Constants of Motion via Poisson Brackets

Every planet in the solar system carries an invisible arrow. It points from the Sun toward the orbit's perihelion, its length is the orbit's eccentricity (times $mk$ ), and for a pure $1/r$ attraction it never moves — the arrow is a conserved quantity, the Laplace–Runge–Lenz vector, and its constancy is the deep reason Kepler orbits are closed ellipses rather than slowly turning rosettes. Nobody discovers such a thing by staring at trajectories. It falls out of an algebraic discipline: write down a candidate function on phase space, bracket it with the Hamiltonian, and ask whether the result vanishes. That discipline is this page. The master equation $\dot f=\{f,H\}+\partial f/\partial t$ (derived in Poisson Brackets: Definition) converts "is $f$ conserved?" from a question about all possible motions into a single partial-differential identity you can check in five lines. The craft has real subtleties: conserved quantities that carry an explicit clock and survive only through a cancellation ( $G=pt-mx$ , the constant everyone forgets exists); the hard ceiling of $2n-1$ independent constants and why "independent" is doing heavy lifting; the difference between constants that merely coexist and constants in involution — the currency that buys integrability; and Poisson's theorem, which sometimes manufactures new conservation laws from old ones and sometimes hands you back a uselessly conserved zero. By the end, the Kepler arrow stops being a curiosity: it is the showcase of every tool here — a bracket-verified invariant, the fifth of five independent constants in a system that only needs three to be integrable, and the classical seed of hydrogen's "accidental" degeneracy.

Key takeaways

A phase-space function is a constant of motion exactly when $\{f,H\}+\partial f/\partial t=0$ ; for time-independent $f$ this is the clean commutation condition $\{f,H\}=0$ .
Some genuine invariants carry an explicit clock: the free-particle $G=pt-mx$ has $\{G,H\}=-p\neq0$ , yet the explicit drift $\partial_tG=+p$ cancels it exactly.
Any function of constants of motion is a constant of motion; a system with $n$ degrees of freedom has at most $2n-1$ functionally independent time-independent constants.
Poisson's theorem (proved in Algebra of Poisson Brackets) says $\{f,g\}$ of two constants is a constant — productive when it yields something new ( $\{L_x,L_y\}=L_z$ ), sterile when it returns zero or old news.
Constants in involution — mutually vanishing brackets, like $L^2$ and $L_z$ — are the currency of integrability; non-commuting constants like $L_x,L_y,L_z$ can never be "used" simultaneously.
The Laplace–Runge–Lenz vector $\mathbf A=\mathbf p\times\mathbf L-mk\hat{\mathbf r}$ is Kepler's extra invariant: it fixes the perihelion direction, encodes the eccentricity, and makes the problem maximally superintegrable.
Key idea: conservation is bracket algebra — a quantity is constant precisely when its bracket with $H$ cancels its explicit clock, and the inventory of such quantities dictates how solvable, and how degenerate, the dynamics is.

The core insight

A quantity is conserved precisely when its Poisson bracket with the Hamiltonian cancels its explicit time dependence — $\{f,H\}+\partial_tf=0$ — and the resulting inventory of constants, counted and sorted by their mutual brackets, determines whether the motion is integrable, superintegrable, or chaotic.

Where this lives today

The inventory-of-constants viewpoint is working infrastructure across physics. Solar-system dynamics still runs on it: modern secular theories track the slow evolution of each planet's eccentricity vector — Laplace's arrow, no longer frozen once planet-planet perturbations act — and the measured $43''$ /century anomalous drift of Mercury's arrow remains general relativity's cleanest classical test, now repeated on millisecond pulsars where the same LRL drift reaches degrees per year. Hydrogen's $\mathfrak{so}(4)$ is textbook quantum mechanics, but its consequences are current technology: Rydberg-atom quantum simulators manipulate states organized by exactly the $(n,\ell,m)$ degeneracy structure the classical algebra predicts, and "circular Rydberg states" — engineered for qubit longevity — are the quantum orbits with maximal $|\mathbf L|$ and vanishing $|\mathbf A|$ . And at the frontier of dynamics, the failure of this page's machinery is the diagnostic: searches for chaos in exoplanet systems, particle accelerators, and galactic potentials all reduce to asking which approximate invariants survive perturbation — the opening question of KAM theory, posed in exactly the bracket language built here.

The physical picture

A constant of motion is a wall in phase space. The level sets of a conserved $f$ are hypersurfaces, and the trajectory, pinned to the value of $f$ it started with, can never cross from one level to another. Each independent constant you find adds another wall, and the trajectory must live in the intersection of all of them. One constant confines a trajectory in a $2n$ -dimensional phase space to a $(2n-1)$ -dimensional sheet; find $2n-1$ independent constants and the intersection is one-dimensional — the trajectory itself, determined without integrating anything.

Figure

Constants of motion as walls — each conserved quantity confines the trajectory to a level surface, and the orbit lives where all the walls intersect

But the walls are not all equally useful, and this is where the bracket earns its keep twice. First it tests candidate constants: $\{f,H\}=0$ or not. Then it sorts the survivors by their brackets with each other. Two constants with $\{f,g\}=0$ — "in involution" — generate flows that commute; they can serve as simultaneous coordinates on the invariant surfaces, and $n$ of them organize the motion onto tori (the Liouville–Arnold structure this page's leads-to threads pick up). Two constants with $\{f,g\}\neq0$ are individually conserved but mutually incompatible as labels — the rotation algebra $\{L_x,L_y\}=L_z$ is the standard case, and its quantum image (you may know $L^2$ and one component of $\hat{\mathbf L}$ ) is the same incompatibility priced in $\hbar$ .

Figure

Compatible versus incompatible constants — involution decides which conserved quantities can label the motion simultaneously

Hold both pictures — walls, and the compatibility ledger — and the rest of the page is systematic: state the criterion, stress-test it on an invariant with a clock, count how many walls can exist, and then cash everything in on the Kepler problem.

The criterion

Everything starts from the master evolution equation of Poisson Brackets: Definition: $\mathrm df/\mathrm dt=\{f,H\}+\partial f/\partial t$ . Demanding that the total derivative vanish identically gives the conservation criterion,

\frac{\partial f}{\partial t}+\{f,H\}=0 \qquad (1)

where: $f(q,p,t)$ — the candidate invariant; $\{f,H\}$ — the dynamical drift, how the flow transports $f$ ; $\partial f/\partial t$ — the explicit drift, how $f$ 's own formula ticks at a frozen phase point. Conservation is their exact cancellation, everywhere in phase space, not just on one lucky orbit (the pointwise/orbitwise equivalence is Derivation A's opening move). Most working constants have no clock in their formula, and the criterion collapses to its famous special case:

\{f,H\}=0 \qquad\text{(for } \partial f/\partial t=0\text{)} \qquad (2)

where: the language " $f$ commutes with $H$ " enters — anticipating the quantum condition $[\hat f,\hat H]=0$ verbatim. Two immediate corollaries set the rules of the game. Antisymmetry gives $\{H,H\}=0$ , so an autonomous Hamiltonian is always conserved — energy conservation costs one line. And because the bracket obeys Leibniz and linearity, any function $F(f_1,\dots,f_k)$ of constants is itself a constant:

\{F(f_1,\dots,f_k),H\}=\sum_{j=1}^{k}\frac{\partial F}{\partial f_j}\{f_j,H\}=0 \qquad (3)

where: the chain rule pushes the bracket through $F$ — so $L^2=L_x^2+L_y^2+L_z^2$ is conserved for free once the components are, and "new" constants must be checked for functional independence before they count. That bookkeeping question — how many independent walls can coexist — has a sharp answer, derived below: at most $2n-1$ .

Derivation AThe conservation criterion and a constant that carries a clock

We are establishing criterion (1) as a pointwise identity and then stress-testing it on the free-particle invariant $G=pt-mx$ — the standard example where the naive test $\{f,H\}=0$ fails while the quantity is genuinely conserved. The strategy: first note the logical equivalence between "constant on every orbit" and "the field $\partial_tf+\{f,H\}$ vanishes everywhere", then run the full criterion on $G$ . Watch the two drifts cancel with opposite signs — that cancellation is the whole content of "conserved with explicit time dependence".

Step 1. Equivalence. Along the orbit through any phase point, the master equation gives $\mathrm df/\mathrm dt=(\partial_tf+\{f,H\})|_{\text{orbit}}$ . If the field $\partial_tf+\{f,H\}$ vanishes identically, every orbit sees $\dot f=0$ . Conversely, if $f$ is constant on every orbit, then since exactly one orbit passes through each phase point (existence–uniqueness of the flow), the field is forced to vanish at every point. Conservation is an identity on phase space, not an on-shell accident.

Step 2. The candidate. For the free particle $H=p^2/2m$ , try $G=pt-mx$ . Explicit drift: differentiate the formula at frozen $(x,p)$ ,

\frac{\partial G}{\partial t}=p

Step 3. Dynamical drift: with $\partial G/\partial x=-m$ and $\partial G/\partial p=t$ ,

\{G,H\}=\frac{\partial G}{\partial x}\frac{\partial H}{\partial p}-\frac{\partial G}{\partial p}\frac{\partial H}{\partial x}=(-m)\frac{p}{m}-t\cdot0=-p

Why the signs oppose: the flow carries $x$ forward ( $\dot x=p/m>0$ for $p>0$ ), which decreases $-mx$ and hence $G$ ; meanwhile the explicit $pt$ term grows at exactly the compensating rate. The invariant is a moving target that the motion chases at matched speed.

Step 4. Total: $\mathrm dG/\mathrm dt=\partial_tG+\{G,H\}=p+(-p)=0$ . Conserved. Its meaning: along the motion $x(t)=x_0+pt/m$ , so $G=pt-m(x_0+pt/m)=-mx_0$ — the invariant is (minus $m$ times) the launch position, the constant that turns "where you started" into a conserved quantity. It is the one-particle shadow of the center-of-mass theorem, and the generator of Galilean boosts (see Symmetries and Conservation Laws in Hamiltonian Form for that side of the story). Verified: matches Goldstein §9.6 and Arnold §40; the explicit solution $x(t)=x_0+pt/m$ gives $pt-mx=-mx_0$ identically, confirming the bracket computation by an independent route.

Go deeper · Derivation B · 2 minWhy two n minus one is the ceiling on independent constants

Derivation BWhy two n minus one is the ceiling on independent constants

We are proving the counting fact quoted in every mechanics course and justified in few: a system with $n$ degrees of freedom admits at most $2n-1$ functionally independent time-independent constants of motion. The strategy is dimensional: constants are coordinates that are constant along the flow, and a one-dimensional trajectory simply has no room for more than $2n-1$ independent labels. The subtlety to watch is "functionally independent" — gradients linearly independent as vectors, not formulas that merely look different.

Step 1. Setup. Phase space has dimension $2n$ . Suppose $f_1,\dots,f_k$ are time-independent constants with linearly independent gradients $\nabla f_1,\dots,\nabla f_k$ in a region (functional independence). Their joint level set $\{f_1=c_1,\dots,f_k=c_k\}$ is then a submanifold of dimension $2n-k$ .

Step 2. The trajectory through any point of that level set stays on it forever — each $f_j$ is constant along the motion. So the level set contains whole trajectories, and must have room for them: its dimension obeys $2n-k\geq1$ wherever the flow does not sit at a fixed point ( $\dot\eta=J\nabla H\neq0$ ).

Step 3. Rearranged: $k\leq2n-1$ . Equality means the level set is the trajectory — the orbit is carved out as an intersection of algebraic surfaces, and the dynamics is solved except for the time parametrization along the curve. Why time-independent matters: clock-carrying invariants like $G=pt-mx$ live on extended phase space $(q,p,t)$ , which has one more dimension; there the count is $2n$ , and indeed the free particle has both $p$ and $G$ — two independent invariants for $n=1$ , one of them wearing a clock.

Step 4. Vocabulary for what the count does not say: it does not promise that $2n-1$ constants exist as globally well-defined, single-valued ("isolating") functions. Generic Hamiltonians have only a handful; chaotic ones have essentially only $H$ itself. The hierarchy — $n$ in involution (integrable) up through $2n-1$ (maximally superintegrable) — measures how exceptional a system is. Kepler sits at the top of it, which is the business of the worked examples. Verified: consistent with Arnold §40 and Goldstein §9.6; sanity check on the 1D oscillator: $n=1$ , ceiling $2n-1=1$ , and indeed $H$ is the only independent time-independent constant — phase along the ellipse is not a single-valued function.

Go deeper · Derivation C · 1 minThe Casimir property of L squared from the algebra alone

Derivation CThe Casimir property of L squared from the algebra alone

We are showing $\{L^2,L_z\}=0$ — and by relabeling, $\{L^2,L_i\}=0$ for every component — using no partial derivatives at all: just bilinearity, the Leibniz rule, and the rotation algebra $\{L_i,L_j\}=\epsilon_{ijk}L_k$ established in Algebra of Poisson Brackets. This is deliberate craft practice: once a bracket table is known, higher brackets are algebra, not calculus. Watch the pairwise cancellation — it is the signature of a Casimir, an element that commutes with the whole algebra.

Step 1. Expand with linearity: $\{L^2,L_z\}=\{L_x^2,L_z\}+\{L_y^2,L_z\}+\{L_z^2,L_z\}$ .

Step 2. Leibniz on each square: $\{L_x^2,L_z\}=L_x\{L_x,L_z\}+\{L_x,L_z\}L_x=2L_x\{L_x,L_z\}$ . From the table $\{L_x,L_z\}=-\{L_z,L_x\}=-L_y$ (cyclic order $z\to x\to y$ ), so $\{L_x^2,L_z\}=-2L_xL_y$ .

Step 3. Likewise $\{L_y^2,L_z\}=2L_y\{L_y,L_z\}=2L_yL_x$ , using $\{L_y,L_z\}=L_x$ . And $\{L_z^2,L_z\}=2L_z\{L_z,L_z\}=0$ by antisymmetry.

Step 4. Sum: $-2L_xL_y+2L_yL_x+0=0$ . (Classical functions commute under multiplication; the cancellation is exact.)

\{L^2,L_i\}=0,\qquad i=x,y,z

Why this matters for the inventory: $L^2$ and any single $L_i$ are constants in involution for a central potential — a compatible pair — while the three components among themselves are not. The maximal commuting inventory for a central-field particle is $\{H,L^2,L_z\}$ : exactly $n=3$ constants in involution, integrability on the nose, and the exact classical skeleton of the quantum labels $(E,\ell,m)$ . Verified: matches Goldstein eq. (9.128) region and Landau–Lifshitz §42; the quantum image $[\hat L^2,\hat L_i]=0$ is the standard Casimir statement of $\mathfrak{so}(3)$ .

Poisson's theorem: factory or mirage

The bracket structure adds one genuinely surprising production rule. Poisson's theorem — proved via the Jacobi identity in Algebra of Poisson Brackets, not re-proved here — states: if $f$ and $g$ are constants of motion, so is $\{f,g\}$ . In one line of why: the Jacobi identity converts $\{\{f,g\},H\}$ into two brackets that each contain a conserved quantity bracketed with $H$ , and both die. The craft question is whether the factory produces anything. Two contrasting cases calibrate expectations:

input constants	bracket	verdict
$L_x$ , $L_y$ (central potential)	$\{L_x,L_y\}=L_z$	productive — a third, functionally independent constant, forced into existence
$L^2$ , $L_z$	$\{L^2,L_z\}=0$	sterile — conserved, yes; informative, no
$H$ , any conserved $f$	$\{f,H\}=0$	sterile by definition — the test itself
$A_x$ , $A_y$ (Kepler LRL components)	$\propto L_z\times(-2H)$	productive-with-a-twist — closes onto old constants, revealing the $\mathfrak{so}(4)$ algebra

The pattern: the factory can only output elements of the (finite-dimensional) symmetry algebra the system already possesses. When the known constants already span it, new brackets return combinations of old ones — that closure is not failure, it is the discovery of the algebra, and reading it off is how one recognizes Kepler's hidden $\mathfrak{so}(4)$ or the oscillator's $\mathfrak{su}(n)$ .

The crown example: Kepler's frozen arrow

For the Kepler Hamiltonian, one conserved vector beyond energy and angular momentum survives all brackets:

\mathbf A=\mathbf p\times\mathbf L-mk\,\hat{\mathbf r},\qquad \{A_i,H\}=0,\qquad H=\frac{p^2}{2m}-\frac{k}{r} \qquad (4)

where: $\mathbf A$ — the Laplace–Runge–Lenz vector; $k$ — the force constant ( $GMm$ gravitationally); $\hat{\mathbf r}$ — the radial unit vector. $\mathbf A$ lies in the orbital plane ( $\mathbf A\cdot\mathbf L=0$ ), points from focus to perihelion, and its magnitude packages the orbit's shape:

e=\frac{|\mathbf A|}{mk},\qquad |\mathbf A|^2=m^2k^2+2mEL^2 \qquad (5)

where: $e$ — the eccentricity; $E$ — the (negative, for bound orbits) energy. The conservation of a direction in the plane is exactly the statement that the ellipse does not precess: perihelion stays put, orbits close. Add a perturbation — an extra $1/r^3$ term, an oblate sun, general relativity — and $\{A_i,H\}\neq0$ ; the arrow begins to drift, and its drift rate is the perihelion precession. Mercury's famous $43''$ per century is a measured $\dot{\mathbf A}$ . The bookkeeping crowns the section. Constants in hand: $H$ (one), $\mathbf L$ (three), $\mathbf A$ (three) — seven, in a system whose ceiling is $2n-1=5$ . The two required relations are exactly $\mathbf A\cdot\mathbf L=0$ and equation (5): five independent constants survive, the maximum. Kepler is maximally superintegrable — the orbit (though not the schedule along it) is completely pinned by algebra — and the bracket table of $\{L_i, A_j\}$ closes into $\mathfrak{so}(4)$ , the four-dimensional rotation algebra, whose quantum representation theory hands over the hydrogen spectrum $E_n\propto-1/n^2$ without solving any differential equation.

Limiting cases

Case	Result	Meaning
$\partial_tf=0$	criterion collapses to $\{f,H\}=0$	the workaday commutation test
$f=H$ , autonomous	$\{H,H\}=0$	energy conservation from antisymmetry alone
cyclic coordinate $q_k$	$\{p_k,H\}=-\partial H/\partial q_k=0$	conserved momentum = the simplest wall (cf. the cyclic-coordinates article)
$F$ (constants)	conserved via chain rule, eq. (3)	not a new wall — check independence before celebrating
$k=2n-1$ constants	level set = the orbit itself	maximal superintegrability; Kepler, isotropic oscillator
chaotic system	only $H$ (and friends) survive	non-integrability = a starved inventory, the KAM story's opening condition

Common trapA nonzero bracket with H does not disqualify a candidate — and Poisson's theorem is not a perpetual-motion machine for conservation laws

Two opposite failure modes, both common. First: computing $\{f,H\}\neq0$ and discarding $f$ without checking the clock term. Genuine invariants with explicit time dependence exist in every course ( $G=pt-mx$ ; the Galilean boost generator; adiabatic-invariant candidates) and are conserved through the cancellation $\partial_tf=-\{f,H\}$ — run the full criterion (1) before rejecting. Second, the mirror error: feeding Poisson's theorem two constants and announcing a "new conservation law" without an independence check. The bracket of two constants may be zero, a number, or a function of constants you already hold — only a result whose gradient is linearly independent of the existing inventory adds a wall. The theorem's guaranteed output is conserved; its useful output is bounded by the symmetry algebra, and once the algebra closes, the factory only recirculates stock.

Worked examples

1 · The constant that remembers where you started (confidence builder)

A free particle of mass $m=1.5\ \mathrm{kg}$ passes $x_0=2.0\ \mathrm m$ at $t=0$ with momentum $p=+3.0\ \mathrm{kg\,m/s}$ . Verify numerically that $G=pt-mx$ is conserved, and interpret its value.

Step 1. At $t=0$ : $G=3.0\times0-1.5\times2.0=\mathbf{-3.0\ \mathrm{kg\,m}}$ .

Step 2. Evolve to $t=2.0\ \mathrm s$ : the particle moves at $v=p/m=2.0\ \mathrm{m/s}$ , so $x=2.0+2.0\times2.0=6.0\ \mathrm m$ , while $p$ stays $3.0$ . Then $G=3.0\times2.0-1.5\times6.0=6.0-9.0=\mathbf{-3.0\ \mathrm{kg\,m}}$ ✓ — same value, clock term and dynamics in exact cancellation.

Step 3. Interpretation: $G=-mx_0$ , i.e. $x_0=-G/m=3.0/1.5=2.0\ \mathrm m$ ✓. The invariant is the launch point promoted to a conserved quantity. For $N$ particles the same construction gives $\mathbf G=\mathbf Pt-M\mathbf R_{\rm cm}$ : the center of mass moves uniformly — the theorem usually quoted without noticing it is a conservation law with a clock in it.

2 · Kepler's arrow, computed and audited (exam level)

A satellite (unit mass $m=1.0$ ) orbits a center with $k=4.0$ (SI units suppressed). At one instant it is at $\mathbf r=(1.0,\,0)$ with $\mathbf p=(0,\,2.5)$ . (a) Compute $\mathbf L$ and $\mathbf A$ . (b) Extract the eccentricity two independent ways. (c) Sketch why $\{A_x,H\}=0$ holds symbolically.

Step 1. Angular momentum: $L_z=xp_y-yp_x=1.0\times2.5-0=\mathbf{2.5}$ . In 2D, $\mathbf p\times\mathbf L=(p_yL_z,\,-p_xL_z)$ .

Step 2. LRL components at $r=1.0$ : $A_x=p_yL_z-mk\,x/r=2.5\times2.5-4.0\times1.0=6.25-4.0=\mathbf{2.25}$ ; $A_y=-p_xL_z-mk\,y/r=0-0=\mathbf{0}$ . The arrow points along $+\hat x$ — perihelion of this orbit lies on the positive $x$ -axis. (Consistency: at this instant $\mathbf r\cdot\mathbf p=0$ and the particle is at $r=1$ ; an apse on the $x$ -axis is exactly where a purely tangential velocity belongs.)

Step 3. Eccentricity, route one: $e=|\mathbf A|/(mk)=2.25/4.0=\mathbf{0.5625}$ .

Step 4. Route two, from $E$ and $L$ : $E=p^2/2m-k/r=\dfrac{2.5^2}{2}-\dfrac{4.0}{1.0}=3.125-4.0=-0.875$ (bound ✓). Then

e=\sqrt{1+\frac{2EL^2}{mk^2}}=\sqrt{1+\frac{2(-0.875)(6.25)}{1.0\times16.0}}=\sqrt{1-0.6836}=\sqrt{0.3164}=\mathbf{0.5625}\ \checkmark

Two structurally different formulas, one number — equation (5) verified digit for digit. (Bonus reading: $r=1.0$ with the apse there and $e=0.5625$ makes this the perihelion itself: $r_{\min}=\frac{L^2/mk}{1+e}=\frac{6.25/4}{1.5625}=1.0$ ✓.)

Step 5. The bracket audit, symbolically: with $A_x=xp_y^2-yp_xp_y-kx/r$ (expanding $p_yL_z-kx/r$ at $m=1$ ), compute $\{A_x,H\}$ term by term against $H=\tfrac12(p_x^2+p_y^2)-k/r$ . The kinetic part of $H$ acting on the $-kx/r$ piece produces $-k(p_xy^2-\dots)/r^3$ -type terms; the potential part of $H$ acting on the momentum-bearing pieces produces their exact negatives. Grouped: the $p_xp_y^2$ terms cancel between the two slots of the bracket, the $kp_xy^2/r^3$ terms cancel between kinetic-on-potential and potential-on-kinetic, and the $kxyp_y/r^3$ terms cancel three ways. Every monomial finds its partner: $\{A_x,H\}=0$ identically — the frozen arrow, certified by algebra. (The full expansion is five lines of partials; the provenance note lists it as a verifier target.)

Origins · 1 minPierre-Simon Laplace and how this was discovered

Pierre-Simon Laplace

1749–1827 · Paris

Laplace spent a career trying to prove the solar system needs no adjustments. Newton had suspected that planetary perturbations would slowly wreck the orbits — divine intervention keeping the clockwork wound. Laplace's Mécanique céleste (five volumes, 1799–1825) answered with theorems: to the orders he could reach, the perturbations average out, eccentricities and inclinations only oscillate. When Napoleon needled him about God's absence from the book, he reportedly replied "I had no need of that hypothesis." Inside volume one sits the conserved vector this page crowns: the eccentricity arrow of the two-body problem, which Jakob Hermann and Johann Bernoulli had glimpsed in 1710 and Laplace established systematically. Its modern name is a comedy of misattribution — Carl Runge put it in a 1919 vector-analysis textbook, Wilhelm Lenz borrowed it for atomic physics in 1924, and "Runge–Lenz" stuck to an object neither man discovered. The arrow had the last laugh twice: Pauli used its quantum version in 1926 to crack the hydrogen spectrum before Schrödinger's equation was solved, and Fock showed in 1935 that it rotates hydrogen's states in a hidden four-dimensional sphere — Laplace's stability obsession, reborn as the $\mathfrak{so}(4)$ symmetry behind the periodic table's first column of magic numbers.

History

1710

— Hermann and Johann Bernoulli identify a fixed direction in the two-body orbit, the arrow's first appearance

1799

— Laplace's Mécanique céleste systematizes the conserved vector and the stability program of celestial mechanics

1809

— Poisson proves his theorem: brackets of constants of motion are constants of motion

1845

— Hamilton rediscovers the arrow as the "eccentricity vector" via his quaternion methods

1926

— Pauli solves hydrogen's spectrum from the quantized Runge–Lenz algebra, months before wave mechanics could

1935

— Fock reveals the four-dimensional rotational symmetry behind the Kepler/Coulomb degeneracy

Going further

References. Goldstein, Poole & Safko, Classical Mechanics, 3rd ed., §9.5–9.6 · Landau & Lifshitz, Mechanics, 3rd ed., §42 · Morin, Introduction to Classical Mechanics, Hamiltonian chapter · Taylor, Classical Mechanics, Ch. 13 · Arnold, Mathematical Methods of Classical Mechanics, §40

Provenance. Conservation criterion and Poisson's theorem from Goldstein §9.5–9.6, Landau-Lifshitz §42, Arnold §40. Verifier should recompute: (1) $\{J_x,J_y\}=J_z$ and the $\mathfrak{so}(3)$ closure from $J_i=\epsilon_{ijk}q_j p_k$ ; (2) the free-particle result $\{G,H\}=-p$ with $G=pt-mx$ , $H=p^2/2m$ , confirming $\mathrm{d}G/\mathrm{d}t=0$ ; (3) the Jacobi-identity proof of Poisson's theorem. All bracket relations match standard textbook conventions $\{q_i,p_j\}=\delta_{ij}$ . The numeric examples (free-particle invariant $G=-3.0$ at both $t=0$ and $t=2.0\ \mathrm s$ ; Kepler point $\mathbf r=(1,0)$ , $\mathbf p=(0,2.5)$ , $k=4.0$ giving $e=0.5625$ by both $|\mathbf A|/mk$ and $\sqrt{1+2EL^2/mk^2}$ ) were computed for this article and cross-checked by independent routes; a verifier should additionally expand $\{A_x,H\}$ in partials and confirm the three-way cancellation.

Builds on: Poisson Brackets: Definition · Poisson-Bracket Formulation of Dynamics · Algebra of Poisson Brackets Leads to: Definition of Canonical Transformations · Symplectic Group and Phase-Space Geometry · Fundamental Poisson Brackets

What exactly does "functionally independent" mean, and how do I check it in practice?

Constants $f_1,\dots,f_k$ are functionally independent on a region when their gradients $\nabla f_1,\dots,\nabla f_k$ (as $2n$ -component phase-space vectors) are linearly independent there — equivalently, no smooth relation $F(f_1,\dots,f_k)=0$ holds identically. The practical check is a rank computation: stack the gradients into a $k\times2n$ matrix and verify rank $k$ at a generic point. Traps to respect: independence can fail on special submanifolds (circular orbits, zero angular momentum) while holding generically; and quantities that look unrelated can be secretly dependent — $H$ , $L^2$ , and $A^2$ for Kepler look like three scalars but equation (5) ties them into two.

Is energy always among the constants of motion?

Only when the Hamiltonian has no explicit time dependence — then $\{H,H\}=0$ does the job. Drive the system (a pumped swing, an RF trap, any $H(q,p,t)$ ) and energy is genuinely not conserved; the criterion shows why: $\mathrm dH/\mathrm dt=\partial H/\partial t\neq0$ . Two useful refinements. Periodically driven systems often conserve a quasi-energy (Floquet invariant) instead — conservation reorganizes rather than disappears. And for slow driving, an adiabatic invariant (the action $J=\oint p\,\mathrm dq$ ) is approximately conserved even while $H$ changes by large amounts — a clock-carrying near-invariant in exactly the spirit of criterion (1).

How do I actually hunt for constants of motion in an unfamiliar Hamiltonian?

A field-tested sequence. First, scan for cyclic coordinates in every chart you can think of — Cartesian, polar, rotating, parabolic; each absent coordinate donates its momentum ( $\phi$ absent $\Rightarrow$ $L_z$ ). Second, read the symmetries physically: rotation invariance, translation invariance, boosts, and write the corresponding generator (the dictionary lives in Momentum and Angular Momentum as Generators). Third, bracket-test polynomial ansätze: assume $f$ quadratic in momenta with coordinate-dependent coefficients and demand $\{f,H\}=0$ — this is how the LRL vector and the Hénon–Heiles non-integrability results were both found. Fourth, feed survivors into Poisson's theorem and map the algebra. If everything fails, suspect chaos: numerically, a Poincaré section that fills area instead of tracing curves is the visual verdict that the inventory is exhausted.

Why does the ceiling sit at 2n minus 1 rather than 2n?

Because a trajectory is one-dimensional and needs one coordinate to move along. Each independent constant flattens the accessible region by one dimension; $2n$ independent constants would pin the "motion" to isolated points — no dynamics at all. The last, $2n$ -th label along the orbit is essentially "time along the trajectory," and it cannot be a time-independent single-valued phase-space function (it would have to increase forever along a bounded orbit, returning to the same phase point with a different value — a contradiction). On extended phase space, where $t$ is a coordinate, the clock-carrying invariants like $G=pt-mx$ legitimately raise the count to $2n$ .

If two components of L force the third via Poisson's theorem, why can I not iterate my way to unlimited constants?

Because the factory's output is confined to the Lie algebra generated by its inputs, and that algebra is finite-dimensional. Starting from $L_x$ and $L_y$ : their bracket gives $L_z$ , but every further bracket ( $\{L_x,L_z\}=-L_y$ , …) lands back inside the span of the three components — $\mathfrak{so}(3)$ has closed. Poisson's theorem recirculates stock once the algebra is complete; genuinely new constants require new physics (a hidden symmetry, as with Kepler's $\mathbf A$ enlarging $\mathfrak{so}(3)$ to $\mathfrak{so}(4)$ ), not more bracketing. The count is also bounded absolutely by the $2n-1$ ceiling regardless of algebraic enthusiasm.

Why is involution the condition for integrability rather than just having n constants?

Integrability needs the constants to serve as simultaneous coordinates on the invariant surfaces, and that requires their flows to commute — which is precisely $\{F_i,F_j\}=0$ (the flows-inherit-the-algebra result of Algebra of Poisson Brackets). With $n$ commuting constants, the joint level set is an $n$ -dimensional surface on which $n$ commuting flows act; the Liouville–Arnold theorem upgrades this to a torus with action-angle coordinates, and the motion becomes straight lines on the torus. Non-commuting constants confine the motion just as much dimensionally, but they cannot be simultaneously "straightened": $L_x,L_y,L_z$ for a central field are three constants, yet the involutive subset you can extract is only $\{L^2,L_z\}$ — which is why superintegrable systems are described as integrable "with extra credit" (degenerate tori, closed orbits) rather than as a different mechanism.

Does the Laplace-Runge-Lenz vector survive quantization?

Yes, with one ordering repair: $\mathbf p\times\mathbf L$ is not Hermitian as written, so Pauli symmetrized it, $\hat{\mathbf A}=\tfrac12(\hat{\mathbf p}\times\hat{\mathbf L}-\hat{\mathbf L}\times\hat{\mathbf p})-mk\hat{\mathbf r}/r$ . The symmetrized operator commutes with $\hat H$ , and together with $\hat{\mathbf L}$ it closes (on each energy shell) into the same $\mathfrak{so}(4)$ the classical brackets exhibit. Pauli's 1926 tour de force ran the representation theory of that algebra and extracted $E_n=-mk^2/(2\hbar^2n^2)$ with the full $n^2$ degeneracy — the hydrogen spectrum from bracket algebra, before the Schrödinger equation had been solved for it. The "accidental" degeneracy of hydrogen ( $\ell$ -levels coinciding) is thus not accidental at all: it is the quantum shadow of the classical frozen arrow.

What happens to all these constants when the Hamiltonian is perturbed?

The generic fate is demotion, in ranked order. Constants protected by exact surviving symmetries stay exact ( $L_z$ if the perturbation is axisymmetric). Constants of broken symmetries become slow variables: their evolution rate is the bracket with the perturbation, $\dot f=\{f,\epsilon V\}=O(\epsilon)$ — the LRL vector under general relativity drifts at Mercury's $43''$ /century rather than jumping. And the integrable structure itself degrades gracefully: KAM theory shows most invariant tori survive small perturbations slightly deformed, while resonant tori shatter into chains of islands and thin chaotic layers. This ranked robustness is why the constants-inventory language remains useful far beyond exactly solvable systems — perturbation theory is bookkeeping for approximately conserved quantities.