic fields are not separate: they are the six independent components of one antisymmetric rank-2 tensor, the Faraday tensor $F_{\mu\nu}$ , which mixes them under Lorentz boosts.
The tensor is the four-dimensional curl of a four-potential, $F_{\mu\nu}=\partial_\mu A_\nu-\partial_\nu A_\mu$ with $A^\mu=(\phi/c,\mathbf{A})$ , which makes the homogeneous Maxwell equations automatic and gauge invariance transparent.
Maxwell's four equations become two: $\partial_\mu F^{\mu\nu}=\mu_0 J^\nu$ (sources) and $\partial_{[\lambda}F_{\mu\nu]}=0$ (the Bianchi identity), and the dynamics of a charge is $\mathrm{d}p^\mu/\mathrm{d}\tau=qF^{\mu}{}_{\nu}u^\nu$ .
Two Lorentz scalars, $F_{\mu\nu}F^{\mu\nu}=2(B^2-E^2/c^2)$ and $F_{\mu\nu}\tilde F^{\mu\nu}\propto\mathbf{E}\cdot\mathbf{B}$ , are the same in every frame, so "purely electric" is observer-dependent but the invariants are not.
Magnetism is the relativistic shadow of electrostatics: a charge's Coulomb field, boosted, is exactly the magnetic field of a current.
Key idea: define $F_{\mu\nu}=\partial_\mu A_\nu-\partial_\nu A_\mu$ ; then all of electromagnetism is $\partial_\mu F^{\mu\nu}=\mu_0 J^\nu$ with $\partial_{[\lambda}F_{\mu\nu]}=0$ , transforming covariantly so that $\mathbf{E}$ and $\mathbf{B}$ are one object seen from different frames.

The core insight

Define the antisymmetric tensor $F_{\mu\nu}=\partial_\mu A_\nu-\partial_\nu A_\mu$ with $A^\mu=(\phi/c,\mathbf{A})$ ; its time-space components are the electric field and its space-space components the magnetic field. All of Maxwell's equations become $\partial_\mu F^{\mu\nu}=\mu_0 J^\nu$ together with the identity $\partial_{[\lambda}F_{\mu\nu]}=0$ , and a charge obeys $m\,\mathrm{d}u^\mu/\mathrm{d}\tau=qF^\mu{}_\nu u^\nu$ — all Lorentz covariant.

Where this lives today

The tensor formulation is not a historical tidying-up; it is the working language wherever fields, motion, and gravity meet. In particle accelerators the covariant force law $\mathrm{d}p^\mu/\mathrm{d}\tau=qF^\mu{}_\nu u^\nu$ is the design equation for every storage ring: bending magnets steer beams without adding energy (magnetic forces do no work), while radio-frequency cavities supply the electric field that accelerates, and the sharply forward-beamed synchrotron light follows from boosting a rest-frame dipole pattern. At the extreme end, magnetars — neutron stars with surface fields near $10^{11}\ \mathrm{T}$ — push the invariants of $F_{\mu\nu}$ into the regime where the quantum vacuum itself becomes birefringent; X-ray polarimetry of such stars (for example the IXPE mission launched in 2021) is now testing this prediction of strong-field QED, which is written entirely in terms of the two scalars $F_{\mu\nu}F^{\mu\nu}$ and $F_{\mu\nu}\tilde F^{\mu\nu}$ . And the tensor is the doorway to gravity: its stress-energy $T^{\mu\nu}=\frac{1}{\mu_0}(F^{\mu\alpha}F^\nu{}_\alpha-\tfrac14\eta^{\mu\nu}F_{\alpha\beta}F^{\alpha\beta})$ sources Einstein's equations for charged (Reissner–Nordström) black holes and for the radiation-dominated early universe, and the minimal-coupling rule $\partial_\mu\to\nabla_\mu$ , $\eta\to g$ promotes flat-space electromagnetism to curved spacetime essentially unchanged — the subject of the article that follows this one.

The physical picture

Start with the counting, because it is the whole secret. A rank-2 antisymmetric tensor in four spacetime dimensions has exactly $\binom{4}{2}=6$ independent components — no more, no fewer. And there are exactly six numbers in $(\mathbf{E},\mathbf{B})$ : three each. That match is not luck. Once you accept that electromagnetism must be built from a potential four-vector $A^\mu$ and must transform properly under Lorentz transformations, the antisymmetric derivative $\partial_\mu A_\nu-\partial_\nu A_\mu$ is forced, and it has to have six slots. Electromagnetism came out this way because Minkowski spacetime left it no choice. Picture the tensor as a single $4\times4$ ledger. The entries that couple a time direction to a space direction, $F_{0i}$ , hold the electric field; the entries that couple two space directions, $F_{ij}$ , hold the magnetic field. A Lorentz boost rotates a time axis into a space axis, so it inevitably shovels entries out of the time-space block and into the space-space block and back — that is the precise sense in which a boost turns $\mathbf{E}$ into $\mathbf{B}$ . A static charge carries only time-space entries (pure $\mathbf{E}$ ); set it moving and the boost paints in space-space entries too (a $\mathbf{B}$ appears). Magnetism is the geometry of moving electricity.

Figure

The Faraday tensor F_μν gathers all six field components into one antisymmetric 4×4 array — three electric entries F_0i = E_i/c along the time row and column, three magnetic entries F_ij = −ε_ijk B_k in the spatial block — so what one observer calls electric or magnetic is just which slice of one object they read

Two combinations of these six numbers survive every boost untouched. One is like a squared length, $B^2-E^2/c^2$ ; its sign says whether the field is "magnetic-dominated" (there is a frame where $\mathbf{E}$ vanishes) or "electric-dominated" (a frame where $\mathbf{B}$ vanishes). The other is like an orientation, $\mathbf{E}\cdot\mathbf{B}$ ; it says whether the two fields can ever be made perpendicular. A plane light wave has both invariants zero — a "null" field, the electromagnetic echo of a light cone, that stays a wave in every frame no matter how you run toward or away from it. The second figure shows the cleanest case of the mixing: a charge whose field is pure Coulomb at rest sprouts a circulating magnetic field the moment you view it in motion.

Figure

Magnetism as relativistic electricity: a charge with only a radial Coulomb field E′ in its rest frame at left acquires a circulating magnetic field in the lab at right once boosted to v = 0.6c with γ = 1.25, following B = v × E/c² — no new force law, just the same tensor F_μν read from a moving frame

The core equations

Everything begins with one definition: the field strength is the four-dimensional curl of the potential. Because the potential $A^\mu$ is a four-vector and the derivative $\partial_\mu$ transforms as a covector, this construction is manifestly a tensor.

F_{\mu\nu}=\partial_\mu A_\nu-\partial_\nu A_\mu \qquad (1)

where: $F_{\mu\nu}$ — Faraday (field-strength) tensor, antisymmetric by construction, $F_{\mu\nu}=-F_{\nu\mu}$ ; $A_\mu$ — electromagnetic four-potential, $A_\mu=(-\phi/c,\mathbf{A})$ in signature $(-,+,+,+)$ ; $\partial_\mu=(\tfrac1c\partial_t,\nabla)$ — the four-gradient. Physical reading: the antisymmetrized gradient of the potentials automatically produces six components, and they turn out to be precisely $(\mathbf{E},\mathbf{B})$ . Writing out those components in matrix form makes the electric and magnetic blocks visible. This is the same array drawn in the first figure.

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} \qquad (2)

where: $E_i/c$ — electric-field components scaled by $1/c$ so they share units with $B$ ; $B_i$ — magnetic-field components filling the spatial $3\times3$ block. Physical reading: the top row and left column carry $\mathbf{E}$ ; the inner block carries $\mathbf{B}$ ; a boost trades entries between the two. The two source-carrying Maxwell equations — Gauss's law and the Ampère–Maxwell law — are a single tensor equation once you contract $F$ with a derivative and set it equal to the four-current.

\partial_\mu F^{\mu\nu}=\mu_0 J^\nu \qquad (3)

where: $J^\nu=(\rho c,\mathbf{J})$ — four-current density; $\mu_0$ — vacuum permeability, with $\mu_0\epsilon_0=1/c^2$ . Physical reading: the $\nu=0$ component is $\nabla\cdot\mathbf{E}=\rho/\epsilon_0$ and the $\nu=i$ components are $\nabla\times\mathbf{B}-\tfrac{1}{c^2}\partial_t\mathbf{E}=\mu_0\mathbf{J}$ ; charge conservation $\partial_\nu J^\nu=0$ is automatic because $\partial_\nu\partial_\mu F^{\mu\nu}=0$ by antisymmetry. The two remaining Maxwell equations — Faraday's law and $\nabla\cdot\mathbf{B}=0$ — carry no sources at all. They are the Bianchi identity, a geometric consequence of $F$ being built from a potential.

\partial_{[\lambda}F_{\mu\nu]}=0 \quad\Longleftrightarrow\quad \partial_\lambda F_{\mu\nu}+\partial_\mu F_{\nu\lambda}+\partial_\nu F_{\lambda\mu}=0 \qquad (4)

where: $[\cdots]$ — total antisymmetrization over the enclosed indices. Physical reading: because $F=\mathrm{d}A$ , this is $\mathrm{d}\mathrm{d}A=0$ , an identity requiring neither sources nor even a metric — the homogeneous equations are "for free". A charge responds to the field through one four-vector law that unifies electric acceleration and magnetic deflection.

\frac{\mathrm{d}p^\mu}{\mathrm{d}\tau}=q\,F^{\mu}{}_{\nu}\,u^\nu \qquad (5)

where: $p^\mu=mu^\mu$ — four-momentum; $u^\nu=\gamma(c,\mathbf{v})$ — four-velocity, normalized $u^\mu u_\mu=-c^2$ ; $\tau$ — proper time; $q$ — charge. Physical reading: the spatial part is $\mathrm{d}\mathbf{p}/\mathrm{d}t=q(\mathbf{E}+\mathbf{v}\times\mathbf{B})$ and the time part is $\mathrm{d}E/\mathrm{d}t=q\,\mathbf{E}\cdot\mathbf{v}$ , so only $\mathbf{E}$ does work. Two scalars built from $F$ take the same value in every inertial frame. They are the physical content that survives the observer-dependent $\mathbf{E}/\mathbf{B}$ split.

\mathcal{I}_1=F_{\mu\nu}F^{\mu\nu}=2\!\left(B^2-\frac{E^2}{c^2}\right),\qquad \mathcal{I}_2=F_{\mu\nu}\tilde{F}^{\mu\nu}=-\frac{4}{c}\,\mathbf{E}\cdot\mathbf{B} \qquad (6)

where: $\mathcal{I}_1$ — scalar invariant whose sign sorts electric- from magnetic-dominated fields; $\mathcal{I}_2$ — pseudoscalar invariant that vanishes exactly when $\mathbf{E}\perp\mathbf{B}$ ; $\tilde{F}^{\mu\nu}=\tfrac12\epsilon^{\mu\nu\rho\sigma}F_{\rho\sigma}$ — the Hodge dual, which sends $\mathbf{E}\to c\mathbf{B}$ , $\mathbf{B}\to-\mathbf{E}/c$ . Physical reading: no boost can change these numbers, so a null field ( $\mathcal{I}_1=\mathcal{I}_2=0$ ) is a wave to everyone. Finally, the explicit boost rules show the mixing directly, and are the working tool for problems.

\mathbf{E}'_\perp=\gamma(\mathbf{E}+\mathbf{v}\times\mathbf{B})_\perp,\qquad \mathbf{B}'_\perp=\gamma\!\left(\mathbf{B}-\tfrac{1}{c^2}\mathbf{v}\times\mathbf{E}\right)_\perp,\qquad \mathbf{E}'_\parallel=\mathbf{E}_\parallel,\quad \mathbf{B}'_\parallel=\mathbf{B}_\parallel \qquad (7)

where: $\parallel,\perp$ — components along and across the boost velocity $\mathbf{v}$ ; $\gamma=1/\sqrt{1-v^2/c^2}$ . Physical reading: fields along the motion are untouched; transverse fields are amplified by $\gamma$ and swapped into one another, which is where the induced $\mathbf{v}\times\mathbf{B}$ and $\mathbf{v}\times\mathbf{E}/c^2$ terms come from.

Derivation AField mixing from the tensor transformation law

The goal is to prove that a boost really does turn a magnetic field into an electric one, not by asserting the rule (7) but by deriving it from first principles: apply the Lorentz transformation of a rank-2 tensor, $F'_{\mu\nu}=\Lambda^\alpha{}_\mu\Lambda^\beta{}_\nu F_{\alpha\beta}$ , to a boost along $x$ and read off how a transverse $E_y$ and $B_z$ get scrambled. Everything follows from the single boost matrix and the component dictionary of equation (2), so the "mysterious" mixing of $\mathbf{E}$ and $\mathbf{B}$ becomes pure bookkeeping in one tensor law.

Step 1. Write the boost matrix along $x$ with $\beta=v/c$ : its nonzero entries are $\Lambda^0{}_0=\Lambda^1{}_1=\gamma$ , $\Lambda^0{}_1=\Lambda^1{}_0=-\gamma\beta$ , and $\Lambda^2{}_2=\Lambda^3{}_3=1$ . Why: this is the same matrix that mixes $t$ and $x$ for coordinates; the field tensor must transform with two copies of it, one per index.

Step 2. Transform the $(0,2)$ entry. Since $\Lambda^\beta{}_2=\delta^\beta_2$ , only $\alpha=0,1$ survive: $F'_{02}=\Lambda^0{}_0 F_{02}+\Lambda^1{}_0 F_{12}=\gamma F_{02}-\gamma\beta F_{12}$ . Why: the $y$ -index is a spectator under an $x$ -boost, so only the time and $x$ slots of the first index contribute — the sum truncates to two terms.

Step 3. Insert the dictionary $F_{02}=-E_y/c$ and $F_{12}=-B_z$ from the matrix: $F'_{02}=\gamma(-E_y/c)-\gamma\beta(-B_z)=-\tfrac{\gamma}{c}E_y+\gamma\beta B_z$ . Reading the left side as $-E'_y/c$ gives $E'_y=\gamma(E_y-vB_z)$ . Why: this is the payoff line — a frame with only a magnetic $B_z$ measures an electric $E'_y=-\gamma vB_z$ ; the magnetic field has become partly electric.

Step 4. Repeat for the $(1,2)$ entry: $F'_{12}=\Lambda^0{}_1\Lambda^2{}_2 F_{02}+\Lambda^1{}_1\Lambda^2{}_2 F_{12}=-\gamma\beta(-E_y/c)+\gamma(-B_z)$ , so $-B'_z=-\gamma B_z+\gamma\beta E_y/c$ , i.e. $B'_z=\gamma(B_z-vE_y/c^2)$ . Why: the reciprocal mixing confirms that $\mathbf{E}$ and $\mathbf{B}$ feed into each other symmetrically, reproducing exactly the general rule (7) restricted to this pair. *Verified: matches Jackson (3rd ed.) Eq. 11.149 and Griffiths Eq. 12.108; the $v\ll c$ limit gives $\mathbf{E}'\approx\mathbf{E}+\mathbf{v}\times\mathbf{B}$ , the familiar motional field.*

Go deeper · Derivation B · 1 minMagnetism as relativistic electrostatics

Derivation BMagnetism as relativistic electrostatics

The goal is the single most persuasive fact in this whole subject: that magnetism is not an independent force but the relativistic consequence of electrostatics plus motion. We take a point charge that carries only a Coulomb field in its own rest frame, with zero magnetic field, boost to the lab where it moves, and show that the boost manufactures a magnetic field of exactly the Biot–Savart form. If this works, then a current — which is just charges in motion — must produce magnetism as a matter of logical necessity, with no separate magnetic law required.

Step 1. In the charge's rest frame $S'$ the field is pure electrostatics: $\mathbf{E}'=\frac{q}{4\pi\epsilon_0}\frac{\hat r'}{r'^2}$ and $\mathbf{B}'=0$ . Why: a charge at rest has no magnetic field by definition, so the second invariant $\mathcal{I}_2\propto\mathbf{E}'\cdot\mathbf{B}'=0$ and the field is manifestly electric-dominated — a clean starting point with nothing magnetic to hide.

Step 2. Boost to the lab frame $S$ where the charge moves at $\mathbf{v}$ . The inverse of rule (7) with $\mathbf{B}'=0$ gives $\mathbf{B}_\perp=\gamma\big(\mathbf{B}'+\tfrac{1}{c^2}\mathbf{v}\times\mathbf{E}'\big)_\perp=\gamma\,\tfrac{1}{c^2}(\mathbf{v}\times\mathbf{E}')_\perp$ . Why: with no magnetic field to start from, every bit of the lab's $\mathbf{B}$ is generated by the motion acting on the electric field — the magnetic field is literally made of boosted electricity.

Step 3. Re-express the boosted field in terms of the lab electric field $\mathbf{E}$ to get the compact exact relation $\mathbf{B}=\tfrac{1}{c^2}\,\mathbf{v}\times\mathbf{E}$ , valid everywhere around a uniformly moving charge. Why: this ties the induced magnetic field directly to the observed electric field and the velocity, independent of the details of the Lorentz-contracted field pattern.

Step 4. In the slow-motion limit, insert the Coulomb $\mathbf{E}$ and use $1/c^2=\mu_0\epsilon_0$ : $\mathbf{B}\approx\frac{\mu_0}{4\pi}\frac{q\,\mathbf{v}\times\hat r}{r^2}$ . Why: this is exactly the Biot–Savart field of a moving point charge — recovering the textbook magnetostatic result from relativity alone, which is the whole point. *Verified: the relation $\mathbf{B}=\mathbf{v}\times\mathbf{E}/c^2$ and its Biot–Savart limit appear in Griffiths Sec. 12.3.1 and Purcell, Electricity and Magnetism, Ch. 5 — the standard demonstration that magnetism is a relativistic effect.*

Go deeper · Derivation C · 1 minThe first Lorentz invariant

Derivation CThe first Lorentz invariant

The goal is to compute $F_{\mu\nu}F^{\mu\nu}$ explicitly and show it equals $2(B^2-E^2/c^2)$ , a genuine scalar identical in every frame. This is the calculation that underwrites the entire classification of fields into electric-dominated, magnetic-dominated, and null, and it is also the most error-prone in the subject because it demands raising indices on an antisymmetric tensor in a mixed signature. Doing it carefully, tracking every sign from $\eta^{00}=-1$ , both delivers the result and inoculates against the commonest mistake in relativistic electrodynamics.

Step 1. Split the double sum by index type: $F_{\mu\nu}F^{\mu\nu}=2F_{0i}F^{0i}+F_{ij}F^{ij}$ , the factor of two because time-space pairs occur as both $0i$ and $i0$ . Why: separating the electric ( $0i$ ) from the magnetic ( $ij$ ) contributions lets us handle their different index-raising signs one block at a time.

Step 2. Raise the electric part. With signature $(-,+,+,+)$ , $F^{0i}=\eta^{00}\eta^{ii}F_{0i}=-F_{0i}$ , so $2F_{0i}F^{0i}=-2F_{0i}F_{0i}=-2E^2/c^2$ , using $F_{0i}=-E_i/c$ . Why: the lone factor $\eta^{00}=-1$ from the time index flips the sign — this is the exact spot where careless raising corrupts the answer, the Lorentz force, and the source equation.

Step 3. Raise the magnetic part. Two spatial factors give no sign, $F^{ij}=F_{ij}$ , and with $F_{ij}=-\epsilon_{ijk}B_k$ the contraction is $F_{ij}F_{ij}=\epsilon_{ijk}\epsilon_{ijl}B_kB_l=2\delta_{kl}B_kB_l=2B^2$ . Why: spatial indices raise cleanly, and the identity $\epsilon_{ijk}\epsilon_{ijl}=2\delta_{kl}$ collapses the double sum into $2B^2$ with no sign subtlety.

Step 4. Combine: $F_{\mu\nu}F^{\mu\nu}=-2E^2/c^2+2B^2=2(B^2-E^2/c^2)$ . Why: the result is manifestly a scalar built with no free indices, so it is the same number in every inertial frame — the invariant that classifies the field and that we will confirm numerically in Worked example 1. *Verified: agrees with Jackson Eq. 11.151 and Landau–Lifshitz §25; the second invariant follows identically with the dual, giving $F_{\mu\nu}\tilde F^{\mu\nu}\propto\mathbf{E}\cdot\mathbf{B}$ .*

Limiting cases

The frame-dependence of $\mathbf{E}$ and $\mathbf{B}$ is governed entirely by the two invariants of equation (6). Their signs sort every possible field into physically distinct classes:

Case	Result	Meaning
Magnetic-dominated	$B^2>E^2/c^2,\ \ \mathbf{E}\cdot\mathbf{B}=0$	A frame exists with $\mathbf{E}'=0$ : the field is purely magnetic there. Example: the field of a steady current.
Electric-dominated	$E^2/c^2>B^2,\ \ \mathbf{E}\cdot\mathbf{B}=0$	A frame exists with $\mathbf{B}'=0$ : the field is purely electric there. Example: the Coulomb field of a charge.
Null (radiation) field	$E^2/c^2=B^2,\ \ \mathbf{E}\cdot\mathbf{B}=0$	Both invariants vanish; $\mathbf{E}\perp\mathbf{B}$ and $\lvert\mathbf{E}\rvert=c\lvert\mathbf{B}\rvert$ in every frame. A plane wave; no boost removes either field.
Generic non-orthogonal	$\mathbf{E}\cdot\mathbf{B}\neq0$	The second invariant is nonzero, so the fields are non-orthogonal in every frame; one can boost to a frame where they are parallel but never eliminate either.
Non-relativistic	$v\ll c\ \Rightarrow\ \mathbf{E}'\approx\mathbf{E}+\mathbf{v}\times\mathbf{B},\ \ \mathbf{B}'\approx\mathbf{B}-\tfrac{1}{c^2}\mathbf{v}\times\mathbf{E}$	Leading-order mixing is the motional field $\mathbf{v}\times\mathbf{B}$ ; the $\mathbf{v}\times\mathbf{E}/c^2$ term is the tiny relativistic correction usually dropped in circuit theory.

Common trapE and B are separate fields that happen to be related

The single deepest misconception is treating $\mathbf{E}$ and $\mathbf{B}$ as two physical fields joined by a coincidence. They are not two fields — they are two projections of one tensor $F_{\mu\nu}$ , and the split between them is as observer-dependent as the split of spacetime into space and time. There is no frame-independent fact about "how much electric field is present". The frame-independent facts are the two invariants. Concretely, this misconception spawns three technical errors. First, index raising: with signature $(-,+,+,+)$ , raising a time index flips the sign, $F^{0i}=-F_{0i}$ while $F^{ij}=F_{ij}$ , and forgetting this corrupts $F_{\mu\nu}F^{\mu\nu}$ , $\partial_\mu F^{\mu\nu}$ , and the Lorentz force. Second, conventions: $F_{0i}=-E_i/c$ in $(-,+,+,+)$ SI units, but many texts set $c=1$ or use Gaussian units, so factor and sign assignments differ — always check. Third, treating $\mathbf{E}$ or $\mathbf{B}$ as a four-vector: neither is; only their union $F_{\mu\nu}$ transforms covariantly, and asking for "the time component of $\mathbf{E}$ " is meaningless.

Worked examples

1 · Boosting a pure electric field, and checking the invariant

In the lab a region holds a uniform electric field $\mathbf{E}=(0,\,3\times10^{6},\,0)\ \mathrm{V/m}$ and no magnetic field. An observer moves along $x$ at $v=0.6c$ . Find the fields she measures, classify the field, and verify the invariant $\mathcal{I}_1$ is unchanged.

Step 1. Lorentz factor: $\gamma=1/\sqrt{1-0.6^2}=1/\sqrt{0.64}=\mathbf{1.25}$ . The field $E_y$ is transverse to the boost, so it is amplified by $\gamma$ : $E'_y=\gamma(E_y-vB_z)=\gamma E_y=1.25\times3\times10^6=\mathbf{3.75\times10^{6}\ V/m}$ .

Step 2. The moving observer also sees a magnetic field, generated purely by her motion through $\mathbf{E}$ : $B'_z=\gamma(B_z-vE_y/c^2)=-\gamma vE_y/c^2$ . With $v=0.6c=1.8\times10^8\ \mathrm{m/s}$ , this is $B'_z=-1.25\times(1.8\times10^8)(3\times10^6)/(9\times10^{16})=\mathbf{-7.5\times10^{-3}\ T}$ . A pure electric field in one frame carries a $7.5\ \mathrm{mT}$ magnetic field in another.

Step 3. Classify using $\mathcal{I}_1=2(B^2-E^2/c^2)$ . In the lab, $B=0$ , so $\mathcal{I}_1=-2E^2/c^2=-2(3\times10^6)^2/(9\times10^{16})=\mathbf{-2\times10^{-4}}$ . Negative means electric-dominated: no boost can ever remove the electric field, consistent with it being the field of static charge.

Step 4. Verify invariance in the moving frame: $B'^2=(7.5\times10^{-3})^2=5.625\times10^{-5}$ and $E'^2/c^2=(3.75\times10^6)^2/(9\times10^{16})=1.5625\times10^{-4}$ , so $\mathcal{I}'_1=2(5.625\times10^{-5}-1.5625\times10^{-4})=\mathbf{-2\times10^{-4}}$ — identical to the lab value. The split into $\mathbf{E}$ and $\mathbf{B}$ changed; the invariant did not.

2 · Finding the frame that kills the electric field

A field has $\mathbf{E}=(0,\,5\times10^{5},\,0)\ \mathrm{V/m}$ and $\mathbf{B}=(0,\,0,\,2\times10^{-3})\ \mathrm{T}$ , so $\mathbf{E}\perp\mathbf{B}$ . Show it is magnetic-dominated, find the boost that makes it purely magnetic, and confirm that boost is sub-luminal.

Step 1. Compare $E/c$ with $B$ : $E/c=(5\times10^5)/(3\times10^8)=1.667\times10^{-3}$ , while $B=2\times10^{-3}$ . Since $B>E/c$ , the field is magnetic-dominated. Confirm with the invariant: $\mathcal{I}_1=2(B^2-E^2/c^2)=2\big((2\times10^{-3})^2-(1.667\times10^{-3})^2\big)=\mathbf{+2.44\times10^{-6}>0}$ .

Step 2. Because $\mathbf{E}\cdot\mathbf{B}=0$ and the field is magnetic-dominated, a boost in the $\mathbf{E}\times\mathbf{B}$ direction (here $+x$ ) at the drift speed $v=E/B$ transforms the electric field to zero. Compute: $v=E/B=(5\times10^5)/(2\times10^{-3})=\mathbf{2.5\times10^{8}\ m/s}$ .

Step 3. Check it is physical: $v/c=(2.5\times10^8)/(3\times10^8)=\mathbf{0.833<1}$ , a legitimate boost. Equivalently $v/c=E/(cB)=1.667\times10^{-3}/2\times10^{-3}=0.833$ ; the drift speed stays below $c$ precisely because the field is magnetic-dominated ( $E<cB$ ).

Step 4. In that drift frame the residual field is $B'=B\sqrt{1-(E/cB)^2}=2\times10^{-3}\times\sqrt{1-0.694}=\mathbf{1.11\times10^{-3}\ T}$ , purely magnetic. Verify the invariant: $\mathcal{I}'_1=2B'^2=2(1.11\times10^{-3})^2=2.44\times10^{-6}$ — matching Step 1 exactly. The field that looked like crossed $\mathbf{E}$ and $\mathbf{B}$ is "really" a $1.11\ \mathrm{mT}$ magnetic field seen from a moving frame.

Physics of the portrait

The field tensor is the moment electromagnetism stopped being a list of four laws and became a single geometric statement — and the person who saw it was not a physicist experimenting with charges but a mathematician reasoning about spacetime.

Origins · 1 minHermann Minkowski and how this was discovered

Hermann Minkowski

1864–1909 · the mathematician who fused space and time and recast electrodynamics as tensor geometry

Hermann Minkowski was, improbably, Einstein's mathematics teacher at the Zurich Polytechnic — and reportedly once called his future-famous student a "lazy dog". In 1907–1908, three years after special relativity, Minkowski gave it the language it still uses: a four-dimensional spacetime in which the Lorentz transformations are simply rotations. His 1908 paper Die Grundgleichungen für die elektromagnetischen Vorgänge in bewegten Körpern introduced the antisymmetric field tensor (he called it a "six-vector") and showed that Maxwell's four equations are two tensor relations, with the covariance that had seemed miraculous now automatic. His 1908 Cologne address Raum und Zeit opened with the famous line that "henceforth space by itself, and time by itself, are doomed to fade away into mere shadows." He died the next year of appendicitis at just 44, before he could see how completely his geometry would remake physics: Einstein, who at first dismissed Minkowski's formalism as "superfluous learnedness", built general relativity directly on it. The field tensor $F_{\mu\nu}$ is his monument — the object that made electromagnetism the first geometric field theory.

History

1820

— Ørsted finds that an electric current deflects a nearby compass needle, linking electricity and magnetism

1831

— Faraday discovers electromagnetic induction, tying a changing magnetic flux to an electric force

1865

— Maxwell publishes his dynamical theory of the electromagnetic field, predicting waves at speed c

1905

— Einstein's electrodynamics of moving bodies elevates Lorentz covariance to a principle of nature

1908

— Minkowski recasts Maxwell's equations as tensor relations on four-dimensional spacetime

Going further

Texts. Jackson, Classical Electrodynamics (3rd ed.), Ch. 11 · Misner, Thorne & Wheeler, Gravitation, Ch. 3–4 · Wald, General Relativity, Ch. 4 · Carroll, Spacetime and Geometry, Sec. 1.9 and 4.2 · Griffiths, Introduction to Electrodynamics (4th ed.), Ch. 12 · Landau & Lifshitz, The Classical Theory of Fields, Ch. 3–4.

Provenance. The component conventions ( $F_{0i}=-E_i/c$ , $F_{ij}=-\epsilon_{ijk}B_k$ in $(-,+,+,+)$ SI units) and all equations are cross-checked against Jackson 3rd ed. Ch. 11, Griffiths 4th ed. Ch. 12, and Carroll Sec. 1.9. A verifier should recompute: (1) the invariant $F_{\mu\nu}F^{\mu\nu}=2(B^2-E^2/c^2)$ by explicit index contraction; (2) the boost mixing $E'_y=\gamma(E_y-vB_z)$ , $B'_z=\gamma(B_z-vE_y/c^2)$ from $F'_{\mu\nu}=\Lambda\Lambda F$ ; (3) that $\partial_\mu F^{\mu\nu}=\mu_0 J^\nu$ yields Gauss and Ampère–Maxwell with $\partial_\nu J^\nu=0$ automatic; (4) the numeric worked examples ( $\gamma=1.25$ , $E'_y=3.75\times10^6$ V/m, $B'_z=-7.5$ mT, $\mathcal{I}_1=-2\times10^{-4}$ invariant; drift speed $v=E/B=2.5\times10^8$ m/s $=0.833c$ ). Sign and factor conventions differ in MTW and Landau–Lifshitz (often $c=1$ , Gaussian units); the relations are convention-independent up to those rescalings. No fabricated constants.

Builds on: Lorentz Transformations · Four-Vectors · Minkowski Metric and Signature · Lorentz Boosts and Rapidity · Minkowski Spacetime. Leads to: Stress-Energy Tensor · Einstein Field Equations · Gauge Theory · Maxwell Equations in Curved Spacetime.

Ask the tutor: Give me any $\mathbf{E}$ and $\mathbf{B}$ and I will build $F_{\mu\nu}$ , boost it to whatever frame you like, and tell you whether it is electric-dominated, magnetic-dominated, or a null wave.

Are the electric and magnetic fields really the same thing?

Not the same thing, but the same object seen differently. $\mathbf{E}$ and $\mathbf{B}$ are two blocks of one tensor $F_{\mu\nu}$ , and a Lorentz boost trades entries between the blocks, so how much of the field looks electric versus magnetic depends entirely on your frame — exactly as the split of spacetime into "space" and "time" depends on your worldline. There is no frame-independent fact about "the amount of electric field present". What every observer agrees on are the two invariants $B^2-E^2/c^2$ and $\mathbf{E}\cdot\mathbf{B}$ . So "the same thing" is too strong, but "two faces of one thing" is exactly right.

Which sign convention should I use for F_0i?

It depends on your metric signature and unit system, and there is no universal answer — which is why conventions are the number-one source of sign errors here. In $(-,+,+,+)$ SI units, the fully-lowered tensor has $F_{0i}=-E_i/c$ and $F_{ij}=-\epsilon_{ijk}B_k$ , which is what this article uses. Raising one index to the mixed form $F^\mu{}_\nu$ flips the time-index sign, giving the matrix in equation (2) with $+E_i/c$ in the top row. Texts using $(+,-,-,-)$ , or setting $c=1$ , or using Gaussian units, place the factors differently. The rule that never fails: fix your signature, derive the components once from $F_{\mu\nu}=\partial_\mu A_\nu-\partial_\nu A_\mu$ , and never mix results across conventions.

Why must the field tensor be antisymmetric?

Because it is built as a curl. The definition $F_{\mu\nu}=\partial_\mu A_\nu-\partial_\nu A_\mu$ is antisymmetric in $\mu\leftrightarrow\nu$ by construction — swap the indices and the expression changes sign. This is not a physical assumption we impose; it is forced the moment we insist the field strength come from a potential four-vector. The antisymmetry then does enormous work: it fixes the number of independent components at six (matching $\mathbf{E}$ and $\mathbf{B}$ exactly), it makes charge conservation automatic through $\partial_\nu\partial_\mu F^{\mu\nu}=0$ , and it guarantees magnetic forces do no work because $F_{\mu\nu}u^\mu u^\nu=0$ when an antisymmetric tensor is contracted with a symmetric pair of velocities.

What does it mean that a plane wave is a null field?

A plane electromagnetic wave has both invariants equal to zero: $E=cB$ makes $B^2-E^2/c^2=0$ , and $\mathbf{E}\perp\mathbf{B}$ makes $\mathbf{E}\cdot\mathbf{B}=0$ . "Null" borrows the word from null spacetime intervals, because such a field is the electromagnetic analogue of a light cone. The physical consequence is striking: since invariants cannot change under a boost, no observer can ever boost a wave into a pure electric field or a pure magnetic field, and no observer can bring it to rest. It looks like a wave, with $\mathbf{E}$ and $\mathbf{B}$ perpendicular and equal in scaled magnitude, in every inertial frame — the field-theory expression of the fact that light moves at $c$ for everyone.

How does the field tensor carry over to curved spacetime?

Almost trivially, which is exactly why it is the right object. The homogeneous equation $\partial_{[\lambda}F_{\mu\nu]}=0$ involves no metric at all — as the exterior-derivative identity $\mathrm{d}F=0$ it holds on any manifold unchanged. The source equation needs only the minimal-coupling substitution $\partial_\mu\to\nabla_\mu$ and $\eta_{\mu\nu}\to g_{\mu\nu}$ , giving $\nabla_\mu F^{\mu\nu}=\mu_0 J^\nu$ . Because $F$ is antisymmetric, even this simplifies: the covariant divergence reduces to $\frac{1}{\sqrt{-g}}\partial_\mu(\sqrt{-g}\,F^{\mu\nu})=\mu_0 J^\nu$ . So electromagnetism in a gravitational field is flat-space electromagnetism with partial derivatives promoted to covariant ones — a one-line upgrade, developed fully in the follow-on article on Maxwell's equations in curved spacetime.

Why is charge conservation automatic in this formulation?

Take the source equation $\partial_\mu F^{\mu\nu}=\mu_0 J^\nu$ and hit it with another derivative: $\partial_\nu\partial_\mu F^{\mu\nu}=\mu_0\partial_\nu J^\nu$ . The left side contracts the symmetric operator $\partial_\nu\partial_\mu$ (partial derivatives commute) with the antisymmetric tensor $F^{\mu\nu}$ , and any symmetric–antisymmetric contraction vanishes identically. Therefore $\partial_\nu J^\nu=0$ — which written out is $\partial_t\rho+\nabla\cdot\mathbf{J}=0$ , the continuity equation. Charge conservation is not an extra postulate bolted onto Maxwell's equations; it is a mathematical consequence of the antisymmetry of $F$ , guaranteed the moment you write the source law in tensor form. This is a recurring theme: the geometry enforces the physics for free.

Is the four-potential physically real if it is gauge-dependent?

The potential $A^\mu$ is not unique — the transformation $A_\mu\to A_\mu+\partial_\mu\Lambda$ leaves $F_{\mu\nu}$ completely unchanged, since the extra term is a symmetric second derivative that cancels in the antisymmetric curl. So different potentials give the same fields and the same forces, and in classical electromagnetism only $F_{\mu\nu}$ is directly measurable. But the potential is not mere bookkeeping: the Aharonov–Bohm effect shows that $A^\mu$ has observable consequences in quantum mechanics even where $F_{\mu\nu}=0$ , and gauge invariance itself — the freedom to choose $\Lambda$ — is the organizing principle of the entire Standard Model. The gauge-dependence of $A^\mu$ is a feature, not a defect: it is the symmetry from which the interaction is built.