Momentum and Angular Momentum as Generators

Ask a beginning student what momentum is and you hear "mass times velocity" — a formula, tied to one kind of particle in one kind of coordinate. Ask a working physicist and you get a stranger, sturdier answer: momentum is the thing that generates translations. Slide every particle of a system sideways by the same small amount and ask which phase-space function drove the slide through the canonical machinery — the answer is the total momentum, always, whatever the system is made of. Angular momentum earns its name the same way: it is whatever generates rotations, and the familiar $x p_y - y p_x$ is just that abstract job description evaluated in Cartesian coordinates. This page takes the generator machinery of Infinitesimal Canonical Transformations and runs it on the two workhorses. The payoffs are concrete. The translation flow of $p$ terminates after one term — finite translations are as exact as infinitesimal ones. The rotation flow of $L_z$ turns out to rotate both the position vector and the momentum vector, a fact with teeth: forget the momentum's co-rotation and rotational invariance of kinetic energy silently breaks. The brackets of the generators among themselves reproduce the geometry of the operations: translations commute, rotations about different axes do not, and $\{L_x,L_y\}=L_z$ is the algebraic residue of the book-flipping experiment every student has done. Most consequentially, "momentum generates translations" is the definition that survives. When quantum mechanics arrives, $\hat p=-i\hbar\,\partial_x$ is not an analogy but the same sentence — and this page is where that sentence is first spoken precisely.

Key takeaways

The flow generated by $G=p_j$ moves $q_j$ at unit rate and touches nothing else: $\{q_i,p_j\}=\delta_{ij}$ is the statement "momentum translates its own coordinate." The flow is exactly linear — finite translations come from one bracket, no series needed.
The flow generated by $G=L_z=xp_y-yp_x$ rotates the $x$ – $y$ components of both $\mathbf r$ and $\mathbf p$ rigidly through the flow angle; $|\mathbf r|$ , $|\mathbf p|$ , $\mathbf r\cdot\mathbf p$ and $L_z$ itself ride along unchanged.
Any triple $v_i$ obeying $\{L_i,v_j\}=\epsilon_{ijk}v_k$ rotates as a vector — position and momentum both qualify, which is what "vector" means in mechanics.
For $N$ particles, total momentum $\mathbf P=\sum_a\mathbf p_a$ generates the simultaneous equal translation of every particle (separations untouched); total $L_z$ rotates the whole configuration rigidly.
Translations commute, $\{p_x,p_y\}=0$ ; rotations do not, $\{L_x,L_y\}=L_z\neq0$ — the bracket of generators encodes the geometry of composing the operations.
Quantum mechanics keeps the job descriptions verbatim: $\hat p=-i\hbar\,\partial_x$ generates translations, $\hat L_z$ generates rotations, with brackets promoted to commutators.
Key idea: momentum and angular momentum are not defined by formulas but by jobs — they are the canonical generators of translations and rotations, and everything about them follows from those jobs.

The core insight

Momentum is the generator of translations and angular momentum the generator of rotations: $\delta f=\epsilon\{f,p_j\}$ shifts the $j$ -th coordinate, $\delta f=\theta\{f,L_z\}$ rotates both position and momentum — the definitions that survive unchanged into quantum mechanics.

Where this lives today

The generator viewpoint is daily engineering language in three fields. Robotics runs on it: the modern screw-theoretic formulation of rigid-body motion (the standard graduate textbooks and the ROS motion-planning stack alike) represents every joint's freedom as an element of the Lie algebra $\mathfrak{se}(3)$ — a linear-momentum-like generator for prismatic joints, an angular-momentum-like generator for revolute ones — and composes finite motions by exponentiating, exactly this article's flow construction with hardware attached. Condensed-matter physics leans on the crystal version: in a lattice, continuous translation symmetry breaks to a discrete subgroup, and the generator's eigenvalue survives as quasi-momentum — conserved only modulo reciprocal-lattice vectors, which is why electrons Bragg-scatter and why "momentum space" in an ARPES measurement is a Brillouin zone rather than a plane. And machine-learned interatomic potentials now build the bracket patterns in by construction: equivariant network architectures constrain their internal features to transform under rotations exactly as equation (5) dictates for vectors (and its higher-rank cousins for tensors), so that predicted forces rotate with the molecule automatically — Weyl's definition of a vector, enforced layer by layer in silicon.

The physical picture

Every phase-space function is an engine, as the flow-meter picture of Poisson Brackets: Definition established: feed $G$ into the canonical machinery and each point acquires the velocity $(\partial G/\partial p,\,-\partial G/\partial q)$ per unit flow parameter. This page simply asks: what do the engines labeled $p$ and $L_z$ do? The momentum engine is almost embarrassingly simple. $G=p_x$ has $\partial G/\partial p_x=1$ and every other derivative zero, so the recipe reads: advance $x$ at unit rate, leave everything else alone. Run it for a finite parameter $a$ and every configuration slides rigidly by $a$ along $x$ — the momenta, which encode how the system is moving, are untouched, because sliding a snapshot sideways changes where things are, not how they move. Conjugacy is exclusive: $p_x$ translates $x$ and only $x$ ; it does not stir $y$ , $z$ , or any momentum. The fundamental table $\{q_i,p_j\}=\delta_{ij}$ (see Fundamental Poisson Brackets) is precisely this exclusivity written as algebra.

Figure

The two workhorse flows — momentum slides every particle equally leaving momenta fixed, while angular momentum rotates positions and momenta together

The angular-momentum engine is subtler in exactly one way. $G=L_z=xp_y-yp_x$ depends on positions and momenta, so its flow moves both: the position vector turns through the flow angle $\theta$ , and the momentum vector turns with it, through the same angle. That co-rotation is not a bonus — it is what "rotating the system" means. Rotate a snapshot of the solar system and you must rotate every velocity arrow too, or you have not rotated the system, you have vandalized it. The lengths $|\mathbf r|$ , $|\mathbf p|$ , the angle between them ( $\mathbf r\cdot\mathbf p$ ), and $L_z$ itself all commute with $L_z$ under the bracket, so the rotation flow carries them unchanged — the full inventory of what a rotation should preserve, delivered automatically by the algebra. The third picture is about composing engines. Run the $p_x$ engine then the $p_y$ engine, or in the opposite order: same final state, because $\{p_x,p_y\}=0$ . Run a rotation and a translation in the two orders and the final configurations differ — rotate-then-slide is not slide-then-rotate, as anyone parallel-parking knows. The bracket of the generators is precisely the residue of swapping the order (the general theory of that statement lives in Generators of Continuous Transformations); this page's Worked Example 2 measures the residue with actual numbers.

Momentum: the translation engine

The flow equations for $G=p_j$ come straight from the bracket recipe $\mathrm{d}f/\mathrm{d}\lambda=\{f,p_j\}$ . Evaluated on the coordinates themselves via the fundamental table:

\frac{\mathrm{d}q_i}{\mathrm{d}\lambda}=\{q_i,p_j\}=\delta_{ij},\qquad \frac{\mathrm{d}p_i}{\mathrm{d}\lambda}=\{p_i,p_j\}=0 \qquad (1)

where: $\lambda$ — the flow parameter, here carrying units of length. One coordinate advances at unit rate; every other variable is inert. Because the right-hand sides are constants, the flow integrates exactly in one line — no Lie series, no approximation:

q_j(\lambda)=q_j+\lambda,\qquad \text{all other } q_i,\,p_i \text{ unchanged} \qquad (2)

where: the finite translation is identical in form to the infinitesimal one. In the Lie-series language of Poisson-Bracket Formulation of Dynamics, the series terminates after its first term: $\{q_j,p_j\}=1$ and the bracket of 1 with anything vanishes. Translations are the rare flow you never need to exponentiate carefully. For a system of $N$ particles the same logic runs on the total momentum, $P_x=\sum_a p_{a,x}$ : each particle's $x_a$ has unit bracket with its own $p_{a,x}$ and zero with the others, so the flow advances every $x_a$ at the same unit rate — a rigid slide of the entire configuration, with all momenta and all separations $\mathbf r_a-\mathbf r_b$ untouched (Derivation A). This is the precise sense in which "total momentum" is the system's collective translation handle, and it is why $\mathbf P$ pairs with the center of mass: $X_{\mathrm{cm}}$ is, up to the total mass, the conjugate coordinate that $P_x$ translates.

Derivation AThe translation flow of momentum, one particle and many

We are deriving the finite translation flow (2) generated by $G=p_j$ , and its $N$ -particle extension generated by total momentum — including the claim, stronger than for generic generators, that the finite flow is exactly the infinitesimal formula with $\epsilon\to\lambda$ . The strategy is to write the flow ODEs from the bracket recipe, observe that their right-hand sides are constants, and integrate trivially; the N-particle case then reduces to bookkeeping over which brackets vanish. The payoff sentence: conjugacy is exclusive and exact — each momentum owns exactly one coordinate.

Step 1. Single particle, $G=p_j$ . The flow ODEs are $\mathrm{d}q_i/\mathrm{d}\lambda=\{q_i,p_j\}$ and $\mathrm{d}p_i/\mathrm{d}\lambda=\{p_i,p_j\}$ . The fundamental table gives $\{q_i,p_j\}=\delta_{ij}$ and $\{p_i,p_j\}=0$ .

Step 2. Integrate: only $q_j$ moves, and it moves at constant unit rate, so $q_j(\lambda)=q_j(0)+\lambda$ exactly. Why the series terminates: the Lie series for the flow is $q_j(\lambda)=q_j+\lambda\{q_j,p_j\}+\tfrac{\lambda^2}{2}\{\{q_j,p_j\},p_j\}+\cdots$ . The first bracket is the constant 1, and every bracket of a constant vanishes — all higher terms are identically zero. Finite translations are infinitesimal translations scaled up, with no correction terms, ever.

Step 3. $N$ particles, $G=P_x=\sum_b p_{b,x}$ . For particle $a$ : $\mathrm{d}x_a/\mathrm{d}\lambda=\{x_a,\sum_b p_{b,x}\}=\sum_b\delta_{ab}=1$ — each coordinate hears only its own momentum in the sum, but every particle's coordinate finds its partner there. All momenta and all $y,z$ coordinates have vanishing bracket with $P_x$ .

Step 4. Integrate: $x_a(\lambda)=x_a(0)+\lambda$ for every $a$ simultaneously; separations transform as $\mathbf r_a-\mathbf r_b\to\mathbf r_a-\mathbf r_b$ — the slide is rigid. Internal structure, relative motion, and all momenta are spectators. Verified: the transformation is canonical by the bracket test — the Jacobian is the identity plus a constant shift, so $\{Q_i,P_j\}=\delta_{ij}$ trivially; matches Goldstein §9.4's translation ICT; Worked Example 2 exhibits the rigid slide numerically with separations preserved.

Angular momentum: the rotation engine

Now feed $G=L_z=xp_y-yp_x$ to the machinery, with flow parameter $\theta$ . Four brackets, each a two-line evaluation from the definition (done honestly in Derivation B), give the flow ODEs:

\frac{\mathrm{d}x}{\mathrm{d}\theta}=-y,\quad \frac{\mathrm{d}y}{\mathrm{d}\theta}=x,\quad \frac{\mathrm{d}p_x}{\mathrm{d}\theta}=-p_y,\quad \frac{\mathrm{d}p_y}{\mathrm{d}\theta}=p_x \qquad (3)

where: the first pair says the position vector turns counterclockwise at unit angular rate; the second pair says the momentum vector obeys the identical equations — it turns with the positions, through the same angle. Solving the paired ODEs (Derivation B) gives the finite flow as one rotation matrix acting twice:

\begin{pmatrix}x(\theta)\\y(\theta)\end{pmatrix}=R(\theta)\begin{pmatrix}x\\y\end{pmatrix},\qquad \begin{pmatrix}p_x(\theta)\\p_y(\theta)\end{pmatrix}=R(\theta)\begin{pmatrix}p_x\\p_y\end{pmatrix},\qquad R(\theta)=\begin{pmatrix}\cos\theta&-\sin\theta\\ \sin\theta&\cos\theta\end{pmatrix} \qquad (4)

where: $R(\theta)$ — the standard active rotation. Everything a rotation should preserve is preserved because it bracket-commutes with the generator: $\{x^2+y^2,L_z\}=0$ , $\{p_x^2+p_y^2,L_z\}=0$ , $\{xp_x+yp_y,L_z\}=0$ , and $\{L_z,L_z\}=0$ — lengths, the position–momentum angle, and the generator itself all ride the flow untouched. The pattern in (3) is worth naming. A triple of phase-space functions $v_j$ that responds to the rotation flow by turning like an arrow satisfies

\{L_i,v_j\}=\epsilon_{ijk}\,v_k \qquad (5)

where: $\epsilon_{ijk}$ — the alternating symbol. This is the definition of a vector in Hamiltonian mechanics: not "three numbers" but "three functions with this bracket pattern against the rotation generators." Position and momentum both satisfy it (Derivation C checks a component of each); so does $\mathbf L$ itself, and setting $v=L$ in (5) reproduces the celebrated closure $\{L_i,L_j\}=\epsilon_{ijk}L_k$ — derived from the fundamentals in the worked examples of Algebra of Poisson Brackets, and reappearing here with its geometric meaning on display: rotations about different axes fail to commute, and the failure is itself a rotation. The same wiring scales up to $N$ particles: total $L_z=\sum_a(x_ap_{a,y}-y_ap_{a,x})$ generates the rigid rotation of the entire configuration — every position and every momentum vector turns by the same $\theta$ about the same axis. And the two engines interlock: $\{L_z,P_x\}=P_y$ , $\{L_z,P_y\}=-P_x$ , so total momentum itself transforms as a vector under the total rotation flow — translations and rotations knit into the Euclidean group, with the brackets as its multiplication table.

Figure

One flow angle, two turning arrows — the L flow rotates position and momentum by the same angle, preserving lengths, the angle between them, and L itself

Go deeper · Derivation B · 1 minThe rotation flow of angular momentum

Derivation BThe rotation flow of angular momentum

We are deriving the flow equations (3) from the bare bracket definition — verifying each of the four signs by explicit partial derivatives, since sign errors here are the single most common failure in student work — and then solving the ODE system to the finite rotation (4). The strategy: evaluate $\{f,L_z\}$ for $f=x,y,p_x,p_y$ using $L_z=xp_y-yp_x$ , then decouple the paired equations by differentiating once more. Watch how the momentum pair's equations come out identical to the position pair's: that identity of form is the co-rotation fact this article keeps insisting on.

Step 1. Brackets of the positions. With one particle in the plane, $\{f,g\}=\sum_{k\in\{x,y\}}(\partial_{q_k}f\,\partial_{p_k}g-\partial_{p_k}f\,\partial_{q_k}g)$ :

\{x,L_z\}=\frac{\partial x}{\partial x}\frac{\partial L_z}{\partial p_x}-0=\frac{\partial L_z}{\partial p_x}=-y,\qquad \{y,L_z\}=\frac{\partial L_z}{\partial p_y}=x

Step 2. Brackets of the momenta — only the second half of the definition survives, since momenta have no position dependence:

\{p_x,L_z\}=-\frac{\partial L_z}{\partial x}=-p_y,\qquad \{p_y,L_z\}=-\frac{\partial L_z}{\partial y}=-(-p_x)=p_x

Why the two pairs match: $L_z$ is antisymmetric under the simultaneous swap $(x,y)\leftrightarrow(p_x,p_y)$ up to relabeling, so the machinery pushes momenta exactly as it pushes positions. Structurally: both $\mathbf r$ and $\mathbf p$ satisfy the vector pattern (5).

Step 3. Solve the position pair. Differentiate $\mathrm{d}x/\mathrm{d}\theta=-y$ once more: $\mathrm{d}^2x/\mathrm{d}\theta^2=-\mathrm{d}y/\mathrm{d}\theta=-x$ — simple harmonic in the flow parameter. With initial data $(x,y)$ at $\theta=0$ : $x(\theta)=x\cos\theta-y\sin\theta$ , $y(\theta)=x\sin\theta+y\cos\theta$ , which is $R(\theta)$ acting on $(x,y)$ . The identical system for $(p_x,p_y)$ gives $R(\theta)$ acting on the momentum.

Step 4. Invariants. By Leibniz (the algebraic rules live in Algebra of Poisson Brackets): $\{x^2+y^2,L_z\}=2x\{x,L_z\}+2y\{y,L_z\}=2x(-y)+2y(x)=0$ ; identically $\{p_x^2+p_y^2,L_z\}=2p_x(-p_y)+2p_y(p_x)=0$ ; and $\{xp_x+yp_y,L_z\}=x(-p_y)+p_x(-y)\cdot 0\ldots$ evaluated by Leibniz termwise: $x\{p_x,L_z\}+p_x\{x,L_z\}+y\{p_y,L_z\}+p_y\{y,L_z\}=-xp_y-yp_x+yp_x+xp_y=0$ . Lengths and the mutual angle are flow constants, and $\{L_z,L_z\}=0$ closes the inventory. Verified: at $\theta=\pi/2$ the flow sends $(x,y)\to(-y,x)$ and $(p_x,p_y)\to(-p_y,p_x)$ — Worked Example 1 confirms numerically, with $L_z=-1.10$ , $|\mathbf r|^2=5.0$ , $|\mathbf p|^2=0.34$ , $\mathbf r\cdot\mathbf p=0.70$ all reproduced exactly after the quarter turn; the matrix in (4) is the standard active rotation, matching Goldstein §9.4.

Go deeper · Derivation C · 1 minMomentum transforms as a vector under the rotation flow

Derivation CMomentum transforms as a vector under the rotation flow

We are establishing the pattern (5) for the momentum triple — concretely, deriving $\{L_z,p_x\}=p_y$ and its cyclic partners — which certifies that $\mathbf p$ is a vector in the bracket-theoretic sense: the rotation flow turns it as an arrow. The strategy is a single honest evaluation from the definition, then an appeal to the antisymmetric structure to fill the rest of the table. This derivation is deliberately short; its value is the pattern, which organizes every rotation computation you will ever do, classical or quantum.

Step 1. Evaluate $\{L_z,p_x\}$ from the definition. Only position-derivatives of $L_z$ can pair with the momentum-derivative of $p_x$ (which is a constant in momentum): $\{L_z,p_x\}=\sum_k \partial_{q_k}L_z\,\partial_{p_k}p_x-\partial_{p_k}L_z\,\partial_{q_k}p_x$ . The second product dies ( $p_x$ has no position dependence); in the first, $\partial_{p_k}p_x=\delta_{kx}$ picks out $k=x$ :

\{L_z,p_x\}=\frac{\partial L_z}{\partial x}=\frac{\partial}{\partial x}(xp_y-yp_x)=p_y

Step 2. Same computation for the other component: $\{L_z,p_y\}=\partial L_z/\partial y=-p_x$ . And trivially $\{L_z,p_z\}=0$ — rotation about $z$ leaves the $z$ -component of any vector alone.

Step 3. Assemble: the three results are exactly $\{L_z,p_j\}=\epsilon_{zjk}p_k$ . Cyclic relabeling of axes (nothing in the computation distinguished $z$ ) gives the full pattern $\{L_i,p_j\}=\epsilon_{ijk}p_k$ , and the identical computation with $r$ in place of $p$ gives $\{L_i,x_j\}=\epsilon_{ijk}x_k$ . Why this is the right definition of "vector": the flow change of $p_j$ under rotation about axis $i$ is $\delta p_j=\theta\{p_j,L_i\}=-\theta\epsilon_{ijk}p_k$ , which is the component form of $\delta\mathbf p=\theta\,\hat{\mathbf n}\times\mathbf p$ — the infinitesimal rotation of an arrow. Any triple obeying the pattern rotates as an arrow; any failing it does not, whatever its indices look like.

Step 4. Interlock with translations: $\{L_z,P_x\}=P_y$ holds for the totals of an $N$ -particle system by summing Step 1 over particles. The generators of translations and rotations close into the algebra of the Euclidean group — the bracket table is the group's structure, which is the general story of Generators of Continuous Transformations. Verified: setting $v=L$ in the pattern reproduces $\{L_i,L_j\}=\epsilon_{ijk}L_k$ , the so(3) closure derived from fundamentals in Algebra of Poisson Brackets; the quantum transcription $[\hat L_i,\hat p_j]=i\hbar\epsilon_{ijk}\hat p_k$ is verbatim.

Limiting cases

Case	Result	Meaning
$G=p_j$ , any $\lambda$	$q_j\to q_j+\lambda$ , all else fixed	translation flow is exact at finite parameter — the Lie series terminates
$G=L_z$ , $\theta\to0$	$\delta x=-\theta y$ , $\delta p_x=-\theta p_y$ , …	infinitesimal rotation of both arrows; the ICT of the general theory
$G=L_z$ , $\theta=2\pi$	identity map	classical rotations are exactly periodic — no spinor surprise without $\hbar$
Central potential, $H=p^2/2m+V(r)$	$\{L_z,H\}=0$	the rotation generator commutes with $H$ : conserved, by the symmetry article's reciprocity
Two translations	$\{p_x,p_y\}=0$	sliding east then north = north then east; flows commute
Two rotations	$\{L_x,L_y\}=L_z\neq0$	order matters; the residue of swapping is itself a rotation
Quantum transcription	$\hat p=-i\hbar\partial_x$ , $[\hat L_i,\hat L_j]=i\hbar\epsilon_{ijk}\hat L_k$	the job descriptions survive; only the currency changes

Common trapThe rotation generated by L only turns the positions — momenta stay where they were

This is the misconception that breaks everything quietly. The flow of $L_z$ moves every phase-space variable with nonvanishing bracket against it, and Derivation B shows the momenta's flow equations are identical in form to the positions'. Rotate positions only, and you have applied a non-canonical map: kinetic energy $p^2/2m$ — which must be blind to orientation — would change relative to the potential's orientation, $\mathbf r\cdot\mathbf p$ would jump, and the bracket test of Poisson Brackets as a Test for Canonicity would fail on the mixed blocks. The same trap has a quantum echo: rotating a wavefunction with $e^{-i\theta\hat L_z/\hbar}$ rotates position and momentum distributions together. When you "rotate a system," you rotate its state — both arrows — or you have not performed a symmetry operation at all.

Worked examples

1 · A quarter turn of one phase point (confidence builder)

A particle in the plane has $(x,y)=(2.0,\,1.0)\ \mathrm{m}$ and $(p_x,p_y)=(0.50,\,-0.30)\ \mathrm{kg\,m/s}$ . Apply the finite $L_z$ flow with $\theta=90^\circ$ . Verify that $L_z$ , $|\mathbf r|^2$ , $|\mathbf p|^2$ and $\mathbf r\cdot\mathbf p$ all survive the turn.

Step 1. Rotate the position by $R(90^\circ)$ : $(x',y')=(x\cos90^\circ-y\sin90^\circ,\;x\sin90^\circ+y\cos90^\circ)=(-1.0,\;2.0)\ \mathrm{m}$ .

Step 2. Rotate the momentum by the same matrix: $(p_x',p_y')=(-p_y,\;p_x)=(\mathbf{0.30},\;\mathbf{0.50})\ \mathrm{kg\,m/s}$ — the co-rotation that the warning box insists on.

Step 3. Angular momentum before: $L_z=xp_y-yp_x=2.0\times(-0.30)-1.0\times0.50=-0.60-0.50=\mathbf{-1.10\ kg\,m^2/s}$ . After: $L_z'=(-1.0)(0.50)-(2.0)(0.30)=-0.50-0.60=\mathbf{-1.10}$ ✓ — the generator rides its own flow unchanged.

Step 4. Lengths and angle: $|\mathbf r|^2=4.0+1.0=5.0$ before, $1.0+4.0=5.0$ after ✓. $|\mathbf p|^2=0.25+0.09=0.34$ both sides ✓. $\mathbf r\cdot\mathbf p=2.0\times0.50+1.0\times(-0.30)=0.70$ before; after: $(-1.0)(0.30)+(2.0)(0.50)=-0.30+1.00=0.70$ ✓. Every invariant on Derivation B's inventory checks out; the negative $L_z$ correctly records that the momentum arrow points clockwise of the position arrow, before and after.

2 · Sliding and turning a two-particle system — and catching the non-commutativity (exam level)

Two particles, $m_1=1.0$ , $m_2=3.0\ \mathrm{kg}$ , sit at $\mathbf r_1=(0,0)$ , $\mathbf r_2=(4.0,\,0)\ \mathrm{m}$ with momenta $\mathbf p_1=(1.0,\,2.0)$ , $\mathbf p_2=(-1.0,\,0.5)\ \mathrm{kg\,m/s}$ . (a) Apply the total- $P_y$ flow with $\lambda=1.2\ \mathrm{m}$ . (b) Then apply the total- $L_z$ flow with $\theta=30^\circ$ and verify total $L_z$ is conserved through both operations. (c) Show by explicit computation that reversing the order changes where particle 1 ends up.

Step 1. Total momentum: $\mathbf P=(1.0-1.0,\;2.0+0.5)=(0,\;2.5)\ \mathrm{kg\,m/s}$ . Total angular momentum: $L_z=(0\cdot2.0-0\cdot1.0)+(4.0\times0.5-0\times(-1.0))=0+2.0=\mathbf{2.0\ kg\,m^2/s}$ .

Step 2. The $P_y$ flow, $\lambda=1.2$ : every $y_a\to y_a+1.2$ , nothing else moves (Derivation A). New positions $\mathbf r_1=(0,\,1.2)$ , $\mathbf r_2=(4.0,\,1.2)$ ; momenta unchanged; separation $\mathbf r_2-\mathbf r_1=(4.0,\,0)$ unchanged — the slide is rigid. Check $L_z$ after the slide: $(0\times2.0-1.2\times1.0)+(4.0\times0.5-1.2\times(-1.0))=-1.2+3.2=\mathbf{2.0}$ ✓. Why it survived: under a $y$ -translation, $\Delta L_z=\lambda\{L_z,P_y\}=-\lambda P_x$ , and this system happens to have $P_x=0$ . Translation-invariance of $L_z$ is a gift of this configuration, not a law.

Step 3. The $L_z$ flow, $\theta=30^\circ$ ( $\cos=0.86603$ , $\sin=0.50000$ ), rotates all four arrows: $\mathbf r_1'=(0\times0.86603-1.2\times0.5,\;0\times0.5+1.2\times0.86603)=(\mathbf{-0.600},\;\mathbf{1.03923})$ ; $\mathbf r_2'=(4.0\times0.86603-1.2\times0.5,\;4.0\times0.5+1.2\times0.86603)=(\mathbf{2.86410},\;\mathbf{3.03923})$ ; $\mathbf p_1'=(1.0\times0.86603-2.0\times0.5,\;1.0\times0.5+2.0\times0.86603)=(\mathbf{-0.13397},\;\mathbf{2.23205})$ ; $\mathbf p_2'=(-1.0\times0.86603-0.5\times0.5,\;-1.0\times0.5+0.5\times0.86603)=(\mathbf{-1.11603},\;\mathbf{-0.06699})$ (all in SI units).

Step 4. Conservation audit: $L_z'=(-0.600)(2.23205)-(1.03923)(-0.13397)+(2.86410)(-0.06699)-(3.03923)(-1.11603)$ . Termwise: $-1.33923+0.13923-0.19186+3.39186=\mathbf{2.0000}$ ✓. The rotation generator is conserved under its own flow through both operations, to the last digit.

Step 5. Order matters. Rotate first (about the origin), translate second: particle 1 starts at the origin, which the rotation fixes, and the slide then puts it at $(0,\;1.2)$ . Translate first, rotate second — Step 3's result: $(-0.600,\;1.03923)$ . The two orderings disagree by $(0.600,\;0.16077)$ m, about $0.62$ m of displacement, even though the ingredient operations were identical. The mismatch is exactly what $\{L_z,P_y\}=-P_x\neq0$ (as an operator statement on the flows) encodes; the systematic theory of such residues — down to why the mismatch is itself a translation — is developed in Generators of Continuous Transformations.

Origins · 1 minHermann Weyl and how this was discovered

Hermann Weyl

1885–1955 · Göttingen, Zürich, Princeton

Weyl is the physicist-mathematician who made "momentum is the generator of translations" the definition rather than a curiosity. His 1928 Gruppentheorie und Quantenmechanik — written in a race with the physicists he politely outran — rebuilt the brand-new quantum theory on group theory: observables like momentum and angular momentum are not primarily quantities but generators of the symmetry operations of space, and their commutation relations are the group's structure constants wearing physical units. The book's viewpoint was so far ahead of its audience that Pauli mockingly called the group-theory influx the Gruppenpest — the "group plague" — before the plague conquered physics completely. Weyl had form: his 1918 attempt to unify gravity and electromagnetism by gauging scale invariance failed as physics but coined the word gauge and the template that, re-aimed at quantum phases, became the architecture of the Standard Model. Exiled from Göttingen in 1933 (his wife was Jewish; he resigned his chair), he settled at the Institute for Advanced Study alongside Einstein. His dictum — "symmetry, as wide or as narrow as you may define its meaning, is one idea by which man through the ages has tried to comprehend and create order" — is this article compressed to a sentence: the generators are the order.

History

1788

— Lagrange's Mécanique analytique pairs each coordinate with its conjugate momentum, the seed of the generator relationship

1809

— Poisson's bracket makes the pairing computational: momentum and position acquire their unit bracket

1873

— Lie builds continuous transformation groups from infinitesimal generators — translations and rotations become flows of functions

1918

— Noether ties each continuous symmetry to its conserved generator, momentum to homogeneity and angular momentum to isotropy of space

1928

— Weyl's Gruppentheorie und Quantenmechanik installs generators as the foundation of quantum kinematics

1939

— Wigner classifies elementary particles by how they respond to the Poincaré generators — mass and spin as generator eigenvalues

Going further

References. Goldstein, Poole & Safko, Classical Mechanics 3rd ed., §9.4–9.6 (infinitesimal canonical transformations and generators) · Landau & Lifshitz, Mechanics Vol. 1, §40–42 (Poisson brackets, conservation laws) · Morin, Introduction to Classical Mechanics, Ch. 15 (Hamiltonian formalism) · Taylor, Classical Mechanics, Ch. 13 (Hamiltonian mechanics) · Arnold, Mathematical Methods of Classical Mechanics, §40 (Hamiltonian phase flows and symmetries)

Provenance. All Poisson-bracket identities ( $\{q_i,p_j\}=\delta_{ij}$ , $\{L_i,L_j\}=\epsilon_{ijk}L_k$ , $\{L_i,p_j\}=\epsilon_{ijk}p_k$ ) and the infinitesimal generator relation $\delta f=\epsilon\{f,G\}$ should be recomputed directly from the definitions; they match Goldstein §9.4–9.6, Landau-Lifshitz Vol.1 §40–42, and Arnold §40. The finite-rotation derivation reproduces the standard active rotation matrix. The canonicity check uses only equality of mixed partial derivatives. No empirical numbers appear; the structure constants $\epsilon_{ijk}$ are exact and define $\mathfrak{so}(3)$ . The worked numerics (quarter-turn of $(2.0,1.0,0.50,-0.30)$ with $L_z=-1.10$ and invariants $5.0$ , $0.34$ , $0.70$ ; the two-particle slide-and-rotate audit with total $L_z=2.0$ conserved and ordering gap $(0.600,\,0.16077)$ m) were computed for this article and cross-checked against the rotation matrices directly.

Builds on: Poisson Brackets: Definition · Fundamental Poisson Brackets · Definition of Canonical Transformations · Type-2 Generating Function F2(q,P,t) Leads to: Poisson-Bracket Formulation of Dynamics · Algebra of Poisson Brackets · Symplectic Group and Phase-Space Geometry

Which comes first — is momentum the generator because p = mv, or is p = mv because momentum generates translations?

The generator role is primary; the formula is derived and provincial. In Cartesian coordinates with $L=\tfrac12 m\dot q^2-V$ the conjugate momentum happens to be $m\dot q$ , but change the setting and the formula changes while the job does not: for a charged particle in a magnetic field the translation generator is $m\mathbf v+e\mathbf A$ , not $m\mathbf v$ ; for a rotating coordinate the generator conjugate to the angle is the angular momentum, with different units entirely. Quantum mechanics settles the priority question empirically — $\hat p=-i\hbar\,\partial_x$ has no " $mv$ " anywhere in it, only the translation job. The safe mental model: "momentum" names whatever function the canonical structure pairs with your coordinate, and the pairing means exactly "generates its displacement."

If the L_z flow rotates momenta too, why does the formula L_z = x p_y − y p_x only mention one particle's worth of variables?

The formula and the flow are two views of one object. $L_z$ as a formula is evaluated at a phase point; $L_z$ as a generator acts on all variables through brackets, and its action on the momenta comes from its own momentum-dependence: $\{p_x,L_z\}=-\partial L_z/\partial x=-p_y$ . Any generator depending on both $q$ 's and $p$ 's moves both — that is the mechanical reason. The conceptual reason is better: a rotation is an operation on states, and a state is position-and-momentum together. A generator that turned only positions would not represent any physical rotation of anything; it would be a different (and non-canonical) operation that merely shares a name.

Is there a boost generator too — something that shifts velocities the way p shifts positions?

Yes: $G=mx-pt$ (per particle; sum for systems, giving $\mathbf G=M\mathbf X_{\rm cm}-\mathbf Pt$ ) generates Galilean boosts. Its flow at fixed $t$ shifts $p$ by $m\,\delta v$ and $x$ by $\delta v\,t$ — exactly "view the system from a frame moving at $-\delta v$ ." It is the odd member of the family: explicitly time-dependent, so its conservation involves the $\partial G/\partial t$ term, and its conservation law is not a frozen number but a statement — the center of mass moves uniformly. Its brackets with the other generators ( $\{G_i,P_j\}=m\delta_{ij}$ , a central term) distinguish the Galilei group from the Poincaré group and are the classical shadow of the mass superselection rule. The worked treatment belongs to Constants of Motion via Poisson Brackets and Symmetries and Conservation Laws in Hamiltonian Form.

Why do I never need the Lie series for translations but always for rotations?

Termination depth of the iterated brackets. For $G=p$ : the first bracket $\{q,p\}=1$ is constant, so the second-order term is the bracket of a constant — zero. Series over after one term; finite = infinitesimal scaled. For $G=L_z$ : $\{x,L_z\}=-y$ , whose bracket with $L_z$ is $-x$ , whose bracket is $+y$ … the series cycles forever with period four, and summing it is exactly how the sine and cosine of the rotation matrix are born (the same mechanism that resums the oscillator flow in Poisson-Bracket Formulation of Dynamics). Rule of thumb: generators linear in the variables they move give terminating flows; generators that feed variables into each other give trigonometric or hyperbolic resummations.

What breaks if I rotate positions but forget to rotate momenta in a simulation?

Three alarms, in order of subtlety. First, energy: for a particle orbiting in a central potential, kinetic energy is orientation-blind, but the relation between $\mathbf p$ and the local orbit direction is scrambled — your circular orbit instantly becomes an eccentric one with different $L_z$ , so the "same" system behaves differently after the "symmetry operation." Second, the canonical structure: the map (rotate $q$ , freeze $p$ ) fails the bracket test on the mixed block — $\{Q_x,P_x\}=\cos\theta\neq1$ — so downstream machinery assuming canonicity (symplectic integrators, perturbation theory) silently degrades. Third, conservation bookkeeping: total $L_z$ computed after the half-rotation differs from before, and you will hunt a nonexistent bug in your force law. The fix is one line: rotate the momentum block with the same matrix.

How does the crystal-lattice quasi-momentum connect to the generator story?

Continuous translation symmetry has generator $p$ with a continuum of conserved values. A crystal keeps only the discrete translations by lattice vectors $\mathbf a$ ; the surviving operation is the finite flow $e^{\mathrm{i}\mathbf a\cdot\hat{\mathbf p}/\hbar}$ (quantum) or the finite canonical map (classical), not the infinitesimal generator itself. Eigenvalues of a finite translation are phases $e^{\mathrm{i}\mathbf k\cdot\mathbf a}$ , so the conserved label $\mathbf k$ is defined only modulo $2\pi/\mathbf a$ — the Brillouin zone. Everything strange about crystal momentum (umklapp scattering, band folding, phonons carrying "momentum" that isn't $mv$ of anything) is the ordinary generator story with "conserved number" replaced by "conserved number modulo the reciprocal lattice." The generator viewpoint makes the modification look inevitable rather than mysterious.

Do the generator jobs of p and L have anything to do with their conservation, or are those separate facts?

Deeply connected, but logically separable — and this article deliberately handles only the job half. Generator status is kinematic: $p$ translates and $L_z$ rotates in every system whatsoever, including ones with walls and anisotropies where neither is conserved. Conservation is dynamic: it holds exactly when the Hamiltonian is blind to the flow — $\{G,H\}=0$ — which for $p$ means a translation-invariant $H$ and for $L_z$ a rotationally symmetric one. The two-way reciprocity (H invariant under G's flow ⟺ G constant along H's flow) is the Hamiltonian form of Noether's theorem, and it is the entire subject of Symmetries and Conservation Laws in Hamiltonian Form. Keep the layers straight and the logic of mechanics gets simpler: first learn what each generator does, then ask whether your particular $H$ lets it keep its value.

In quantum mechanics the momentum operator involves a derivative — where did the derivative come from, classically?

It was always there, disguised as a bracket. Classically, the change of any observable under the translation flow is $\delta f=\lambda\{f,p\}=\lambda\,\partial f/\partial q$ — bracketing with $p$ is differentiation with respect to position (one of the limiting-case rows of Poisson Brackets: Definition). Quantization sends brackets to commutators, $\{f,p\}\to[\hat f,\hat p]/i\hbar$ , so the statement "commuting with $\hat p$ differentiates" forces the representation $\hat p=-i\hbar\,\partial_x$ on wavefunctions. The derivative is not a quantum novelty; it is the classical translation flow wearing operator clothes. This is the cleanest instance of the article's slogan: the formula for momentum changed between 1833 and 1926, the job never did.