Hidden Symmetry and Orbit Closure

Take the textbook picture of a planet on its ellipse and notice one quiet miracle: the orbit returns. Every revolution, the planet retraces exactly the same curve, perihelion in the same place, forever. That is not what central forces generically do. Nudge the force law away from inverse-square — $1/r^{2.1}$ , say — and the ellipse fails to close by a little each turn, pivoting into a rosette that smears over a whole annulus. Energy and angular momentum are conserved for every central force, so they cannot be what holds the ellipse shut. Something extra must be conserved. That something is the Laplace–Runge–Lenz vector for the $1/r$ potential and the Fradkin tensor for the isotropic oscillator $\tfrac12kr^2$ — the only two central potentials, by Bertrand's theorem, whose bound orbits all close. Each extra conserved quantity pins the orientation of the orbit, so the perihelion cannot drift. The symmetry behind it is real but hidden: its generators mix positions and momenta, invisible in ordinary space, revealing themselves only in phase space, where they enlarge the rotation algebra to $SO(4)$ for bound Kepler motion and $SU(3)$ for the oscillator. The payoff reaches beyond celestial mechanics: the same hidden $SO(4)$ is why hydrogen's levels carry their "accidental" $n^2$ degeneracy — Pauli used it to crack the hydrogen spectrum in 1926 — and the breaking of the symmetry is what astronomers measure when a real orbit precesses.

Key takeaways

Energy and angular momentum are conserved for every central force, yet generic central forces give bound orbits that never close — precessing rosettes.
Exactly two potentials close every bound orbit: Kepler $V=-k/r$ (apsidal angle $\pi$ ) and the isotropic oscillator $V=\tfrac12kr^2$ (apsidal angle $\pi/2$ ) — Bertrand's theorem.
Each hides an extra conserved quantity — the Laplace–Runge–Lenz vector for Kepler, the Fradkin tensor for the oscillator — pinning the orbit's orientation and forbidding precession.
These generate a dynamical symmetry in phase space: $SO(4)$ for bound Kepler motion, $SO(3,1)$ for scattering, $SU(3)$ for the oscillator; the quantum fingerprint is hydrogen's $n^2$ and the oscillator's $(N+1)(N+2)/2$ degeneracies.
Key idea: closed orbits signal a conserved quantity beyond energy and angular momentum — a hidden dynamical symmetry that fixes the orbit's orientation, possessed only by the inverse-square and harmonic forces.

The core insight

Closed orbits signal a conserved quantity beyond energy and angular momentum. The inverse-square and harmonic forces each possess a hidden dynamical symmetry — an extra conserved vector/tensor that fixes the orbit's orientation — and Bertrand's theorem proves these are the only two central forces with this property.

Where this lives today

The cleanest modern measurement of hidden-symmetry breaking orbits the black hole at the centre of our galaxy. The star S2 rides a 16-year, $e=0.88$ ellipse around Sagittarius A*, and the GRAVITY interferometer has watched its orbit fail to close: the apsidal line swings forward by about 12 arcminutes per revolution — the Schwarzschild precession predicted for a $4\times10^6\,M_\odot$ black hole. In this article's language, GRAVITY measures $\mathrm{d}\mathbf A/\mathrm{d}t\neq0$ ; the GRAVITY+ upgrade is extending the test to fainter stars deeper in the potential, where any non-Kepler mass would leave its own signature in the precession. The oscillator side is just as alive: ion traps and optical tweezers are engineered to be isotropically harmonic precisely because the $SU(3)$ symmetry makes every orbit closed and every mode degenerate — anharmonicity shows up as dephasing, the laboratory rosette.

The physical picture

Every bound orbit in a central field is two motions stitched together: the radius breathes between $r_{\min}$ and $r_{\max}$ while the angle winds forward. The orbit closes only if the two clocks are commensurate. For a generic force law they are not:

Figure

Same conservation laws, different fates — a generic force weaves a rosette while the inverse-square force closes an ellipse

What could hold an ellipse shut? Think of the orbit as a rigid wire: the hidden symmetry is the statement that the wire's orientation within the plane is dynamically locked. One extra conserved direction is exactly what that takes — and for the $1/r$ potential it exists: the Laplace–Runge–Lenz vector, pointing permanently from focus toward perihelion. Perturb the force law and the vector begins to rotate; its slow rotation is the precession of the rosette. The two charmed potentials are the two fixed points of the closure condition: Kepler closes with apsidal angle $\pi$ (force centre at a focus), Hooke with $\pi/2$ (force centre at the centre). And hold on to what "hidden" means: the extra symmetry genuinely commutes with the Hamiltonian, but its generators mix positions and momenta — no rotation of space produces it. It lives in phase space, where $\mathbf A$ and $\mathbf L$ close into the algebra of $SO(4)$ .

The apsidal angle measures closure

To make "does the orbit close?" quantitative, track the angle swept during one radial half-cycle, perihelion to aphelion — the apsidal angle. Dividing the angular rate $\dot\phi=L/mr^2$ by the radial speed from energy conservation and integrating between turning points gives it directly:

\Delta\phi = \int_{r_{\min}}^{r_{\max}} \frac{(L/r^2)\,\mathrm{d}r}{\sqrt{2m\,[E-V(r)] - L^2/r^2}} \qquad (1)

where: $\Delta\phi$ — apsidal angle, the azimuth swept from $r_{\min}$ to $r_{\max}$ ; $r_{\min},r_{\max}$ — the turning points, roots of the radicand; $E,L$ — conserved energy and angular momentum; $m$ — the (reduced) mass. Successive perihelia are separated by $2\Delta\phi$ , so the orbit closes precisely when $\Delta\phi$ is a rational multiple of $\pi$ — and for the whole family to close, the same rational at every $E$ and $L$ . That double demand is the engine of Bertrand's theorem. Near a circular orbit the integral collapses: a small radial wobble at frequency $\omega_r$ rides on a revolution at frequency $\omega_\phi$ , and the apsidal angle is $\pi$ times their ratio — Derivation A works this out from the curvature of the effective potential:

\Delta\phi = \pi\left[\,3+\frac{r_0\,V''(r_0)}{V'(r_0)}\,\right]^{-1/2},\qquad V=\alpha r^{n}:\;\; \Delta\phi=\frac{\pi}{\sqrt{\,n+2\,}} \qquad (2)

where: $r_0$ — the circular-orbit radius, where $V_{\mathrm{eff}}'(r_0)=0$ ; $n$ — the exponent of the potential (Kepler $n=-1$ , oscillator $n=2$ ); the logarithmic potential is the $n\to0$ member with $\Delta\phi=\pi/\sqrt2$ . The champions give $\pi$ and $\pi/2$ — rational multiples of $\pi$ , independent of $r_0$ , exactly as closure demands. Convention alert: for the force $F\propto r^{m}$ the same result reads $\Delta\phi=\pi/\sqrt{3+m}$ with $m=n-1$ — the form the circular-orbit stability and apsidal precession articles use.

The extra conserved quantity

Now the deeper structure behind $\Delta\phi=\pi$ . For the Kepler potential — and no other — the following combination is a constant of the motion:

\mathbf{A} = \mathbf{p}\times\mathbf{L} - mk\,\hat{\mathbf{r}} \qquad (3)

where: $\mathbf A$ — the Laplace–Runge–Lenz (LRL) vector; $\mathbf p=m\dot{\mathbf r}$ — momentum; $\mathbf L=\mathbf r\times\mathbf p$ — angular momentum; $k$ — the force strength in $\mathbf F=-k\hat{\mathbf r}/r^2$ . Proving $\dot{\mathbf A}=0$ is a satisfying cancellation, spelled out in the dedicated Laplace–Runge–Lenz article; here we take the conservation as given and spend it. Dotting with $\mathbf L$ locates the vector in the orbital plane, and squaring calibrates its length:

\mathbf A\cdot\mathbf L = 0, \qquad A^2 = m^2k^2 + 2mEL^2 \;\Rightarrow\; A = mk\,e \qquad (4)

where: $A=|\mathbf A|$ ; $e$ — the eccentricity, $e=\sqrt{1+2EL^2/mk^2}$ . The LRL vector is the eccentricity vector scaled by $mk$ : it points from focus toward perihelion, and a circle has $\mathbf A=\mathbf 0$ . The closure argument now takes two lines: projecting, $\mathbf r\cdot\mathbf A=(\mathbf r\times\mathbf p)\cdot\mathbf L-mkr=L^2-mkr$ , and reading the left side as $rA\cos\phi$ gives the focal conic $1/r=(mk/L^2)(1+e\cos\phi)$ with perihelion along $\hat{\mathbf A}$ : the orbit is chained to a frozen arrow and must close. The harmonic oscillator plays the same game with a tensor. For $V=\tfrac12m\omega^2r^2$ the conserved object is the Fradkin tensor:

T_{ij} = \frac{p_i p_j}{m} + m\omega^2 x_i x_j \qquad (5)

where: $T_{ij}$ — a symmetric $3\times3$ tensor; $x_i,p_i$ — Cartesian position and momentum components. Its eigenvectors are the principal axes of the orbital ellipse, so its conservation locks the orientation just as $\mathbf A$ does — but a tensor cannot tell an axis from its reverse, which is why the oscillator ellipse is centred on the force centre and closes in half a revolution.

The hidden rotation algebra

Conserved quantities do more than sit still — under Poisson brackets they generate symmetries, and bracketing them together reveals the group:

\{L_i,L_j\}=\epsilon_{ijk}L_k,\qquad \{L_i,A_j\}=\epsilon_{ijk}A_k,\qquad \{A_i,A_j\}=-2mE\,\epsilon_{ijk}L_k \qquad (6)

where: $\{\,,\,\}$ — the Poisson bracket; $\epsilon_{ijk}$ — the Levi-Civita symbol; $E$ — the conserved energy. The first relation is ordinary rotations; the second says $\mathbf A$ transforms as a vector; the third is the surprise — two components of $\mathbf A$ bracket back onto $\mathbf L$ , with a coefficient set by the energy. Rescaling $\mathbf D=\mathbf A/\sqrt{-2mE}$ on a bound orbit absorbs the coefficient, and $(\mathbf L,\mathbf D)$ close into $so(4)$ — Derivation B assembles this and splits it into the two $su(2)$ blocks Pauli exploited. Scattering orbits give the Lorentz algebra $so(3,1)$ , the boundary $E=0$ the Euclidean $e(3)$ ; the oscillator's Fradkin tensor generates the larger $SU(3)$ . Quantum mechanics is where the hidden symmetry prints measurable numbers — degeneracy is symmetry's signature, and a bigger group means bigger degenerate families:

Figure

Degeneracy is the quantum fingerprint — hydrogen stacks all its same-n levels at one energy, a screened atom splits them apart

Hydrogen's "accidental" $n^2$ degeneracy is the quantum shadow of orbit closure: the Casimirs of $SO(4)$ fix $E_n\propto-1/n^2$ with every $\ell<n$ riding along. Screen the potential, as sodium's inner electrons do, and the classical orbits precess while the quantum levels split — one symmetry breaking, two languages. The oscillator's $SU(3)$ likewise stacks $(N+1)(N+2)/2$ states in level $N$ , the ladder beneath the nuclear shell model's magic numbers.

Derivation AThe apsidal angle for nearly circular orbits

We want equation (2), the master formula $\Delta\phi=\pi[3+r_0V''/V']^{-1/2}$ , which converts the closure question into a single number computed from the potential. The strategy is the standard equilibrium analysis transplanted to the radial channel: a near-circular orbit is a small radial oscillation about the minimum of the effective potential, wobbling harmonically at a frequency set by the well's curvature while the angle advances at the circular rate; the apsidal angle is $\pi$ times the frequency ratio. Watch two things: the bracket must be positive or there is no stable circular orbit to expand about, and the result holds only to leading order in the wobble amplitude.

Step 1. Locate the circular orbit. With $V_{\mathrm{eff}}(r)=V(r)+L^2/2mr^2$ , the circular radius satisfies $V_{\mathrm{eff}}'(r_0)=0$ , i.e. $L^2=mr_0^3V'(r_0)$ . Why: for an attractive potential $V'>0$ , so every $L$ picks out a genuine $r_0$ .

Step 2. Compute the curvature of the well. Differentiating again and eliminating $L^2$ with Step 1: $V_{\mathrm{eff}}''(r_0)=V''(r_0)+3L^2/mr_0^4=V''(r_0)+\dfrac{3V'(r_0)}{r_0}$ . Why: the circular condition trades $L$ for the potential's own derivatives.

Step 3. Identify the two frequencies. A small displacement obeys $m\ddot\eta=-V_{\mathrm{eff}}''(r_0)\eta$ , so $\omega_r^2=V_{\mathrm{eff}}''(r_0)/m$ ; the angular rate gives $\omega_\phi^2=V'(r_0)/mr_0$ , and the ratio is $\dfrac{\omega_r^2}{\omega_\phi^2}=3+\dfrac{r_0V''(r_0)}{V'(r_0)}$ . Why: masses and angular momenta cancel — closure near the circle is decided by one dimensionless number, the local logarithmic steepness of the potential.

Step 4. Convert to the apsidal angle. Perihelion to aphelion is half a radial period, $t=\pi/\omega_r$ , during which the azimuth advances $\Delta\phi=\omega_\phi t=\pi\,\omega_\phi/\omega_r$ — equation (2). For $V=\alpha r^n$ , $r_0V''/V'=n-1$ gives $\Delta\phi=\pi/\sqrt{n+2}$ , the same at every $r_0$ . Why: power laws are exactly the potentials with constant logarithmic steepness — the observation that launches Bertrand's theorem. *As a check: $n=-1$ gives $\Delta\phi=\pi$ (Kepler) and $n=2$ gives $\pi/2$ (centred ellipse), both matching the exact solutions.*

Go deeper · Derivation B · 1 minThe hidden rotation algebra of the Kepler problem

Derivation BThe hidden rotation algebra of the Kepler problem

We assemble the six conserved quantities of bound Kepler motion into a Lie algebra and identify the group — the result is $SO(4)$ , with the clean split $so(4)\cong su(2)\oplus su(2)$ that makes the quantum problem solvable by pure algebra. The strategy: start from the three bracket relations (6), then rescale $\mathbf A$ to absorb the energy-dependent coefficient. The step to watch is the rescaling: dividing by $\sqrt{-2mE}$ is legal only because $E$ is itself conserved, and the sign of $E$ changes the answer.

Step 1. Start from the brackets. Direct computation with $\{x_i,p_j\}=\delta_{ij}$ gives the three relations (6). Why: the first two only say $\mathbf L$ generates rotations and $\mathbf A$ is a vector; the third is the surprise — the six quantities close among themselves, the defining property of a Lie algebra.

Step 2. Absorb the energy. On the bound family $E<0$ , define $\mathbf D=\mathbf A/\sqrt{-2mE}$ ; then $\{D_i,D_j\}=\epsilon_{ijk}L_k$ with the other brackets unchanged. Why: the rescaling depends only on the conserved $E$ , so $\mathbf D$ is still a constant of the motion and the structure constants become pure numbers.

Step 3. Name the algebra. Six generators with these brackets form the rotation algebra of four-dimensional Euclidean space, $so(4)$ : $\mathbf L$ rotates our three dimensions, $\mathbf D$ rotates the fictitious fourth axis into the real three. Why: the count and the vector-pair structure under the $SO(3)$ subgroup match $SO(4)$ precisely.

Step 4. Split into two blocks. Define $\mathbf J^{\pm}=\tfrac12(\mathbf L\pm\mathbf D)$ : the brackets decouple into two commuting copies of $su(2)$ . Why: this is the isomorphism $so(4)\cong su(2)\oplus su(2)$ ; two independent "angular momenta" mean two independent quantum numbers.

Step 5. Harvest. For $E>0$ the rescaling $\mathbf A/\sqrt{2mE}$ flips a sign: the algebra becomes the Lorentz $so(3,1)$ ; at $E=0$ it contracts to Euclidean $e(3)$ . Quantum mechanically, the Casimirs of the two $su(2)$ s constrain $\hat A^2=m^2k^2+2m\hat H(\hat L^2+\hbar^2)$ , giving $E_n=-mk^2/2\hbar^2n^2$ with degeneracy $n^2$ . Why: Pauli ran exactly this argument in 1926 — the hydrogen spectrum from symmetry alone, before wave mechanics reached the atom. *As a check: the Casimir route reproduces the Bohr formula $E_n\propto-1/n^2$ with $n^2$ degeneracy.*

Limiting cases

Case	Result	Meaning
Kepler, $V=-k/r$	$\Delta\phi=\pi$ ; $\mathbf A$ conserved; $SO(4)$	Focus-centred ellipse; perihelion pinned.
Harmonic, $V=\tfrac12kr^2$	$\Delta\phi=\pi/2$ ; Fradkin tensor; $SU(3)$	Centre-centred ellipse; closes in half a revolution.
Power law $V=\alpha r^n$	$\Delta\phi=\pi/\sqrt{n+2}$ near-circular	Generically irrational multiple of $\pi$ : rosette.
Logarithmic, $V=\beta\ln r$	$\Delta\phi=\pi/\sqrt2\approx127.3°$	Flat-rotation-curve potential; never closes.
Inverse-cube boundary, $n\to-2$	bracket $\to0$ , $\Delta\phi\to\infty$	No stable circular orbit; trajectories spiral.

Worked examples

1 · The apsidal angle of a linear potential (confidence builder)

A particle orbits in the attractive linear potential $V(r)=\alpha r$ (constant inward force — the classic confinement model). Find the apsidal angle of near-circular orbits and decide whether the orbit closes.

Step 1. The potential is a power law with $n=1$ , so equation (2) applies directly: $\Delta\phi=\pi/\sqrt{n+2}=\pi/\sqrt3=180°/\sqrt3=\boxed{103.9°}$ .

Step 2. Successive perihelia are separated by $2\Delta\phi=207.8°$ — each radial cycle overshoots a half-turn by $27.8°$ : a pivoting rosette.

Step 3. Closure would need $1/\sqrt3$ rational, which it is not: the orbit never closes. After five radial cycles the perihelion has moved $5\times207.8°=1039.2°$ — two turns plus $319.2°$ , still not home.

2 · The Sun's rosette around the Milky Way (exam level)

The Galaxy's rotation curve is flat near the Sun: $v_c=235\ \mathrm{km/s}$ at $R_0=8.2\ \mathrm{kpc}$ , i.e. the logarithmic potential $V=mv_c^2\ln r$ . Find the apsidal angle, the two periods, the apsidal drift per radial cycle, and the number of radial oscillations in the Sun's 4.6-Gyr life.

Step 1. For $V=mv_c^2\ln r$ : $V'=mv_c^2/r$ , $V''=-mv_c^2/r^2$ , so $r_0V''/V'=-1$ and equation (2) gives $\Delta\phi=\pi/\sqrt{3-1}=\pi/\sqrt2=\boxed{127.3°}$ . Equivalently $\omega_r=\sqrt2\,\omega_\phi$ — galactic dynamicists call $\omega_r$ the epicyclic frequency $\kappa$ , and for a flat curve $\kappa=\sqrt2\,\Omega$ exactly.

Step 2. Azimuthal period: $T_\phi=2\pi R_0/v_c=2\pi\times(8.2\times3.086\times10^{19}\ \mathrm{m})/(2.35\times10^5\ \mathrm{m/s})=6.77\times10^{15}\ \mathrm{s}=\boxed{214\ \mathrm{Myr}}$ — the "galactic year".

Step 3. Radial period: $T_r=T_\phi/\sqrt2=214/1.414=\boxed{152\ \mathrm{Myr}}$ . The radius breathes faster than the Sun revolves.

Step 4. Per radial cycle the azimuth advances $2\Delta\phi=254.6°$ instead of $360°$ : the line of apsides regresses by $105.4°$ per radial period, and since $1/\sqrt2$ is irrational the orbit is a permanently open rosette.

Step 5. Radial oscillations completed: $4600/152\approx\boxed{30}$ — thirty radial breaths on an orbit that has never once closed. There is no conserved LRL vector in the Galaxy.

Origins · 1 minWolfgang Pauli and how this was discovered

Wolfgang Pauli

1900–1958 · Hamburg and Zurich

Pauli was quantum theory's most feared critic — "not even wrong" is his phrase — and one of its most incisive builders. In January 1926, before Schrödinger's wave equation had been applied to the atom, Pauli derived the complete hydrogen spectrum purely algebraically: he promoted the Laplace–Runge–Lenz vector to a quantum operator, showed that with angular momentum it closes into what we now call $SO(4)$ , and extracted $E_n\propto-1/n^2$ with its full $n^2$ degeneracy from the Casimirs alone. It was the first exact solution of a real quantum system and remains the model for solving problems by symmetry rather than differential equations. The exclusion principle (1925) and the neutrino hypothesis (1930) bracket the same decade.

History

1799

— Laplace identifies the conserved orbital vector that fixes the orientation of planetary ellipses.

1873

— Bertrand proves that only the inverse-square and linear force laws close all bounded orbits.

1926

— Pauli solves the hydrogen atom algebraically from the quantum LRL operator, exposing the

n^2

degeneracy as symmetry.

1935

— Fock shows hydrogen's bound states are rotations of a sphere in four dimensions:

SO(4)

made geometric.

1965

— Fradkin exhibits the conserved tensor and

SU(3)

structure of the isotropic oscillator, completing the pair.

Going further

References. Goldstein, Poole & Safko, Classical Mechanics (3rd ed.), §3.6 (Bertrand) and §3.9 (Runge–Lenz) · Landau & Lifshitz, Mechanics (Vol. 1), §14–15 and problems on the LRL vector · Morin, Introduction to Classical Mechanics, Ch. 7 · Taylor, Classical Mechanics, Ch. 8 · Bertrand, J. (1873), Comptes Rendus 77, 849 · Fradkin, D. M. (1965), Am. J. Phys. 33, 207 — hidden symmetry of the oscillator Key results to keep in mind: $\dot{\mathbf A}=0$ follows from the BAC–CAB identity for $V=-k/r$ ; the LRL magnitude satisfies $A^2=m^2k^2+2mEL^2$ with eccentricity $e=\sqrt{1+2EL^2/mk^2}$ ; projecting $\mathbf r\cdot\mathbf A$ yields the orbit equation directly; the near-circular apsidal angle is $\Delta\phi=\pi(3+rV''/V')^{-1/2}$ , giving $\pi/\sqrt{n+2}$ for power laws and confirming $\pi$ for Kepler and $\pi/2$ for the oscillator; the $SO(4)$ Poisson-bracket algebra closes after rescaling $\mathbf A$ by $\sqrt{-2mE}$ . Anchor numbers: Mercury perihelion advance $43''$ /century (GR); hydrogen $n^2$ degeneracy (Pauli 1926); $\pi/\sqrt3=103.9°$ ; $\pi/\sqrt2=127.3°$ ; $T_\phi=214$ Myr, $T_r=152$ Myr, apsidal regression $105.4°$ per radial cycle for $v_c=235$ km/s, $R_0=8.2$ kpc.

Builds on: The Effective Potential · The Orbit Equation and Binet's Equation · Angular Momentum Conservation in Central Fields · The Kepler Problem Leads to: Kepler's First Law: Elliptical Orbits · Conic-Section Orbits · Classification of Orbits by Energy

If the hidden symmetry never moves a single point of space, in what sense is it a symmetry at all?

In exactly the sense Noether's theorem cares about: it is a continuous family of canonical transformations that leave the Hamiltonian unchanged. Ordinary rotations act on positions and drag momenta along; the transformations generated by $\mathbf A$ mix positions and momenta into each other, so they move points of phase space while acting invisibly on any snapshot of real space. The conserved quantity attached to them is $\mathbf A$ itself, on the same footing as $\mathbf L$ is attached to rotations. The difference is only visibility: the hidden symmetry must be inferred from its consequences — orbits that close, perihelia that refuse to drift, spectra that stack degenerate.

Why exactly two special potentials — why not just Kepler, or twenty force laws?

The near-circular analysis (Derivation A) already narrows the field to power laws with $\sqrt{n+2}$ rational — still infinitely many candidates. What kills all but two is demanding closure at every amplitude: the higher-order expansion in eccentricity yields a selection rule only $n=-1$ and $n=2$ survive — the full argument is the subject of the Bertrand's theorem article. That the survivors come as a pair is no accident: the Bohlin–Arnold duality $z\mapsto z^2$ maps harmonic orbits onto Kepler orbits — one structure seen through two lenses.

How can hydrogen's electron know about closed classical orbits — there are no orbits in quantum mechanics?

The orbits do not survive quantization, but the symmetry does, and symmetry is what both phenomena express. Classically, the conserved $\mathbf A$ forbids the ellipse from precessing; quantum mechanically, the same operator commutes with the Hamiltonian and connects states of different $\ell$ within one shell, forcing 2s to sit exactly on 2p. One conserved object, two manifestations: closed orbits in the classical limit, $n^2$ -fold degeneracy in the spectrum. The correspondence is exact rather than an analogy — which is why screening, which classically makes orbits precess, quantum mechanically splits sodium's levels.

The oscillator's conserved object is a tensor while Kepler's is a vector — why the difference?

Trace it to where the ellipse sits. A Kepler ellipse has its force centre at a focus, which singles out one genuine direction — focus toward perihelion — so a conserved vector is the right object. The oscillator ellipse is centred on the force centre: the perihelion direction and its opposite are equivalent, since the orbit hits $r_{\min}$ twice per revolution. An object encoding an axis-without-arrow is precisely a symmetric tensor, whose eigenvectors define lines, not arrows. The Fradkin tensor's eigenvectors are the ellipse's principal axes, and its conservation pins them exactly as $\mathbf A$ pins the Kepler apsidal line.

What happens to the SO(4) symmetry when the orbit is unbound?

The algebra survives but changes character with the sign of the energy, because normalizing $\mathbf A$ involves $\sqrt{-2mE}$ . Bound orbits give compact $SO(4)$ , whose finite-dimensional representations produce the discrete $n^2$ -degenerate levels. Scattering orbits force the rescaling by $\sqrt{2mE}$ , flipping a sign in the brackets: the group becomes the non-compact Lorentz group $SO(3,1)$ , matching a continuous spectrum. At $E=0$ the algebra contracts to the Euclidean group $E(3)$ . One family of conserved quantities, three groups — the symmetry reads the energy before deciding what it is.

If the LRL vector is conserved, why does Mercury's perihelion precess at all?

Because the Sun's potential at Mercury is not exactly $-k/r$ : the other planets add their pulls, and general relativity adds a small $1/r^3$ term. Each deviation breaks the exact cancellation in $\dot{\mathbf A}=0$ , so $\mathbf A$ rotates slowly — about $532''$ per century from planetary perturbations and a further $43''$ per century from GR. The conservation law is exactly as fragile as the force law is exact, which makes precession a sensitive instrument; the apsidal angles and precession article works out these rates quantitatively.