Three-Body Lab · Guide · glidepathscience.org
GuideThree-Body Lab
Contents
This Guide explains the ideas and equations behind every Walkthrough step, Concept Challenge question and Predict round, in a textbook's order. Part 1 prepares the Intro track and Part 2 the Advanced track; each takes about twenty minutes to read. Each section ends with a worked example that is not one of the sim's own problems.
Part 1 · The essentials
Math you’ll need: algebra, square roots, adding vectors tip to tail, and exponentials and natural logarithms; no calculus.
Two bodies and their center of mass
Gravity comes in pairs: the Moon pulls the Earth as hard as the Earth pulls the Moon, so neither sits still while the other moves. Newton solved this two-body problem completely in 1687, and every harder problem in this sim is measured against his solution.
The key is the center of mass, the mass-weighted average of the positions. For two bodies of masses $m_1$ and $m_2$ on a line,
$x_{cm} = (m_1 x_1 + m_2 x_2)/(m_1 + m_2)$ $m_1 r_1 = m_2 r_2$
where $r_1$ and $r_2$ are the bodies' distances from it. The forces between the bodies are equal and opposite, so the total momentum never changes and the center of mass moves at constant velocity; in its own frame it stands still. Each body follows a conic section, a circle, ellipse, parabola or hyperbola, with the center of mass at a focus. The two ellipses have the same eccentricity and stay on opposite sides of the center of mass, and their sizes are in the inverse ratio of the masses: the lighter body swings on the larger ellipse.
The usual mistake is to put the center of mass nearer the lighter body, or halfway between. It sits nearer the heavier one, because $r_1/r_2 = m_2/m_1$.
Knowing where the bodies circle, we can ask how fast they must move to do it.
Circular and escape speeds
A body on a circular orbit is falling all the time; it just keeps missing. Gravity supplies exactly the centripetal force a circle of radius r needs, $GMm/r^2 = mv^2/r$, and the orbiting mass m cancels. That leaves the circular speed, which depends only on the central mass M and the distance r:
$v_c = \sqrt{GM/r}$ $v_{esc} = \sqrt{2GM/r}$
Strictly, the two bodies circle their common center, and M should be the total mass; for the Earth around the Sun the difference is a few parts in a million, for Jupiter five parts in ten thousand. The second formula is the escape speed. The total energy of the orbiting body is $E = \tfrac{1}{2}mv^2 - GMm/r$. While E is negative the body is bound and its orbit closes; at E = 0 it can just coast away to infinity, which happens at $v_{esc} = \sqrt{2}\,v_c$. Launched sideways at k times the circular speed, a body follows an ellipse for k below √2, a parabola at √2 and a hyperbola beyond.
Two mistakes are common. One is forgetting the √2, which gives the circular speed again. The other is measuring r from the surface instead of from the center: an altitude is not a distance from the center.
A circle has one speed. An ellipse has many, and the next section finds the rule that links them.
Equal areas, from angular momentum
Kepler's second law says that the line from the Sun to a planet sweeps out equal areas in equal times. It is conservation of angular momentum in disguise. Gravity points straight at the Sun, so it exerts no torque about the Sun, and the angular momentum per unit mass, $r v_\perp$, stays fixed, where $v_\perp$ is the part of the velocity perpendicular to the line to the Sun. In a short time Δt the line sweeps a thin triangle of area $\tfrac{1}{2} r v_\perp\,\Delta t$, so the area swept per unit time is constant too. At perihelion (closest) and aphelion (farthest) the velocity is exactly perpendicular to the radius, and the law takes its simplest form:
$r_p v_p = r_a v_a$ $\Delta A/\Delta t = \tfrac{1}{2} r v_\perp$
The speed falls in proportion to the distance, not its square; the square belongs to the force. Kepler's third law completes the set: for any ellipse of semi-major axis a, half the long axis, the period is $T^2 = 4\pi^2 a^3/GM$, however stretched the ellipse, so orbits with the same a have the same period. Energy conservation gives the speed anywhere on the ellipse, the vis-viva relation $v^2 = GM(2/r - 1/a)$; on a circle r = a, and it gives $v_c$ again.
With two bodies everything closes neatly. Add a third, and it does not.
Three bodies: no formula, and bound is not safe
Three bodies obey the same law of gravity as two. Each feels the sum of the other two pulls, each of size $Gm_im_j/r_{ij}^2$ and aimed at the body pulling, and energy, momentum and angular momentum are conserved exactly. What fails is the formula. In a plane, two bodies need 8 numbers, their positions and velocities, and the conservation laws fix all but one, the place along the orbit. Three bodies need 12 numbers, and the conservation laws still fix only 6. Poincaré proved in 1890 that no further conserved quantities come to the rescue and that the motion is usually chaotic, so no formula exists. The motion is found the way this sim finds it, by a computer stepping it forward a short time at a time.
The energy of three bodies at rest is all potential energy, a sum over the three pairs:
$E = K + U$ $U = -G\left(\frac{m_1 m_2}{r_{12}} + \frac{m_1 m_3}{r_{13}} + \frac{m_2 m_3}{r_{23}}\right)$
A negative E means the three cannot all fly apart, but not that all three stay. Two of them can fall into a tight binary, whose own energy $-Gm_im_j/2a$ grows more negative the closer the pair, without limit, and the energy the binary gives up can carry the third body away for good. Reading "E < 0" as "nobody escapes" is the mistake.
If three bodies are usually chaotic, are there any arrangements that are not? Lagrange found one.
Lagrange's triangle, and whether it survives
In 1772 Lagrange showed that any three masses at the corners of an equilateral triangle can turn together rigidly, the triangle keeping its shape and size. At each corner the pulls of the other two add to a force aimed exactly at the center of mass, and exactly as strong as circular motion around it needs. The rotation rate depends only on the side a and the total mass M:
$\omega^2 = GM/a^3$ $T = 2\pi/\omega$
This is Kepler's third law again, with the total mass. Using one mass instead of M is the usual slip; it makes the period too long by $\sqrt{M/m}$. An exact solution is useful only if small nudges stay small, and Gascheau found in 1843 that Lagrange's triangle is stable against small nudges only if, with $M = m_1 + m_2 + m_3$,
$27(m_1 m_2 + m_2 m_3 + m_3 m_1) < M^2$
The left side is large unless one mass dominates the other two. An unstable triangle tears apart within a few turns of a tiny nudge: an exact solution can be as useless as a pencil balanced on its point.
That third body, tiny beside the other two, is the starting point of the next sections.
The rotating frame: five points, passes and hilltops
When the third body is too light to move the other two, the problem simplifies: the two large bodies, the primaries, circle each other as in the first section, and only the small body's motion is unknown. This is the restricted three-body problem, best watched in a frame that turns with the primaries, where they stand still. Their mass ratio μ is the smaller primary's share of the total.
A turning frame adds two fictitious forces: a centrifugal push outward from the axis, and the Coriolis force, a sideways push at right angles to a body's velocity in the frame, of size 2mωv for a body moving at v in a frame turning at ω. The centrifugal push and the two gravities make a landscape of potential energy, with deep wells at the primaries and five flat spots, the Lagrange points, where a craft at rest in the frame stays at rest. L₁ lies between the primaries, L₂ beyond the smaller one and L₃ beyond the larger; L₄ and L₅ make equilateral triangles with them. L₁ and L₂ sit about one Hill radius from the smaller primary:
$\mu = m_2/(m_1 + m_2)$ $r_{Hill} \approx a(\mu/3)^{1/3}$
Leaving out the 3, or measuring from the larger body, are the common slips. In the landscape, L₁ and L₂ are mountain passes, the gates a craft must cross to travel between regions, and L₄ and L₅ are hilltops. A ball set on a hilltop rolls off, yet Trojan asteroids have stayed near Jupiter's L₄ and L₅ for billions of years. The Coriolis force is the reason: as an asteroid starts downhill it speeds up, and the sideways push bends its path around the hill into a slow tadpole, or a horseshoe that encloses L₄, L₅ and L₃. This works only while the larger primary holds nearly all the mass, more than about 96 % of it. The passes have no such help, so L₁, L₂ and L₃ are unstable, and a spacecraft parked near one must fire small thrusters regularly.
A small body near a moving planet also explains how spacecraft borrow speed.
A gravity assist
A spacecraft flying past a planet cannot gain energy relative to that planet. In the planet's frame the craft falls in, speeds up, swings around and climbs out again, and once it is far away it leaves at the speed it arrived with, $v_\infty$; only its direction has changed. The Sun sees something different. Relative to the Sun the planet moves at velocity $\vec V$, and the craft's velocity is the planet's plus its own relative to the planet:
$\vec{v} = \vec{V} + \vec{v}_\infty$
Turning $\vec v_\infty$ toward the planet's direction of motion lengthens the sum, so the craft leaves faster relative to the Sun than it came. The planet pays, slowing immeasurably. This is the patched-conic picture, which leaves out the Sun's pull during the encounter, an excellent approximation close to the planet. The mistake to avoid is mixing frames: the speed relative to the planet is unchanged, the speed relative to the Sun is not.
Gravity assists are orderly. The general three-body problem is not, and the next section asks how far ahead it can be predicted.
Order inside chaos, and the horizon of prediction
Chaos is not the whole story. Besides Lagrange's triangle and Euler's three collinear rigid rotations, there is the figure-eight: three equal masses chasing one another around one curve, each a third of a period behind the one ahead, found numerically by Cris Moore in 1993 and proved to exist by Chenciner and Montgomery in 2000. Its angular momentum, the sum of every body's $m r v_\perp$ about the center of mass, counted positive for counterclockwise turning and negative for clockwise, is conserved, so it cannot change sign: watch which way each lobe of the eight is traced, and decide what the sum must be.
Most starts are chaotic, and chaos has a precise meaning: nearby starts separate exponentially. If two copies of a system begin a tiny distance $\delta_0$ apart, the gap grows on average like
$\delta(t) \approx \delta_0\,e^{t/\tau}$ $t = \tau\ln(\Delta/\delta_0)$
where τ is the e-folding time. The second form is the predictability horizon, the time for an initial uncertainty $\delta_0$ to grow to a size Δ at which the prediction is useless. Because it depends on a logarithm, better data buy little: a $\delta_0$ a thousand times smaller adds only τ ln 1000 ≈ 6.9τ. Treating the growth as linear, or counting factors of 10 instead of factors of e, are the usual slips. In a real triple, close encounters multiply the gap in bursts, so τ is an average.
Every prediction here comes from a computer stepping the equations of motion, and the step itself can mislead.
Stepping an orbit
To compute an orbit, take the position and velocity now, find them a short time h later, and repeat. The simplest rule is Euler's: move with the old velocity, and change the velocity with the old acceleration. A one-line change, pushing first and then moving with the new velocity, gives the Euler–Cromer rule:
Euler: $x_{n+1} = x_n + h\,v_n$, $v_{n+1} = v_n + h\,a(x_n)$ Euler–Cromer: $v_{n+1} = v_n + h\,a(x_n)$, $x_{n+1} = x_n + h\,v_{n+1}$
Euler's rule moves along the tangent, and a tangent always lands outside a curve that bends toward the Sun. The old acceleration, from the old position, then pulls too weakly to bring the planet back, so the planet gains energy on every orbit and spirals outward. Round-off is not the cause; it is about $10^{-16}$ per step. Euler–Cromer makes errors of the same size within each orbit, but they cancel over the orbit instead of adding up, so its energy error stays bounded however long it runs.
Part 2 · Going further
Math you’ll need: cross products, time derivatives and the gradient, linearized equations with exponential solutions, and complex numbers in the last section.
An orbit from energy and angular momentum
Any state of two bodies, a relative position $\vec r$ and velocity $\vec v$, fixes the whole orbit. The specific energy ε fixes its size, and the specific angular momentum h its shape:
$\varepsilon = v^2/2 - GM/r$, $h = |\vec{r}\times\vec{v}|$ $a = -GM/2\varepsilon$, $e = \sqrt{1 + 2\varepsilon h^2/(GM)^2}$
Here M is the total mass of the pair. A bound orbit has ε < 0 and so a positive semi-major axis. Energy counts all of the speed, but h only the part perpendicular to $\vec r$: using the whole speed in h is the usual slip, and it makes the orbit look rounder than it is. For Part 1's sideways launch at k times the circular speed, $h^2 = GMr\,k^2$ and $2\varepsilon h^2/(GM)^2 = k^4 - 2k^2$, so $e = |k^2 - 1|$: the craft launched near Mars at 0.8 and 1.2 times $v_c$ follow ellipses with e = 0.36 and 0.44. The effective potential $-GM/r + h^2/2r^2$ turns the motion into a one-dimensional problem between two turning points, $r_p = a(1 - e)$ and $r_a = a(1 + e)$. A third conserved vector, the Laplace–Runge–Lenz vector $\vec v\times\vec h/GM - \hat r$, has length e and points at periapsis, which is why a two-body ellipse never turns.
Two bodies reduce to one relative orbit. Three bodies need a systematic way to remove their center of mass.
Jacobi coordinates and the Lagrange–Jacobi identity
Part 1's count of numbers has a history. For two bodies the one number the conservation laws leave free, the place along the orbit, follows from Kepler's equation. For three, Bruns (1887) proved that no further algebraic conserved quantities exist and Poincaré (1890) no further analytic ones; Sundman's series (1912) converges for all time but needs of order $10^{8\,000\,000}$ terms. What can be done is to remove the center of mass systematically.
In the center-of-mass frame the kinetic energy of three bodies can be written with two Jacobi vectors: $\vec\rho_1 = \vec r_2 - \vec r_1$, from the first body to the second, and $\vec\rho_2$, from the center of mass of that pair to the third body. Each carries its own reduced mass,
$\mu_1 = m_1 m_2/(m_1 + m_2)$ $\mu_2 = m_3(m_1 + m_2)/(m_1 + m_2 + m_3)$
and then $K = \tfrac{1}{2}\mu_1|\dot{\vec\rho}_1|^2 + \tfrac{1}{2}\mu_2|\dot{\vec\rho}_2|^2$. In Part 1's 3-4-5 example the escaping body and the binary carry the freed 1.717 as $\tfrac{1}{2}\mu_2 v_\infty^2$, with $\mu_2$ = 2/3, so they separate at $v_\infty = \sqrt{2 \times 1.717/(2/3)}$ = 2.27 units. The same vectors give the moment of inertia about the center of mass, $I = \sum m_i|\vec r_i - \vec R_{cm}|^2 = \mu_1\rho_1^2 + \mu_2\rho_2^2$, one measure of the system's size. The potential energy is homogeneous of degree −1 (double every distance and U halves), so Euler's theorem gives $\sum\vec r_i\cdot\nabla_i U = -U$, and two derivatives of I reduce to the Lagrange–Jacobi identity:
$\ddot{I} = 4K + 2U = 4E - 2U$
If E ≥ 0, then $\ddot I \ge -2U > 0$ and something must escape; if the motion stays bounded, $\ddot I$ averages to zero, the virial theorem $\langle 2K\rangle = -\langle U\rangle$. At rest K = 0 and $\ddot I = 2U$; forgetting the 2, or the sign of U, are the slips.
Some arrangements of three bodies never change shape at all, and they lead straight back to the Lagrange points.
From central configurations to Routh's criterion
A central configuration is an arrangement in which every acceleration points at the center of mass in proportion to its distance, so the shape can turn rigidly. For three bodies Euler and Lagrange found all of them: three collinear shapes, one for each body in the middle, and two equilateral triangles, one of each handedness. Shrink the third mass toward zero and they become the five Lagrange points: Euler's shapes become L₁, L₂ and L₃, and Lagrange's triangles L₄ and L₅. Their stability carries over: setting $m_3 = 0$, $m_1 = 1 - \mu$ and $m_2 = \mu$ in Gascheau's condition gives Routh's criterion, and the always-unstable collinear shapes become saddles. Linearizing about L₄ agrees: small motions go as $e^{\lambda t}$, where
$\lambda^4 + \lambda^2 + \tfrac{27}{4}\mu(1-\mu) = 0$ $27\mu(1-\mu) < 1$ $\nu_L \approx \sqrt{27\mu/4}$
As a quadratic in $\lambda^2$ the roots are $[-1 \pm \sqrt{1 - 27\mu(1-\mu)}]/2$. The motion is stable when both are real, negative and distinct, the middle condition (at equality a nudge grows in proportion to t). Solving $27\mu(1-\mu) = 1$ for its smaller root gives the critical mass ratio $\mu_R$; the larger root is no use, since μ is at most ½. For small μ the two frequencies separate into a fast epicycle near the orbital frequency and the slow libration $\nu_L$, in units of the orbital frequency, so a libration takes $1/\nu_L$ orbits.
Stability tells whether a craft stays near a point. A conserved quantity tells where it can go at all.
The Jacobi constant and the gates
In the restricted problem the primaries move, so they do work on the craft: neither its energy nor its angular momentum in the inertial frame is conserved. In the rotating frame, though, the landscape is steady, and the energy there, centrifugal term included, is constant. Multiplied by −2 and shifted, it is the Jacobi constant. In units where the separation, the total mass and G are 1:
$C = 2\Omega - v^2$, $\Omega = \tfrac{1}{2}(x^2+y^2) + \frac{1-\mu}{r_1} + \frac{\mu}{r_2} + \tfrac{1}{2}\mu(1-\mu)$
Here v is the speed in the rotating frame, and $r_1$ and $r_2$ are the distances from the primaries at (−μ, 0) and (1 − μ, 0). Since $v^2 \ge 0$, a craft can only be where $2\Omega \ge C$; the boundary, a zero-velocity curve, fences in the allowed region. A high C means little energy and a small region, and lowering C opens the gates in order, L₁, L₂, L₃, until below C(L₄) nothing in the plane is forbidden. This lab adds ½μ(1 − μ) to Ω so that C(L₄) = 3; many texts leave it out and quote every C lower by μ(1 − μ), so check the convention before comparing. The speed enters with a minus sign, so a faster craft has a lower C.
L₁ and L₂ are gates, but they are also places to park, and their instability has a precise form.
The collinear points: a saddle and a center
Near L₁, L₂ or L₃ the landscape curves differently along and across the axis. With $A = (1-\mu)/r_1^3 + \mu/r_2^3$ at the point, the second derivatives of Ω are $\Omega_{xx} = 1 + 2A$ and $\Omega_{yy} = 1 - A$, and small motions grow or oscillate as $e^{\lambda t}$ with
$\lambda^4 + (2 - A)\lambda^2 + (1 + 2A)(1 - A) = 0$ $\lambda^2 = \left[A - 2 + \sqrt{9A^2 - 8A}\right]/2$
A is greater than 1 at every collinear point, so the two roots for $\lambda^2$ have opposite signs. The positive one gives ±λ, a saddle that pushes a craft away along one direction; the negative one gives ±iν, with $\nu^2 = [2 - A + \sqrt{9A^2 - 8A}]/2$, a center about which small orbits circulate. A parked craft's error grows by a factor of e every 1/λ time units (one unit is the primaries' period over 2π); mixing up λ with ν, or forgetting to convert to days, are the slips. A small planar Lyapunov orbit circulates on the center direction with period near 2π/ν. Differential correction finds it: start on the axis with no speed along it, and adjust the starting speed until the next axis crossing is perpendicular too. The equations are unchanged by (x, y, t) → (x, −y, −t), so the second half mirrors the first, run backward, and the orbit closes.
A flyby, too, is best seen in a frame of its own.
A flyby turns the approach velocity
In the planet's frame a flyby is a hyperbola. Far away the craft moves at $v_\infty$; at periapsis, a distance $r_p$ from the planet's center, it moves faster, at $v_p = \sqrt{v_\infty^2 + 2GM/r_p}$. The asymptotes turn the approach velocity through an angle δ:
$\sin(\delta/2) = 1/(1 + r_p v_\infty^2/GM)$ $|\Delta\vec{v}| = 2v_\infty\sin(\delta/2)$
The incoming and outgoing velocities have the same length and differ by δ in direction, so with their difference they form an isosceles triangle, which gives the second formula. $\Delta\vec v$ is a change of vector, not of speed: relative to the planet the speed does not change at all. The same vector, added in the Sun's frame, does change the speed. A closer pass or a slower approach turns $\vec v_\infty$ farther. In the restricted Sun–planet problem the Jacobi constant survives the flyby though the energy relative to the Sun does not; far from the planet C becomes, approximately, Tisserand's parameter, by which astronomers recognize a comet a planet has deflected.
Every result so far assumes the computed orbit is right. The next section asks when it is.
Integrators you can trust
A method of order p makes a global error, after a fixed time, proportional to the step to the power p: error ∝ $h^p$. In the limit of small h, halving the step divides the error by $2^p$, and dividing it by k divides the error by $k^p$. Euler and Euler–Cromer are first order, leapfrog (Euler–Cromer made symmetric) second, and RK4 and Yoshida-4 fourth. A symplectic method such as Euler–Cromer (also called symplectic Euler), leapfrog or Yoshida-4 conserves, up to exponentially small terms, a "shadow" energy $\tilde H = H + h^pH_p + \ldots$ close to the true one, so its energy error oscillates and never drifts. RK4's error, smaller at first, grows a little at every periapsis, so over millions of orbits a symplectic method wins.
Close encounters need tiny steps there and only there. Adaptive methods shrink h where a time scale is short, $h = \min(h_{max}, \eta\,\tau_{min})$; regularized methods change the clock itself. LogH steps in a fictitious time s with $dt = ds/|U|$, so the time step shrinks automatically as the bodies approach and U grows. A small energy error is required of a good run, but it proves little: an integrator can keep the books balanced while putting the bodies in the wrong places, since an error in the timing of one close passage changes the whole future without changing E. In a chaotic system the test of a computed ending is convergence: halve the step or the tolerance, and trust the answer only if it holds.
With trustworthy runs, chaos itself can be measured.
Measuring chaos: Lyapunov exponents and Poincaré sections
The largest Lyapunov exponent λ is the long-run growth rate of a small separation between two copies of a system:
$\lambda = \lim (1/t)\ln(\delta(t)/\delta_0)$ $\log_{10}\delta = \log_{10}\delta_0 + (\lambda/\ln 10)\,t$
On a plot of $\log_{10}\delta$ against t the exponential stretch is a straight line of slope λ/ln 10, so λ is ln 10 times the slope; reading the slope itself as λ, or dividing by ln 10 instead of multiplying, are the slips. Finite-time estimates depend on the stretch fitted, so Benettin's method renormalizes the twin to $\delta_0$ at regular intervals and averages the log of the stretch. Following an unstable orbit for a time t needs about λt/ln 10 significant digits.
A Poincaré section shows the structure behind the numbers. The restricted problem's state has four numbers, (x, y, ẋ, ẏ); fixing C leaves a three-dimensional shell, and recording (x, ẋ) only when the orbit crosses y = 0 with ẏ > 0 leaves a two-dimensional map. A quasi-periodic orbit lies on a torus, and its crossings fall on a smooth closed curve; by KAM theory most such tori survive a small perturbation. Between them, around broken resonances, chaotic orbits scatter their crossings over whole regions, the chaotic sea.
Sections remove one dimension. The last section removes everything but the triangle's shape.
The shape sphere
Take away the position of the center of mass, the orientation and the size of a triangle of three bodies, and only its shape is left. Every shape is then one point on a sphere, the shape sphere. With the mass-weighted Jacobi vectors $z_1 = \sqrt{\mu_1}\,\rho_1$ and $z_2 = \sqrt{\mu_2}\,\rho_2$ written as complex numbers, the Hopf map gives the point
$\vec{w} = (|z_1|^2 - |z_2|^2,\; 2\,\mathrm{Re}\,\bar{z}_1 z_2,\; 2\,\mathrm{Im}\,\bar{z}_1 z_2)/I$
with $I = |z_1|^2 + |z_2|^2$, so $\vec w$ has length 1. Its third component is proportional to the triangle's signed area. The equator, $w_3 = 0$, holds the collinear shapes, among them the three Euler configurations and the three points where two bodies collide; for three equal masses the two poles are the equilateral triangles, one of each handedness (with unequal masses the equilateral shapes sit near the poles, not on them). A motion of three bodies traces a path on the sphere, and the special points it meets characterize the orbit; the figure-eight's existence proof and the 2013 families of periodic orbits were both found on this sphere.
Reference
Common mistakes
Part 1
- Putting the center of mass nearer the lighter body. $m_1 r_1 = m_2 r_2$: the heavier body's distance is the smaller.
- Dropping the √2, or measuring from the surface. $v_{esc} = \sqrt{2}\,v_c$, and r is from the center.
- Squaring the ratio of distances. $v_a = v_p r_p/r_a$ at the ends of an ellipse.
- Reading "E < 0" as "nobody escapes". A tightening binary can pay for one body to leave.
- Using one mass in Lagrange's rate. $\omega^2 = GM/a^3$ takes the total mass.
- Confusing exact with stable. An exact triangle can tear apart from a tiny nudge.
- Mixing frames in a gravity assist. The speed relative to the planet is unchanged; relative to the Sun it is not.
- Counting factors of 10 for factors of e. The horizon is $\tau\ln(\Delta/\delta_0)$, a natural logarithm.
Part 2
- Using the whole speed in h. Only the perpendicular part counts: $h = |\vec r\times\vec v|$.
- Mixing conventions for C. This lab's C includes μ(1 − μ), so that C(L₄) = 3.
Equation sheet
| Part 1 | |
|---|---|
| center of mass | $m_1 r_1 = m_2 r_2$ |
| circular, escape | $v_c = \sqrt{GM/r}$, $v_{esc} = \sqrt{2GM/r}$ |
| equal areas | $r_p v_p = r_a v_a$; $T^2 = 4\pi^2 a^3/GM$ |
| energy | $E = K + U$, $U = -\sum Gm_im_j/r_{ij}$ |
| Lagrange | $\omega^2 = GM/a^3$, $T = 2\pi/\omega$ |
| Gascheau | $27(m_1 m_2 + m_2 m_3 + m_3 m_1) < M^2$ |
| L₁, L₂ | $r_{Hill} \approx a(\mu/3)^{1/3}$ |
| gravity assist | $\vec{v} = \vec{V} + \vec{v}_\infty$ |
| the horizon | $t = \tau\ln(\Delta/\delta_0)$ |
| Part 2 | |
|---|---|
| elements | $a = -GM/2\varepsilon$, $e = \sqrt{1 + 2\varepsilon h^2/(GM)^2}$ |
| Lagrange–Jacobi | $\ddot{I} = 4K + 2U = 4E - 2U$ |
| Jacobi constant | $C = 2\Omega - v^2$, $\Omega = \tfrac{1}{2}(x^2+y^2) + \frac{1-\mu}{r_1} + \frac{\mu}{r_2} + \tfrac{1}{2}\mu(1-\mu)$ |
| L₄ | $27\mu(1-\mu) < 1$ Routh; $\nu_L \approx \sqrt{27\mu/4}$ |
| L₁, L₂, L₃ | $\lambda^2 = \left[A - 2 + \sqrt{9A^2 - 8A}\right]/2$, $A = (1-\mu)/r_1^3 + \mu/r_2^3$ |
| flyby | $\sin(\delta/2) = 1/(1 + r_p v_\infty^2/GM)$, $|\Delta\vec{v}| = 2v_\infty\sin(\delta/2)$ |
| integrators | error ∝ $h^p$; symplectic: a shadow $\tilde H$ conserved |
Symbols: M total (or central) mass · a semi-major axis, or the triangle's side · μ the smaller primary's mass fraction · $\vec v_\infty$ velocity relative to the planet, far from it · τ e-folding time · ε, h specific energy and angular momentum · I moment of inertia · C Jacobi constant · $\nu_L$ libration frequency (orbital frequency = 1) · δ turning angle · p an integrator's order. G = 1 in scaled scenes; SI with km, km/s and days for named systems.
Further reading
For a first course, OpenStax, University Physics Volume 1, chapter 13, "Gravitation" (an open textbook licensed CC BY-NC-SA 4.0). For the restricted problem, Lagrange points and Tisserand's parameter, C. D. Murray and S. F. Dermott, Solar System Dynamics (Cambridge, 1999); for the integrators, E. Hairer, C. Lubich and G. Wanner, Geometric Numerical Integration (Springer, 2006); for the three-body problem itself, M. Valtonen and H. Karttunen, The Three-Body Problem (Cambridge, 2006), and R. Montgomery's survey of the figure-eight, Notices of the American Mathematical Society 48, 471 (2001). The Field Manual lists every source.
Three-Body Lab v1.5 · © 2026 Dr. David Meier · CC BY-NC-SA 4.0 · glidepathscience.org