The Canonical Stress-Energy Tensor

Feed spacetime translations — the statement that physics is the same here as there, today as tomorrow — into Noether's machine and out comes not one conservation law but four, packaged as a single two-index object. The canonical stress-energy tensor $T^\mu{}_\nu$ is the master ledger of classical field theory: sixteen entries that answer, for every point of spacetime, the questions how much energy is here, how much momentum, which way are they flowing, and what is pushing on what. Energy density, momentum density, energy flux, pressure, shear — quantities that the mechanics chapters treated as separate characters — turn out to be entries of one tensor, and their conservation laws four components of one equation, $\partial_\mu T^\mu{}_\nu = 0$ . The Noether article derived the formula; this article learns to read it. The reading pays immediately: the $00$ entry reproduces exactly the Hamiltonian density of the Hamiltonian-formulation article (Noether's theorem and the Legendre transform agree on what energy is); the $0i$ entries reveal that a traveling wave carries momentum in strict proportion to its energy — the field-theory root of radiation pressure; and the $ij$ block is the relativistic completion of the Cauchy stress tensor from the continuum-mechanics chapter. The article ends on the tensor's honest defects: for fields with spin it is not symmetric, for gauge fields not even gauge-invariant — precise diseases with a precise cure, the Belinfante-Rosenfeld improvement, which is the next-but-one article's subject and the reason general relativity can use this object as the source of gravity.

Key takeaways

$T^\mu{}_\nu = \frac{\partial\mathcal{L}}{\partial(\partial_\mu\phi_a)}\partial_\nu\phi_a - \delta^\mu{}_\nu\mathcal{L}$ is the Noether current of the four spacetime translations; $\partial_\mu T^\mu{}_\nu = 0$ is four continuity equations in one.
Read the indices as a ledger: the column $\nu$ says what flows (energy for $\nu = 0$ , $i$ -momentum for $\nu = i$ ); the row $\mu$ says which way (through a time slice for $\mu = 0$ : a density; through a spatial face for $\mu = i$ : a flux).
$T^{00} = \pi_a\dot\phi_a - \mathcal{L} = \mathcal{H}$ : Noether's energy density is exactly the Hamiltonian density — two constructions, one answer.
$T^{0i}$ is momentum density and $T^{i0}$ energy flux; $T^{ij}$ is the stress (pressure and shear), the relativistic upgrade of the Cauchy tensor.
The conserved charges are the four-momentum $P_\nu = \int\mathrm{d}^3x\,T^0{}_\nu$ , with the flux ledger $\dot P_\nu = -\oint T^i{}_\nu\,\mathrm{d}S_i$ .
For the free scalar, $T^{00}$ is a manifest sum of squares — energy is positive — and a traveling wave carries momentum $k/\omega$ per unit energy.
The canonical tensor is not unique ( $T^{\mu\nu} + \partial_\lambda K^{[\lambda\mu]\nu}$ , same charges), generally not symmetric, and for gauge fields not gauge-invariant — the Belinfante-Rosenfeld article repairs all three.
Key idea: translation symmetry makes energy-momentum a locally conserved four-current of four-currents — one tensor, the whole mechanical ledger of a field.

The core insight

$T^\mu{}_\nu = \frac{\partial\mathcal{L}}{\partial(\partial_\mu\phi_a)}\partial_\nu\phi_a - \delta^\mu{}_\nu\,\mathcal{L}$ is the Noether current for spacetime translations; $\partial_\mu T^\mu{}_\nu = 0$ encodes local conservation of energy ( $\nu = 0$ ) and momentum ( $\nu = i$ ).

Where this lives today

The stress-energy tensor is the busiest object in physics. In general relativity it is the source: $G_{\mu\nu} = 8\pi G\,T_{\mu\nu}$ couples spacetime curvature to exactly this ledger (in its symmetric, improved form), so every gravitational computation — LIGO waveform templates, neutron-star structure, cosmological expansion — begins by writing down somebody's $T^{\mu\nu}$ . Cosmology's inventory is literally a decomposition of the universe's stress-energy into equations of state read off this article's table: dust ( $p = 0$ , the $k \to 0$ wave), radiation ( $p = \rho/3$ , the traceless case), and dark energy ( $p = -\rho$ , a pure $-\eta^{\mu\nu}\mathcal{L}$ -type term — the tensor of a constant, which is what makes the cosmological-constant problem a stress-energy problem). Neutron-star physics is an ongoing measurement of $T^{ij}$ at supranuclear density: the equation of state connecting pressure to energy density is being pinned between NICER's radius measurements and LIGO's tidal deformabilities. In quantum field theory the tensor survives as an operator whose expectation values violate classical positivity ( $\langle T^{00}\rangle < 0$ between Casimir plates — measured), fueling the quantum-energy-inequality program that polices wormholes and warp drives; the trace acquires an anomaly that governs conformal field theory; and in the fluid-gravity and holographic programs, $T^{\mu\nu}$ is the primary dictionary entry between boundary field theories and bulk spacetimes. Even materials science speaks it now: the momentum-flux reading of $T^{ij}$ underlies the optical forces of laser tweezers (2018 Nobel) and the acoustic radiation pressure of ultrasound tractor beams. Within the chapter: the angular-momentum article takes moments of this tensor and discovers the spin current; the Belinfante-Rosenfeld article performs the repair (4) demands; and the four charges $P_\nu$ join the bracket algebra as the translation generators of the Poincaré set.

The physical picture

Think of $T^\mu{}_\nu$ as a 4×4 bookkeeping table, and commit the index grammar to memory now — it prevents every standard confusion later. The column index $\nu$ names the conserved commodity: the $\nu = 0$ column tracks energy, the $\nu = i$ columns track the three components of momentum. The row index $\mu$ names the direction of flow: row $\mu = 0$ entries are densities (commodity per volume, what a constant-time slice counts), rows $\mu = i$ are fluxes (commodity per area per time, what a detector face collects). The single tensor equation $\partial_\mu T^\mu{}_\nu = 0$ is then four separate continuity equations of the conserved-currents article — one per column, each with its own density-and-flux pair.

Figure

The stress-energy ledger — sixteen entries, four continuity equations, one tensor

Why must the ledger exist at all? In particle mechanics, time-translation symmetry conserved one number ( $E$ ) and space-translation three more ( $\mathbf{p}$ ). A field has energy and momentum at every point, so "conserved" can only mean locally accounted: the energy in a cell changes only by flux through its faces, the momentum in a cell changes only by the stresses acting on its faces. The mixed entries — flux of energy, density of momentum — are not bureaucratic extras; they are what make the local balance close. And one of the deepest facts in the table is the near-triviality of a symmetry: for well-behaved theories $T^{0i}$ and $T^{i0}$ — momentum density and energy flux, two conceptually distinct quantities — turn out equal. That identity (manifest for the scalar field, restored by improvement in general) is the field-theoretic face of $E = mc^2$ : flowing energy is momentum, inertia belongs to energy flux, and a beam of light pushes on what absorbs it. The last picture to install is the tensor's non-uniqueness, inherited from the conserved-currents article: adding $\partial_\lambda K^{\lambda\mu}{}_\nu$ with $K$ antisymmetric in $\lambda\mu$ is like superposing a curl on the ledger — it redistributes where energy and momentum are attributed locally while changing no total and no conservation law. Whether nature cares about the attribution is a real question with a real answer — gravity couples to a specific tensor — and it is deferred to the Belinfante-Rosenfeld article, where the freedom becomes a tool.

The tensor and its charges

The Noether article derived, from invariance under $x^\mu \to x^\mu + a^\mu$ , the canonical stress-energy tensor; we recall the result in its mixed-index form, which is the safest for bookkeeping:

T^\mu{}_\nu = \frac{\partial\mathcal{L}}{\partial(\partial_\mu\phi_a)}\,\partial_\nu\phi_a - \delta^\mu{}_\nu\,\mathcal{L}, \qquad \partial_\mu T^\mu{}_\nu = 0 \quad\text{(on-shell)} \qquad (1)

where: the sum runs over all fields $\phi_a$ in the theory; the first term is "canonical momentum flowing in direction $\mu$ " times "field gradient in direction $\nu$ ", and the $-\delta^\mu{}_\nu\mathcal{L}$ term is the dragged-density contribution $K^\mu$ that the translation's non-strict invariance forces (forget it and the energy density comes out wrong — the Noether article's warning). Raising $\nu$ with $\eta^{\nu\rho}$ gives $T^{\mu\nu}$ , the form that displays symmetry properties. Integrating the density row over a time slice gives the conserved charges — the field's total four-momentum:

P_\nu = \int_\Sigma \mathrm{d}^3x\; T^0{}_\nu, \qquad \frac{\mathrm{d}P_\nu}{\mathrm{d}t} = -\oint_{\partial\Sigma} T^i{}_\nu\,\mathrm{d}S_i \qquad (2)

where: the flux form is the conserved-currents ledger applied per column: for a localized field the boundary term dies and all four $P_\nu$ are constants; for a radiating system the boundary integral is the radiated power ( $\nu = 0$ ) and radiated momentum ( $\nu = i$ ). Everything proved in the conserved-currents article — slice-independence, frame consistency, the $1/r^2$ fine print — applies to each column verbatim. One identification deserves its own display, because two different formalisms converge on it. Set $\mu = \nu = 0$ in (1): the canonical momentum is $\pi_a = \partial\mathcal{L}/\partial\dot\phi_a$ , and

T^0{}_0 = \pi_a\,\dot\phi_a - \mathcal{L} = \mathcal{H} \qquad (3)

where: the right-hand side is precisely the Legendre transform that defined the Hamiltonian density in the Hamiltonian-formulation article. Noether's theorem (symmetry route) and the Legendre transform (canonical route) manufacture the same energy density — the consistency check that ties this chapter's two formalisms together, and the reason $P_0 = H$ generates time evolution in the Poisson-bracket article while being the time-translation charge here.

Derivation ADirect verification: the divergence of T vanishes on-shell

We verify $\partial_\mu T^\mu{}_\nu = 0$ by brute-force differentiation of the definition (1) — no variational argument, just calculus and the field equations. The value of doing this once explicitly is twofold: it shows exactly where the Euler-Lagrange equations enter (Noether conservation is on-shell, and here you watch the on-shell input consumed), and it is the index-gymnastics warm-up for every tensor manipulation in the rest of the chapter. Strategy: expand the divergence by the product rule, expand $\partial_\nu\mathcal{L}$ by the chain rule, and watch four terms annihilate in pairs.

Step 1. Differentiate the first term of (1) by the product rule:

\partial_\mu\!\left(\frac{\partial\mathcal{L}}{\partial(\partial_\mu\phi_a)}\,\partial_\nu\phi_a\right) = \underbrace{\left[\partial_\mu\frac{\partial\mathcal{L}}{\partial(\partial_\mu\phi_a)}\right]}_{(A)}\partial_\nu\phi_a + \frac{\partial\mathcal{L}}{\partial(\partial_\mu\phi_a)}\,\underbrace{\partial_\mu\partial_\nu\phi_a}_{(B)}

Step 2. Differentiate the second term. Because $\mathcal{L}$ has no explicit $x$ -dependence (the hypothesis of translation symmetry), the chain rule gives

\partial_\mu\big(\delta^\mu{}_\nu\mathcal{L}\big) = \partial_\nu\mathcal{L} = \frac{\partial\mathcal{L}}{\partial\phi_a}\,\partial_\nu\phi_a + \frac{\partial\mathcal{L}}{\partial(\partial_\mu\phi_a)}\,\partial_\nu\partial_\mu\phi_a

Why explicit x-dependence would ruin it: an external background (a spacetime-dependent mass, a driving term) adds $(\partial_\nu\mathcal{L})_{\text{explicit}}$ to this line, and the final result becomes $\partial_\mu T^\mu{}_\nu = -(\partial_\nu\mathcal{L})_{\text{explicit}}$ — energy-momentum is pumped by the background at a computable rate. Conservation fails exactly when, and exactly as much as, the symmetry does.

Step 3. Subtract Step 2 from Step 1. The second derivatives (B) cancel against their twins (partial derivatives commute), leaving

\partial_\mu T^\mu{}_\nu = \left[\partial_\mu\frac{\partial\mathcal{L}}{\partial(\partial_\mu\phi_a)} - \frac{\partial\mathcal{L}}{\partial\phi_a}\right]\partial_\nu\phi_a

Step 4. The bracket is the Euler-Lagrange expression: zero on solutions. Hence $\partial_\mu T^\mu{}_\nu = 0$ on-shell, for every $\nu$ — four conservation laws, delivered by one line of index algebra. Off-shell, the identity quantifies the leak, exactly as Noether's off-shell identity promised. Verified: matches the Noether-route derivation of the Noether article (the two must agree — this is the same current); the explicit-dependence caveat cross-checked against Goldstein §13.3's treatment of open systems.

Reading the entries: the free scalar as exhibit

Go deeper · Derivation B · 1 minThe stress-energy of the free scalar field, entry by entry

Derivation BThe stress-energy of the free scalar field, entry by entry

We evaluate every block of $T^{\mu\nu}$ for the free real scalar, $\mathcal{L} = \tfrac12\partial_\mu\phi\,\partial^\mu\phi - \tfrac12 m^2\phi^2$ (signature $+,-,-,-$ ), and read off the physics: positive energy density, the momentum-density/energy-flux identity, and the stress block. This is the exhibit case — symmetric, gauge-trivial, every pathology absent — against which the diseased cases (EM below, spin later) will be diagnosed. Strategy: compute the canonical momentum, substitute into (1), and evaluate the four blocks separately, keeping the physical reading beside each.

Step 1. Canonical momentum four-vector: $\frac{\partial\mathcal{L}}{\partial(\partial_\mu\phi)} = \partial^\mu\phi$ . The tensor, with both indices up:

T^{\mu\nu} = \partial^\mu\phi\,\partial^\nu\phi - \eta^{\mu\nu}\mathcal{L}

Manifestly symmetric in $\mu\nu$ — for the scalar, the canonical tensor needs no repair. (This is special: it happens because a scalar carries no spin index for translations to entangle, as the angular-momentum article will explain.)

Step 2. The $00$ entry:

T^{00} = \dot\phi^2 - \mathcal{L} = \tfrac{1}{2}\dot\phi^2 + \tfrac{1}{2}(\nabla\phi)^2 + \tfrac{1}{2}m^2\phi^2

A manifest sum of squares: the energy density is non-negative, every term accounted (kinetic, gradient-spring, mass-well — the anatomy the Hamiltonian article established), and (3) is confirmed explicitly.

Step 3. The mixed entries:

T^{0i} = \dot\phi\;\partial^i\phi = -\dot\phi\,\partial_i\phi = T^{i0}

One expression, two readings: as $T^{0i}$ it is the density of $i$ -momentum; as $T^{i0}$ it is the flux of energy in direction $i$ . Their equality — automatic here — is the "inertia of energy flow" statement: wherever energy streams, momentum resides, in the exact ratio $c^{-2}$ (invisible in natural units). A standing wave has $\dot\phi$ and $\nabla\phi$ ninety degrees out of phase: its time-averaged $T^{0i}$ vanishes — energy sloshes but does not travel.

Step 4. The stress block:

T^{ij} = \partial^i\phi\,\partial^j\phi + \delta^{ij}\mathcal{L}

Diagonal entries are pressures (momentum flux through a face perpendicular to itself), off-diagonal entries shear stresses — this is the Cauchy stress tensor of the continuum chapter, generated now from a Lagrangian rather than postulated from force balance. For a static configuration ( $\dot\phi = 0$ ), $\partial_i T^{ij} = 0$ reduces to the continuum chapter's equilibrium equation: field statics is elastostatics with $\mathcal{L}$ -derived stresses. Verified: $T^{00}$ against the Hamiltonian article's $\mathcal{H}$ ; symmetry manifest; the static reduction against the Cauchy-relation article of the continuum chapter; matches Peskin & Schroeder eqs. 2.17–2.18 forms.

Figure

The ledger of a traveling wave, computed — energy density and momentum density ride together, locked in ratio k over omega

The one diseased case worth meeting now, because it motivates the two articles that follow: run formula (1) on electromagnetism, $\mathcal{L} = -\tfrac14 F_{\alpha\beta}F^{\alpha\beta}$ with fields $A_\alpha$ . The canonical output is

T^{\mu\nu}_{\text{can}} = -F^{\mu\alpha}\,\partial^\nu A_\alpha + \tfrac{1}{4}\eta^{\mu\nu}F^2 \qquad (4)

where: the offending structure is $\partial^\nu A_\alpha$ — a bare potential gradient, not assembled into the field strength $F$ . Consequences: (4) is not symmetric ( $\mu$ enters through $F$ , $\nu$ through $\partial A$ — different footings), and worse, not gauge-invariant (shift $A_\alpha \to A_\alpha + \partial_\alpha\lambda$ and $T_{\text{can}}$ changes by more than an improvement). Yet its charges are correct, and adding one cleverly chosen improvement term $\partial_\lambda(F^{\mu\lambda}A^\nu)$ — legal, by the conserved-currents article's freedom — produces the symmetric, gauge-invariant tensor whose $T^{0i}$ is the Poynting vector $\mathbf{E}\times\mathbf{B}$ . That repair, done systematically rather than by cleverness, is the Belinfante-Rosenfeld construction; the reason the disease strikes fields with indices and spares scalars is the spin current, next article. Here the lesson is diagnostic: the canonical formula is always right about the totals and not always right about the local attribution.

Limiting cases

Case	Result	Meaning
free real scalar	$T^{\mu\nu}$ symmetric, gauge-trivial	the clean exhibit; canonical = improved
$\mu = \nu = 0$	$T^{00} = \mathcal{H}$	Noether energy = Legendre energy
standing wave	$\langle T^{0i}\rangle = 0$	energy sloshes in place; no net transport
traveling wave	$\langle T^{0i}\rangle/\langle T^{00}\rangle = k^i/\omega$	momentum rides with energy at $v_g$
static field	$\partial_j T^{ji} = 0$	elastostatics: the continuum chapter's equilibrium equation
massless conformal field (EM in 4D)	$T^\mu{}_\mu = 0$	tracelessness = scale invariance; radiation $p = \rho/3$
electromagnetism, canonical	asymmetric, gauge-dependent	needs Belinfante repair; totals still correct
explicit $x$ -dependence	$\partial_\mu T^\mu{}_\nu = -(\partial_\nu\mathcal{L})_{\text{expl}}$	backgrounds pump energy-momentum at a computable rate

Common trapDo not assume T is symmetric — and do not confuse the row with the column

Two failure modes, one structural and one notational. Structural: the canonical tensor is symmetric only for scalars. The moment fields carry indices — vectors, spinors, anything with spin — $T^{\mu\nu}_{\text{can}} \neq T^{\nu\mu}_{\text{can}}$ , and for gauge fields the canonical object is not even gauge-invariant, as (4) exhibits. Consequences are not cosmetic: an asymmetric $T$ cannot balance the angular-momentum books ( $\partial_\mu M^{\mu\nu\lambda} = 0$ forces $T^{[\nu\lambda]} = -\tfrac12\partial_\mu S^{\mu\nu\lambda}$ , the next article's central identity), and general relativity's field equations demand a symmetric source. Never quote "the" stress tensor of a spinful theory without saying which one — canonical or Belinfante-improved — you mean; they differ by terms that redistribute density while preserving every charge. Notational: keep the index roles straight. In $T^\mu{}_\nu$ , the lower (column) index labels the conserved current, the upper (row) labels the flow direction; $\partial_\mu$ contracts the row. Swapping them in a computation produces objects like $\partial_\nu T^\mu{}_\nu$ (nonsense) or conservation claims for the wrong index — and in the asymmetric case, where $T^{0i} \neq T^{i0}$ matters, the swap silently exchanges momentum density with energy flux. The mixed-index form (1) plus the ledger reading (column = what, row = which way) is the error-proof idiom.

Worked examples

1 · Energy and momentum of a traveling wave (confidence builder)

A free scalar traveling wave $\phi = A\cos(kx - \omega t)$ with $A = 0.20$ , $k = 0.75$ , $m = 1.00$ (so $\omega = 1.25$ ). Compute the cycle-averaged energy density $\langle T^{00}\rangle$ and momentum density $\langle T^{0x}\rangle$ , and their ratio.

Step 1. Derivatives: $\dot\phi = A\omega\sin(kx - \omega t)$ , $\partial_x\phi = -Ak\sin(kx - \omega t)$ ; averages $\langle\sin^2\rangle = \langle\cos^2\rangle = \tfrac12$ .

Step 2. Energy density: $\langle T^{00}\rangle = \tfrac14 A^2\omega^2 + \tfrac14 A^2k^2 + \tfrac14 A^2m^2 = \tfrac14 A^2(\omega^2 + \omega^2) = \tfrac12 A^2\omega^2$ using $k^2 + m^2 = \omega^2$ . Numerically: $\tfrac12(0.04)(1.5625) = \mathbf{0.03125}$ .

Step 3. Momentum density: $T^{0x} = -\dot\phi\,\partial_x\phi = A^2\omega k\sin^2(kx - \omega t)$ , so $\langle T^{0x}\rangle = \tfrac12 A^2\omega k = \tfrac12(0.04)(0.9375) = \mathbf{0.01875}$ .

Step 4. The ratio: $\langle T^{0x}\rangle/\langle T^{00}\rangle = k/\omega = 0.75/1.25 = \mathbf{0.600} = v_g$ — the wave carries momentum in exact proportion to its energy, transported at the group velocity, the same $0.600$ the charge rode at in the Noether article's example. Absorb this wave and you absorb momentum: fields push. In the massless limit $k/\omega \to 1$ : momentum equals energy (in $c = 1$ units), the extreme case realized by light and the reason radiation pressure was the first laboratory proof that fields carry mechanical momentum.

2 · The full ledger of the wave, and its pressure (exam level)

For the same wave, (i) compute the stress entry $\langle T^{xx}\rangle$ (the pressure the wave exerts along its direction of travel), (ii) verify the $x$ -momentum continuity equation $\partial_t T^{0x} + \partial_x T^{xx} = 0$ pointwise, and (iii) compare the pressure-to-energy ratio with the photon-gas value.

Step 1. The Lagrangian on the wave: $\mathcal{L} = \tfrac12\dot\phi^2 - \tfrac12(\partial_x\phi)^2 - \tfrac12 m^2\phi^2 = \tfrac12 A^2(\omega^2 - k^2)\sin^2 - \tfrac12 A^2 m^2\cos^2 = \tfrac12 A^2 m^2(\sin^2 - \cos^2)$ , whose average is zero — for a free wave, kinetic and potential actions balance on cycle average.

Step 2. Pressure: $T^{xx} = (\partial_x\phi)(\partial_x\phi) + \eta^{xx}\mathcal{L}\cdot(-1) = (\partial_x\phi)^2 + \mathcal{L}$ (with $\eta^{xx} = -1$ , $T^{xx} = \partial^x\phi\partial^x\phi - \eta^{xx}\mathcal{L} = (\partial_x\phi)^2 + \mathcal{L}$ ). Averaging: $\langle T^{xx}\rangle = \tfrac12 A^2k^2 + 0 = \tfrac12(0.04)(0.5625) = \mathbf{0.01125}$ .

Step 3. Continuity check, pointwise. With $u = kx - \omega t$ : $T^{0x} = A^2\omega k\sin^2 u$ and $T^{xx} = A^2k^2\sin^2 u + \tfrac12 A^2m^2(\sin^2 u - \cos^2 u)$ . Then $\partial_t T^{0x} = -2A^2\omega^2 k\sin u\cos u$ and $\partial_x T^{xx} = 2A^2k^3\sin u\cos u + 2A^2m^2k\sin u\cos u = 2A^2k(k^2 + m^2)\sin u\cos u = 2A^2k\omega^2\sin u\cos u$ . Sum: $\mathbf{0}$ , identically — the dispersion relation is once again the conservation mechanism, now for momentum.

Step 4. The equation-of-state ratio: $\langle T^{xx}\rangle/\langle T^{00}\rangle = k^2/\omega^2 = \mathbf{0.36}$ . A massive field's wave is "soft": pressure is only 36% of energy density here, and $\to 0$ for a wave at rest ( $k \to 0$ : pure mass-energy, pressureless — the field-theory picture of cold matter). A massless field's unidirectional wave has $p = \rho$ exactly; averaged over isotropic directions in 3D this becomes the photon-gas $p = \rho/3$ , the trace-zero radiation equation of state. One tensor, read three ways, spans dust, radiation, and everything between — which is precisely how cosmology parametrizes the contents of the universe.

Origins · 1 minMax von Laue and how this was discovered

Max von Laue

1879–1960 · Berlin

Laue is remembered for the 1912 discovery that crystals diffract X-rays — Einstein called it one of the most beautiful experiments in physics, and it won the 1914 Nobel Prize — but his deepest theoretical legacy is this article's object. In 1911 he wrote the first textbook on relativity ever published, and in it worked out the complete dynamics of continuous media in special relativity: the "world tensor" $T^{\mu\nu}$ , its role as the carrier of energy density, energy flux, momentum density and stress, and the theorem that a closed static system's stresses must integrate to zero (the Laue theorem) — resolving the notorious " $4/3$ problem" of the electron's electromagnetic mass, where energy and momentum bookkeeping of the self-field disagreed until the stresses were counted. Laue understood before almost anyone that energy flux carries inertia and that the symmetric tensor, not force laws, is the fundamental mechanical object — the conceptual step this chapter's ledger reading inherits. He was also one of German science's rare profiles in courage: an open opponent of the Nazi regime who defended Einstein's physics when it was denounced as "Jewish," carried Planck's scientific conscience through the war years (Einstein said of him, "of them all, only Laue remained upright"), and helped rebuild German science afterward. He died at eighty, injured in a car crash while driving to his laboratory.

History

1822

— Cauchy formulates the stress tensor of continuum mechanics: force per area as a linear map on surface normals.

1884

— Poynting identifies the energy flux of the electromagnetic field — the first mixed entry of the ledger measured.

1908

— Minkowski unifies electromagnetic energy, momentum and stress into one four-tensor on spacetime.

1911

— Laue's relativity monograph builds the complete stress-energy dynamics of continua; the Laue theorem settles the electron's 4/3 problem.

1915

— Einstein places

T^{\mu\nu}

on the right-hand side of the field equations: the ledger becomes the source of gravity.

1918

— Noether derives

T^{\mu\nu}

from translation symmetry: the tensor's existence becomes a theorem.

Going further

References. Goldstein, Poole & Safko, Classical Mechanics, 3rd ed., §13.3 and §13.7 · Landau & Lifshitz, The Classical Theory of Fields, §32–35 · Peskin & Schroeder, An Introduction to QFT, §2.2 (eqs. 2.17–2.18) · Misner, Thorne & Wheeler, Gravitation, Ch. 5 (the stress-energy tensor as geometric object)

Provenance. Canonical formula and Noether derivation should be re-verified against Goldstein §13.3, Peskin–Schroeder §2.2 (Eqs. 2.17–2.18), and Landau–Lifshitz Vol.2 §32. Verifier should: (1) reproduce $T^\mu{}_\nu$ from the translation Noether current; (2) confirm on-shell $\partial_\mu T^\mu{}_\nu=0$ using Euler–Lagrange; (3) check the free-scalar $T^{\mu\nu}$ and positive energy density; (4) confirm the EM improvement gives Poynting density $\mathbf{E}\times\mathbf{B}$ and Maxwell stress. No numerical constants are asserted beyond standard $8\pi G$ in Einstein's equation (Wald/MTW). Additionally for this upgrade: worked-example numbers recomputed directly ( $\langle T^{00}\rangle = \tfrac12 A^2\omega^2 = 0.03125$ ; $\langle T^{0x}\rangle = \tfrac12 A^2\omega k = 0.01875$ ; ratio $k/\omega = 0.600$ ; $\langle T^{xx}\rangle = \tfrac12 A^2 k^2 = 0.01125$ ; $p/\rho = k^2/\omega^2 = 0.36$ ; pointwise continuity check via $2A^2k(k^2 + m^2 - \omega^2)\sin u\cos u = 0$ ). Scope note: the translation Noether derivation is owned by the Noether article (recalled, not re-derived); the classical Cauchy stress by the continuum chapter's stress-tensor article; the symmetrization procedure by the Belinfante-Rosenfeld article; the angular-momentum identity by the spin-current article.

Builds on: The Lagrangian Density and Field Action · Covariant Euler-Lagrange Equations for Fields · Noether's Theorem for Continuous Symmetries · Conserved Currents and Charges Leads to: Symmetrization and the Belinfante-Rosenfeld Tensor · Angular Momentum and Spin Current in Fields · Hamiltonian Formulation for Fields · Lorentz Covariance of Classical Field Theory

Why is the energy density the same whether I use Noether's theorem or the Legendre transform?

Because both constructions answer the same question — "what does time-translation symmetry conserve?" — through different doors, and consistency of mechanics requires one answer. The Legendre transform builds $\mathcal{H} = \pi\dot\phi - \mathcal{L}$ as the generator of time evolution in the canonical formalism; Noether builds $T^{00}$ as the conserved density of the time-translation symmetry. The identity (3) is algebraically instant — set $\nu = 0$ in the canonical formula and the definition of $\pi$ appears — but its structure is worth savoring: the $-\delta^\mu{}_\nu\mathcal{L}$ term of the Noether current is precisely the $-\mathcal{L}$ of the Legendre transform. That is, the "dragged density" correction that the Noether article's warning insists on is the Legendre subtraction. The agreement generalizes: $P_i = \int T^0{}_i$ equals the canonical field momentum $-\int\pi\,\partial_i\phi$ that generates spatial translations in the bracket article, and the whole Poincaré charge set can be assembled from either formalism. When the two constructions disagree in some exotic setting (higher-derivative theories, for instance), that disagreement is itself diagnostic — usually of an incomplete Legendre analysis with constraints.

Is T⁰ⁱ = Tⁱ⁰ a definition, a theorem, or an accident of the scalar field?

A theorem with hypotheses — and one of the deepest single equations in physics when read physically. For the scalar field the equality is manifest in the algebra (Step 3 of the exhibit derivation). In general the canonical tensor does not satisfy it; what is true is that the improved (Belinfante) tensor is symmetric, and gravity and angular-momentum bookkeeping both select that tensor, so the physically attributed momentum density and energy flux are equal. The physical content is the inertia of energy: $T^{0i} = T^{i0}$ says (restoring units) $\text{momentum density} = \text{energy flux}/c^2$ , which is $E = mc^2$ in local, streaming form — energy in transit has momentum, hence exerts pressure when absorbed, hence gravitates as a flow. Planck stated this as the "inertia of energy" principle in 1908; Laue built relativistic continuum mechanics on it. Its experimental face is everywhere from radiation pressure (first measured by Lebedev in 1901, now the operating principle of laser cooling and optical tweezers) to the momentum budget of solar sails. When the canonical tensor violates the symmetry — spinful fields — the violation is exactly compensated by the spin current, which is the next article's central identity.

What exactly do the diagonal and off-diagonal entries of the spatial block measure?

$T^{ij}$ is the flux of $j$ -momentum through a face with normal $i$ — equivalently, the $j$ -component of force per unit area transmitted across that face. Diagonal entries are normal stresses: $T^{xx} > 0$ means the field on one side of a $yz$ -face pushes the other side along $+x$ — a pressure; $T^{xx} < 0$ is a tension (field lines "pulling," as electrostatic field lines do along their length in the EM track's Maxwell stress tensor). Off-diagonal entries are shears: $T^{xy}$ transmits $y$ -force across an $x$ -face, the signature of momentum flowing obliquely. This is exactly the Cauchy stress tensor of the continuum-mechanics chapter — same object, same interpretation — with two upgrades: it is now derived from the Lagrangian rather than postulated, and it sits inside a spacetime tensor so that boosts mix it with densities and fluxes (a moving fluid's pressure contributes to its energy flux — the enthalpy structure of relativistic hydrodynamics, and ultimately of the Euler equations of the fluids articles). For statics the connection is verbatim: $\partial_j T^{ij} = 0$ with no time dependence is the continuum chapter's equilibrium condition, and a scalar field's static stress block gives field-line tension/pressure exactly analogous to the electromagnetic case.

What does the trace of the stress-energy tensor mean physically?

The trace $T^\mu{}_\mu = T^{00} - \sum_i T^{ii} = \rho - 3p$ (isotropic case) measures how far the theory is from scale invariance, and it organizes an astonishing amount of physics. For the free scalar, $T^\mu{}_\mu = -\,\partial_\mu\phi\,\partial^\mu\phi \cdot 0 + \ldots$ works out proportional to $m^2\phi^2$ (plus improvement-sensitive pieces): the mass term is the obstruction — mass sets a scale, and the trace knows it. Massless electromagnetism in four dimensions has $T^\mu{}_\mu = 0$ identically, which is why radiation obeys $p = \rho/3$ and why a photon gas redshifts as $a^{-4}$ in cosmology while pressureless dust ( $p = 0$ , trace $= \rho$ ) dilutes as $a^{-3}$ ; dark energy sits at the opposite extreme, $p = -\rho$ , trace $= 4\rho$ with the sign that accelerates expansion. In the quantum theory the trace becomes a celebrity: classically traceless theories acquire a trace anomaly proportional to the beta function — QCD's proton gets most of its mass from exactly this term (the quarks' Higgs masses supply only a few percent; the rest is anomalous gluon field energy) — and in conformal field theory the trace's central charges count degrees of freedom. A one-line dictionary: trace = mass scale content; zero trace = scale-free radiation; negative-pressure trace = vacuum-like.

If improvements can move energy density around, is the local energy density of a field even measurable?

Within an isolated field theory, only up to the improvement class — but nature provides an external arbiter, and it is gravity. The equivalence principle says everything gravitates: spacetime curvature responds to the local stress-energy, so two candidate tensors differing by an improvement predict (in principle) different gravitational fields sourced by the same matter, and the ambiguity becomes an experimental question. The answer selected by consistency and observation is the symmetric tensor obtained by varying the action with respect to the metric — the Hilbert tensor — which the Belinfante-Rosenfeld article shows coincides with the improved canonical tensor. At laboratory scale the discrimination is hopeless (the differences are total derivatives whose integrated moments are tiny), but the principle resolves the philosophy: local energy density is physical because gravity probes it. There is a quantum coda: the Casimir effect measures energy-density differences between boundary conditions, and gravitational coupling of Casimir energy is actively discussed — the improvement question, wearing modern clothes. Within this chapter's scope, the operational rule is simpler: quote totals (unambiguous), quote densities with the tensor named, and use the symmetric tensor whenever angular momentum or gravity is in play.

How does the stress-energy tensor of a field relate to the energy-momentum of the particles that QFT will find in it?

Beautifully directly, and the mode language of the Hamiltonian article makes it exact. Expand the free field in Fourier modes: the charges assemble into $P^\mu = \sum_{\mathbf{k}} (\text{mode occupancy}) \times (\omega_k, \mathbf{k})$ — each mode contributes four-momentum proportional to $(\omega_k, \mathbf{k})$ , which is precisely the four-momentum $(\hbar\omega_k, \hbar\mathbf{k})$ of the particle quantization will assign to that mode (factors of $\hbar$ arriving with the quanta). The dispersion relation $\omega_k^2 = k^2 + m^2$ is then literally $E^2 = p^2 + m^2$ — the SR chapter's mass-shell relation — announcing that the field's quanta will be relativistic particles of mass $m$ before quantum mechanics is switched on. The worked example's ratios preview the kinematics: momentum/energy $= k/\omega = v_g$ is the particle's velocity, and the pressure/energy ratio $k^2/\omega^2 = v^2$ interpolates between dust and radiation exactly as a gas of particles of speed $v$ does. This is the chapter's recurring miracle in stress-energy form: the classical field already carries the complete particle kinematics; quantization only makes the bookkeeping integer-valued.