Noether's Theorem

The claim

If the action of a system is invariant under a continuous transformation of its variables — for instance, shifting all clocks forward by ε seconds, or rotating the whole apparatus by ε radians — then there is a quantity built out of the dynamical variables and their derivatives that does not change with time. Noether's theorem makes this rigorous. The conserved quantity is computable directly from the symmetry transformation and the Lagrangian; the recipe is mechanical.

The result was published in 1918 by Amalie Emmy Noether, working in Göttingen alongside Hilbert and Klein, who had asked her to clarify how energy could be conserved in general relativity. Her answer — published as "Invariante Variationsprobleme" — solved that problem and accidentally produced the most-cited theorem in 20th-century theoretical physics. Albert Einstein wrote to Hilbert that she was "the most significant creative mathematical genius thus far produced since the higher education of women began."

The setup

Start with a Lagrangian L(q, q̇, t) and the action S = ∫L dt over a path q(t). Consider a one-parameter family of transformations

q(t) → q(t) + ε·F(q, q̇, t) + O(ε²)
t   → t   + ε·T(q, q̇, t) + O(ε²)

If the action is invariant — δS = 0 — for arbitrary path (not just on-shell), the transformation is a symmetry. Noether's theorem says: if δS = 0 to first order in ε for all q(t), then on the actual classical trajectory (where the Euler-Lagrange equations hold), the quantity

Q = (∂L/∂q̇)·F − [(∂L/∂q̇)·q̇ − L]·T

satisfies dQ/dt = 0. Q is the Noether charge. The first term is the "internal" part for transformations of q; the second is the "spacetime" part for transformations of t. For pure space-of-state transformations (T = 0), it simplifies to Q = (∂L/∂q̇)·F − Λ where Λ accounts for the case L → L + dΛ/dt under the transformation (a "quasi-symmetry").

Three classic applications

Time translation gives energy. Take T = 1, F = 0: shift t → t + ε. If L does not depend on t explicitly, S is invariant. The conserved charge is Q = (∂L/∂q̇)·q̇ − L = H, the Hamiltonian. So energy conservation is exactly time-translation invariance.

Space translation gives momentum. Take F_i = δ_i,k, T = 0: shift the k-th coordinate by ε. If L is independent of q_k, S is invariant. The conserved charge is Q = ∂L/∂q̇_k = p_k, the conjugate momentum. Translation invariance is momentum conservation.

Rotation gives angular momentum. For a particle in 3D with rotation R(ε)·r ≈ r + ε·(n̂ × r), the conserved charge is Q = r × p · n̂ = L · n̂, the component of angular momentum along the rotation axis. Rotational invariance is angular-momentum conservation.

Three of the most fundamental conservation laws in classical mechanics fall out of three lines of algebra applied to three obvious symmetries of empty space.

Worked example: deriving the conserved current

Consider the 2D harmonic oscillator with isotropic Lagrangian:

L = ½m(ẋ² + ẏ²) − ½k(x² + y²)

This is rotationally symmetric: x → x cos ε − y sin ε, y → x sin ε + y cos ε. To first order in ε, δx = −ε y, δy = ε x. So F_x = −y, F_y = x. Verify δL = 0:

δL = m(ẋ·δẋ + ẏ·δẏ) − k(x·δx + y·δy)
   = m(ẋ·(−ẏ) + ẏ·ẋ) − k(x·(−y) + y·x)
   = 0 + 0 = 0  ✓

The action is invariant. The Noether charge is

Q = (∂L/∂ẋ)·F_x + (∂L/∂ẏ)·F_y
  = mẋ·(−y) + mẏ·(x)
  = m(x ẏ − y ẋ)
  = L_z

The angular momentum L_z. Take a numerical case: m = 1 kg, k = 4 N/m (so ω = 2 rad/s), with initial conditions x = 0.5 m, y = 0, ẋ = 0, ẏ = 1.0 m/s. The trajectory is an ellipse, x(t) = 0.5 cos(2t), y(t) = 0.5 sin(2t). At any time t, m(x ẏ − y ẋ) = 1·[0.5cos(2t)·1.0cos(2t) − 0.5sin(2t)·(−1.0sin(2t))] = 0.5 (cos² + sin²) = 0.5 kg·m²/s. Constant for all t, as Noether predicts.

Symmetry-conservation correspondence

Symmetry	Group	Conserved quantity	Domain	Experimental test
Time translation	ℝ	Energy E	Universal	Calorimetry to ppm
Space translation	ℝ³	Momentum p	Universal	Newton's cradle, billiards
Rotation	SO(3)	Angular momentum L	Universal	Gyroscope precession
Galilean boost	ℝ³	Center-of-mass theorem	Non-relativistic	Kinematics of collisions
Lorentz boost	SO(3,1)	Center-of-energy 4-vector	Relativistic	Particle accelerators
U(1) phase rotation	U(1)	Electric charge Q	Electromagnetism	Charge conservation in β decay
SU(2) isospin	SU(2)	Isospin (approximate)	Strong force	Proton/neutron mass near-equality
SU(3) color	SU(3)	Color charge	QCD	Quark confinement, gluon jets
Conformal scale	ℝ⁺	Dilatation current	Massless theories	Approximate at high E in QCD

Each row of this table is a Noether application. The bottom four are gauge symmetries — internal symmetries of the field, not spacetime — and they generate the conservation laws that make the Standard Model self-consistent.

Gauge symmetry and charge conservation

The complex Klein-Gordon Lagrangian L = ∂_μφ* ∂^μ φ − m²|φ|² is invariant under the global U(1) phase rotation φ → e^(iα)φ. Applying Noether's theorem with F = iφ:

j^μ = i (φ* ∂^μ φ − φ ∂^μ φ*)

This is the conserved 4-current: ∂_μj^μ = 0. The integrated charge Q = ∫j⁰ d³x is the electric charge (after multiplying by the coupling e). Gauge invariance — promoting α to be spacetime-dependent α(x) — then requires the introduction of the photon A_μ as the gauge field that compensates the spacetime-dependent phase. The whole machinery of QED falls out of demanding U(1) gauge invariance and applying Noether's first theorem.

Same procedure for SU(2): the symmetry is a 3-parameter group, so there are three Noether currents corresponding to the W±, W³ gauge bosons (which mix with the U(1) hypercharge to produce the photon, W⁺, W⁻, and Z⁰ of the electroweak sector). For SU(3) you get eight currents — the eight gluons of QCD. Every gauge boson in nature is a Noether current dressed up as a gauge field.

Where Noether's theorem shows up

• The 13.8-billion-year stability of fundamental constants. The non-observation of charge non-conservation puts experimental bounds on the U(1) symmetry violation at < 10⁻⁶⁰. The MAJORANA Demonstrator and KamLAND-Zen searches for neutrinoless double-beta decay test lepton-number conservation (an approximate Noether charge) to lifetimes > 10²⁶ years.
• JWST and the cosmological energy budget. The James Webb Space Telescope measures the redshift of light from galaxies up to z ≈ 14. The CMB photons we see were emitted at z ≈ 1100 with peak energy ~kT ≈ 0.3 eV; today they peak at ~2.7 K thermal radiation, energy ~10⁻⁴ eV — they have lost a factor of (1+z) = 1100 in energy. Noether says energy is not globally conserved in an expanding spacetime because time-translation invariance is broken by the FRW metric.
• The muon g−2 experiment at Fermilab. The April 2025 final result of E989 measured the muon's anomalous magnetic moment to 127 ppb. The discrepancy from the Standard Model prediction probes hypothetical broken symmetries (supersymmetry, extra U(1)s) whose Noether currents would manifest as new contributions to (g−2). Each potential new symmetry implies a new conservation law that has to be hunted experimentally.
• The CKM matrix in flavor physics. Approximate SU(3)_flavor symmetry of u, d, s quarks gives an approximate Noether-conserved isospin and strangeness. The deviations from exact conservation (described by the Cabibbo-Kobayashi-Maskawa matrix) explain quark mixing and CP violation. The 2008 Nobel Prize honored Kobayashi and Maskawa for predicting that three generations of quarks were necessary to embed CP violation in the Noether structure of the weak interaction.
• Symplectic numerical integrators. Algorithms used in molecular dynamics, planetary simulations, and accelerator physics (leapfrog, Forest-Ruth, Yoshida-4) are designed to preserve the symplectic 2-form ω = Σ dq ∧ dp, which is the geometric structure underlying Noether's first theorem. They preserve a nearby Hamiltonian and so respect Noether-conservation to machine precision over 10⁹ time steps.

The second Noether theorem

Noether actually proved two theorems in 1918. The first concerns global symmetries — those parametrized by a constant ε. The second concerns local or gauge symmetries, where ε(x) depends on spacetime. For each gauge parameter, Noether's second theorem produces a constraint on the Euler-Lagrange equations themselves: not all field equations are independent.

This is what makes general relativity self-consistent. The Einstein-Hilbert action is invariant under arbitrary diffeomorphisms (4 gauge parameters per spacetime point). Noether's second theorem then says four of the ten Einstein equations are redundant — they follow from the other six via the Bianchi identities ∇_μG^μ^ν = 0. The "missing" four equations are precisely what the diffeomorphism freedom lets you set: they correspond to coordinate choices.

The same structure appears in every gauge theory. In QED the four Maxwell equations are not all independent — the gauge invariance δA_μ = ∂_μα implies that ∂_μj^μ = 0 is automatic, so charge conservation is built into the equations of motion rather than being an additional constraint.

Variants and extensions

• Field-theoretic Noether's theorem. For a Lagrangian density ℒ(φ, ∂_μφ), each continuous symmetry produces a conserved 4-current j^μ with ∂_μj^μ = 0. Integrating j⁰ over space gives the conserved charge. The recipe powers the entire Standard Model.
• Energy-momentum tensor. Spacetime-translation invariance gives the canonical energy-momentum tensor T^μν, the object that sources gravity in general relativity. The symmetric Belinfante-Rosenfeld form is the version that makes ∇_μT^μν = 0 manifestly covariant.
• Ward identities. The quantum-mechanical version of Noether conservation. They constrain n-point correlation functions and govern renormalization. The Ward identity for the U(1) current in QED prevents the photon from picking up a mass.
• Anomalies. When a classical Noether symmetry fails to survive quantization, you get an anomaly — the chiral anomaly, the conformal anomaly, the gravitational anomaly. The chiral anomaly correctly predicts the rate of π⁰ → 2γ decay and is essential for consistency of the Standard Model.
• Approximate and spontaneously broken symmetries. An approximate symmetry (like SU(2) isospin) gives an approximately conserved current. A spontaneously broken continuous symmetry (like the chiral symmetry of QCD) gives massless Goldstone bosons (the pions). Noether's framework still applies; the implications are richer.
• Generalized symmetries. A 21st-century development. Higher-form symmetries act on extended objects (lines, surfaces) instead of points. Their Noether charges are integrals over higher-dimensional submanifolds. They are essential for understanding confinement, topological phases of matter, and aspects of gauge theory dualities.

Common pitfalls

• Confusing "symmetry of the equations" with "symmetry of the action." Noether requires the action to be invariant up to a total derivative, which is a stronger condition. Lots of equations of motion have spurious "symmetries" that do not lift to the action and hence do not produce a Noether charge.
• Forgetting the Lagrangian-multiplier piece for quasi-symmetries. If the Lagrangian transforms as L → L + dΛ/dt under the symmetry (instead of L → L), then the conserved charge is Q = pF − Λ, with the −Λ correction. Galilean boosts in non-relativistic mechanics are the textbook example; Λ is a function of t multiplying the boost velocity.
• Misidentifying H with energy when time symmetry is broken. If L = L(q, q̇, t) has explicit t dependence, then dH/dt = ∂L/∂t ≠ 0; H is not conserved. Energy conservation requires both Lagrangian formulation and explicit time-translation invariance.
• Treating discrete symmetries as Noether symmetries. Parity, time-reversal, charge-conjugation are discrete; they have multiplicatively conserved quantum numbers (P-parity, T-parity, C-parity) but no Noether currents. CPT is a general theorem, not a Noether consequence.
• Forgetting that gauge symmetries do not give "new" conservation laws. Noether's first theorem applied to a gauge symmetry produces a current that is automatically conserved as a consequence of the equations of motion — but the corresponding charge transforms trivially under the gauge symmetry, so it is not a label for inequivalent states. The genuine conservation content of a gauge theory comes from the second Noether theorem (Bianchi identities) plus global subgroups of the gauge group.