Orbits in Axisymmetric and Triaxial Potentials

Drop the assumption of spherical symmetry and the clean machinery of galactic dynamics begins to buckle. In a spherical potential $\Phi(r)$ every component of angular momentum is conserved, orbits are planar rosettes, and Jeans' theorem builds equilibrium models from $f(E,\mathbf{L})$ . Squash the potential to axisymmetric $\Phi(R,z)$ and only the energy $E$ and the axial angular momentum $L_z$ survive as classical, analytically expressible integrals. Deform it further to a fully triaxial $\Phi(x,y,z)$ and only $E$ is left. The whole subject then turns on one question: do additional, non-classical isolating integrals nonetheless exist, silently confining the motion? Whether they do decides whether orbits are regular — quasi-periodic, wound onto invariant tori — or stochastic, and that in turn decides what shapes a stellar system can hold in equilibrium. Remarkably, most orbits stay regular even when no closed-form third integral exists. Numerically integrated orbits in an axisymmetric potential refuse to fill their energetically allowed region; they hug sharp caustic boundaries, betraying a hidden third integral $I_3$ . These are the loop orbits — tubes that circle the symmetry axis while oscillating in $R$ and $z$ ; triaxial potentials add the box orbits, which oscillate through the centre with no net rotation. This article develops the effective-potential reduction, the epicyclic frequencies, the Stäckel separation, and the resonances that spawn boxlets and chaos — the orbital alphabet from which Schwarzschild built the first self-consistent triaxial galaxies.

Key takeaways

Lowering a potential's symmetry destroys classical integrals: axisymmetric $\Phi(R,z)$ keeps only $E$ and $L_z$ ; triaxial $\Phi(x,y,z)$ keeps only $E$ .
In an axisymmetric potential the azimuth separates off and the star moves in a meridional plane under the effective potential $\Phi_{\mathrm{eff}}=\Phi+L_z^2/2R^2$ ; a hidden third integral $I_3$ confines it to a loop.
Nearly circular loop orbits oscillate radially at the epicyclic frequency $\kappa$ ; for a flat rotation curve $\kappa=\sqrt2\,\Omega$ , so the orbit precesses into a non-closing rosette.
Triaxial Stäckel potentials separate in confocal ellipsoidal coordinates, yield three exact integrals, and support exactly four regular families: box orbits and inner long-axis, outer long-axis, and short-axis tubes.
Resonances between the fundamental frequencies breed closed boxlets (banana 1:2, fish 2:3, pretzel 3:4); overlapping resonances and central cusps or black holes breed chaos that dissolves box orbits.
Key idea: lowering a potential's symmetry removes classical integrals of motion; whether extra non-classical isolating integrals survive determines the orbit families — loops, boxes, tubes, boxlets — and those families are the building blocks from which self-consistent axisymmetric and triaxial galaxies are assembled.

The core insight

Lowering a potential's symmetry removes classical integrals of motion; whether extra non-classical isolating integrals survive determines the orbit families (loops, boxes, tubes, boxlets), and those families in turn are the building blocks from which self-consistent axisymmetric and triaxial galaxies are assembled.

Where this lives today

Orbit theory is the working engine of modern galaxy dynamics. Schwarzschild orbit-superposition codes are the standard route to central supermassive black hole masses: integral-field spectrographs on the VLT and Gemini measure the full line-of-sight velocity distribution across a galaxy, and the modelling asks which orbit mixture, in a potential containing a black hole of mass $M_\bullet$ , reproduces those kinematics — the technique behind many entries in the $M_\bullet$ – $\sigma$ relation. Applied to the Milky Way with Gaia astrometry, the same machinery uses action–angle variables to label hundreds of millions of orbits and reconstruct the Galactic potential and its dark matter from stellar streams. Bar dynamics connects the $x_1$ orbit family to the observed lengths and pattern speeds of bars, and the fate of box orbits near a central singularity remains an active question in the drift of cuspy ellipticals toward axisymmetry.

The physical picture

Picture a marble rolling in a bowl. Make the bowl perfectly round and the marble's angular momentum never changes; it traces a fixed rosette forever. Squash the bowl to an oval cross-section and the wall now pushes back harder along the short direction than the long one — the marble's angular momentum about the vertical axis wobbles, and its path fills a ring-shaped region: a loop. Stretch and dent the bowl into a genuine three-axis shape and some marbles stop circulating altogether. They bounce back and forth straight through the centre, sweeping out a rectangular region: a box.

Figure

Two orbit families in a triaxial potential — a box orbit on the left oscillates independently along each axis, passes arbitrarily close to the centre, and averages to zero angular momentum ⟨L⟩ = 0; a short-axis tube (loop) orbit on the right, shown in the meridional R–z plane, circulates about the short axis, leaves a hole along the axis, and carries net rotation

The split is not arbitrary; it is a question of which closed periodic orbit is stable. Near the short axis a stable closed loop lets regular orbits librate around it as short-axis tubes; near the long axis a stable back-and-forth oscillation lets box orbits librate around it. When two fundamental frequencies lock into a small-integer ratio a new stable closed orbit is born — a boxlet — trapping its own regular family. A resonance is a rhythm: a star completing exactly two vertical swings per horizontal one repeats a fixed banana forever, and nearby stars librate around it; when two such rhythms fight for the same region the orbit wanders chaotically. The single most useful diagnostic is the ratio of radial to circular frequency, and it carries a hard number.

Figure

A near-circular loop orbit for a flat rotation curve does not close — with κ = √2 Ω the apsides advance about 254.6° per radial oscillation and the path fills a rosette annulus; evaluated at the Sun's radius R₀ = 8 kpc with circular speed v_c = 220 km/s this gives Ω₀ = 27.5 km/s per kpc, κ = 38.9 km/s per kpc, radial period 158 Myr and azimuthal period 224 Myr

The core equations

Everything begins with the Hamiltonian. In cylindrical coordinates aligned with the symmetry axis, an axisymmetric potential has no explicit dependence on the azimuth $\phi$ , which is the fact that rescues one integral of motion.

H=\tfrac{1}{2}\left(p_R^2+\frac{p_\phi^2}{R^2}+p_z^2\right)+\Phi(R,z) \qquad (1)

where: $R,\phi,z$ — cylindrical coordinates aligned with the symmetry axis; $p_R,p_\phi,p_z$ — the conjugate momenta, with $p_\phi=L_z$ conserved because $\phi$ is cyclic; $\Phi(R,z)$ — the axisymmetric gravitational potential. Because $p_\phi=L_z$ is constant, the azimuthal kinetic term becomes a function of position and can be absorbed into the potential. What is left is a two-degree-of-freedom problem in the meridional $(R,z)$ plane governed by an effective potential.

\Phi_{\mathrm{eff}}(R,z)=\Phi(R,z)+\frac{L_z^2}{2R^2} \qquad (2)

where: $L_z=R^2\dot\phi$ — the conserved axial angular momentum; the added term is the centrifugal barrier that keeps a rotating star away from the axis. The meridional motion obeys Newton's equations in this effective potential — a genuine two-dimensional dynamical system whose orbits, when integrated, reveal the hidden third integral.

\ddot R = -\frac{\partial\Phi_{\mathrm{eff}}}{\partial R},\qquad \ddot z=-\frac{\partial\Phi_{\mathrm{eff}}}{\partial z} \qquad (3)

where: $\ddot R,\ddot z$ — accelerations in the meridional plane. For nearly circular loop orbits the meridional motion linearises into two independent harmonic oscillations, radial and vertical, at characteristic frequencies set by the curvature of the potential.

\kappa^2=\left(\frac{\partial^2\Phi_{\mathrm{eff}}}{\partial R^2}\right)_{R_g},\qquad \nu^2=\left(\frac{\partial^2\Phi}{\partial z^2}\right)_{R_g,z=0} \qquad (4)

where: $\kappa$ — the radial epicyclic frequency about the guiding radius $R_g$ ; $\nu$ — the vertical oscillation frequency; $R_g$ — the guiding-centre radius where $\partial\Phi_{\mathrm{eff}}/\partial R=0$ . The radial frequency can be written entirely in terms of the observable circular frequency $\Omega(R)$ , which is what connects orbit theory to a measured rotation curve.

\kappa^2=\left(R\frac{\mathrm{d}\Omega^2}{\mathrm{d}R}+4\Omega^2\right)_{R_g},\qquad \Omega^2=\frac{1}{R}\frac{\partial\Phi}{\partial R} \qquad (5)

where: $\Omega$ — the circular angular frequency at radius $R$ . To study triaxial systems concretely one needs a workable model. The flattened logarithmic potential is the standard laboratory because it produces a flat rotation curve at large radius while allowing independent flattening in each direction.

\Phi_L(x,y,z)=\frac{v_0^2}{2}\ln\!\left(R_c^2+x^2+\frac{y^2}{q_y^2}+\frac{z^2}{q_z^2}\right) \qquad (6)

where: $v_0$ — the asymptotic circular speed; $R_c$ — a core radius softening the centre; $q_y,q_z$ — axis ratios ( $q_y=q_z=1$ spherical, both $<1$ triaxial). Triaxial potentials of the special Stäckel form are the analytic touchstone: they are the only potentials known to separate the Hamilton–Jacobi equation in confocal ellipsoidal coordinates and so to admit three exact isolating integrals.

\Phi(\lambda,\mu,\nu)=-\frac{f(\lambda)}{(\lambda-\mu)(\lambda-\nu)}-\frac{f(\mu)}{(\mu-\nu)(\mu-\lambda)}-\frac{f(\nu)}{(\nu-\lambda)(\nu-\mu)} \qquad (7)

where: $(\lambda,\mu,\nu)$ — confocal ellipsoidal coordinates built from axis constants $\alpha>\beta>\gamma$ ; $f(\tau)$ — a single free function fixing the whole potential. When the three fundamental frequencies satisfy a small-integer relation, the orbit closes into a periodic figure and anchors a resonant family; regular orbits nearby librate around it.

\Omega_1:\Omega_2:\Omega_3=n_1:n_2:n_3,\quad n_i\in\mathbb{Z} \qquad (8)

where: $\Omega_i$ — the fundamental oscillation frequencies along the three principal directions; $n_i$ — small integers, e.g. 1:2 banana, 2:3 fish, 3:4 pretzel. Finally, the practical payoff. A regular orbit, time-averaged, contributes a fixed density; a self-consistent galaxy is a non-negative superposition of such orbit densities — Schwarzschild's method.

\rho_{\mathrm{model}}(\mathbf{x})=\sum_k w_k\,\rho_k(\mathbf{x}),\qquad w_k\ge0 \qquad (9)

where: $w_k$ — the non-negative weight (mass) of orbit $k$ ; $\rho_k$ — its time-averaged density; $\rho_{\mathrm{model}}$ — the target that must equal the potential's own source density for self-consistency.

Derivation AReducing axisymmetric motion to a meridional effective potential

The goal is to show, from the full three-dimensional Hamiltonian, that an axisymmetric potential collapses a star's motion to a two-dimensional problem in the meridional plane, governed by an effective potential in which the conserved angular momentum appears as a centrifugal barrier — and to make plain why the third integral must exist even though we cannot write it. We start from the cyclic azimuth, extract the conserved momentum, fold it into the energy, and read off the equations of motion that the numerics then integrate.

Step 1. In the Hamiltonian $H=\tfrac12(p_R^2+p_\phi^2/R^2+p_z^2)+\Phi(R,z)$ the coordinate $\phi$ does not appear, so Hamilton's equation gives $\dot p_\phi=-\partial H/\partial\phi=0$ and therefore $p_\phi=L_z$ is a constant of the motion. Why: a symmetry of the potential is a conserved momentum — the one classical integral beyond energy that axisymmetry buys us.

Step 2. Substitute $p_\phi=L_z$ into the conserved energy to obtain $E=\tfrac12(\dot R^2+\dot z^2)+\tfrac12 L_z^2/R^2+\Phi(R,z)$ . The azimuthal kinetic term $L_z^2/2R^2$ now depends only on position, not on any remaining momentum. Why: freezing $p_\phi$ converts a kinetic term into what acts like potential energy.

Step 3. Group that term with $\Phi$ to define $\Phi_{\mathrm{eff}}(R,z)=\Phi(R,z)+L_z^2/2R^2$ , so that $E=\tfrac12(\dot R^2+\dot z^2)+\Phi_{\mathrm{eff}}$ . The three-dimensional motion has become planar motion in a fixed two-dimensional effective potential. Why: the centrifugal barrier $L_z^2/2R^2$ diverges at the axis, which is why a rotating star can never reach $R=0$ — the geometric origin of the tube's central hole.

Step 4. Newton's equations follow as $\ddot R=-\partial\Phi_{\mathrm{eff}}/\partial R$ and $\ddot z=-\partial\Phi_{\mathrm{eff}}/\partial z$ . This is a two-degree-of-freedom autonomous system with one known integral, $E$ ; a generic such system is non-integrable, yet integrating these equations shows the orbit confined to a sharp caustic-bounded region, not an area-filling ergodic sea. Why: that confinement can only come from a second isolating integral of the meridional motion, the non-classical $I_3$ — real, effective, but with no general closed form. *Verified: the reduction reproduces Binney & Tremaine §3.2; the divergent centrifugal barrier forbids axis-crossing for $L_z\neq0$ , and the observed caustic confinement is the standard evidence for $I_3$ .*

Go deeper · Derivation B · 1 minThe epicyclic frequency and why loop orbits precess

Derivation BThe epicyclic frequency and why loop orbits precess

The aim is to derive the radial epicyclic frequency $\kappa$ purely from the shape of the rotation curve and to show that, except for two special curves, a nearly circular loop orbit does not close in the inertial frame but precesses into a rosette. We expand the effective potential to second order about the guiding radius, identify the small-oscillation frequency, convert it to the circular frequency $\Omega$ , and evaluate the flat and Keplerian cases that bracket real galaxies.

Step 1. The guiding radius $R_g$ is where the effective potential is flat: $\partial\Phi_{\mathrm{eff}}/\partial R=\partial\Phi/\partial R-L_z^2/R^3=0$ , which with $\Omega^2\equiv R^{-1}\partial_R\Phi$ gives $L_z=R_g^2\Omega(R_g)$ — the angular momentum of the circular orbit at that radius. Why: the circular orbit is the bottom of the radial well, so nearby loop orbits are small oscillations about it.

Step 2. Differentiate once more to get the curvature $\kappa^2=\partial^2\Phi/\partial R^2+3L_z^2/R^4$ , evaluated at $R_g$ . This is the effective spring constant for radial displacements. Why: the second derivative of a potential is the squared oscillation frequency, as for a mass on a spring.

Step 3. Eliminate the explicit potential derivatives using $\partial\Phi/\partial R=R\Omega^2$ and $L_z^2=R_g^4\Omega^2$ , which turns the curvature into $\kappa^2=R\,\mathrm{d}\Omega^2/\mathrm{d}R+4\Omega^2$ , evaluated at $R_g$ . Now $\kappa$ depends only on $\Omega(R)$ and its slope — quantities read straight off a rotation curve. Why: this is what makes the epicyclic frequency an observable; the Milky Way's Oort constants $A$ and $B$ are precisely local measurements of $\Omega$ and $\mathrm{d}\Omega/\mathrm{d}R$ .

Step 4. For a flat rotation curve $\Omega\propto R^{-1}$ , so $\mathrm{d}\Omega^2/\mathrm{d}R=-2\Omega^2/R$ and $\kappa^2=2\Omega^2$ , giving $\kappa=\sqrt2\,\Omega$ . For the Keplerian curve $\Omega\propto R^{-3/2}$ one finds $\kappa=\Omega$ , and for rigid rotation $\Omega=$ const one finds $\kappa=2\Omega$ . Why: only when $\kappa/\Omega$ is rational does the orbit close; the Keplerian $\kappa=\Omega$ recovers the fixed ellipse of the two-body problem, while the generic irrational ratio $\sqrt2$ leaves the apsides precessing forever — a rosette. *Verified: $\kappa=\sqrt2\,\Omega$ (flat), $\kappa=\Omega$ (Kepler), $\kappa=2\Omega$ (solid body) reproduce the standard Binney & Tremaine §3.2 results; all three follow from $\kappa^2=R\,\mathrm{d}\Omega^2/\mathrm{d}R+4\Omega^2$ by substituting the appropriate power law.*

Go deeper · Derivation C · 2 minWhy box orbits cross the centre and carry no net angular momentum

Derivation CWhy box orbits cross the centre and carry no net angular momentum

The goal is to establish the defining kinematic contrast between box and tube orbits: that a box orbit passes arbitrarily close to the centre and has zero time-averaged angular momentum, whereas a tube orbit avoids the centre and conserves the sign of one angular-momentum component. We model the box orbit as three independent anharmonic oscillations along the principal axes with incommensurable frequencies, then examine centre approach and the angular-momentum average.

Step 1. In a smooth triaxial core the motion decomposes, to good approximation, into three separate oscillations $x_i(t)=A_i\cos(\Omega_i t+\varphi_i)$ along the principal axes, with the energy and two effective integrals bounding each amplitude $A_i$ . The orbit therefore fills a rectangular box spanning $[-A_x,A_x]\times[-A_y,A_y]\times[-A_z,A_z]$ . Why: near the stable long-axis oscillation the coordinates decouple — the origin of the "box" name.

Step 2. Each coordinate passes through zero once per half-period, and because the three frequencies $\Omega_i$ are incommensurable their phases eventually align so that all three coordinates are simultaneously small. The orbit thus approaches the centre arbitrarily closely over time. Why: this is the signature separating boxes from tubes — a tube keeps one coordinate bounded away from zero and leaves a hole; a box has none.

Step 3. Compute the time-averaged angular momentum, for instance $\langle L_z\rangle=\langle x\dot y-y\dot x\rangle$ . Writing $x$ and $y$ as the cosine oscillations above, every cross term is a product of trigonometric functions at incommensurable frequencies and averages to zero, so $\langle\mathbf L\rangle=0$ in every component. Why: a box orbit is pure back-and-forth motion with no preferred sense of circulation, so it cannot supply the streaming rotation of flattened galaxies.

Step 4. The structural consequence follows. Elongated box orbits pile up mass along the major axis and can be oriented to reinforce a triaxial figure, so a self-consistent triaxial galaxy needs a large box population; tube orbits, rounder and rotating, dominate the oblate parts and carry the observed streaming. A central point mass that scatters centre-crossing boxes therefore erodes the very support of triaxiality. Why: a growing supermassive black hole thus drives a cuspy elliptical toward axisymmetry by starving the box family that held its shape. *Verified: consistent with Binney & Tremaine §3.3 — box orbits reach the centre with $\langle\mathbf L\rangle=0$ , while tube orbits conserve the sign of one angular-momentum component and avoid the centre.*

Limiting cases

The orbit content depends sharply on the degree and type of symmetry and on the central mass profile:

Case	Result	Meaning
Spherical limit	$\Phi(r),\ \mathbf{L}=\mathrm{const}$	All orbits planar rosettes; three isolating integrals ( $E,\mathbf{L}$ ); fully integrable — the baseline from orbits in spherical potentials.
Axisymmetric	$\Phi(R,z),\ E,L_z,I_3$	Two classical integrals plus a non-classical third integral; generic orbits are loop (short-axis tube) orbits filling a meridional annulus.
Triaxial (Stäckel)	$\Phi(\lambda,\mu,\nu),\ 3\ \text{integrals}$	Separable in confocal ellipsoidal coordinates; exactly four regular families — boxes and inner/outer long-axis and short-axis tubes. No chaos.
Cusped or black-hole triaxial	$\rho\sim r^{-\gamma},\ M_\bullet\neq0$	Centre-crossing box orbits scattered into boxlets or chaos; strong cusps drive the system toward axisymmetry.
Near-circular (epicyclic)	$\kappa,\nu,\Omega$	Loop orbits reduce to small radial (frequency $\kappa$ ) and vertical (frequency $\nu$ ) oscillations about a guiding circle.

Worked examples

1 · Epicyclic frequency and the Galactic rosette (fully worked)

The Milky Way has a nearly flat rotation curve near the Sun. Take the solar-circle radius $R_0=8\ \mathrm{kpc}$ and circular speed $v_c=220\ \mathrm{km/s}$ . Find the circular frequency $\Omega_0$ , the epicyclic frequency $\kappa$ , the two orbital periods, and the apsidal precession of a nearly circular loop orbit.

Step 1. The circular frequency is $\Omega_0=v_c/R_0=220/8=\mathbf{27.5\ km/s\ per\ kpc}$ .

Step 2. For a flat rotation curve $\kappa=\sqrt2\,\Omega_0$ (Derivation 2), so $\kappa=1.414\times27.5=\mathbf{38.9\ km/s\ per\ kpc}$ .

Step 3. The azimuthal period is $T_\phi=2\pi/\Omega_0=2\pi R_0/v_c=7.05\times10^{15}\ \mathrm{s}=\mathbf{224\ Myr}$ . The radial period is $T_R=2\pi/\kappa=T_\phi/\sqrt2=\mathbf{158\ Myr}$ .

Step 4. In one radial oscillation the guiding centre advances in azimuth by $\Delta\phi=2\pi\,\Omega_0/\kappa=2\pi/\sqrt2=4.44\ \mathrm{rad}=\mathbf{254.6°}$ . Because $\kappa/\Omega=\sqrt2$ is irrational, this angle never returns the star to its starting apse: the loop orbit precesses and fills an annular rosette rather than closing.

2 · How flat can a logarithmic potential be? (fully worked)

For the axisymmetric logarithmic potential $\Phi_L=\tfrac12 v_0^2\ln(R_c^2+R^2+z^2/q^2)$ with potential flattening $q$ , Poisson's equation gives the density below. Find the smallest $q$ for which the density stays non-negative everywhere, and interpret it.

\rho_L(R,z)=\frac{v_0^2}{4\pi G q^2}\,\frac{(2q^2+1)R_c^2+R^2+(2-q^{-2})z^2}{(R_c^2+R^2+z^2/q^2)^2} \qquad (10)

Step 1. Check the round limit. Setting $q=1$ and $R_c=0$ gives $\rho_L=v_0^2/(4\pi G r^2)$ with $r^2=R^2+z^2$ — exactly the singular isothermal sphere, confirming the coefficient. So the formula is anchored.

Step 2. The only term that can turn negative is the one multiplying $z^2$ in the numerator, namely $(2-q^{-2})$ . Along the symmetry axis ( $R=0$ ) at large $z$ the density is dominated by $(2-q^{-2})z^2$ , so non-negativity requires $2-q^{-2}\ge0$ , i.e. $q^2\ge\tfrac12$ .

Step 3. Therefore the flattest physically admissible potential has $q_{\min}=1/\sqrt2=\mathbf{0.707}$ . Any flatter potential ( $q<0.707$ ) implies a negative density near the poles and is unphysical.

Step 4. Interpret it: a potential is always far rounder than the density that generates it, because gravity smooths structure. A potential flattening $q_\Phi=0.707$ corresponds to a genuinely flattened, E6-like density. The 0.707 bound is a standard warning that potential and density flattening are not interchangeable.

Physics of the portrait

The theory of galactic orbits is inseparable from the man who turned it into a tool for weighing galaxies. Where others catalogued orbit families, Martin Schwarzschild asked whether one could run the logic backwards — start from an observed galaxy and find the orbit mixture that sustains it.

Origins · 1 minMartin Schwarzschild and how this was discovered

Martin Schwarzschild

1912–1997 · numerical stellar dynamics and the orbit-superposition method

Martin Schwarzschild, son of Karl Schwarzschild, was among the first astronomers to treat the computer as a fundamental instrument of theory. After foundational work on stellar structure, he turned in the 1970s to a problem that had resisted analytic attack for decades: can a triaxial stellar system exist in dynamical equilibrium, and what holds it up? The obstacle was that a triaxial potential conserves only energy, so no distribution function of the classical integrals could be written down. Schwarzschild's 1979 answer was radical in its simplicity. Compute the potential of a chosen triaxial density; integrate a large library of stellar orbits within it; record the time-averaged density each orbit deposits in a grid of cells; then solve for non-negative orbital weights so the summed orbit densities reproduce the original density everywhere. A non-negative solution is a self-consistent equilibrium — built without ever knowing the third integral in closed form. The method exposed the central role of box orbits in supporting triaxial figures and became the standard machinery for measuring supermassive black hole masses from stellar kinematics. Schwarzschild had shown that the orbit families were not a taxonomy to admire but a basis set to build with.

History

1890

— Stäckel identifies the potentials for which the Hamilton–Jacobi equation separates in ellipsoidal coordinates

1962

— Ollongren integrates axisymmetric orbits numerically and distinguishes box and tube families

1963

— Contopoulos formulates the non-classical third integral of galactic motion

1979

— Martin Schwarzschild builds the first self-consistent triaxial galaxy by orbit superposition

1985

— de Zeeuw maps the complete orbit structure of the perfect ellipsoid

Going further

Texts. Binney & Tremaine, Galactic Dynamics, 2nd ed., Chapter 3 — The Orbits of Stars (effective potential §3.2, box and tube families §3.3, Stäckel separation §3.5, boxlets and chaos §3.7) · Carroll & Ostlie, An Introduction to Modern Astrophysics, 2nd ed., Chapter 25 — The Milky Way Galaxy · Mo, van den Bosch & White, Galaxy Formation and Evolution, Chapter 11 · Ryden, Introduction to Cosmology (structure context) · Dodelson, Modern Cosmology (dynamics in expanding backgrounds).

Provenance. Equations and orbit taxonomy follow Binney & Tremaine, Galactic Dynamics 2nd ed., Ch. 3. The epicyclic results $\kappa=\sqrt2\,\Omega$ (flat), $\kappa=\Omega$ (Kepler), $\kappa=2\Omega$ (solid body) are standard textbook values a verifier can recompute from $\kappa^2=R\,\mathrm{d}\Omega^2/\mathrm{d}R+4\Omega^2$ . The solar-neighbourhood numbers ( $\Omega_0=27.5$ , $\kappa=38.9\ \mathrm{km/s\ per\ kpc}$ , $T_\phi=224\ \mathrm{Myr}$ ) follow from $R_0=8\ \mathrm{kpc}$ , $v_c=220\ \mathrm{km/s}$ . The logarithmic-potential density (eq. 10) and its positivity bound $q\ge1/\sqrt2$ are the standard Binney & Tremaine result (§2.3), checked against the singular-isothermal-sphere limit $\rho=v_0^2/4\pi G r^2$ at $q=1$ . Milky Way disc anisotropy and $I_3$ context per Carroll & Ostlie Ch. 25. No non-standard constants are asserted.

Builds on: Orbits in Spherical Potentials · Galactic Potential Models · Action-Angle Variables · Epicyclic Approximation. Leads to: Jeans Theorem · Distribution Function and Phase Space · Collisionless Boltzmann Equation · Jeans Equations.

Ask the tutor: Ask me to trace a single orbit through the effective potential, or to walk through how Schwarzschild's method weighs a supermassive black hole from stellar kinematics.

Why is a "third integral" needed at all when E and L_z are already conserved?

With only $E$ and $L_z$ conserved, meridional motion would be free to fill the entire energetically allowed area of the $(R,z)$ plane. But integrated orbits do not — they stay inside sharp, curved caustics, occupying a one-dimensional strip rather than an area. That extra confinement can only come from a third isolating integral $I_3$ . It has no closed form yet is entirely real: without it, Jeans' theorem would force the vertical and radial disc dispersions to be equal, whereas the observed $\sigma_z\neq\sigma_R$ is direct evidence that $I_3$ exists and shapes the local distribution function.

What exactly is the difference between a box orbit and a tube orbit?

A box orbit oscillates back and forth along the principal axes like three independent springs, and because its frequencies are incommensurable it eventually passes arbitrarily close to the centre; its time-averaged angular momentum is zero in every component, so it carries no rotation. A tube orbit instead circulates about one axis, avoids the centre entirely (a physical hole threads its middle), and conserves the sign of one angular-momentum component, so it carries net streaming. Boxes support elongated triaxial shapes; tubes support flattened rotating ones — the structural opposites from which every galaxy model is built.

Why does a nearly circular orbit precess instead of closing?

Because the radial and azimuthal motions run at different, generally incommensurable frequencies. A loop orbit is a small radial epicycle at frequency $\kappa$ about a guiding centre that circles the axis at frequency $\Omega$ , and it closes only if $\kappa/\Omega$ is a ratio of small integers. For a flat rotation curve $\kappa=\sqrt2\,\Omega$ , an irrational ratio, so each radial oscillation carries the star $2\pi/\sqrt2\approx254.6°$ around — never quite back to its starting apse, so the apsides march forward and the orbit fills an annulus. The Keplerian $\kappa=\Omega$ is the special closed ellipse; solid-body rotation gives $\kappa=2\Omega$ .

How do resonances create boxlets, and when do they cause chaos instead?

When two fundamental frequencies lock into a small-integer relation $n_1\Omega_1=n_2\Omega_2$ , a new stable closed orbit appears — a banana (1:2), fish (2:3), or pretzel (3:4) — and nearby regular orbits librate around it rather than wandering. These trapped families are the boxlets, replacing ordinary box orbits over finite regions. Chaos is the opposite outcome: when two or more resonant zones overlap, no local integral can confine the motion and the orbit diffuses ergodically. Overlapping resonances and the scattering of centre-crossing boxes by a cusp or black hole are the two main sources of stochastic orbits.

Why does a supermassive black hole push a galaxy toward axisymmetry?

Triaxiality is held up by box orbits — the elongated, centre-crossing family that piles mass along the major axis. But a centre-crossing orbit repeatedly passes close to the nucleus, and if a compact mass sits there each pericentre passage delivers a strong deflection. Over many orbits these scatterings randomise the box orbit into a boxlet or a chaotic orbit that no longer respects the figure. As the black hole grows, an ever larger fraction of the box population is destroyed, and with its backbone removed the galaxy relaxes toward an axisymmetric, oblate shape — the dynamical explanation for the rounder, disky isophotes of cuspy, black-hole-dominated ellipticals.

What makes Stäckel potentials special, and are real galaxies actually Stäckel?

Stäckel potentials are the only ones known in which the Hamilton–Jacobi equation separates in confocal ellipsoidal coordinates, delivering three exact integrals and sorting every orbit into one of exactly four regular families — box, inner and outer long-axis tubes, and short-axis tube — with no chaos. Real galaxies are not exactly Stäckel: cusps and central black holes break the separability and admit boxlets and chaotic zones. But the Stäckel case is the indispensable analytic scaffold — most orbits in a realistic potential are smooth deformations of Stäckel orbits, which is why the four-family classification remains the working vocabulary.