No-Nonsense Books

Noether's theorem: annotated proof and how symmetries create conservation laws

By Jakob Schwichtenberg, author of No-Nonsense Classical Mechanics (393 pages), No-Nonsense Quantum Field Theory (642 pages), and Physics from Symmetry (Springer). Page references match the current editions.

Published by Emmy Noether in 1918, Noether's theorem is arguably the single deepest bridge in theoretical physics:

Noether's theorem (plain English): Every continuous symmetry of a system's action $S$ automatically produces a quantity $Q$ that stays constant in time along every physical trajectory—and gives you an explicit formula to compute $Q$.

Before seeing Noether's theorem, conservation of energy, linear momentum, angular momentum, and electric charge look like four separate miracles you have to check case by case. Afterwards, they are four instances of the same three-line calculation:

Continuous symmetryInfinitesimal transformationConserved Noether charge / current
Space translation invariance$\mathbf{x} \to \mathbf{x} + \epsilon\,\mathbf{n}$Linear momentum $\mathbf{p} = m\dot{\mathbf{x}}$ (along $\mathbf{n}$)
Rotation invariance$\mathbf{x} \to \mathbf{x} + \epsilon\,(\mathbf{n} \times \mathbf{x})$Angular momentum $\mathbf{L} = \mathbf{x} \times \mathbf{p}$ (about axis $\mathbf{n}$)
Time translation invariance$t \to t + \epsilon$ (or $\delta q = \epsilon\dot{q}$)Total energy / Hamiltonian $E = \sum_i p_i\dot{q}_i - L$
Global $U(1)$ phase invariance$\phi(x) \to e^{-i\epsilon}\phi(x) \approx (1 - i\epsilon)\phi(x)$Conserved 4-current $j^\mu$ ($\partial_\mu j^\mu = 0$, electric / particle charge)

1. Crucial distinction: active symmetry vs. passive coordinate relabeling

Students frequently trip over a subtle trap right at the start (explained in detail in Chapter 10, p223 and Appendix E, p373 of No-Nonsense Classical Mechanics): if the Euler–Lagrange equation works in any coordinate system, why isn't every coordinate change a symmetry?

  • A passive transformation leaves the ball where it is and moves your ruler 1 meter to the left, adjusting the formula for $V$ to compensate. That is mere bookkeeping—it holds even if there is a brick wall sitting at $x = 0$, where momentum is obviously not conserved when the ball hits the wall!
  • An active transformation keeps your ruler fixed and physically shifts the ball's path $q(t) \to q(t) + \epsilon\,K(q(t))$ while evaluating the same Lagrangian function $L$. If a brick wall sits at $x=0$, shifting the ball changes $V(x)$ and breaks the symmetry; if empty space looks the same everywhere ($V = \text{const}$), $L$ is unchanged and momentum is conserved.

2. Annotated proof of Noether's theorem (Lagrangian formalism)

Let us prove Noether's theorem in its full, practical form—including symmetries (like time translation!) where $L$ changes by a total time derivative $\epsilon\,\frac{dF}{dt}$, which leaves the equations of motion unchanged (Section 10.3, p242 of No-Nonsense Classical Mechanics). Consider a continuous family of active path transformations parameterized by a small number $\epsilon$:

$$q_i(t) \;\longrightarrow\; q_i^\epsilon(t) = q_i(t) + \epsilon\,K_i(q,\dot{q},t) + \mathcal{O}(\epsilon^2),$$

where the functions $K_i$ specify the "direction" of the transformation (its generator). We say this transformation is a continuous symmetry of the action (sometimes called a quasi-symmetry) if, for any path $q(t)$—even off-shell paths that do not satisfy the equations of motion!—the Lagrangian changes at first order in $\epsilon$ by at most a total time derivative $\frac{dF}{dt}$ (with $F = 0$ for the simplest symmetries):

$$\left.\frac{dL(q^\epsilon, \dot{q}^\epsilon, t)}{d\epsilon}\right|_{\epsilon=0} = \frac{dF(q,\dot{q},t)}{dt}.$$

Now watch what happens when we evaluate the left-hand side along an actual physical solution $q(t)$ obeying the Euler–Lagrange equation:

Annotated proof — Noether's theorem in Lagrangian mechanics
$$\frac{dF}{dt} = \left.\frac{dL(q^\epsilon, \dot{q}^\epsilon, t)}{d\epsilon}\right|_{\epsilon=0} = \sum_{i=1}^{N} \left(\frac{\partial L}{\partial q_i}\,K_i + \frac{\partial L}{\partial \dot{q}_i}\,\dot{K}_i\right)$$
Expand $\left.\frac{dL}{d\epsilon}\right|_{\epsilon=0}$ using the multivariable chain rule, with $\left.\frac{\partial q_i^\epsilon}{\partial\epsilon}\right|_0 = K_i$ and $\left.\frac{\partial \dot{q}_i^\epsilon}{\partial\epsilon}\right|_0 = \frac{dK_i}{dt} \equiv \dot{K}_i$.
$$\frac{\partial L}{\partial q_i} = \frac{d}{dt}\!\left(\frac{\partial L}{\partial \dot{q}_i}\right) \qquad (\text{along physical trajectories})$$
On-shell step: Restrict $q(t)$ to a physical trajectory so the Euler–Lagrange equations hold for each $i$.
$$\frac{dF}{dt} = \sum_{i=1}^{N} \left[\frac{d}{dt}\!\left(\frac{\partial L}{\partial \dot{q}_i}\right) K_i + \frac{\partial L}{\partial \dot{q}_i}\,\frac{dK_i}{dt}\right]$$
Replace $\frac{\partial L}{\partial q_i}$ by $\frac{d}{dt}\!\left(\frac{\partial L}{\partial\dot{q}_i}\right)$ in the first term of the sum.
$$\frac{dF}{dt} = \frac{d}{dt}\!\left(\sum_{i=1}^{N} \frac{\partial L}{\partial \dot{q}_i}\,K_i\right)$$
Recognize the bracket as the reverse product rule: $\dot{u}\,v + u\,\dot{v} = \frac{d}{dt}(uv)$ with $u = \frac{\partial L}{\partial\dot{q}_i}$ and $v = K_i$.
$$\frac{dQ}{dt} = 0, \qquad Q \equiv \sum_{i=1}^{N} \frac{\partial L}{\partial \dot{q}_i}\,K_i - F$$
Bring $\frac{dF}{dt}$ to the right-hand side. The quantity $Q$ (the Noether charge) is conserved along every physical motion!

Notice why we needed both assumptions: the symmetry identity $\left.\frac{dL}{d\epsilon}\right|_0 = \frac{dF}{dt}$ holds for all paths without using the equations of motion, while the Euler–Lagrange substitution holds only on physical trajectories. Putting them together collapses the whole sum into one total time derivative!

3. Three classic conservation laws in 3 lines each

Example A: Spatial translation $\longrightarrow$ Linear momentum ($F = 0$)

Shift a particle (or an entire closed $N$-body system with interaction potential $V(\mathbf{x}_a - \mathbf{x}_b)$) along a unit vector $\mathbf{n}$: $\mathbf{x} \to \mathbf{x} + \epsilon\,\mathbf{n}$, so $K_i = n_i$ and $\dot{K}_i = 0$. If $L$ does not change ($F = 0$), Noether's formula immediately gives

$$Q = \sum_{i=1}^{3} \frac{\partial L}{\partial \dot{x}_i}\,n_i = \mathbf{p}\cdot\mathbf{n} = \text{const.}$$

Since $\mathbf{n}$ is any direction in which space is homogeneous, the corresponding component of linear momentum $\mathbf{p}$ is conserved.

Example B: Spatial rotation $\longrightarrow$ Angular momentum ($F = 0$)

Rotate $\mathbf{x}$ by an infinitesimal angle $\epsilon$ around unit axis $\mathbf{n}$: $\mathbf{x}^\epsilon = \mathbf{x} + \epsilon\,(\mathbf{n}\times\mathbf{x})$, so $\mathbf{K} = \mathbf{n}\times\mathbf{x}$. For any rotationally invariant Lagrangian ($T = \frac{1}{2}m\dot{\mathbf{x}}^2$ and $V = V(|\mathbf{x}|)$, so $F = 0$), the cyclic triple-product identity $\mathbf{p}\cdot(\mathbf{n}\times\mathbf{x}) = \mathbf{n}\cdot(\mathbf{x}\times\mathbf{p})$ gives

$$Q = \mathbf{p}\cdot(\mathbf{n}\times\mathbf{x}) = \mathbf{n}\cdot(\mathbf{x}\times\mathbf{p}) = \mathbf{n}\cdot\mathbf{L} = \text{const.}$$

Example C: Time translation $\longrightarrow$ Total energy ($F = L$)

What if $\frac{\partial L}{\partial t} = 0$ (the laws of the system do not change from Tuesday to Wednesday)? Shifting paths in time, $q_i^\epsilon(t) \equiv q_i(t + \epsilon) = q_i(t) + \epsilon\,\dot{q}_i(t) + \mathcal{O}(\epsilon^2)$, has generator $K_i = \dot{q}_i$. And when $\partial L/\partial t = 0$, $L(q(t+\epsilon), \dot{q}(t+\epsilon)) = L(t+\epsilon) = L(t) + \epsilon\,\frac{dL}{dt} + \mathcal{O}(\epsilon^2)$, so $F = L$ (non-zero!). Plugging $K_i = \dot{q}_i$ and $F = L$ into Noether's master formula yields:

$$Q = \sum_{i=1}^{N} \frac{\partial L}{\partial \dot{q}_i}\,\dot{q}_i - L = \sum_{i=1}^{N} p_i\dot{q}_i - L \equiv H = \text{const},$$

which is the Hamiltonian (total energy $T + V$ for standard natural systems)!

4. One-line bonus: the Hamiltonian / Poisson-bracket view

In Hamiltonian mechanics (Section 10.2, p232 of No-Nonsense Classical Mechanics), Noether's theorem is even more symmetrical—it becomes a two-way street! Any phase-space observable $G(q,p)$ (with $\partial G/\partial t = 0$) evolves via the Poisson bracket:

$$\frac{dG}{dt} = \{G, H\} = -\{H, G\}.$$

Read that equation both ways: $G$ is conserved ($\frac{dG}{dt} = 0$) if and only if $\{G, H\} = 0$, which is true if and only if $\{H, G\} = 0$—meaning $H$ is invariant under the infinitesimal canonical transformation generated by $G$! Passing to quantum mechanics replaces $\{G, H\} = 0$ by the commutator $[\hat{G}, \hat{H}] = 0$.

5. Noether's theorem in field theory: conserved currents $j^\mu$

In electrodynamics and quantum field theory (Section 4.4, p155 of No-Nonsense Quantum Field Theory), varying fields $\phi_a(x) \to \phi_a(x) + \epsilon\,K_a(\phi)$ under an internal symmetry with $\delta\mathcal{L} = \epsilon\,\partial_\mu F^\mu$ promotes $\frac{dQ}{dt} = 0$ to the local continuity equation:

$$\partial_\mu j^\mu = \frac{\partial j^0}{\partial t} + \nabla\cdot\mathbf{j} = 0, \qquad j^\mu \equiv \sum_a \frac{\partial\mathcal{L}}{\partial(\partial_\mu\phi_a)}\,K_a - F^\mu.$$

For a complex scalar field $\mathcal{L} = (\partial_\mu\phi^*)(\partial^\mu\phi) - m^2\phi^*\phi$, global $U(1)$ phase rotation $\phi \to e^{-i\epsilon}\phi$ ($K_\phi = -i\phi$, $K_{\phi^*} = +i\phi^*$, $F^\mu = 0$) immediately produces the conserved electromagnetic / particle-number current $j^\mu = i\big(\phi^*\partial^\mu\phi - (\partial^\mu\phi^*)\phi\big)$, whose spatial integral $Q = \int d^3x\,j^0$ is the conserved charge!

Common questions

Why doesn't parity ($x \to -x$) give a conserved Noether quantity in classical mechanics?

Because you cannot reflect a right-handed glove continuously into a left-handed glove by dialing a small real parameter $\epsilon$ smoothly away from $0$—there is no derivative $\left.\frac{d}{d\epsilon}\right|_{\epsilon=0}$ to take! In quantum mechanics, however, the unitary parity operator $\hat{P}$ still commutes with a reflection-symmetric $\hat{H}$ ($[\hat{P},\hat{H}]=0$), so its eigenvalues $\pm 1$ are conserved.

What about local gauge symmetries?

Promoting a global $U(1)$ or $SU(N)$ symmetry to a local transformation $\epsilon = \epsilon(x)$ requires introducing gauge fields like the electromagnetic four-potential $A_\mu$ in Maxwell's equations (Noether's second theorem). Local gauge symmetry expresses a redundancy in our mathematical description whose consistency enforces strict conservation of the coupled electric/color current.

Which book should I read to see symmetry drive all of physics?

For classical mechanics and both proofs (Hamiltonian + Lagrangian) with annotated steps, start with No-Nonsense Classical Mechanics. To see Lorentz symmetry and Lie group theory derive relativistic quantum mechanics, electrodynamics, and QFT from the ground up, read my Springer textbook Physics from Symmetry alongside No-Nonsense Quantum Field Theory.

CM Table of Contents (PDF) · Related concept guides: Lagrangian mechanics · Euler–Lagrange equation · Maxwell's equations.