Assignment 03 — Programming Assignment: Optimal Impulsive Manoeuvres, from Theory to Chart

In Assignments 01 and 02 you derived. Here you derive, compute, prove the computation correct, and read an answer off a figure you made yourself. The nine problems below are modelled on the summary charts of the 1968 NASA survey of impulsive trajectories (F. W. Gobetz and J. R. Doll, NASA CR-61999). Each chart hides a small piece of theory, and each can be rebuilt in a few hours with Python. Every problem is built the same way and is meant to cost about the same effort: 8 to 10 hours, of which roughly half is thinking on paper.

Notation follows the lecture slides: \(\mu\) is the gravitational parameter, \(r\) a radius, \(e\) the eccentricity, \(p=a(1-e^2)\) the semi-latus rectum, \(\nu\) the true anomaly, \(\gamma\) the flight path angle (the angle of the velocity above the local horizontal), \(P\) and \(A\) the pericentre and apocentre radii. A burn is an impulsive velocity change; “time open” means the time of flight is free.

1 Overview

1.1 What each problem asks of you

Every problem has the same five stages, and each stage is marked.

  1. The question. A concrete engineering question, stated first. You answer it at the end, from a figure produced by your own code, and confirm the number with your code to the digits asked.
  2. Theory (T). Derivations to be done on paper (or in LaTeX) and included in the report. A derivation is a chain of justified steps, not a formula copied from a book; where the problem says “you should obtain …”, that is a check on your result, not a substitute for the derivation.
  3. Code (C). Functions with a prescribed interface, written by you.
  4. Validation (V). Numbered checks with target values. Your code must reproduce each target within the stated tolerance. Each problem also requires one independent check (a calculation that shares no code and no formula with the function it checks: brute-force search, numerical integration, or an algebraically different route) and one sensitivity test (you deliberately break the code and show that your checks notice).
  5. Figure and answer (F). The figure specified, the answer read from it, and two sentences of engineering interpretation.

1.2 At a glance

# Topic Theory core Computational core Independent check
1 Single-impulse intercept a one-parameter family of conics; a quartic polynomial roots; root-finding for \(\rho^{*}\) dense scan of the cost; closed-form limits
2 Disorbit to a prescribed entry a constrained minimum with two regimes closed forms; the regime boundary brute-force constrained scan; propagated entry angle
3 Escape and the Oberth effect vis-viva at depth; a cancellation identity three escape modes; a crossing the identity to \(10^{-9}\); the limit table
4 Hohmann against bi-elliptic three-burn costs; asymptotics; thresholds thresholds by root-finding; savings a worked km/s example; the asymptotic coefficient
5 Transfer with a plane change optimal split of the turn (Lagrange); a local-minimum test nested optimisation over the split and the conjunction radius pure-plane-change limit; large-\(k\) expansion
6 Circle to ellipse the apsidal-burn lemma; three routes; boundary asymptotics route map; boundaries; a lemma on ellipse-to-ellipse \(10^{5}\) random cases; random initial ellipses
7 Congruent ellipses symmetry; the mirror transfer three-parameter optimisation; a boundary curve an unrestricted two-impulse optimiser
8 The useful angle the primer vector as a gradient of the elements admissible-direction window by scan and bisection contact conditions; limits for small \(e\)
9 Rendezvous phasing the phase relation; a bi-elliptic transfer that absorbs a phase offset inverse \(r_i(\Delta\beta)\) numerical integration of the two-body problem

1.3 Conventions

  • Normalised units. Unless a problem says otherwise, \(\mu=1\) and the radius of the initial circular orbit is \(r_1=1\), so the circular speed there is \(V_{c_1}=\sqrt{\mu/r_1}=1\). Every \(\Delta V\) is quoted divided by \(V_{c_1}\) (written \(\Delta V/V_{c_1}\), or just “\(\Delta V\)” in the validation tables). Radii are in units of \(r_1\) unless a problem says otherwise.
  • Angles. Degrees in input and output, radians inside the code. Say which in every function’s docstring.
  • Python 3.9+ with NumPy, SciPy and Matplotlib. Allowed from SciPy: root finders and minimisers (brentq, minimize, minimize_scalar), and solve_ivp where a problem asks for numerical integration. MATLAB or Julia is acceptable if every deliverable below is met.

2 Your parameters

Let \(s\) be the last two digits of your roll number (\(00\) to \(99\)). Your parameters for each problem are read from the lists below by a zero-based index that the function reduces with a modulus. Use the function rather than reading the lists by eye; the report must print your \(s\) and your parameter dictionary on its first page.

L1  = [40, 45, 50, 55, 65, 75, 80, 100, 110, 130, 140]          # Problem 1: central angle theta, deg
L2  = [-3, -5, -7, -9, -12, -15, -20, -25, -30, -40, -50]        # Problem 2: entry flight path angle, deg
L3D = [3, 5, 6, 10, 12, 16]                                      # Problem 3: depth D = r1/r_ii
L3K = [4, 6, 8, 10, 15, 20]                                      # Problem 3: outer apoapsis K = r_i/r1
L3X = [1.8, 2.2, 2.6, 3.0, 3.5, 4.0]                             # Problem 3: V_inf / Vc1
L5  = [40, 42, 44, 46, 48, 50, 52, 54]                           # Problem 5: inclination, deg
L6  = [13, 15, 18, 20, 25, 30, 40, 50, 70, 100]                  # Problem 6: apocentre ratio A
L7  = [12, 20, 25, 30, 45, 50, 75, 90, 100, 120, 150]            # Problem 7: pericentre rotation omega, deg
L8E = [0.35, 0.45, 0.55, 0.65, 0.75, 0.85, 0.88, 0.91]          # Problem 8: eccentricity
L8N = [40, 70, 110, 130]                                         # Problem 8: true anomaly of the burn, deg
L9K = [1.2, 1.3, 1.4, 1.5, 1.6, 1.7, 1.8, 1.9, 2.0]              # Problem 9: radius ratio k

def my_parameters(s):
    s = int(s) % 100
    return dict(q1_theta=L1[s % 11], q2_gamma=L2[(s + 3) % 11],
                q3_D=L3D[s % 6], q3_K=L3K[(s // 6) % 6], q3_x=L3X[(s + 2*(s // 6) + 1) % 6],
                q4_R=round(12.0 + 0.1*(s % 35), 1), q5_i=L5[(s + 2) % 8], q6_A=L6[(s + 5) % 10],
                q7_om=L7[(s + 7) % 11], q8_e=L8E[s % 8], q8_nu=L8N[(s // 8) % 4], q9_k=L9K[(s + 4) % 9])

Your parameters are not secret, but your answers are personal: a correct figure for the wrong parameters earns no credit for the answer.

3 Rules for every problem

3.1 The validation block

Every script must end with a validation block that prints one line per check, in exactly this form, so that it can be read at a glance and by a grader’s grep:

def check(tag, what, got, want, tol):
    ok = abs(got - want) <= tol
    print(f"[{tag}] {what:<60s} got {got:<14.7g} want {want:<12.7g} tol {tol:<8.1g} {'PASS' if ok else 'FAIL'}")
    return ok

and then a final line, ANSWER Qn name=value, name=value, ..., with the numbers that answer the question. A FAIL left in the submitted log, without a note explaining it, loses the marks for that check.

  • Targets are values obtained by someone else’s careful calculation (a closed form, a published number, or a reference run). Where a tolerance is given, it is what a correct implementation achieves comfortably; a larger error is a bug, not noise.
  • The independent check must not call the function under test, nor reuse its formulas. Say in one sentence what makes it independent.
  • The sensitivity test. Make one deliberate, plausible mistake in a copy of your code (drop an absolute value, swap arctan2 for arccos, use \(\cos\) for \(\sin\), use degrees where radians are expected, …). Run your checks on the broken copy and report which check fails. If none does, your suite is too weak, and you should add a check that does.

3.2 Figures

Label both axes with the quantity and the unit (or “normalised by \(V_{c_1}\)”); title each figure with the problem number and your parameters; include a legend; use text of at least 9 pt; use colours that remain distinguishable in greyscale (vary line style as well); save a PNG (200 dpi) and a PDF into figures/. A figure that cannot be read without the code is not finished.

3.3 What to hand in

One archive containing:

  • report.pdf — your name, roll number, \(s\) and parameters on the first page; then, for each problem, the derivations (T), a short description of the code and its checks (C, V), the figure(s), and the answer (F). At most 6 pages per problem.
  • q1.py … q9.py — one script per problem. Each must run on a clean machine with python qN.py, in under 5 minutes, using fixed random seeds, write its figures to figures/, and print its validation block and ANSWER line.
  • README.md — Python and package versions, how to run everything, run time, and a list of any tools you used other than the three packages above (including AI assistants) and for what.
  • validation_log.txt — the output of running all nine scripts.

4 Marking and integrity

100 marks. Each problem: Theory 4, Code 2, Validation 2, Figure and answer 2 = 10 marks, for 90 in all. The remaining 10 marks are for the report as a whole: clarity, reproducibility, and a clean archive.

Integrity. You may discuss ideas with other students; derivations, code and figures must be your own. Parameters differ between students, so identical numbers in two reports will be noticed. You are responsible for every line you submit, and any part of your work may be examined orally.

5 The problems

5.1 Problem 1 — One impulse to a point

Source chart: survey Fig. 1.

5.1.1 The question

A spacecraft coasts on a circular orbit of radius \(r_1\). With one impulse, and with the flight time free, it must pass through a point at radius \(\rho\,r_1\) that lies a central angle \(\theta\) ahead of the burn point (measured at the focus, in the direction of motion). Take \(\theta=\theta_s=\) q1_theta.

  1. Up to what radius ratio \(\rho^{*}(\theta_s)\) can the cheapest intercept still be an ellipse? Beyond it the optimal arc is a hyperbola.
  2. What does the burn cost at \(\rho^{*}\), at what angle to the local horizon does the burn leave the circle, and by what factor does the cost exceed the \(\theta=180^\circ\) intercept of the same radius (the first burn of a Hohmann transfer)?
  3. The usual belief is that a larger central angle is always cheaper. Does your chart say the same thing about staying elliptic?

5.1.2 A. Theory

T1. Let \(\nu_1\) be the true anomaly of the burn point on the transfer conic. Show that passing through both points gives \(p/r_1=1+e\cos\nu_1\) and \(p/r_2=1+e\cos(\nu_1+\theta)\). With \(x=p/r_1\), \(A=e\cos\nu_1\) and \(B=e\sin\nu_1\), express \(A\) and \(B\) in terms of \(x\), \(\rho\) and \(\theta\), and explain why every \(x>0\) gives one conic: one free parameter survives.

T2. Before the burn the velocity is purely transverse and equal to \(V_{c_1}\). Find the radial and transverse components of the velocity just after the burn on the conic of T1, and show that the cost obeys \(J(x)\equiv(\Delta V/V_{c_1})^2=B^2/x+(\sqrt x-1)^2\). Say which term is waste and when it vanishes.

T3. Minimise \(J(x)\). With \(s=\sqrt x\), \(\alpha=\cos\theta-1/\rho\) and \(\beta=1-\cos\theta\), show that the stationary condition becomes a quartic in \(s\) with no \(s^2\) and no \(s^1\) term, and give its coefficients. Explain which roots you keep.

T4. Special cases. Show that \(\theta=180^\circ\) gives the Hohmann transfer (find \(x\) and \(\Delta V\) in closed form, and show the burn is tangential); that \(\rho=1\) gives \(\Delta V=0\); and, by an energy argument, that as \(\theta\to0\) with \(\rho>1\) the cost tends to \(\sqrt{3-2/\rho}\) (the best you can do is to kill the transverse velocity and fall radially).

T5. Prove \(\Delta V(\theta)=\Delta V(360^\circ-\theta)\).

T6. Give \(e\) and the flight path angle \(\gamma^{+}\) of the post-burn velocity (above the local horizontal) in terms of the root. For \(e>1\) the arc must not cross an asymptote: state the condition on \(\nu\) and say when it can bind.

5.1.3 B. Code

  • intercept(rho, theta) returns \(\Delta V\), \(x\), \(e\), \(\nu_1\), \(\gamma^{+}\) and a flag that says whether the arc is physical. Use numpy.roots on the quartic of T3 (no numerical minimiser), keep the positive real roots, take the smallest \(J\), and handle \(\sin\theta=0\) explicitly.
  • rho_star(theta) returns the smallest \(\rho>1\) at which the optimal conic has \(e=1\): bracket the sign change of \(e(\rho)-1\) on a logarithmic grid, then refine with brentq.
  • brute(rho, theta) finds the minimum of \(J(x)\) by a dense scan in \(x\) followed by a local refinement. It is used for validation only.

5.1.4 C. Validation

tag check target tol
V1 \(\theta=180^\circ\), \(\rho=0.5,\ 2,\ 10\) \(0.183503,\ 0.154701,\ 0.348400\) \(5\times10^{-7}\)
V2 \(\theta=0\), \(\rho=0.5,\ 3\) \(1.000000,\ 1.527525\) \(5\times10^{-7}\)
V3 \(\rho=1\), any \(\theta\) \(0\) \(10^{-9}\)
V4 \(\rho=6\), \(\theta=60^\circ\): \(s\), \(x\), \(e\), \(\nu_1\), \(a/r_1\), \(\Delta V\), \(\gamma^{+}\) \(1.093192\), \(1.195069\), \(1.05551\), \(79.35^\circ\), \(-10.473\), \(0.95347\), \(40.96^\circ\) \(5\times10^{-6}\), \(5\times10^{-6}\), \(2\times10^{-5}\), \(0.01^\circ\), \(10^{-3}\), \(10^{-5}\), \(0.01^\circ\)
V5 \(\Delta V(70^\circ)-\Delta V(290^\circ)\) at \(\rho=4\) \(0\) \(10^{-12}\)
V6 independent: quartic against brute at \((\rho,\theta)=(3,100^\circ)\) equal \(10^{-9}\)
V7 \(\rho^{*}\) at \(\theta=30^\circ,\ 60^\circ,\ 90^\circ,\ 120^\circ\) \(19.70,\ 4.19,\ 4.43,\ 8.25\) \(0.05,\ 0.02,\ 0.02,\ 0.05\)
V8 \(\min_\theta\rho^{*}(\theta)\) and where it occurs \(3.85\) near \(70^\circ\) \(0.02\); \(2^\circ\)
V9 spot values \((\theta,\rho)\to\Delta V\): \((30^\circ,0.5)\), \((60^\circ,2)\), \((90^\circ,4)\), \((120^\circ,2)\), \((60^\circ,10)\) \(0.639,\ 0.521,\ 0.534,\ 0.207,\ 1.036\) \(2\times10^{-3}\)
V10 sensitivity test one check fails —

5.1.5 D. Figure and answer

Figure 1a. \(\Delta V/V_{c_1}\) against \(\rho\) (logarithmic axis, \(0.1\) to \(10\); vertical range \(0\) to \(2\)) for \(\theta\in\{0^\circ,30^\circ,60^\circ,90^\circ,120^\circ,180^\circ\}\) and for \(\theta_s\) (heavier line). Draw a curve solid where the optimal conic is an ellipse and dashed where it is not, and mark \(\rho^{*}\) on each curve with an open circle.

Figure 1b. \(\rho^{*}(\theta)\) against \(\theta\) for \(20^\circ\le\theta\le170^\circ\), with its minimum marked.

Answer. Report, for \(\theta_s\): \(\rho^{*}\), \(\Delta V(\rho^{*})\), \(\gamma^{+}(\rho^{*})\), the \(\theta=180^\circ\) cost at the same \(\rho^{*}\), and their ratio. For (c): at which central angle is it hardest to remain elliptic, what is \(\rho^{*}\) there, and what does that say about “more central angle is always cheaper”?

5.1.6 Notes

  • The optimal burn is not tangential except at \(\theta=180^\circ\): do not assume \(B=0\).
  • numpy.roots returns complex numbers; filter on a small imaginary part, not on exact equality. Recover \(\nu_1\) with arctan2, not arccos.
  • If \(e\) is not monotone in \(\rho\) near \(1\), your bracket for \(\rho^{*}\) is wrong; plot \(e(\rho)\) before root-finding.

Marks: Theory 4 · Code 2 · Validation 2 · Figure and answer 2.

5.2 Problem 2 — Deorbit to a prescribed entry

Source chart: survey Fig. 2.

5.2.1 The question

A capsule on a circular orbit of radius \(r_1\) must reach the entry radius \(r_2<r_1\) with a prescribed entry flight path angle \(\gamma_2<0\), after one impulse. Where around the globe it comes down is free. Define \(n=r_1/r_2\ge1\) and take \(\gamma_2=\gamma_s=\) q2_gamma.

  1. What parking altitude \(n_{\rm best}\) minimises the cost, and what is \(\Delta V_{\min}\)?
  2. For which \(n\) is the best burn exactly horizontal, and for which is it tilted below the horizon? Where is the switch, \(n_b(\gamma_s)\)?
  3. With the parking orbit at \(n_s=1+\tfrac12\,(n_b(\gamma_s)-1)\), what is the burn direction and cost, and how much dearer is it than the best?

5.2.2 A. Theory

T1. Write the post-burn state at \(r_1\) as \((w,u)=(v_r,v_\theta)/V_{c_1}\), with \(w\le0\). The pre-burn state is \((0,1)\); show \((\Delta V/V_{c_1})^2=w^2+(u-1)^2\).

T2. Carry energy and angular momentum from \(r_1\) to \(r_2\) to find \(v_2^2\) and \(v_{\theta_2}\) in terms of \((u,w,n)\). Impose \(\cos\gamma_2=v_{\theta_2}/v_2\) and show that the constraint collapses to \(w^2=Ku^2-M\). Find \(K(n,\gamma_2)\) and \(M(n)\).

T3. Eliminate \(w\) and obtain the cost as a parabola in \(u\). Find its vertex. State the feasibility condition \(w^2\ge0\) and the two regimes: the interior optimum (impulse tilted below the horizon) and the constrained optimum (impulse exactly horizontal, \(w=0\)). Define the thrust angle \(\phi=\arctan2(|w|,1-u)\) of the retro-burn below the local horizontal.

T4. Closed forms. Show that \(\Delta V=|\sin\gamma_2|\) at \(n=1\); that in the horizontal regime \(\Delta V=1-\cos\gamma_2\sqrt{2(n-1)/(n^2-\cos^2\gamma_2)}\); and, by maximising the square root, find \(n_{\rm best}\) and \(\Delta V_{\min}\) in closed form in terms of \(\gamma_2\).

T5. Find the equation that \(n=n_b(\gamma_2)\) must satisfy when the vertex of T3 just becomes infeasible. Describe in words how \(n_b\) behaves as \(\gamma_2\to0\) and as \(\gamma_2\to-90^\circ\), and say why a maximum over \(\gamma_2\) must exist.

5.2.3 B. Code

  • disorbit(n, gamma) returns \(\Delta V\), \(u\), \(w\), \(\phi\) and the regime (“tilted” or “horizontal”).
  • n_boundary(gamma) returns \(n_b\) by bracketing and brentq.
  • entry_angle(u, w, n) takes the post-burn state, carries it to \(r_2\) using only energy and angular momentum, and returns the flight path angle there. It is the independent check on T2.
  • brute(n, gamma) minimises \(w^2+(u-1)^2\) by a dense scan in \(u\), using the constraint from T2 to supply \(w\).

5.2.4 C. Validation

tag check target tol
V1 \(\Delta V\) at \(n=1\) for \(\gamma_2=-2^\circ,\ -10^\circ\) \(0.034899,\ 0.173648\) \(2\times10^{-6}\)
V2 \(n_{\rm best}\) and \(\Delta V_{\min}\) for \(\gamma_2=-2^\circ\) \(1.0349,\ 0.017605\) \(2\times10^{-4}\), \(3\times10^{-6}\)
V3 the same for \(\gamma_2=-10^\circ\) \(1.1736,\ 0.090961\) \(2\times10^{-4}\), \(3\times10^{-6}\)
V4 independent: brute against the closed form at \(n_{\rm best}\), both angles equal \(3\times10^{-5}\)
V5 \(n_b(-2^\circ)\), \(n_b(-10^\circ)\) \(1.0129,\ 1.0577\) \(5\times10^{-4}\)
V6 \(\max_{\gamma_2}n_b\) and the \(\gamma_2\) where it occurs \(1.1250\) at \(-37.3^\circ\) \(5\times10^{-4}\); \(0.6^\circ\)
V7 \(\gamma_2=-10^\circ\): \(n=1.03\) is tilted with \(\phi=57.72^\circ\); \(n=1.10\) is horizontal with \(\phi=0\) as stated \(0.01^\circ\)
V8 independent: entry_angle for the optimum at \(\gamma_2=-10^\circ\), \(n=1.03\) \(-10^\circ\) \(10^{-6}\) deg
V9 sensitivity test one check fails —

The value in V6 is the constant behind a statement of the survey: “if \(r_1/r_2>1.125\) the impulse is always horizontal.” Say in your report what your computation shows about that statement.

5.2.5 D. Figure and answer

Figure 2a. \(\Delta V/V_{c_1}\) against \(n\) for \(1\le n\le3\) and \(\gamma_2\in\{-2^\circ,-5^\circ,-10^\circ,-20^\circ,\gamma_s\}\). Draw each curve dashed where the burn is tilted and solid where it is horizontal, and mark the minimum of each with an open circle.

Figure 2b. \(n_b(\gamma_2)\) for \(-89^\circ\le\gamma_2\le-1^\circ\), with the maximum marked and the line \(n=1.125\) drawn.

Answer. For \(\gamma_s\): \(n_{\rm best}\), \(\Delta V_{\min}\), \(n_b(\gamma_s)\); at \(n_s\): the regime, \(\phi\), \(\Delta V\), and the penalty over \(\Delta V_{\min}\) in per cent. One sentence: is the tilted regime ever optimal if you are free to choose the altitude? Support it with Figure 2a.

5.2.6 Notes

  • \(w\le0\) throughout (descending); take the negative root \(w=-\sqrt{Ku^2-M}\).
  • \(K>0\) for every \(n\ge1\) and \(\gamma_2\ne0\); at \(n=1\) it equals \(\tan^2\gamma_2\).
  • Do not locate \(n_{\rm best}\) by an optimiser: the point of T4 is that you do not need one, and V4 is where you compare.

Marks: Theory 4 · Code 2 · Validation 2 · Figure and answer 2.

5.3 Problem 3 — Escape, and why it pays to dive

Source charts: survey Figs. 3 and 4.

5.3.1 The question

A spacecraft on a circular orbit of radius \(r_1\) must leave on a hyperbola of excess speed \(V_\infty=x\,V_{c_1}\). Three ways to do it:

  • one impulse at \(r_1\);
  • two impulses: a retro burn that drops the pericentre to \(r_{ii}=r_1/D\), then escape from there;
  • three impulses: a prograde burn that raises the apoapsis to \(r_i=Kr_1\), a retro burn there that drops the pericentre to \(r_{ii}=r_1/D\), and escape from \(r_{ii}\).

For \((D,K,x)=(D_s,K_s,x_s)=\) (q3_D, q3_K, q3_x):

  1. Which mode is cheapest, and by what percentage does it beat the single impulse?
  2. At what \(V_\infty/V_{c_1}\) does the three-impulse mode first beat the single impulse? Does it depend on the depth \(D\)? Explain what you see.

5.3.2 A. Theory

T1. Show that a single tangential impulse costs \(\Delta V_1/V_{c_1}=\sqrt{x^2+2}-1\).

T2. Derive the two-impulse cost \(\Delta V_2(x,k)\) for a dive to \(r_i=kr_1\) (\(k<1\)): the cost of dropping the pericentre, plus the cost of escaping from \(r_i\) with excess speed \(x\) (all speeds are apsidal, so vis-viva with the two apsides is enough).

T3. Show that for \(x=\sqrt2\) (that is, \(V_\infty=V_{\rm esc}(r_1)\)) the cost \(\Delta V_2\) equals \(1\) for every \(k\). Explain why this makes \(x=\sqrt2\) the watershed between the one- and two-impulse modes.

T4. Derive the three-impulse cost \(\Delta V_3(x,K,k_{ii})\) with \(k_{ii}=r_{ii}/r_1\). Show that burns 2 and 3 together do not depend on \(k_{ii}\) when \(V_\infty=\sqrt{2\mu/r_i}\), i.e. \(x=\sqrt{2/K}\). Use this to show that \(\Delta V_3=\Delta V_1\) at \(x=\sqrt{2/K}\) for any depth.

T5. Let \(K\to\infty\) and \(k_{ii}\to0\). Show that \(\Delta V_3\to\sqrt2-1\) for any \(x\). What does it cost in other resources?

5.3.3 B. Code

dv1(x), dv2(x, k), dv3(x, K, kii); a function crossing(K, D) that finds, by scanning and brentq, the smallest \(x\) at which \(\Delta V_3=\Delta V_1\); and a sweep helper that evaluates each mode over a grid of \(x\). In the three-impulse mode burn 2 is a retro burn: take its magnitude.

5.3.4 C. Validation

tag check target tol
V1 \(x=1.5\): one impulse; two impulses with \(k=0.45\); three impulses with \(K=3\), \(k_{ii}=0.25\) \(1.0616,\ 1.0488,\ 0.8906\) \(10^{-4}\)
V2 \(x=\sqrt2\): \(\Delta V_2\) for \(k=0.9,\ 0.5,\ 0.1,\ 0.01\) \(1.0000000000\) each \(10^{-10}\)
V3 \(\Delta V_3\) with \(K=10^{3},\ 10^{6}\) and \(k_{ii}=1/K\): \(x=1.5\) \(0.4400,\ 0.4150\) \(10^{-4}\)
V4 the same for \(x=3\), and for \(x=6\) at \(K=10^{6}\) \(0.5154,\ 0.4174\); \(0.4269\) \(10^{-4}\)
V5 depth \(D=8\), \(x=2\), \(K=10\): one, two, three impulses \(1.450,\ 1.230,\ 0.931\) \(2\times10^{-3}\)
V6 independent: the crossing of \(\Delta V_3\) and \(\Delta V_1\) for \((K,D)=(4,3),\ (10,8),\ (20,16)\) equals \(\sqrt{2/K}\) \(0.70711,\ 0.44721,\ 0.31623\) \(10^{-8}\)
V7 at \(K=2\), \(x=1\): spread of \(\Delta V_3\) over \(k_{ii}\in[0.02,0.5]\) \(0\) \(10^{-9}\)
V8 sensitivity test one check fails —

5.3.5 D. Figure and answer

Figure 3. \(\Delta V/V_{c_1}\) against \(x\in[0.2,4]\) at depth \(D_s\): the one-impulse curve, the two-impulse curve, and the three-impulse curves for \(K=K_s\) and \(K=3K_s\). Draw dotted verticals at \(x=\sqrt2\) and at \(x=\sqrt{2/K_s}\), and ring each crossing with a one-impulse curve.

Answer. At \(x_s\): the three costs, the cheapest mode, and its saving over one impulse in per cent. From your figure, the value of \(x\) at which the three-impulse curve for \(K_s\) crosses the single impulse (to 3 digits), compared with \(\sqrt{2/K_s}\). State in one sentence what the independence from \(D\) means physically.

5.3.6 Notes

  • A dive needs \(k<1\): guard against a call with \(k\ge1\).
  • All three modes use apsidal burns, so there is no flight path angle to track; vis-viva with the two apsides gives every speed.
  • The three-impulse saving is bought with a trajectory that passes inside \(r_1\) and takes a long time. Say so in the report; do not leave the impression that it is free.

Marks: Theory 4 · Code 2 · Validation 2 · Figure and answer 2.

5.4 Problem 4 — Hohmann against bi-elliptic

Source charts: survey Figs. 5 and 6.

5.4.1 The question

An outward transfer between coplanar circular orbits has radius ratio \(R=r_2/r_1\). A bi-elliptic transfer goes out to a conjunction radius \(r_i=kr_1\) before dropping the pericentre to \(r_2\). For \(R=R_s=\) q4_R (which lies between the two famous thresholds, \(11.94\) and \(15.58\)):

  1. What is the smallest \(k=r_i/r_1\) (and \(r_i/r_2\)) for which the bi-elliptic transfer beats Hohmann?
  2. How much does it save, as a percentage of the Hohmann cost, at \(r_i=10R_sr_1\) and in the limit \(r_i\to\infty\)?
  3. Why is the threshold so extreme for your \(R_s\)?

5.4.2 A. Theory

T1. Write the two Hohmann burns and the total, with absolute values so that one expression covers \(R>1\) and \(R<1\).

T2. Derive the three bi-elliptic burns \(\Delta v_1,\Delta v_2,\Delta v_3\) for \(k\ge1\) (units \(V_{c_1}=1\); use the speed at an apsis of an ellipse with apsides \((r,s)\), \(v(r;s)=\sqrt{2\mu s/(r(r+s))}\)). Show that \(\Delta v_3\) changes sign at \(k=R\) and say what that means physically.

T3. Landmarks. Show that \(k=R\) and \(k=1\) both reproduce the Hohmann cost, and that \(k\to\infty\) gives \((\sqrt2-1)(1+R^{-1/2})\).

T4. Expand each burn for large \(k\) and show that \[\frac{\Delta V}{V_{c_1}}=\frac{\Delta V_\infty}{V_{c_1}}+\frac{\sqrt2}{2}\,\frac{\sqrt R-3}{k}+O(k^{-2}).\] What does the sign of the coefficient tell you for \(R<9\) and \(R>9\)?

T5. Two thresholds. (a) Show that the \(R\) at which Hohmann and the bi-parabolic transfer cost the same is the root of a single equation, and write it down. (b) For \(k\) just above \(R\), \(\Delta v_3\) is a braking burn; differentiate the total cost with respect to \(k\) at \(k=R^{+}\), set it to zero, and write the resulting equation for \(R\). Explain what happens to the sign of the slope for \(R\) above and below this root.

T6. For \(1<k<R\) (an interior conjunction) the cost returns to the Hohmann value at both ends. Argue why it cannot dip below that value in between, or show numerically in V7 that it does not.

5.4.3 B. Code

burns(R, k) returning the three burns, dv_hohmann(R), dv_inf(R), R_inf() and R_touch() for the two thresholds of T5, and threshold_ratio(R) for the smallest \(r_i/r_2\) at which the transfer beats Hohmann (bracket on a logarithmic grid, then brentq).

5.4.4 C. Validation

tag check target tol
V1 \(R=2,\ 5,\ 10,\ 20,\ 100\): Hohmann cost \(0.28446,\ 0.48001,\ 0.52979,\ 0.53473,\ 0.49312\) \(6\times10^{-6}\)
V2 the same \(R\): bi-parabolic cost \(0.70711,\ 0.59946,\ 0.54520,\ 0.50683,\ 0.45563\) \(6\times10^{-6}\)
V3 \(\Delta V(k=1)\) equals the Hohmann cost, same five \(R\) equal \(6\times10^{-6}\)
V4 the two thresholds of T5 \(11.93877,\ 15.58172\) \(2\times10^{-5}\); \(3\times10^{-4}\)
V5 smallest \(r_i/r_2\) at \(R=12,\ 13,\ 14,\ 15\) \(68.0,\ 3.76,\ 1.86,\ 1.21\) \(1\%\)
V6 independent: \(k(\Delta V-\Delta V_\infty)\) at \(k=10^{5}\) for \(R=2,\ 5,\ 9,\ 20,\ 100\) against the T4 coefficient \(-1.121,\ -0.540,\ 0,\ 1.041,\ 4.948\) \(3\times10^{-3}\)
V7 interior crest of the cost, \(R=2\) \(0.2907\) at \(k=1.40\) \(10^{-4}\); \(0.02\)
V8 worked example: LEO \(r_1=6678.137\) km, \(\mu=398600\) km\(^3\)/s\(^2\), \(R=15\) (\(r_2=15r_1\)), \(r_i=400\,000\) km: the three burns, km/s \(3.1100,\ 0.4509,\ 0.5280\); total \(4.0889\) \(5\times10^{-4}\)
V9 largest saving of the bi-parabolic transfer over Hohmann \(7.99\%\) at \(R=57.2\) \(0.01\); \(0.3\)
V10 sensitivity test one check fails —

5.4.5 D. Figure and answer

Figure 4a. \(\Delta V/V_{c_1}\) against \(k=r_i/r_1\) (logarithmic axis, \(1\) to \(10^{4}\)) for \(R\in\{2,5,10,20,100,R_s\}\). Mark the Hohmann locus \(k=R\) (a dashed curve through the points where each exterior curve is born), the bi-parabolic asymptote of each curve (dotted), and the crest of each interior curve.

Figure 4b. The threshold \(r_i/r_2\) against \(R\) for \(11.9\le R\le15.6\) (logarithmic vertical axis), with the two thresholds of T5 as vertical lines.

Answer. For \(R_s\): \(k_{\min}=r_i/r_1\) and \(r_i/r_2\) to 3 significant figures; the saving at \(r_i=10R_sr_1\) and as \(r_i\to\infty\) (per cent of the Hohmann cost; the first may be negative, and then you must say what that means); and a two-sentence explanation of (c) that uses T4 and T5.

5.4.6 Notes

  • Use the same function for \(\Delta v_3\) on both sides of \(k=R\): with an absolute value it is continuous, without one your interior branch is wrong.
  • The threshold search must start at \(k\) slightly above \(R\), not at \(R\) itself, where the cost equals the Hohmann cost by construction.
  • Near \(R=11.94\) the threshold is enormous; use a logarithmic grid reaching at least \(10^{4}R\).

Marks: Theory 4 · Code 2 · Validation 2 · Figure and answer 2.

5.5 Problem 5 — Transfer with a plane change

Source chart: survey Fig. 7.

5.5.1 The question

Two circular orbits with \(r_2=2r_1\) (\(R=2\)) are inclined to each other by \(i_s=\) q5_i. Every burn is made at an apse on the line of nodes and is circumferential (it changes the speed and turns the plane, never the radial velocity). The flight time is free. There are three candidates: a generalised Hohmann (two burns, the turn shared between them), a three-impulse (bi-elliptic, three burns, the turn shared among them) transfer, and a transfer through infinity.

  1. Which is optimal for \(i_s\), and with which conjunction radius \(r_i/r_1\) and which split \((\alpha_1,\alpha_2,\alpha_3)\) of the turn?
  2. What does it save, in per cent, against each of the other two modes?
  3. For which inclinations does each mode win at \(R=2\)? Give both crossover angles to \(0.1^\circ\).

5.5.2 A. Theory

T1. A burn that changes the speed from \(a\) to \(b\) and turns the plane through \(\alpha\) costs \(\Delta v=\sqrt{a^2+b^2-2ab\cos\alpha}\). Prove it. Show that \(\partial\Delta v/\partial\alpha=ab\sin\alpha/\Delta v\) is zero at \(\alpha=0\) unless \(a=b\), and explain what that implies for tacking a small turn onto a burn that already changes the speed.

T2. Minimise \(\sum_j\Delta v_j\) subject to \(\sum\alpha_j=i\) with a Lagrange multiplier. Show that every burn that takes a turn has the same marginal price \(a_jb_j\sin\alpha_j/\Delta v_j=\lambda\). Which burns take most of the turn, and why?

T3. Express the three bi-elliptic burns’ speeds \((a_j,b_j)\) in terms of \(R\) and \(k=r_i/r_1\) and write the total cost \(\Delta V(k,i)\) as a minimum over the split. Show that \(k=R\) is the generalised Hohmann transfer and \(k\to\infty\) the transfer through infinity.

T4. Show that through infinity the plane change is made entirely at the third apse by an infinitesimal impulse and that \[\frac{\Delta V_\infty}{V_{c_1}}=(\sqrt2-1)\left(1+\sqrt{r_1/r_2}\right)\quad\text{independently of }i.\]

T5. Expand \(\Delta V(k,i)-\Delta V_\infty\) for large \(k\) with the whole turn at burn 2 and obtain a coefficient \(c(R,i)/k\). Show that through infinity is not even a local minimum when \(c<0\), and that \(c\ge0\) is equivalent to \(\cos i\le\left[3\left(\sqrt x+1/\sqrt x\right)-2\right]/8\) with \(x=r_1/r_2\).

5.5.3 B. Code

  • split3(R, k, i) returns the minimum of the three-burn cost over the split and the angles \((\alpha_1,\alpha_2,\alpha_3)\). The burn cost is convex for small \(\alpha\) and concave for large, so one local optimiser from one starting point is not enough: use a grid of step \(0.25^\circ\) or finer on two of the angles, then refine locally.
  • split2(R, i) for the generalised Hohmann; dv_inf(R); best_finite(R, i) minimising over \(k>R\) (grid in \(k\), then a bounded refinement); mode(R, i) returning the best mode and all three costs; i_s(x) from T5.

5.5.4 C. Validation

tag check target tol
V1 \(R=1\) (two circles of equal radius), \(i=10^\circ,\ 30^\circ,\ 50^\circ\): \(\Delta V\) and optimal \(r_i/r_1\) \(0.17369,\ 0.50015,\ 0.75653\); \(1.02,\ 1.18,\ 1.91\) \(2\times10^{-4}\); \(0.02\), \(0.03\), \(0.05\)
V2 \(R=2\), \(i=30^\circ\), generalised Hohmann: \(\Delta V\) and the turn at burn 1 \(0.48639\); \(5.1^\circ\) \(2\times10^{-5}\); \(0.1^\circ\)
V3 \(R=2\), \(i=50^\circ\), \(k=3.07\): \(\Delta V\) and the three turns \(0.66855\); \(4.6^\circ,\ 41.9^\circ,\ 3.5^\circ\) \(2\times10^{-4}\); \(0.1^\circ\)
V4 \(R=2\): through infinity \(0.70711\) \(2\times10^{-5}\)
V5 \(R=2\): tie of the three-impulse mode with through infinity \(57.37^\circ\) \(0.05^\circ\)
V6 \(i_s(x)\) of T5 at \(x=1,\ 0.5,\ 0.25,\ 1/6,\ 1/8\) \(60.00^\circ,\ 56.94^\circ,\ 46.57^\circ,\ 34.75^\circ,\ 19.40^\circ\) \(0.02^\circ\)
V7 independent: pure plane change \(a=b=1\), \(\alpha=40^\circ\) against \(2\sin(\alpha/2)\) equal \(10^{-12}\)
V8 independent: \(k(\Delta V-\Delta V_\infty)\) at \(k=10^{5}\) against \(c(R,i)\) of T5 for \((R,i)=(1,50^\circ),\ (4,30^\circ),\ (4,50^\circ),\ (12,10^\circ)\) \(-0.21887,\ -0.36867,\ +0.08270,\ +0.35824\) \(2\times10^{-5}\)
V9 sensitivity test one check fails —

5.5.5 D. Figure and answer

Figure 5a. At \(R=2\), \(\Delta V/V_{c_1}\) against \(i\in[0^\circ,70^\circ]\) for the generalised Hohmann, the best three-impulse transfer (with \(k\) free) and through infinity; shade the region of each mode’s optimality and mark the two crossover angles.

Figure 5b. Inside the three-impulse range: the optimal \(r_i/r_1\) and the three turns \(\alpha_1,\alpha_2,\alpha_3\) against \(i\).

Answer. For \(i_s\): the optimal mode, \(\Delta V\), \(r_i/r_1\), \((\alpha_1,\alpha_2,\alpha_3)\) to \(0.1^\circ\), the savings in per cent over the other two modes, and the two crossover angles. One sentence on where the plane change is made and why (T2).

5.5.6 Notes

  • Keep \(k\) strictly above \(R\) in the search for the finite optimum; \(k=R\) is the Hohmann mode and is evaluated separately.
  • A finite \(k\) should only be called optimal if it wins by more than a small tolerance (\(10^{-7}\)), or numerical noise will invent a three-impulse region.
  • Speed: precompute the burn-cost arrays on the angle grid once per \(k\).

Marks: Theory 4 · Code 2 · Validation 2 · Figure and answer 2.

5.6 Problem 6 — Circle to ellipse

Source charts: survey Figs. 8, 9 and 11.

5.6.1 The question

A spacecraft on a circular orbit of radius \(r_1\) must reach an ellipse with pericentre radius \(P\,r_1\) and apocentre radius \(A\,r_1\). The orientation of the ellipse is free and the time is free. Burns are tangential and at apsides; there are three candidate routes: P first (transfer ellipse with apsides \((1,P)\)), A first (apsides \((1,A)\)) and through infinity. Take \(P=10\) and \(A=A_s=\) q6_A.

  1. Which route is cheapest, and what are the three costs?
  2. At what pericentre ratio \(P^{*}(A_s)\) does through infinity take over? Does it win at \(P=10\)?
  3. If the initial orbit were an ellipse rather than a circle, would that boundary move?

5.6.2 A. Theory

T1. Lemma. A tangential burn at an apse of radius \(r\), making the speed there \(w\), leaves \(r\) an apse of the new orbit and moves only the other apse. Show that the new other apse is \(s=r^2w^2/(2-rw^2)\). Conclude that in the \((P,A)\) plane every burn is a horizontal or vertical step.

T2. Write the cost of the P-first and A-first routes and of through infinity, \[\frac{\Delta V_\infty}{V_{c_1}}=(\sqrt2-1)+\sqrt{\frac2P}\left(1-\sqrt{\frac{A}{A+P}}\right).\] Derive the last one (burn 1 to a parabola; free switch at infinity; one braking burn at the pericentre).

T3. Show that \(\Delta V_\infty\to\sqrt2-1\) as \(A\to\infty\) for every \(P\), and that on the diagonal \(P=A=R\) it reduces to the bi-parabolic cost of Problem 4.

T4. The rule: the P-first route is cheaper when the ellipse lies inside the circle (\(A\le1\)); the A-first route is cheaper otherwise. Prove that arriving at the pericentre is never dearer than at the apocentre for a parabolic arrival: with \(u=\sqrt{P/(A+P)}\), \(w=\sqrt{A/(A+P)}\) (\(u^2+w^2=1\), \(u\le w\)), compare \(\sqrt2(1-w)/\big(u\sqrt{A+P}\big)\) with \(\sqrt2(1-u)/\big(w\sqrt{A+P}\big)\).

T5. Boundaries. For \(A\to\infty\), expand the A-first and through-infinity costs and show that \[A\,(\Delta V_A-\Delta V_\infty)\to\tfrac{\sqrt2}{2}(\sqrt P-3)\ (P>1),\qquad\tfrac{\sqrt2}{2}(1-3\sqrt P)\ (P<1),\] which changes sign at \(P=9\) and at \(P=1/9\).

T6. Coaxial lemma. For aligned coaxial ellipses (initial \((P_1,A_1)\), final \((P_2,A_2)\), \(P_2\ge P_1\), \(A_{\max}=\max(A_1,A_2)\)) show that the difference \(\Delta V_{\rm Hohmann}-\Delta V_\infty\) does not contain the smaller apocentre: the term common to both costs cancels. Conclude that the boundary is a curve in the plane \(\big(P_2/A_{\max},\,P_2/P_1\big)\), whatever the initial orbit.

5.6.3 B. Code

v(r, s); hohmann_P, hohmann_A, through_inf; best(P, A) returning the cheapest route and all three costs (only the pair \(\{P,A\}\) matters, so sort it); P_right(A) and P_left(A) for the two boundaries by brentq; F(P1, P2, Amax); boundary(y) giving \(P_2/P_1\) at \(P_2/A_{\max}=y\) from \(F=0\); and boundary_direct(A1, y) giving the same from the full aligned-axes costs (the two apsidal transfers \(P_1\leftrightarrow A_2\) and \(A_1\leftrightarrow P_2\), whichever is cheaper, against through infinity), without using \(F\).

5.6.4 C. Validation

tag check target tol
V1 circle-to-circle LEO to GEO, \(P=A=6.31374\) \(0.50384\) \(2\times10^{-5}\)
V2 \(\sqrt P=0.432\), \(\sqrt A=0.452\): P first, A first, infinity \(1.07778,\ 1.10681,\ 1.32128\) \(2\times10^{-5}\)
V3 \((P,A)\to\Delta V\): \((0.3,1),\ (0.3,5),\ (0.1,20),\ (2,2),\ (1,2)\) \(0.321,\ 0.399,\ 0.425,\ 0.284,\ 0.155\) \(2\times10^{-3}\)
V4 left boundary \(P^{*}(A)\) at \(A=0.5,\ 2,\ 10\) \(0.083761,\ 0.094781,\ 0.107206\) \(2\times10^{-6}\)
V5 right boundary \(P^{*}(A)\) at \(A=20,\ 50,\ 100\) \(10.644,\ 9.622,\ 9.305\) \(10^{-3}\)
V6 \(A(\Delta V_A-\Delta V_\infty)\) at \(A=10^{7}\) for \(P=0.05,\ 0.2,\ 5,\ 12\) \(+0.23277,\ -0.24158,\ -0.54018,\ +0.32817\) \(10^{-4}\)
V7 \(F(1,10,40)\) for \(A_1=1,\ 2,\ 5,\ 20,\ 40\) \(0.000511919\) each \(10^{-9}\)
V8 independent: boundary_direct for \(y=0.5\) and \(A_1=1,\ 1.5,\ 3,\ 8\) \(10.5506\) each \(5\times10^{-4}\)
V9 independent: the route rule of T4 on \(10^{5}\) random \((P,A)\) (\(P\) log-uniform over \(e^{\pm2}\), \(A/P\) over \([1,e^{2.5}]\), fixed seed) \(0\) violations \(0\)
V10 sensitivity test one check fails —

5.6.5 D. Figure and answer

Figure 6a. At \(A=A_s\): \(\Delta V/V_{c_1}\) against \(P\) (\(0.01\) to \(30\), logarithmic) for the three routes, with the optimum as a heavy grey line; ring the points where through infinity takes over.

Figure 6b. The boundary in the plane \((P_2/A_{\max},\,P_2/P_1)\) from \(P_2/A_{\max}=0\) to \(1\) (from boundary), with the points from the circle-to-ellipse boundary \(P^{*}(A)\) and from boundary_direct for five random initial ellipses overplotted. Mark \((0,9)\) and \((1,11.938)\).

Answer. For \((P,A)=(10,A_s)\): the cheapest route and the three costs; \(P^{*}(A_s)\) and whether through infinity wins at \(P=10\); and, for (c), the largest deviation between the boundary points of the random initial ellipses and the circle-to-ellipse curve. State the lemma of T6 in one sentence.

5.6.6 Notes

  • In the P-first route the second burn is at \(P\), in the A-first route at \(A\); the sign of each burn changes with the configuration, so keep the absolute values.
  • The right boundary exists only for \(A\ge11.94\) (it starts on the diagonal); your \(A_s\) is inside that range.
  • Plot the three costs on a logarithmic \(P\) axis before writing the root-finders; the crossings are easier to bracket after you have seen them.

Marks: Theory 4 · Code 2 · Validation 2 · Figure and answer 2.

5.7 Problem 7 — Congruent ellipses

Source chart: survey Fig. 14.

5.7.1 The question

A satellite on an ellipse of eccentricity \(e\) must rotate its apse line through the angle \(\omega\) in the same plane, keeping the size and shape of the orbit: the final orbit is congruent to the initial one. The flight time is free. Two modes compete: the symmetric two-impulse transfer and the transfer through infinity. Take \(\omega=\omega_s=\) q7_om and let \(e_t=e^{*}(\omega_s)-0.05\).

  1. Find the boundary eccentricity \(e^{*}(\omega_s)\) above which through infinity is cheaper.
  2. For a satellite with \(e=e_t\) and \(\omega=\omega_s\): the cost of both modes (units \(\sqrt{\mu/a}\)), the true anomaly on the transfer ellipse at which the burn is made, and the eccentricity \(e_T\) of the transfer ellipse.
  3. How does \(1-e^{*}\) behave as \(\omega\to0\)?

5.7.2 A. Theory

T1. Units \(\mu=1\), \(a=1\) for both orbits. Take the pericentre of orbit 1 along \(\theta=0\) and that of orbit 2 along \(\theta=\omega\). Give \(p\), the position and the velocity components \((v_r,v_\theta)\) at polar angle \(\theta\) on each orbit.

T2. Mirror symmetry. Explain why the optimal two-impulse transfer between two congruent ellipses is its own mirror image in the bisector of the two apse lines, with burn points at \(\theta=\psi\pm\nu\), where \(\psi=\omega/2\) or \(\omega/2+\pi\) is the direction of the bisector, and with the transfer’s apse line on the bisector. The unknowns are then the true anomaly \(\nu\) on the transfer conic at the burn and its eccentricity \(e_T\). Derive the total cost \(\Delta V(\nu,e_T;e,\omega,\psi)\) (two equal burns). Say what \(\nu=0\) means and show that, for the right \(e_T\), it is the one-impulse transfer, with cost \(2e\sin(\omega/2)/\sqrt{1-e^2}\); conclude that the symmetric optimum is never dearer than one impulse.

T3. Through infinity: show that the cost is \(2(\sqrt2-\sqrt{1+e})/\sqrt{1-e}\), independent of \(\omega\).

T4. At \(\omega=180^\circ\) the transfer is the circle through the two apocentres. Find its cost in closed form and show that the tie with through infinity happens where \(\sqrt{1-e}+2e=\sqrt{2(1+e)}\).

T5. For \(\omega\to0\) both costs vanish and \(e^{*}\to1\). Explain why the boundary must approach \(e=1\), and estimate numerically (V5) the leading behaviour of \(1-e^{*}\).

5.7.3 B. Code

  • sym_cost(e, om, nu, eT, psi) for T2 (return \(+\infty\) for an invalid conic, i.e. \(1+e_T\cos\nu\le0\), or for \(e_T\ge1\) with \(\nu>0\)).
  • dv_sym(e, om) minimises over \((\nu,e_T)\) for both \(\psi\): evaluate on a grid of at least \(180\times100\) points, then polish the best few with scipy.optimize.minimize (Nelder–Mead). Return the cost, \(\nu\), \(e_T\) and \(\psi\).
  • dv_inf(e); e_star(om) by bracketing and brentq on \(\Delta V_{\rm sym}-\Delta V_\infty\).
  • unrestricted(e, om): an independent optimiser for the general coplanar two-impulse transfer. Its variables are \((\theta_1,\theta_2,p_T)\): burn points on orbits 1 and 2 and the semi-latus rectum of the transfer conic through them. For a conic through \((r_1,\theta_1)\) and \((r_2,\theta_2)\) sweeping \(\Delta=\theta_2-\theta_1\) (mod \(2\pi\)), \[e\cos\nu_1=\frac{p_T}{r_1}-1,\quad e\sin\nu_1=\frac{(p_T/r_1-1)\cos\Delta-(p_T/r_2-1)}{\sin\Delta},\quad e\cos\nu_2=\frac{p_T}{r_2}-1,\quad e\sin\nu_2=\frac{(p_T/r_1-1)-(p_T/r_2-1)\cos\Delta}{\sin\Delta},\] with \(v_r=e\sin\nu/\sqrt{p_T}\) and \(v_\theta=\sqrt{p_T}/r\) at each end. Use at least 60 random starts, elliptic transfers only.

5.7.4 C. Validation

tag check target tol
V1 \(\Delta V_{\rm sym}\) for \((e,\omega)=(0.5,70^\circ),\ (0.6,60^\circ),\ (0.8,40^\circ),\ (0.3,100^\circ),\ (0.7,120^\circ)\) \(0.28993,\ 0.31510,\ 0.33264,\ 0.22334,\ 0.61523\) \(10^{-5}\)
V2 \((e,\omega)=(0.5,70^\circ)\): \(e_T\) and the burn’s true anomaly \(0.4107\); \(106.91^\circ\) \(5\times10^{-4}\); \(0.05^\circ\)
V3 \(e^{*}\) at \(\omega=16^\circ,\ 34^\circ,\ 65^\circ\) \(0.923,\ 0.824,\ 0.697\) \(10^{-3}\)
V4 \(e^{*}(180^\circ)\) from your code and from the closed form of T4 \(0.535334\) \(2\times10^{-6}\)
V5 \((1-e^{*})/(2\omega^2)\) (\(\omega\) in radians) at \(\omega=2^\circ\) and \(1^\circ\) \(0.973,\ 0.997\) \(0.01,\ 0.004\)
V6 independent: unrestricted(0.5, 70°) against dv_sym equal \(2\times10^{-6}\)
V7 one impulse is dearer than the symmetric transfer at \((0.5,70^\circ)\) \(0.6623>0.2899\) —
V8 sensitivity test one check fails —

5.7.5 D. Figure and answer

Figure 7. The \((\omega,e)\) plane, \(5^\circ\le\omega\le180^\circ\) (at least 25 values of \(\omega\)): the boundary curve \(e^{*}(\omega)\) with the regions labelled “two impulses” below and “through infinity” above; mark \((\omega_s,e_t)\) and the minimum \(e^{*}(180^\circ)\). Add the small-\(\omega\) law \(1-e=2\omega^2\) as a dashed curve.

Answer. \(e^{*}(\omega_s)\); and at \((e_t,\omega_s)\): \(\Delta V_{\rm sym}\), \(\Delta V_\infty\), the saving of the cheaper one in per cent, the burn’s true anomaly \(\nu\) (degrees) and \(e_T\). For (c), the limiting value of \((1-e^{*})/(2\omega^2)\) and one sentence on why the boundary falls steeply from \(e=1\).

5.7.6 Notes

  • A grid is not enough: the minimum is flat in some directions, and the grid values differ from the polished ones in the fifth digit. Compare with V1 only after polishing.
  • Both choices of \(\psi\) must be tried; the better one depends on \((e,\omega)\).
  • unrestricted is slow and noisy; it is a check, not part of the figure.

Marks: Theory 4 · Code 2 · Validation 2 · Figure and answer 2.

5.8 Problem 8 — The useful angle

Source chart: survey Fig. 16.

5.8.1 The question

On a coasting ellipse of eccentricity \(e\), the impulse that starts or ends an optimal two-impulse arc cannot point just anywhere: its direction must lie in a narrow wedge between the local horizontal and the velocity, the useful angle. At its two edges, \(\phi_i\) and \(\phi_s\), a partner burn appears elsewhere on the same ellipse. For \(e=e_s=\) q8_e and a burn at true anomaly \(\nu_b=\) q8_nu degrees:

  1. What are \(\phi_i\), \(\phi_s\) and the width of the wedge, compared with the flight path angle \(\gamma\)?
  2. Where is the partner burn for each edge, and on which side of the apse line?
  3. At which true anomaly is the wedge widest, and how wide is it there?

5.8.2 Background: the primer vector in a paragraph

For fuel-optimal impulsive trajectories with free final time, Lawden showed that there is a vector function \(\mathbf p(t)\), the primer, such that: (i) \(\lvert\mathbf p\rvert\le1\) along the whole trajectory; (ii) each impulse is made where \(\lvert\mathbf p\rvert=1\), in the direction of \(\mathbf p\); (iii) at such an instant \(\dot{\mathbf p}\cdot\mathbf p=0\) (a maximum of \(\lvert\mathbf p\rvert\)); (iv) between impulses \(\ddot{\mathbf p}=G(\mathbf r)\mathbf p\) with \(G=\partial\mathbf g/\partial\mathbf r\). In the planar Kepler problem with free time and an orbit (not a position) as the target, the cost-to-go depends on the orbit only through \((a,e,\omega)\), so its costate is a constant row \((\lambda_a,\lambda_e,\lambda_\omega)\) and the primer is the gradient with respect to the velocity of \(\lambda_aa+\lambda_ee+\lambda_\omega\omega\).

5.8.3 A. Theory

T1. Justify the last sentence of the Background in a few lines: why the costates of \((a,e,\omega)\) are constant on a coast arc, why the phase along the orbit carries none, and why the primer is the velocity-gradient of \(\lambda\cdot(a,e,\omega)\). Then use Gauss’s equations (from the Gauss variational equations slides) for an impulse \((\delta v_r,\delta v_\theta)\) to obtain \(\nabla_va\), \(\nabla_ve\), \(\nabla_v\omega\). In units \(\mu=1\), \(p=1\), with \(q=1+e\cos\nu\), \(r=1/q\), you should obtain a \(2\times3\) matrix \(B(\nu)\) with \(\mathbf p(\nu)=B(\nu)\boldsymbol\lambda\) whose entries are elementary functions of \(\nu\).

T2. Let \(\phi\) be the angle of the thrust direction above the local horizontal, measured towards the radial direction, so that \(\hat{\mathbf p}=(\sin\phi,\cos\phi)\) in the \((\hat r,\hat\theta)\) basis. At a burn at \(\nu_1\) demand \(\mathbf p(\nu_1)=(\sin\phi,\cos\phi)\) and \(\mathbf p\cdot\mathbf p'=0\). Show that these three conditions fix \(\boldsymbol\lambda(\phi)\) uniquely: a particular solution of \(B(\nu_1)\boldsymbol\lambda=\hat{\mathbf p}\) plus a multiple of the null vector of \(B(\nu_1)\), the multiple being fixed by \(\mathbf p\cdot\mathbf p'=0\) (give the formula).

T3. A direction \(\phi\) is admissible at \(\nu_1\) if \(\lvert\mathbf p(\nu)\rvert^2\le1\) for all \(\nu\) on the orbit. Explain why this is the right condition for a direction to be usable on this ellipse (use (i)–(iii)). Explain why, at an edge of the admissible set, \(\lvert\mathbf p\rvert\) must touch \(1\) a second time at a partner anomaly \(\nu_2\), and why a two-impulse arc on this ellipse must therefore have both burns on edges.

T4. Show that \(\tan\gamma=e\sin\nu/(1+e\cos\nu)\), and show that a thrust exactly along the velocity (\(\phi=\gamma\)) is never admissible: what happens to \(\lambda(\phi)\) as \(\phi\to\gamma\)?

5.8.4 B. Code

  • B(nu, e) and dB(nu, e), lam_of(nu1, e, phi) from T1–T2.
  • excess(nu1, e, phi): \(\max_\nu\lvert\mathbf p\rvert^2-1\) over a grid of at least \(2000\) points round the orbit.
  • window(nu1, e) returns \((\phi_i,\phi_s,\gamma)\) or None. The window can be as narrow as \(0.01^\circ\) for large \(e\), so scan \(\phi\in(0,\gamma)\) finely, locate the minimum of excess by golden-section search, and only then find each edge by bisection (admissible means excess \(\le10^{-9}\)).
  • partner(nu1, e, phi) returns the anomaly outside a \(2^\circ\) neighbourhood of \(\nu_1\) at which \(\lvert\mathbf p\rvert^2\) is largest.

5.8.5 C. Validation

tag check target tol
V1 \(e=0.3\): \([\phi_i,\phi_s]\) at \(\nu=60^\circ\) and \(150^\circ\) \([4.774^\circ,\,6.564^\circ]\); \([3.156^\circ,\,5.246^\circ]\) \(3\times10^{-3}\) deg
V2 \(e=0.3\), \(\nu=150^\circ\): partner of \(\phi_i\) and of \(\phi_s\) \(316.1^\circ\); \(41.8^\circ\) \(0.3^\circ\)
V3 independent: at an edge, \(\lvert\mathbf p(\nu_1)\rvert=1\), \(\mathbf p\cdot\mathbf p'(\nu_1)=0\), \(\lvert\mathbf p(\nu_2)\rvert=1\), \(\mathbf p\cdot\mathbf p'(\nu_2)=0\) \(1,\ 0,\ 1,\ 0\) \(10^{-8}\)
V4 \(e=0.01\), \(\nu=90^\circ\): \(\phi_i/\gamma\) and \(\phi_s/\gamma\) \(1/3,\ 1/2\) \(2\times10^{-3}\)
V5 \(e=0.5\), \(\nu=0.5^\circ\): \(\phi_i/\gamma\) and \(\phi_s/\gamma\) \(3/7=0.42857,\ 9/17=0.52941\) \(5\times10^{-4}\)
V6 a window exists at \((e,\nu)=(0.92,122^\circ)\) and not at \((0.93,122^\circ)\) yes; no —
V7 property: the width \(\phi_s-\phi_i\) is positive and \(\phi_s<\gamma\) for 20 random \((e,\nu)\) with \(e<0.9\) (fixed seed) all 20 —
V8 sensitivity test one check fails —

5.8.6 D. Figure and answer

Figure 8a. For \(0<\nu<180^\circ\) (steps of \(5^\circ\) or finer): \(\phi_i(\nu)\) solid, \(\phi_s(\nu)\) dashed and \(\gamma(\nu)\) dotted, for \(e\in\{0.3,0.7,0.9,e_s\}\), shading the wedge between \(\phi_i\) and \(\phi_s\).

Figure 8b. At your \((e_s,\nu_b)\): \(\lvert\mathbf p(\nu)\rvert^2\) round the whole orbit (\(0\) to \(360^\circ\)) for \(\phi\) slightly below \(\phi_i\), at \(\phi_i\), at the middle of the wedge, at \(\phi_s\) and slightly above \(\phi_s\). Mark the contacts with \(1\).

Answer. For \((e_s,\nu_b)\): \(\gamma\), \(\phi_i\), \(\phi_s\) and the width (degrees, 3 decimals); the two partner anomalies and the side of the apse line each lies on; the true anomaly at which the width is largest (to \(5^\circ\)) and that width. One sentence each on what happens to the wedge as \(e\to0\) and as \(e\) rises towards \(0.925\).

5.8.7 Notes

  • Cross-check with Problem 7 (optional, not marked): the burn of the symmetric transfer at \((e,\omega)=(0.5,70^\circ)\) has \(e_T=0.4107\) at \(\nu=106.91^\circ\); show that its thrust angle above the local horizontal equals \(\phi_i\) of that transfer ellipse, \(8.10^\circ\).
  • The forward and rearward directions \(\phi\) and \(\phi+180^\circ\) are equivalent; work with forward only. For \(\pi<\nu<2\pi\) use mirror symmetry.
  • \(\lambda(\phi)\) becomes badly conditioned near \(\phi=\gamma\) and near the apses; keep \(\phi\) away from \(\gamma\) by a small fraction.

Marks: Theory 4 · Code 2 · Validation 2 · Figure and answer 2.

5.9 Problem 9 — Rendezvous: fitting a transfer to a phase angle

Source charts: survey Figs. 19 and 20.

5.9.1 The question

An interceptor is on a circular orbit of radius \(r_1\); the target is on a circular orbit of radius \(r_2=kr_1\), \(k=k_s=\) q9_k. The target is ahead of the interceptor by the phase angle \(\beta=\beta_H+\Delta\beta\), where \(\beta_H\) is the phase that suits a Hohmann transfer. For \(\Delta\beta=\Delta\beta_s\), defined as the value midway between \(\Delta\beta^{*}(k_s)\) (see T4) and \(360^\circ\), rounded to the nearest \(5^\circ\):

  1. A Hohmann rendezvous waits on the initial orbit for the phase to come right. How long does the whole rendezvous take?
  2. A bi-elliptic transfer with no waiting can absorb the offset through its intermediate radius. Find \(r_i/r_1\), the rendezvous time and the \(\Delta V\).
  3. How much faster is the bi-elliptic rendezvous, and how much dearer? Give the times in minutes for \(r_1=6678.137\) km, \(\mu=398600\) km\(^3\)/s\(^2\).

5.9.2 A. Theory

T1. A transfer that sweeps the interceptor through the angle \(\Theta\) in the time \(t\) arrives when the target has moved \(\omega_2t\). Show that rendezvous requires \(\beta+\omega_2t=\Theta\pmod{2\pi}\). In units \(r_1=1\), \(\mu=1\) the angular speeds are \(\omega_1=1\) and \(\omega_2=k^{-3/2}\). For a Hohmann transfer find \(t_H\) and the Hohmann phase angle \(\beta_H=\pi\left[1-\left(\frac{1+k}{2k}\right)^{3/2}\right]\); show that \(\beta_H\le\pi(1-2^{-3/2})=116.36^\circ\).

T2. If \(\Delta\beta\ne0\) a Hohmann rendezvous must wait. Show that the wait is \(\Delta\beta/(\omega_1-\omega_2)\) and that the longest wait is one synodic period. Write the total Hohmann rendezvous time \(t_H+\Delta\beta/(\omega_1-\omega_2)\).

T3. A bi-elliptic transfer through \(r_i\) flies half of an ellipse with apsides \((1,r_i)\) and half of one with apsides \((r_i,k)\), so \(\Theta=2\pi\). Write \(t_{bi}(r_i)\). Show that the offset it absorbs with no waiting is \[\Delta\beta=3\pi-\omega_2(t_{bi}-t_H)\ \ (r_i\ge k,\ \text{outer}),\qquad\Delta\beta=\pi-\omega_2(t_{bi}-t_H)\ \ (1\le r_i\le k,\ \text{inner}),\] by choosing the branch of “\(\pmod{2\pi}\)” that puts \(\Delta\beta\) in \([0,2\pi]\). Find the range of \(\Delta\beta\) and the rendezvous time at the two ends of each branch.

T4. Equal-time points. Show that the inner bi-elliptic and the waiting Hohmann transfer take the same total time at \(\Delta\beta=\pi(1-k^{-3/2})\), and the outer one and the waiting Hohmann at \(\Delta\beta^{*}(k)=3\pi(1-k^{-3/2})\). Show that \(\Delta\beta^{*}\ge2\pi\) exactly when \(k\ge3^{2/3}\), and say what that means for \(k=k_s\).

T5. Write the total \(\Delta V\) of the three tangential burns of the bi-elliptic transfer through \(r_i\) (any \(r_i>0\)).

5.9.3 B. Code

  • beta_H(k), t_H(k), t_bi(k, ri), dbeta(k, ri) (the two branches of T3), ri_of(k, db, branch) by brentq, dv_bi(k, ri), and time_hohmann(k, db).
  • fly(k, ri, beta): the independent check. Integrate the two-body equations with scipy.integrate.solve_ivp (relative and absolute tolerances \(10^{-12}\) or tighter): start on the unit circle at angle \(0\), apply burn 1 (tangential, to the apse speed of the ellipse with apsides \((1,r_i)\)), integrate to the next apse (an event on \(\mathbf r\cdot\mathbf v=0\); take a short first step so that the event does not fire at \(t=0\)), apply burn 2, integrate to the next apse, and compare the interceptor’s position with the target’s at that instant (the target is at angle \(\beta+\omega_2t\) on the circle of radius \(k\)). Return the miss distance and \(t\).

5.9.4 C. Validation

tag check target tol
V1 \(\beta_H\) at \(k\to\infty\); at \(k=2\); at \(k=10\) \(116.36^\circ,\ 63.09^\circ,\ 106.58^\circ\) \(0.005^\circ,\ 0.01^\circ,\ 0.01^\circ\)
V2 outer \(r_i/r_1\) at \(\Delta\beta=\pi(1-k^{-3/2})\) for \(k=2,\ 3,\ 4,\ 5\) \(3.733,\ 5.361,\ 7.017,\ 8.686\) \(2\times10^{-3}\)
V3 the corner \(k\to1\), \(\Delta\beta\to0\) (outer): \(r_i/r_1\) \(2^{5/3}-1=2.1748\) \(10^{-4}\)
V4 smallest \(k\) for which \(\Delta\beta^{*}\ge2\pi\) \(3^{2/3}=2.08008\) \(10^{-8}\)
V5 time at \(r_i=r_2\) and at \(\Delta\beta\to0\) (\(k=3\)), \((t_{bi}-t_H)/T_2\) \(0.5,\ 1.5\) \(10^{-8}\)
V6 independent: fly for \(k=3\) at \(\Delta\beta=1\) rad on both branches: miss distance (units \(r_1\)) \(0\) \(10^{-8}\)
V7 independent: fly for the Hohmann transfer at \(\beta=\beta_H\) and at \(\beta=\beta_H+0.5\) rad \(0\); \(\ne0\) \(10^{-8}\); —
V8 break-even times from your own time functions, solved for \(\Delta\beta\) at \(k=2,\ 3,\ 10\) (inner) \(116.36^\circ,\ 145.36^\circ,\ 174.31^\circ\) \(0.01^\circ\)
V9 sensitivity test one check fails —

5.9.5 D. Figure and answer

Figure 9a. \(r_i/r_1\) (logarithmic axis, \(1\) to \(10\)) against \(\Delta\beta\): the outer branch for \(\Delta\beta\) from \(0^\circ\) to \(360^\circ\) on the left of a vertical line labelled “360/0”, the inner branch from \(0^\circ\) to \(\pi(1-k^{-3/2})\) (where \(r_i=r_1\)) on the right, for \(k\in\{1,2,3,5,10\}\). Add the two dashed loci of T4 (the outer curves evaluated at \(\pi(1-k^{-3/2})\) and at \(3\pi(1-k^{-3/2})\)) and hatch the region where the outer transfer beats the waiting Hohmann transfer.

Figure 9b. For \(k=k_s\): the total rendezvous time against \(\Delta\beta\in[0,360^\circ]\) (in units of the target period) for the waiting Hohmann transfer and the outer bi-elliptic transfer; mark \(\Delta\beta^{*}(k_s)\) and \(\Delta\beta_s\).

Answer. For \((k_s,\Delta\beta_s)\): \(\Delta\beta^{*}(k_s)\) and \(\Delta\beta_s\) (degrees); \(r_i/r_1\) for the bi-elliptic transfer; the two rendezvous times in minutes, and the percentage saved; the two \(\Delta V\) values in units of \(V_{c_1}\) and the percentage extra; and the miss distance from fly. One sentence: when is buying time with \(\Delta V\) worth it?

5.9.6 Notes

  • The inner branch lies in \(1\le r_i\le k\), the outer in \(r_i\ge k\); at exactly \(r_i=k\) the two meet. Evaluate dbeta just on the correct side when bracketing.
  • Both bi-elliptic burns 2 and 3 change sign between the inner and outer branches. Use magnitudes in the cost, and take care in the integrator that burn 2 and burn 3 set the speed to the apse speed of the next ellipse.
  • The time of flight is the time to the second apse crossing; do not add the waiting time for the bi-elliptic transfer: it has none.

Marks: Theory 4 · Code 2 · Validation 2 · Figure and answer 2.

6 Final checklist