Noether's Theorem and Conservation Laws

Every conservation law you have used — energy, linear momentum, angular momentum — has been derived so far as a lucky feature of a particular Lagrangian: a coordinate happened to be missing, or $L$ happened to have no explicit time. Noether's theorem, published by Emmy Noether in 1918, replaces luck with a principle: to every continuous symmetry of the action there corresponds a conserved quantity, and the correspondence is explicit — the symmetry's infinitesimal generator, contracted with the canonical momenta, is the constant of the motion.

This chapter states and proves the theorem for point transformations, then works through the three classical cases (time translation ↔ energy, space translation ↔ momentum, rotation ↔ angular momentum) as corollaries of the single result. It then covers the two subtleties that trip people up: symmetries that change $L$ by a total time derivative (still symmetries, "quasi-symmetries"), and symmetries that are not geometrically obvious (the Laplace–Runge–Lenz vector of the Kepler problem). The converse — a conserved quantity generates a symmetry — is the doorway to the Hamiltonian mechanics of the next chapters.

How to read this chapter. Learn the master formula (Section 3) and then treat Sections 4–6 as substitutions into it. The proof is short and worth doing once. The payoff is conceptual: after this chapter you never look for a conserved quantity — you look for a symmetry, and the conserved quantity is then automatic. This reframing is the organising idea of all of modern physics.

1. Symmetry, Invariance and the Action

A symmetry of a mechanical system is a transformation of the coordinates (and possibly the time) that maps solutions of the equations of motion to other solutions. The cleanest way to detect one is through the action

A transformation is a symmetry of the action if it leaves $S$ unchanged (or changes it only by boundary terms). Because the equations of motion are exactly the condition $\delta S = 0$, any transformation that preserves $S$ automatically preserves the set of solutions.

We restrict to continuous symmetries: families of transformations $q_j \to Q_j(\vec q,t;\varepsilon)$ labelled by a real parameter $\varepsilon$, with $Q_j(\vec q,t;0) = q_j$. Discrete symmetries (parity, time reversal) are important but do not give conservation laws through Noether's theorem — the theorem needs the parameter to be turned on infinitesimally.

2. Infinitesimal Generators

Expand a continuous transformation to first order in $\varepsilon$:

The functions $\tau$ and $\eta_j$ are the generators of the transformation — they encode "which way" the transformation pushes the time and the coordinates. Examples:

• Time translation: $\tau = 1$, $\eta_j = 0$ (shift the clock).
• Translation along $\hat x$: $\tau = 0$, $\eta_j = \delta_{jx}$ (move every particle by the same amount).
• Rotation about $\hat z$: $\tau = 0$, $\eta_x = -y$, $\eta_y = x$ (infinitesimal rotation).
• Galilean boost along $\hat x$: $\tau = 0$, $\eta_j = t\,\delta_{jx}$ (add a uniform velocity).

3. Noether's Theorem

Statement. If the action is invariant under the one-parameter family generated by $(\tau,\eta_j)$ — meaning $\delta S = 0$ to first order in $\varepsilon$ for every path, not only solutions — then the quantity

is constant along every solution of the equations of motion. The bracketed combination is the Jacobi energy function $h = \sum_j p_j\dot q_j - L$, so $I = \sum_j p_j\,\eta_j - h\,\tau$.

Proof. The total first-order change in the action under the transformation is

where $\bar\eta_j = \eta_j - \dot q_j\,\tau$ is the change in $q_j$ at fixed $t$. On a solution the first bracket (the Euler–Lagrange expression) vanishes. Invariance means $\delta S = 0$ for the transformation, so the remaining total derivative integrates to zero for arbitrary endpoints, which forces its integrand to vanish:

Expanding $\bar\eta_j = \eta_j - \dot q_j\tau$ gives $\dfrac{dI}{dt} = 0$ with $I$ as boxed. $\blacksquare$

4. Time-Translation Symmetry ↔ Energy

Take $\tau = 1$, $\eta_j = 0$. The action is invariant under $t \to t + \varepsilon$ exactly when $L$ has no explicit time dependence, $\partial L/\partial t = 0$. Noether's quantity is

So $h$ is conserved. For a scleronomic system with $T$ quadratic in the velocities, Euler's theorem on homogeneous functions gives $\sum_j (\partial T/\partial\dot q_j)\dot q_j = 2T$, hence $h = 2T - (T - V) = T + V = E$: time-translation invariance is conservation of energy. For a rheonomic system $h = T_2 - T_0 + V$ is still conserved but is not the mechanical energy.

5. Space-Translation Symmetry ↔ Linear Momentum

Take $\tau = 0$ and $\eta_j$ = the components of a fixed unit vector $\hat n$ applied identically to every particle: $\delta\vec r_i = \varepsilon\,\hat n$ for all $i$. The kinetic energy is unchanged (it depends on velocities), so the action is invariant iff $V$ is unchanged by a rigid shift, i.e. iff $V$ depends only on relative positions $\vec r_i - \vec r_j$. Noether's quantity is

Since $\hat n$ is arbitrary, the total linear momentum $\vec P$ is conserved. Translational invariance of the potential is Newton's third law in disguise: internal forces come in equal-and-opposite pairs precisely because $V$ cannot tell where the whole system sits.

6. Rotational Symmetry ↔ Angular Momentum

Take $\tau = 0$ and an infinitesimal rotation by angle $\varepsilon$ about the axis $\hat n$: $\delta\vec r_i = \varepsilon\,\hat n\times\vec r_i$. Again $T$ is invariant (rotations preserve $|\dot{\vec r}_i|$), so the action is invariant iff $V$ is unchanged by a rigid rotation — iff $V$ depends only on distances $|\vec r_i - \vec r_j|$ and $|\vec r_i|$. Noether's quantity is

For arbitrary $\hat n$, the total angular momentum $\vec L$ is conserved. If $V$ is invariant only under rotations about one particular axis (a system in an axial external field), only that component of $\vec L$ is conserved — exactly the cyclic-azimuth result of the last chapter, now seen as a symmetry statement.

7. Quasi-Symmetries and the Divergence Term

Some transformations change the Lagrangian by a total time derivative,

which changes the action only by the boundary term $\varepsilon[F]_{t_1}^{t_2}$ and so is still a symmetry (a quasi-symmetry). The conserved quantity picks up a compensating term:

Worked example — Galilean boost. For a free particle, $L = \tfrac12 m\dot x^2$. Under $x \to x + \varepsilon t$ (a boost), $\dot x \to \dot x + \varepsilon$, so $L \to \tfrac12 m(\dot x + \varepsilon)^2 = L + \varepsilon m\dot x + O(\varepsilon^2) = L + \varepsilon\,\dfrac{d}{dt}(m x) + \cdots$. Thus $F = m x$, $\eta = t$, $\tau = 0$, and Noether's quantity is $I = m\dot x\cdot t - m x = m(\dot x\,t - x) = -m x_0$: the initial position, or equivalently the statement that $x(t) - \dot x\,t$ is constant. The boost symmetry conserves the centre-of-mass "initial position", the fourth Galilean conservation law.

8. Hidden Symmetries: the Laplace–Runge–Lenz Vector

Not every conserved quantity comes from a geometrically obvious symmetry. The Kepler problem $V = -k/r$ conserves, in addition to $\vec L$ and $E$, the Laplace–Runge–Lenz vector

which points along the major axis of the ellipse and has magnitude $mke$. Its conservation is why Kepler orbits do not precess. The corresponding symmetry is a rotation not in ordinary space but in a four-dimensional space (the $SO(4)$ symmetry of the bound Kepler problem), invisible in the configuration-space picture but perfectly real. The lesson: a system can have more conserved quantities than manifest symmetries, and finding them ("superintegrability") is a live research technique.

9. The Converse and the Road to Hamiltonian Mechanics

Noether's theorem has a converse: a conserved quantity generates a symmetry. Given a constant of the motion $G(\vec q,\vec p)$, the flow it generates through the Poisson bracket, $\delta q_j = \varepsilon\{q_j,G\}$, $\delta p_j = \varepsilon\{p_j,G\}$, is a symmetry of the dynamics. This two-way street — symmetries $\leftrightarrow$ conserved quantities — is cleanest in the Hamiltonian formalism, where both are elements of the same algebraic structure (the Poisson-bracket algebra). Momentum generates translations, angular momentum generates rotations, and the Hamiltonian generates time evolution itself. Chapters 5 and 6 build that machinery.

Common Pitfalls

• Requiring $\delta S = 0$ only on solutions. Noether's theorem needs the action to be invariant for every path, at least to first order; invariance merely on solutions is automatic and yields nothing.
• Forgetting the $F$ term for a quasi-symmetry. If the transformation changes $L$ by $dF/dt$, the conserved quantity is $\sum_j p_j\eta_j - h\tau - F$; dropping $F$ gives a quantity that is not conserved (the Galilean boost is the standard trap).
• Confusing $h$ with $E$. Time-translation symmetry conserves the Jacobi function $h$; it equals the mechanical energy only for scleronomic systems with $T$ quadratic in the velocities.
• Thinking a broken symmetry conserves nothing. If $V$ is invariant under rotation about one axis only, that one component of $\vec L$ is still conserved.
• Assuming every conserved quantity has a visible symmetry. The Runge–Lenz vector and other "hidden" (dynamical) symmetries act in enlarged spaces; absence of an obvious geometric symmetry does not mean absence of a conservation law.
• Applying the theorem to discrete symmetries. Parity and time reversal give selection rules, not Noether charges; the parameter $\varepsilon$ must be continuous.

Summary

• Noether's theorem: every continuous symmetry of the action — generated by $(\tau,\eta_j)$ with $t\to t+\varepsilon\tau$, $q_j\to q_j+\varepsilon\eta_j$ — yields the conserved quantity $I = \sum_j p_j\,\eta_j - h\,\tau$, with $p_j = \partial L/\partial\dot q_j$ and $h = \sum_j p_j\dot q_j - L$.
• Time translation ($\tau=1$, $\eta_j=0$), valid when $\partial L/\partial t = 0$: conserves $h$; for scleronomic $T$-quadratic systems $h = E$.
• Space translation ($\eta_i = \hat n$), valid when $V$ depends only on relative positions: conserves total linear momentum $\vec P$.
• Rotation ($\delta\vec r_i = \hat n\times\vec r_i$), valid when $V$ depends only on distances: conserves total angular momentum $\vec L$ (or one component under an axial symmetry).
• Quasi-symmetries change $L$ by $dF/dt$; the conserved quantity becomes $\sum_j p_j\eta_j - h\tau - F$. The Galilean boost is the canonical example, conserving the centre-of-mass initial position.
• Hidden symmetries (e.g. the Laplace–Runge–Lenz vector for Kepler, from $SO(4)$) give conservation laws with no configuration-space symmetry. The converse — conserved quantity generates symmetry via the Poisson bracket — motivates the Hamiltonian formalism.