Optimal Spacecraft Trajectories John E.

Prussing
Visit to download the full and correct content document:
https://ebookmass.com/product/optimal-spacecraft-trajectories-john-e-prussing/
O P T I M A L S PA C E C R A F T T R A J E C T O R I E S
Optimal Spacecraft
Trajectories

john e. prussing
Professor Emeritus
Department of Aerospace Engineering
University of Illinois at Urbana-Champaign
Urbana, Illinois, USA

3
 3
Great Clarendon Street, Oxford, OX2 6DP,
United Kingdom
Oxford University Press is a department of the University of Oxford.
It furthers the University’s objective of excellence in research, scholarship,
and education by publishing worldwide. Oxford is a registered trade mark of
Oxford University Press in the UK and in certain other countries
© John E. Prussing 2018
The moral rights of the author have been asserted
First Edition published in 2018
Impression: 1
All rights reserved. No part of this publication may be reproduced, stored in
a retrieval system, or transmitted, in any form or by any means, without the
prior permission in writing of Oxford University Press, or as expressly permitted
by law, by licence or under terms agreed with the appropriate reprographics
rights organization. Enquiries concerning reproduction outside the scope of the
above should be sent to the Rights Department, Oxford University Press, at the
address above
You must not circulate this work in any other form
and you must impose this same condition on any acquirer
Published in the United States of America by Oxford University Press
198 Madison Avenue, New York, NY 10016, United States of America
British Library Cataloguing in Publication Data
Data available
Library of Congress Control Number: 2017959440
ISBN 978–0–19–881108–4 (hbk.)
ISBN 978–0–19–881111–4 (pbk.)
Printed and bound by
CPI Group (UK) Ltd, Croydon, CR0 4YY
Preface

T his text takes its title from an elective course at the University of Illinois at Urbana-
Champaign that has been taught to graduate students for the past 29 years. The
book assumes the reader has a background of undergraduate engineering (aerospace,
mechanical, or related), physics, or applied mathematics. Parts of the book rely heav-
ily on Optimal Control Theory, but an entire chapter on that topic is provided for those
unfamiliar with the subject.
Books on the field of optimal spacecraft trajectories are very few in number and date
back to the 1960s, and it has been nearly 40 years since a comprehensive new book has
appeared. Hence this volume.
Some classical results are presented using modern formulations and both impulsive-
thrust and continuous-thrust trajectories are treated. Also included are topics not
normally covered, such as cooperative rendezvous and second-order conditions. An
unusually large number of appendices (7) is provided to supplement the main text.
The book is suitable for a one-semester graduate-level university course. It is also a
scholarly reference book. Problems are included at the ends of the chapters and some of
the appendices. They either illustrate the subject matter covered or extend it. Solving
them will enhance the reader’s understanding of the material, especially if they are
assigned by your instructor as homework!
I am indebted to all the students who have taken my graduate course and to the many
colleagues I have interacted with over the years.
I have greatly benefitted from interacting with some of the giants in optimal trajec-
tories: Derek F. Lawden, Theodore N. Edelbaum, John V. Breakwell, and Richard H.
Battin. These fine gentlemen have passed on, but I hope this book helps convey their
contributions, both directly and indirectly.
J.E.P.
Urbana. Illinois, USA
February, 2017
Introduction

Performance index for an optimal trajectory

T he performance index for an optimal spacecraft trajectory is typically minimum

propellant consumption, not because of the monetary cost of the propellant, but
because for each kilogram of propellant saved an additional kilogram of payload is
delivered to the final destination. The term “minimum fuel” is often used in place of
“minimum propellant”, even for chemical rockets in which the propellant is composed
of both fuel and oxidizer. As is often the case, the term “fuel” is used rather than “pro-
pellant” because it’s shorter – having one syllable instead of three. In some applications
other performance indices are used, such as minimum time, maximum range, etc.
In optimizing spacecraft trajectories the two major branches of optimization theory
are used: parameter optimization and optimal control theory. In parameter optimization
the parameters are constants and we minimize a function of a finite number of param-
eters. An example is to represent the propellant consumed as the sum of a finite number
of velocity changes caused by rocket thrusts. By contrast, optimal control is a problem of
infinite dimensions where the variables are functions of time. An example of this is the
fuel consumed by continuously varying the thrust magnitude for a finite duration rocket
thrust.

Types of orbital maneuvers

The three types of orbital maneuvers are orbital interception, orbital rendezvous, and
orbit transfer. In orbital interception at the final time tf the position vector of the space-
craft r(tf ) and the position vector of a target body r# (tf ) are equal: r(tf ) = r# (tf ). A word
of explanation of the term “target” would be helpful here. The term is more general (and
more benign) than describing a military target to be eliminated. For a planetary flyby the
target body is the flyby planet, such as Jupiter during the Cassini Mission to Saturn. At
the flyby time the spacecraft position vector and the flyby planet position vector are equal
and the spacecraft has intercepted the flyby planet.

Optimal Spacecraft Trajectories. John E. Prussing.

© John E. Prussing 2018. Published 2018 by Oxford University Press.
2 | Introduction

By contrast, in an orbital rendezvous both the position vector and the velocity vector
of the spacecraft at the final time are equal to those of the target body: r(tf ) = r# (tf ) and
v(tf ) = v# (tf ). Putting a lander on Mars is an example of a rendezvous of a spacecraft with
a planet.
Lastly, in an orbit transfer there is no target body. The desired final condition is a
specific orbit, characterized by values of semimajor axis, eccentricity, inclination, etc.
Any one of these maneuvers can be either time-fixed (constrained, closed, finite or
fixed horizon) or time-open (unconstrained, open, infinite horizon). Note that in a well-
posed time-open optimization problem the value of tf will be optimized as part of the
solution.

Notation and preliminaries

A bold-faced symbol denotes a vector (column matrix) x having components (elements)
xi . An example is the n-vector
⎡ ⎤
x1
⎢x ⎥
⎢ 2⎥
⎢ ⎥
x=⎢ · ⎥ (1)
⎢ ⎥
⎣ · ⎦
xn

A symbol x without the bold face is the magnitude of the vector x:

1 1
x = | x | = (x · x) 2 = (xT x) 2 (2)

We will encounter a scalar function of a vector φ(x), e.g., a gravitational potential

function, and our convention is that the gradient ∂φ
∂x
is a row (1 × n) vector:

∂φ ∂φ ∂φ ∂φ
= ... (3)
∂x ∂x1 ∂x2 ∂xn

One advantage of this convention is that

∂φ
dφ = dx (4)
∂x

with no transpose symbol being needed. The gradient appearing in Eq. (4) corresponds
T
to the gradient vector g = ∂φ∂x
, which is the vector normal to the surface φ = constant.
We will also encounter a Jacobian matrix, which is the derivative of a vector F with
respect to another vector x. If F is a p-vector and x is an n-vector the Jacobian is an p × n
matrix with ij element equal to ∂F
∂xj
i
, where i is the row index and j is the column index:
 Notation and preliminaries | 3
⎡ ∂F ∂F ∂F1 ⎤
1 1
·
⎢ ∂x1 ∂x2 ∂xn ⎥
⎢ ⎥
⎢ ∂F ∂F ∂F2 ⎥
⎢ 2 2
· ⎥
∂F ⎢⎢ ∂x ∂x ∂x
⎥
⎥
= 1 2 n
∂x ⎢ ⎥ (5)
⎢ · · · · ⎥
⎢ ⎥
⎢ ⎥
⎣ ∂F ∂F ∂Fp ⎦
p p
·
∂x1 ∂x2 ∂xn

Finally, we distinguish between a function and a functional. A function maps numbers

into numbers. An example is f (x) = x2 which for f (2) = 4. By contrast a functional maps
1
functions into numbers. An example is J = f (x)dx, which for f (x) = x2 yields J = 1/3.
0
1 Parameter Optimization

1.1 Unconstrained parameter optimization

P arameter optimization utilizes the theory of ordinary maxima and minima. In our
analysis, we will use the notation in Chapter 1 of Ref. [1.1]. The unconstrained prob-
lem is to determine the value of the m-vector u of independent parameters (decision
variables) to minimize the function

L(u) (1.1)

where the scalar L is termed a performance index or cost function. We note that a max-
imum value of L can be determined by minimizing –L, so we will treat only minimization
conditions.
If the ui are independent (no constraints) and the first and second partial derivatives of
L are continuous, then a stationary solution u∗ satisfies the necessary conditions (NC) that

∂L
= 0T (1.2a)
∂u
where 0 is the zero vector (every element equal to zero) and that the Hessian matrix

∂ 2L
≥0 (1.2b)
∂u2 u∗

which is a shorthand notation that the m × m matrix having elements ∂u∂ i ∂u

2
L
j
evaluated
∗
at u must be positive semidefinite (all eigenvalues zero or positive). Equation (1.2a)
represents m equations that determine the values of the m variables u∗i .
Sufficient conditions (SC) for a local minimum are the stationary condition (1.2b) and
that the Hessian matrix of (1.2b) be positive definite (all eigenvalues positive). (The

Optimal Spacecraft Trajectories. John E. Prussing.

© John E. Prussing 2018. Published 2018 by Oxford University Press.
 Unconstrained parameter optimization | 5

semidefinite NC is simply a statement that for the stationary point to be a minimum,

it is necessary that it is not a maximum!)
In the general case of m parameters we can write the first variation of our function as

∂L ∂L
δL = δu1 + · · · + δum = 0 (1.3)
∂u1 ∂um

and conclude, using the conditions of Eq. (1.2a), that at a stationary point δL = 0. In a
series expansion about the stationary point:

L(u) = L(u∗ ) + δL + δ 2 L + · · · (1.4)

the term δ 2 L is the second variation. A zero value for the first variation and a positive
value for the second variation are the SC for a local minimum.
As an example consider u to be a 2-vector. The NC (1.2a) for a minimum provides
two equations that determine the values of u∗1 and u∗2 .
The SC is
⎡ 2 ⎤
∂ L ∂ 2L
2  ⎢ ∂u ∂u ∂u ∂u ⎥ 
T ∂ L ⎢ 1 1 1 2 ⎥ δu1
δ L = δu
2
δu = [δu1 δu2 ] ⎢ ⎥ >0 (1.5)
∂u2 u∗ ⎣ ∂ 2L ∂ L
2 ⎦ δu2
∂u2 ∂u1 ∂u2 ∂u2 u∗

where δu1 ≡ u1 – u∗1 and δu2 ≡ u2 – u∗2 are arbitrary infinitesimal variations away from
their stationary values. If the strict inequality in Eq. (1.5) is satisfied for all nonzero δu1
and δu2 the matrix is positive definite. A simple test is that a matrix is positive definite if
all its leading principal minors are positive. For the 2 × 2 matrix in Eq. (1.5) the required
conditions are that the first principal minor is positive:

∂ 2L
>0 (1.6a)
∂u1 ∂u1

and that the second principal minor (the determinant of the matrix in this case) is
positive:
2
∂ 2L ∂ 2L ∂ 2L
– >0 (1.6b)
∂u1 ∂u1 ∂u2 ∂u2 ∂u1 ∂u2

As a very simple example consider

L(u) = 2u21 + u22 + u1 u2 (1.7)

The reader can verify that the NC (1.2a) for this example has the unique solution
u1 = u2 = 0 and that Eqs (1.6a) and (1.6b) are also satisfied because the values of the
leading principal minors are 4 and 7. So the function has a minimum at u = 0.
6 | Parameter Optimization

1.2 Parameter optimization with equality constraints

If there are one or more constraints the values of the parameters are not independent, as
we assumed previously. As an example, if a point (x1 , x2 ) is constrained to lie on a circle
of radius R centered at the origin we have the constraint

x21 + x22 – R2 = 0 (1.8)

The constrained minimum problem is to determine the values of m-decision parameters u

that minimize the scalar cost L(x, u) of n + m parameters, where the n-state parameters x
are determined by the decision parameters through a set of n equality constraints:

f(x, u) = 0 (1.9)

where f is an n-vector. Note that the number of state parameters n is by definition equal to
the number of constraints. In general, n > m so there are n – m independent variables.
(If m = n the problem is overconstrained and there is no optimization possible; there is
either only one solution or no solution.)

1.2.1 Lagrange multipliers

In the constrained problem the parameters can be treated as independent if we introduce

n additional variables λ1 , λ2 , . . . , λn , called Lagrange multipliers. We use these Lagrange
multipliers to adjoin the constraints to the cost function by forming an augmented
function H

H(x, u, λ) = L(x, u) + λT f(x, u) (1.10)

where λT ≡ [λ1 λ2 · · · λn ]. From Eq. (1.10) we see that if the constraints are satisfied,
f = 0, H is equal to L, and for first-order NC we can treat the problem as the unconstrained
optimization δH = 0. The necessary conditions are

∂H
= 0T (1.11a)
∂x
∂H
= 0T (1.11b)
∂u
and the constraint
∂H
= fT = 0T (1.11c)
∂λ
Equations (1.11a–c) are 2n + m equations in 2n + m unknowns (x, λ, and u).
Introducing the n additional variables λ seems to be counterproductive, because we
now have more unknowns to solve for. Instead of introducing the Lagrange multi-
pliers we could, in principle, solve the constraints f = 0 for the n state variables x in
 Parameter optimization with equality constraints | 7

terms of the decision variables u, leaving only m equations to solve for the m decision
variables. However, those equations, while fewer, are usually more complicated than
Eqs (1.11a–c), so it is easier to solve the simpler 2n + m equations (see Problem 1.2).
The Lagrange multiplier λ has an interesting and useful interpretation, which makes it
worth solving for. It provides the sensitivity of the stationary cost to small changes in the
constants in the constraint equations. So, we can obtain the change in the stationary cost
due to small changes in these constants without resolving the problem. This can be very
useful in a complicated problem that is solved numerically.
This interpretation is discussed in Appendix A. To summarize, write the constraints as

f(x, u, c) = 0 (1.12)

where c is a q-dimensional vector of constants in the constraint equations. Due to changes

in these constants, dc, the change in the stationary cost is given by

dL∗ = λT fc dc (1.13)

∂f
where fc = ∂c
is an n × q matrix.
∂2L
The SC analogous to the Hessian matrix condition ∂u2 u∗
> 0 for the unconstrained
problem is given in Section 1.3 of Ref. [1.1] as

∂ 2L
x fu – fu fx Hxu + fu fx Hxx fx fu > 0
T –T T –T
= Huu – Hux f–1 –1
(1.14)
∂u2 f=0

where the symbol f–Tx is shorthand for the transpose of the inverse, which is equal to the
inverse of the transpose (see Problem 1.3).

Example 1.1 Minimization using a Lagrange multiplier

Determine the rectangle of maximum perimeter that can be inscribed in a circle of

radius R. In this example, there are two variables x and y and a single constraint—
the equation of the circle. So n = m = 1 and we arbitrarily assign y to be the decision
variable u. Let the center of the circle lie at (x, u) = (0, 0) and denote the corner of
the rectangle in the first quadrant as (x, u). To maximize the perimeter, we minimize
its negative and form

L(x, u) = –4(x + u) (1.15)

subject to the constraint

f (x, u) = x2 + u2 – R2 = 0 (1.16)

where x and u are scalars.

8 | Parameter Optimization

We construct the augmented function H as

H(x, u, λ) = L(x, u) + λf (x, u)

(1.17)
= –4(x + u) + λ(x2 + u2 – R2 )

Applying the NC of Eqs. (1.11a–c):

∂H
Hx = = –4 + 2λx = 0 (1.18a)
∂x
∂H
Hu = = –4 + 2λu = 0 (1.18b)
∂u
and the constraint
∂H
Hλ = = x2 + u2 – R2 = 0 (1.18c)
∂λ
where we use subscript notation for partial derivatives. Equations (1.18a–c) are easily
solved to yield
√
∗ ∗ 2
x =u = R (1.19)
2
√
2 2
λ= (1.20)
R
where Eq. (1.19) indicates that the rectangle of maximum perimeter is a square. To
verify that our stationary solution is indeed a minimum we use Eq. (1.14):

∂ 2L
= Huu – Hux fx–1 fu – fu fx–1 Hxu + fu fx–1 Hxx fx–1 fu
∂u2 f =0
√ √ √ (1.21)
4 2 4 2 8 2
= –0–0+ = >0
R R R
and this constrained second derivative√being positive satisfies the SC for a minimum,
resulting in a maximum perimeter of 4 2R = 5.657R.
Applying Eq. (1.13) with the single constant c being R:
√
2 2 √
dL∗ = λfR dR = (–2R)dR = – 4 2dR (1.22)
R
Consider a nominal value R = 1 and a change in the radius
√ of dR = 0.1. Using Eq.
(1.22) the change in the minimum cost is equal to –0.4 2 = –0.566, resulting in a
maximum perimeter
√ of 5.657 + 0.566 = 6.223. This compares very favorably with the
exact value of 4 2(1.1) = 6.223
 Parameter optimization with an inequality constraint | 9

For this simple example using the interpretation of the Lagrange multiplier is not very
profound, but in a complicated problem, especially if the solution is obtained numerically
rather than analytically, a good approximation to the cost of a neighboring stationary
solution can be obtained without completely resolving the problem.
Appendix B provides a more complicated constrained minimum example: the
Hohmann transfer as a minimum v solution.

1.3 Parameter optimization with an inequality

constraint
Consider the simple case of a scalar cost L(y) subject to a constraint of the form f (y) ≤ 0,
where y is p dimensional and f is a scalar. We will consider only the simple case of
a single constraint, because that is all we need for the single application that we treat
in Section 7.2. Consideration of the general case results in the Karush–Kuhn–Tucker
conditions and is discussed in Section 1.7 of Ref. [1.1] (as the Kuhn–Tucker conditions).
As in the equality constraint case in Section 1.1 a Lagrange multiplier λ is introduced
and a scalar H is defined as:

H(y, λ) = L(y) + λf (y) (1.23)

Note that there is a total of p + 1 unknowns, namely the elements of y and λ. As in the
equality constraint case an NC can be written as

∂H
= 0T (1.24)
∂y

but, unlike the equality constraint case, the inequality constraint is either active or inactive,
resulting in the NC:

λ≥0 if f (y) = 0 (active) (1.25)

λ=0 if f (y) < 0 (inactive) (1.26)

Note that in either case the product λf = 0, resulting in H being equal to L.

Expanding Eq. (1.24):

∂H ∂L ∂f
= +λ = 0T (1.27)
∂y ∂y ∂y
or
∂L ∂f
= –λ (1.28)
∂y ∂y

which provides the geometrical interpretation that when the constraint is active the gra-
∂f
dients ∂L
∂y
and ∂y are oppositely directed (see Fig. 1.1 for p = 2). From that figure we see
10 | Parameter Optimization

f<0
Constant L fy
y2 Ly
fy Active
Ly constraint
f=0
Ly
Linareacy fy f=0
y1
f<0
y2 Unconstrained
Min L Inactive
constraint
f<0

y1
¶L ¶f
=–λ = 0T because
¶y ¶y λ=0

Figure 1.1 An active constraint and an inactive constraint

that at any point on the constraint not at the constrained minimum (where the gradients
are oppositely directed) we can slide along the constraint and decrease the cost until we
reach the constrained minimum.
This is even more evident when we note that the SC for the constraint being active is
λ > 0, indicating that the only way to decrease the cost is to violate the constraint.
If the constraint is inactive the constrained minimum is the unconstrained minimum.
In that case ∂L
∂y
= 0T as seen in Eq. (1.28) with λ = 0.

Example 1.2 Simple inequality constraint

In order to show how the NC are applied, consider the very simple example to
minimize
1
L(y) = y2 (1.29)
2
subject to the constraint

f (y) = y – a ≤ 0 (1.30)

where y and a are scalars. Use Eq. (1.23) to form H:

1
H(y, λ) = y2 + λ(y – a) (1.31)
2
Equation (1.24) provides the NC:
∂H
=y+λ=0 (1.32)
∂y
 Problems | 11

where from Eqs. (1.25) and (1.26)

λ≥0 if y = a (active) (1.33)

λ=0 if y < a (inactive) (1.34)

We next assume the constraint is active or inactive and investigate whether the NC
are satisfied. Consider two separate cases, a = 1 and a = –1.
Case I a = 1
Assume the constraint is active. Then y = 1 and, from Eq. (1.32), λ = –1, which
violates NC (1.33). Assuming the constraint is inactive yields λ = 0 and y = 0, which
satisfy the NC. So, the constrained minimum is at y = 0 and is the same as the uncon-
strained minimum with a cost L = 0. Because the constraint is inactive the SC that
∂ 2 L/∂y2 = 1 > 0 is satisfied.
Case II a = –1
Assume the constraint is inactive. Then λ = 0 and y = 0, which violates the con-
straint y < –1. Assuming the constraint is active yields y = –1 and λ = 1, which satisfy
the NC. So the constrained minimum occurs at y = –1 with a cost L = 1/2. Note also
that the SC λ > 0 is satisfied.
Section 7.2 provides a more complicated example of an inequality constraint
minimization, namely, the terminal maneuver for an optimal cooperative impulsive
rendezvous.

Problems

1.1 a) Determine the value of the constant parameter k that “best” approximates a
given function f (x) on the interval a ≤ x ≤ b. Use as a cost function to be
minimized
b
1
J(k) = [k – f (x)]2 dx
2
a

This is the integral-squared error, which is the continuous version of the familiar
least-squares approximation in the finite-dimensional case.
Verify that a global minimum is achieved for your solution.
b) For a = 0, b = 1, and f (x) = x3 evaluate the cost J for (i) the optimal value of k
and (ii) another value of k that you choose.
1.2 Solve the problem in Example 1.1 without using a Lagrange multiplier.
12 | Parameter Optimization

1.3 Show that for a nonsingular square matrix the inverse of its transpose is equal to the
transpose of its inverse.
1.4* Minimize L(y) = 12 (y21 + y22 ) subject to the constraints y1 + y2 = 1 and y1 ≥ 3/4.
a) Attempt to satisfy the NC by assuming the inequality constraint is inactive.
b) Attempt to satisfy the NC by assuming the inequality constraint is active.
c) Determine whether the solution to the NC satisfies the SC.

Reference
[1.1] Bryson, A.E., Jr., and Ho, Y-C, Applied Optimal Control. Hemisphere Publishing, 1975.
2 Rocket Trajectories

2.1 Equations of motion

T he equation of motion of a spacecraft which is thrusting in a gravitational field can

be expressed in terms of the orbital radius vector r as:
r̈ = g(r) + ;  = u (2.1)

The variable  is the thrust acceleration vector. The scalar  is the magnitude of the
thrust acceleration defined as the thrust (force) T divided by the mass of the vehicle m.
The variable u is a unit vector in the thrust direction, and g(r) is the gravitational accel-
eration vector. Equation (2.1) is somewhat deceptive and looks like a simple statement
of “F equals ma” with the thrust term appearing on the right-hand side as if it were an
external force like gravity. In actuality, F does not equal ma (but it does equal the time
derivative of the linear momentum mv), because the mass is changing due to the gen-
eration of thrust and because the thrust is an internal force in the system defined as the
combination of the vehicle and the exhausted particles. A careful derivation of Eq. (2.1)
requires deriving the so-called rocket equation by equating the net external force (such as
gravity) to the time rate of change of the linear momentum of the vehicle-exhaust particle
system (see Sections 6.1–6.4 of Ref. [2.1]).
An additional equation expresses the change in mass of the spacecraft due to the
generation of thrust:
ṁ = –b; b≥0 (2.2)

In Eq. (2.2) b is the (nonnegative) mass flow rate. The thrust magnitude T is given by
T = bc, where c is the effective exhaust velocity of the engine. The word “effective” applies
in the case of high-thrust chemical engines where the exhaust gases may not be fully
expanded at the nozzle exit. In this case an additional contribution to the thrust exists
and the effective exhaust velocity is
Ae
c = ca + (pe – p∞ ) (2.3)
b

Optimal Spacecraft Trajectories. John E. Prussing.

© John E. Prussing 2018. Published 2018 by Oxford University Press.
14 | Rocket Trajectories

In Eq. (2.3) the subscript e refers to conditions at the nozzle exit, ca is the actual (as
opposed to effective) exhaust velocity, and p∞ is the ambient pressure. If the gases are
exhausted into the vacuum of space, p∞ = 0.
An alternative to specifying the effective exhaust velocity is to describe the engine in
terms of its specific impulse, defined to be:

(bc)t c
Isp = = (2.4)
(bt)g g

where g is the gravitational attraction at the earth surface, equal to 9.80665 m/s2 . The spe-
cific impulse is obtained by dividing the mechanical impulse delivered to the vehicle by
the weight (on the surface of the earth) of propellant consumed. The mechanical impulse
provided by the thrust force over a time t is simply (bc)t, and, in the absence of other
forces acting on the vehicle is equal to the change in its linear momentum. The weight
(measured on the surface of the earth) of propellant consumed during that same time
interval is (bt)g, as shown in Eq. (2.4). Note that if instead one divided by the mass
of the propellant (which is the fundamental measure of the amount of substance), the
specific impulse would be identical to the exhaust velocity. However, the definition in
Eq. (2.4) is in standard use with the value typically expressed in units of seconds.

2.2 High-thrust and low-thrust engines

A distinction between high- and low-thrust engines can be made based on the value of
the nondimensional ratio max /g. For high-thrust devices this ratio is greater than unity
and thus these engines can be used to launch vehicles from the surface of the earth. This
ratio extends as high as perhaps 100. The corresponding range of specific impulse val-
ues is between 200 and approximately 850 sec., with the lower values corresponding to
chemical rockets, both solid and liquid, and the higher values corresponding to nuclear
thermal rockets.
For low thrust devices the ratio max /g is quite small, ranging from approximately
10–2 down to 10–5 . These values are typical of electric rocket engines such as magneto-
hydrodynamic (MHD), plasma arc and ion devices, and solar sails. The electric engines
typically require separate power generators such as a nuclear radioisotope generator or
solar cells. The ratio for solar sails is of the order of 10–5 .

2.3 Constant-specific-impulse (CSI)

and variable-specific-impulse (VSI) engines
Two basic types of engines exist: CSI and VSI, also called power-limited engines. The
CSI category includes both high- and low- thrust devices. The mass flow rate b in some
cases can be continuously varied but is limited by a maximum value bmax . For this reason,
this type of engine is also described as a thrust-limited engine, with 0 ≤  ≤  max .
 Constant-specific-impulse (CSI) and variable-specific-impulse | 15

The VSI category includes those low thrust engines which need a separate power
source to run the engine, such as an ion engine. For these engines, the power is limited
by a maximum value Pmax , but the specific impulse can be varied over a range of values.
The fuel expenditure for the CSI and VSI categories is handled separately.
The equation of motion (1) can be expressed as:

cb cb
v̇ = u + g(r); ≡  (2.5)
m m

For the CSI case Eq. (2.5) is solved using the fact that c is constant as follows:

cb
dv = u dt + g(r) dt (2.6)
m

Using Eq. (2.2),

dm
dv = –c u + g(r) dt (2.7)
m

This can be integrated (assuming constant u) to yield:

tf
v = v(tf ) – v(to ) = –cu(ln mf – ln mo ) + g(r) dt (2.8)
to

  tf
mo
v = c u ln + g(r) dt (2.9)
mf
to

which correctly indicates that, in the absence of gravity, the velocity change would be in
the thrust direction u. The actual velocity change achieved also depends on the gravita-
tional acceleration g(r) which is acting during the thrust period. The term in Eq. (2.9)
involving the gravitational acceleration g(r) is called the gravity loss. Note there is no
gravity loss due to an impulsive thrust.
If one ignores the gravity loss term for the time being, a cost functional representing
propellant consumed can be formulated. As will be seen, minimizing this cost functional
is equivalent to maximizing the final mass of the vehicle. Utilizing the fact that the thrust
is equal to the product of the mass flow rate b and the exhaust velocity c, one can write:

–m
ṁ = –b = (2.10)
c
dm 
= – dt (2.11)
m c
16 | Rocket Trajectories

For the CSI case the exhaust velocity c is constant and Eq. (2.11) can be integrated to
yield

 tf
mf 1
ln =–  dt (2.12)
mo c
to

or

 tf
mo
c ln =  dt ≡ JCSI (2.13)
mf
to

JCSI is referred to as the characteristic velocity of the maneuver or the V (pronounced

“delta vee”) and it is clear from Eq. (2.13) that minimizing JCSI is equivalent to maximiz-
ing the final mass mf .
In the impulsive thrust approximation for the unbounded thrust case (max → ∞)
the vector thrust acceleration is represented by


n
(t) = vk δ(t – tk ) (2.14)
k=1

with to ≤ t1 < t2 < · · · < tn ≤ tf representing the times of the n thrust impulses. (See
Sections 6.1–3 of Ref. [2.1].) Using the definition of a unit impulse,

tk
+

δ(t – tk ) dt = 1 (2.15)
tk–

and tk± ≡ lim ε → 0(tk ± ε); ε > 0.

Using Eq. (2.14) in Eq. (2.13):

tf 
n
JCSI = dt = vk (2.16)
to k=1

and the total propellant cost is given by the sum of the magnitudes of the velocity changes.
The corresponding cost functional for the VSI case is obtained differently. The exhaust
power (stream or beam power) is half of the product of the thrust and the exhaust
velocity:

1 1 1
P = Tc = mc = bc2 (2.17)
2 2 2
 Constant-specific-impulse (CSI) and variable-specific-impulse | 17

Using this along with


b –ṁ d 1
= = (2.18)
m2 m2 dt m

results in

d 1 2
= (2.19)
dt m 2P

which integrates to

tf
1 1 1 2
– = dt (2.20)
mf mo 2 P
to

Maximizing mf for a given value of mo regardless of whether  is optimal or not is

obtained by running the engine at maximum power P = Pmax . This is not as obvious as it
looks in Eq. (2.20), because the value of  might be different for different values of P. To
see that the engine should be run at maximum power note that for a specified trajectory
r(t) the required vector thrust acceleration is given by Eq. (2.1) as

(t) = r̈(t) – g[r(t)] (2.21)

Thus, for a given trajectory r(t) (optimal or not), the final mass in Eq. (2.20) is
maximized by running the engine at maximum power.
For this reason the VSI cost functional can be taken to be

tf
1
JVSI =  2 dt (2.22)
2
to

To summarize, the cost functionals representing minimum propellant expenditure are

given by

tf
JCSI =  dt (2.23)
to

and

tf
1
JVSI =  2 dt (2.24)
2
to
18 | Rocket Trajectories

As seen in Eqs. (2.23) and (2.24) the minimum-propellant cost can be written in terms
of the control (t) rather than introducing the mass as an additional state variable whose
final value is to be maximized.

Problems

2.1 For an impulsive thrust consider first-order changes in J = v and in the final
mass mf . Obtain an expression for the marginal (fractional) change δmf /mf in terms
of the marginal change δJ/J. Assume the values of mo and c are specified and there is
no gravity loss.
2.2 Consider a constant thrust T = To in a CSI propulsion system.
a) Determine an expression for ˙ in terms of (only) the exhaust velocity c and the
thrust acceleration magnitude .
b) Solve the ODE in part (a) for (t). Express your answer in terms of To divided
by a time-varying mass.
2.3 Show how Eq. (2.19) follows from Eq. (2.18).

Reference
[2.1] Prussing, J.E. and Conway, B.A., Orbital Mechanics, Oxford University Press, 2nd Edition,
2012.
3 Optimal Control Theory

3.1 Equation of motion and cost functional

C onsider a dynamic system that operates between a specified initial time to and a
final time tf , which may be specified or unspecified. The equation of motion of the
system can be written as

ẋ = f(x, u, t); x(to ) = xo (specified) (3.1)

where x(t) is an n × 1 state vector and u(t) is an m × 1 control vector. Given an initial
state x(to ), the state x(t) is determined by the control u(t) through the equation of
motion (3.1).
The first-order form of Eq. (3.1) is completely general because any higher-order
equation can be transformed into first-order form. As an example consider Eq. (2.1):

r̈ = g(r) +  (2.1)

To transform to first-order form let

ṙ = v (3.2a)
v̇ = g +  (3.2b)

Then
 
ṙ v
ẋ = = (3.3)
v̇ g+

and a second-order equation for a 3-vector r has been transformed into a first-order
equation for a 6-vector x.

Optimal Spacecraft Trajectories. John E. Prussing.

© John E. Prussing 2018. Published 2018 by Oxford University Press.
20 | Optimal Control Theory

To formulate an optimal control problem we define a cost functional

tf
J = φ[x(tf ), tf ] + L[x(t), u(t), t] dt (3.4)
to

to be minimized by choice of a control u(t). The first term in Eq. (3.4) is sometimes
called the “terminal cost” and the second term the “path cost”.
As a simple example let φ = tf and L = 0. Then the cost is J = tf and we have a
minimum time problem.
As noted elsewhere, if we wish to maximize a performance index J̃, we minimize J = –J̃.
There are three basic forms of the cost functional, named after pioneers in the Calculus
of Variations:

Mayer form: J = φ[x(tf ), tf ]

tf
Lagrange form: J = L(x, u, t) dt
to
tf
Bolza form: J = φ[x(tf ), tf ] + L(x, u, t) dt [Eq. (3.4)]
to

There is also a special case of the Mayer form: J = xn (tf ), i.e., the cost is the final value of
the last component of the state vector.
Although the Bolza form appears to be the most general, all the forms have equal gen-
erality. To see this we will transform the Bolza form into the special case of the Mayer
form.

Example 3.1 Cost transformation

Given that x is an n-vector, define ẋn+1 = L with xn+1 (to ) = 0. Then

tf
L dt = xn+1 (tf ) – xn+1 (to ) = xn+1 (tf ).
to

Then,

J = φ[x(tf ), tf ] + xn+1 (tf ) ≡ xn+2 (tf ) (3.5)

 General problem | 21

3.2 General problem

Over the time interval to ≤ t ≤ tf , where the final time can be specified or unspecified
(fixed or free, closed or open) there is the dynamic constraint of the equation of motion:

ẋ = f(x, u, t); x(to ) = xo (specified) (3.1)

and a cost to be minimized

tf
J = φ[x(tf ), tf ] + L[x(t), u(t), t] dt (3.4)
to

The ODE constraint of Eq. (3.1) is what makes Optimal Control Theory a generalization
of the Calculus of Variations.
There may be terminal constraints ψ[x(tf ), tf ] = 0, for example, r(tf ) – r# (tf ) = 0,
where r is the spacecraft position vector and r# is a target body position vector. This con-
straint then represents an orbital interception. And there may be control constraints such
as a bounded control | u(t) | ≤ M.
First consider the simplest optimal control problem: fixed final time, no terminal or
control constraints. We will derive first-order necessary conditions (NC) that must be
satisfied to determine the control u * (t) that minimizes the cost J. This is treated in more
detail in Section 3.3 of Ref. [3.1] and Section 2.3 of Ref. [3.2].
Before we begin our derivation of the NC, a fundamental point needs to be made. We
will be assuming that an optimal control u * (t) exists and deriving conditions that must
be satisfied. If no optimal control exists the NC are meaningless.
Peron’s Paradox illustrates this point and goes as follows: Let N = the largest integer.
We will prove by contradiction that N = 1.
If N > 1, N 2 > N, which is a contradiction because N is the largest integer. Therefore,
N = 1, a nonsensical result because there is no largest integer.
Assuming an optimal control u * (t) exists for the general problem, let’s proceed. First,
we adjoin the equation of motion constraint to the cost to form

tf
J̄ = φ[x(tf ), tf ] + L(x, u, t) + λT [f(x, u, t) – ẋ] dt (3.6)
to

Where λ(t) is an n × 1 vector variously called the “adjoint” vector, the “costate” vector,
or a “time-varying Lagrange multiplier” vector. Note that if the constraint equation of
motion (3.1) is satisfied, we have added zero to the cost, in the same spirit as the use of a
Lagrange multiplier in Section 2.2.1.
We will form the variation in the cost δ J̄ and derive NC that make δ J̄ = 0 due to a
small variation in the control about the optimal control δu(t) = u(t) – u∗ (t). Note that
22 | Optimal Control Theory

the variation is equal to the difference at a common value of time t and for this reason it
is sometimes called a contemporaneous variation, for which δt ≡ 0.
For convenience define the Hamiltonian

H(x, u, λ, t) ≡ L(x, u, t) + λT (t)f(x, u, t) (3.7)

(The terminology comes from classical mechanics, in which H = pT q̇ – L, where L is the

Lagrangian, p are the generalized momenta, and q̇ are the generalized velocities.)
Using the definition of the Hamiltonian, J̄ in Eq. (3.6) is equal to J̄ = φ +
tf
[H – λT ẋ] dt.
to
tf
Before forming the variation δ J̄ we will deal with the last term in J̄, namely – λT ẋ dt.
to
First, we derive a vector integration by parts formula by using d(λT x) = λT dx + dλT x.
T
Then d(λT x) = λT x = λT dx + dλT x. Using dx = ẋdt and dλT = λ̇ dt we obtain
T
λT ẋdt = λT x – λ̇ xdt.
tf
Applying this to the definite integral term – λT ẋ dt.
to

tf tf
T
– λT ẋdt = –λT (tf )x(tf ) + λT (to )x(to ) + λ̇ (t)x(t) dt (3.8)
to to

Using this result we rewrite Eq. (3.6) as

tf
T
J̄ = φ + λT (to )x(to ) – λT (tf )x(tf ) + [H + λ̇ x]dt (3.9)
to

Forming the cost variation δ J̄ due to a control variation δu(t):

δ J̄ = δφ + δ[λT (to )x(to )] – δ[λT (tf )x(tf )]

tf
T
+δ [H + λ̇ x] dt
to

∂φ
= δx(tf ) + λT (to )δx(to ) – λT (tf )δx(tf )
∂x(tf )
tf 
∂H ∂H T
+ δx + δu + λ̇ δx dt (3.10)
∂x ∂u
to
 General problem | 23

Combining coefficients:

∂φ
δ J̄ = – λT (tf ) δx(tf ) + λT (to )δx(to )
∂x(tf )
tf   (3.11)
∂H T ∂H
+ + λ̇ δx + δu dt
∂x ∂u
to

Because δx(t) and δu(t) are arbitrary, for δ J̄ = 0 the NC are obtained from Eq. (3.11) by
setting individual coefficients equal to zero:

∂H
T ∂L ∂f
λ̇ = – =– – λT (3.12)
∂x ∂x ∂x
∂φ
λT (tf ) = (3.13)
∂x(tf )
∂H
= 0T (3.14)
∂u(t)
either δxk (to ) = 0 or λk (to ) = 0, k = 1, 2, . . . , n (3.15)

where δxk (to ) will be equal to 0 if the component xk (to ) is specified, but if xk (to ) is
unspecified, λk (to ) = 0. This represents a generalization of the initial condition x(to ) =
xo to allow some of the initial states to be unspecified. An additional NC is that the
constraint Eq. (3.1) must be satisfied: ẋ = f with initial conditions given by Eq. (3.15).
Equation (3.13) has a simple geometrical interpretation as a gradient vector. The final
value of the adjoint vector is orthogonal to the line or surface φ = constant.
Note that there are 2n + m unknowns: x(t), λ(t), and u(t). Equation (3.1) with initial
conditions given by (3.15) provide n equations; Eqs. (3.12) and (3.13) provide n equa-
tions; and Eq. (3.14) provides m equations, for a total of 2n + m NC equations in that
same number of unknowns. So, in principle, the optimal control problem is solved! But
there are several details to be worked out and observations to be made.
First note that even if the system equation ẋ = f is nonlinear the adjoint equation
(3.12) for λ(t) is linear because L and f are not functions of λ. And if the cost is of the
Mayer form (no L), the adjoint equation is homogeneous. But also note that the bound-
ary condition (3.13) is at the final time in contrast to the conditions in Eq. (3.15) at the
initial time. So, we have a split Two-Point Boundary Value Problem (2PBVP), which
indicates a fundamental difficulty in solving the NC for an optimal control problem,
because not all boundary conditions are given at the same value of time.
Note that the adjoint equation (3.12) and the system equation (3.1) are differential
equations, but what is often called the optimality condition (3.14) is an algebraic equation
that applies at all values of to ≤ t ≤ tf . It requires that the optimal control result in a
Another random document with
no related content on Scribd:
 Madidu Elaui Agola Badjehun Karikari
(present Chief) (dead) (dead)

Assalm El Musa Mursa Djamarata Imuhadjil 1 son (?)

i Mekki

ALIMSAR

Durrata Azuhur Fihirun 3 other sons (?)

(dead)

Aneirum. 2 other 2 sons 1 son

sons (?) (?)

Here too is a list of the tribes making up the Awellimiden

Confederation, with the names of their present chiefs.

THE NOBLE OR IHAGGAREN TRIBES.

Kel Kumeden—Chief Madidu.
Kel Ahara—Chief El Yasan.
Kel Tedjiuane—Chief Arreian.
Iderragagen.
Tarkaitamut.
Tahabanat.
Ibehauen—Chief Sar’adu.
Ifoghas—Chief Waruziga.
Ihegaren—Chief El Auedech.
Kel Tekeniuen—Chief Burhan.
Kel Takabut—Chief Aluania.
Teradabeben—Chief Sidauat.
Tenguereguedeche—Chief Warigoru.
Tademeket—Chief Yunès.
Idalbabu—Chief Ihuar.
Ahianallan.

SERFS OR IMRADS.
Kel Gossi—Chief Ur illies. Tar’ahil—Chief Ekerech.
Irreganaten—Chief Ur orda. Ikairiraen—Chief Ezemek.
Iueraruarar’en—Chief Mahamud. Erkaten—Chief Elanusi.
Imideddar’en—Chief Huberzan. Ikawellaten—Chief Ibunafan.
Ibongitan—Chief Allabi. Ihaiauen—Chief Abba.
 Tafagagat—Chief Karrabau. Kel R’ezafan—Chief Amachecha.

To these tribes making up the actual Confederation must be

added the following, who were brought into it by force, and have long
since submitted with a good grace to be under the protection of the
Awellimiden:
Wadalen—Chief Niugi. Eratafan—Chief Yoba.
Cheibatan—Chief Rafiek. Ibendasan.
Logomaten—Chief Bokar Wandieïdu. Ahiananurde—Chief Amadida.
Tabotan—Chief Muley.

Subject to each of these last-named tribes are imrads, but I only

know the name of one of their tribes, that of the Ekono, vassals of
the Wadalen.
In addition to their predatory excursions the Tuaregs on the right
and left bank of the Niger make two annual migrations, the time of
which is generally the same.
During the dry season, from December to May, the higher districts
are sterile and dry, the ponds and wells empty of water. Then the
Tuaregs move down to the river-banks and their flocks and herds
graze on the coarse weeds which line them. To avoid the sickness
amongst the camels which results from eating damp food, and to
which I alluded in speaking of Timbuktu, they generally leave them a
little further inland. It is at this time that the negroes pay their tribute
of maize and tobacco, and it is also during this same season that
warlike expeditions are generally undertaken.
For the rest of the year the rain pours down in torrents in the
riverside districts, and although its fall is not so constant or so heavy
in the higher lands, they too are fertilized by the filling up of the
ponds and the wells, many of which even overflow.
Then the nomad tribes go back again to their old haunts, and
settle down for the winter in their camps about the wâdies,
resembling those of Algeria, which begin near Gao.
These wâdies are such very characteristic features of Central
Africa, that a description of one of them may be useful. The word
wâdy means the channel of a watercourse which is dry except in the
rainy season, but there is water in the upper portion of that of Gao in
every season. Its source is far away in the north, and it seems to be
identical with the Igharghar of the south, alluded to by Duveyrier, the
Astapus of the ancients, which comes down from the Atakor or
Ahaggar.
This would confirm Barth’s suggestion, that the marshy
depressions which debouch on the Ngiti Sokoto do not extend
beyond the district of Air.
My own opinion is that the Gao Wâdy, before it became choked
up with sand, was a tributary of the Niger when the course of that
river was far more rapid than it is now.
An examination of its banks does in fact lead to the conclusion,
that nearly if not quite all along them a line of cliffs, eroded by the
action of water, marks what was once the bed of part of the old
Niger. In their annual migration the Awellimiden go up as far as the
districts near Air, where they come in contact with their enemies the
Kel Gheres. Probably competition for the ownership of the
pasturages yielding food in the dry season, was the original cause of
the feud between the two races, which dates from centuries ago.
The tribes from the left bank of the Niger also move into the kind
of islet formed by the bend of the river, advancing to near Dori,
where they find a series of ponds and lakes known as Oursi Beli,
etc., an idea of which I have tried to give in the map accompanying
this volume, but I do not know how far I have succeeded.
There are many very curious and interesting hydrographical
problems connected with this bend of the Niger reserved for the
future explorer to solve.
Well, what do my readers think of the Tuaregs after the picture I
have endeavoured to give of them? I certainly have not represented
them as saints, living in a kind of Utopia, where all is well, where the
men have no vices and the women no faults.
You will perhaps, however, agree with me that they have very
decided characters, and many fine qualities, if also many defects.
Their intelligence is certainly great, making it well worth while to try
and win them to a better mode of life, and one more conducive to the
comfort of their neighbours.
I do not of course fail to recognize what hard work it is to row
against the current or to contend against pre-conceived ideas. It is
always difficult, and sometimes dangerous.
In 1859 a young Frenchman, not more than twenty years old at
the most, disembarked at Constantine. He spent three years
travelling about the Algerian Sahara, and under the powerful
protection of the Emir Ikhenukhen, chief of the Azguers, he lived for
more than a year amongst the Tuaregs.
After his return an expedition was sent out by the Governor of
Algeria, and the treaty of Rhadamès was signed.
Then, in accordance with the traditional French policy in matters
colonial, instead of profiting by the results already acquired,
absolutely nothing further was done. Duveyrier described the
Tuaregs as he had found them, just as I have tried to do; he spoke
quite frankly of their faults as well as of their virtues, and insisted on
the possibility of treating with them on favourable terms. He might
well do so, for he had already succeeded in that direction himself.
When twenty years later Flatters was assassinated, Duveyrier
was accused of mendacious optimism, and every one was ready to
cast a stone at him.
As a matter of fact, however, Flatters was killed by the Hoggars,
and Duveyrier had mentioned that they were living in a state of
anarchy, which seemed likely to get worse and worse rather than to
improve. Flatters insisted on going through their territory, although
the Amrar had told him he could not protect him. Now Duveyrier had
made a special point of never going into any district without first
securing an efficacious safe-conduct, yet in spite of all this he is
made responsible for the disaster.
A fitting epilogue ensued, for Duveyrier, disquieted at the
accusations brought against him, weakened by fever contracted in
his journey, and cut to the heart by the ingratitude of his fellow-
countrymen, committed suicide by shooting himself with his revolver,
in the hope perchance of finding the justice denied him here in
another world, if there be indeed such a thing as justice anywhere.
The English would have made him
a peer, and put up statues in his
honour; the ignorance of the French, I
will not use a harsher word, drove him
to commit suicide.
The example is certainly not
encouraging to us later explorers.
I should have been more likely to
win applause if I had pictured the
TUAREGS.
Tuaregs as irreclaimable savages,
relating a thousand entanglements
with them, such as imaginary conflicts with their armed bands, where
my own presence of mind and the courage of my party saved the
expedition from massacre.
I have preferred in the interests of my country to tell the truth, the
whole truth, and nothing but the truth.
Even as I write these words, I hear of the death of two young
officers and their men, who were killed near Timbuktu in a fight with
a Hoggar razzi. The Hoggars again!
This does but confirm what I insisted on when I was at Timbuktu,
that we shall never succeed in getting en rapport with the nomad
tribes except with the aid of those tribes themselves.
We must first subjugate certain tribes, and then form from
amongst them auxiliary levies, or, as the natives call them, maghzen,
which will aid us, at a minimum cost to ourselves, to establish French
influence over the Tuaregs.
Amongst the tribes who would best lend themselves to this
purpose, I place the Awellimiden in the very first rank, and they are
the hereditary enemies of the Hoggars. Or perhaps I should rather
have said, if we wish to bring about a complete pacification of the
country, and at the same time win the friendship of the Awellimiden
chief, we ought to strengthen his hands.
 With this idea in my mind I make the following suggestions. We
should arm the Awellimiden with a hundred or a couple of hundreds
of percussion rifles, with very large nipples, which would only admit
of the use of special caps turned out in French manufactories.
With one hundred such guns the Awellimiden would be invincible,
and could soon butcher all their enemies, whether Kel Gheres or
Hoggars.
The absolute necessity of having French percussion caps would
place them entirely in our hands, and by doling out the ammunition
needed little by little, we should force them to submit to and serve
us. We should, moreover, have it in our power to break up their
strength directly they showed any reluctance to fall in with our
wishes.
In return for a service such as this supply of fire-arms, the
Amenokal would protect our traders; he has already in fact promised
to do so, not only by word of mouth but in writing.
These traders must, however, act with prudence and
circumspection. I am quite convinced that I and my companions
might fearlessly return to the Awellimiden because they know us
now. I have suggested to our Government that we should return, but
I have not been more successful in that direction than I have in
getting the rifles I asked for.
Strangers must not attempt with a light heart to penetrate into the
Tuareg districts, without having secured the formal protection of the
chief.
What would you have? When a Grand Duke announces his
intention of visiting the wine-shops of the outlying boulevards, don’t
we always take care to send an habitué of those boulevards with him
to look after him? A Jaume or a Rossignol[8] is always in attendance.
And if a protector is useful in Paris, can we not well understand that
one would be indispensable in the Sahara?
When Madidu has once said to a traveller “Yes, come,” or “You
can go,” I am convinced that no danger would be run in the districts
subject to him.
 With the Awellimiden on our side we could conquer the Sahara,
and the Tuaregs would help us to push on towards Lake Tchad, Air,
Tunis and Algeria. He would find it to his own advantage to do so,
and the conditions of his existence would be manifestly ameliorated.
Do you imagine that these Tuaregs are stupid enough to miss a
chance of getting stuffs for clothes, coverlids, glass beads, and all
the things they covet? If the men were sufficiently blind to their own
interests, I’ll warrant you their wives would not be.
The Tuareg race will be tamed at last, their faults, all the result of
the fierce struggle for existence, will disappear, and modern
civilization will have conquered a new district in Africa!
One afterthought does, however, occur to me. Will the change be
a good thing for the Tuaregs themselves?
When I think of their wandering life, free from all restraint, when I
remember their courage, which to them is the highest of virtues,
when I consider how truly equal all those worthy of equality are, I ask
myself whether after all they are not happier than we Europeans?
Their life is a hard one, and their habits are frugal, but has not
custom made this life natural to them, and are they really sensible of
its privations?
Good fortune with them is the reward of the brave who know how
to win the victory, and it is in razzis that the victory is gained. To spoil
the vanquished is also to wash out the stain of an hereditary injury,
for the vendetta is not confined to Italy, but often makes friendship
impossible between certain tribes in Africa. The goods of him who
perishes by the sword are the property of the wielder of that sword,
and the death of the vanquished avenges some pillaged or
massacred ancestor, as well as enriches the conqueror.
A rough rendering is given below of the Song of R’Otman, quoted
by Duveyrier, who justly calls it the Tuareg Marseillaise, which is
chanted in defiance of the Chambas by the Azgueurs, who are their
hereditary enemies.
 Death to thy mother! Ma’atalla the devil is in thee!
Call’st thou the Tuaregs traitors, the men of the plain?
Ha! but they know how to travel, to fight in the battle,
Sally at morn and return in the evening again!
Aye, and they know how to fall on the enemy sleeping—
Sleeping at ease in the tent with his flocks at his side,
Lapped in his fine woollen garments, his curtains and carpets
Spreading full length in the shade of the canopy wide.
What though with milk newly-drawn from the udders of camels,
What though with meat and with butter his paunch he has filled,
Straight as a nail to the ground pins the lance of the victor,
Out with a shriek and a yell flies the soul of the killed!
Sunk in despair lies the heart-broken wife of the victim,
Scattered and vanished their goods like as water o’erspilled!

Wild manners truly do these lines describe, but they also express
proud and heroic sentiments. What will the Tuaregs gain by their
transformation into civilized people?
In a few centuries, where the tents of the Amezzar are pitched
there will be permanent towns. The descendants of the Ihaggaren of
the present day will be citizens. There will be nothing about them to
remind their contemporaries of the wild knights of the desert.
No more will they go to war; no more will they lead razzis to
ravage the camps of their neighbours, for they will have given up
pillage altogether; but perhaps in a bank, which will take the place of
the tent of their Amenokal, they will try to float rotten companies, and
mines which exist nowhere but in the imagination of their chiefs.
What will they be then? Not pillagers but thieves!
Truth to tell, I think I prefer my marauders, who fall on their prey
like the lion Ahar!
AN AFRICAN CAMEL.
AN ISOLATED TREE AT FAFA.
 CHAPTER VI

FROM FAFA TO SAY

Our dread of the passage of the river at Fafa may have seemed
almost childish, and we have since had experience of many another
like it, but for a first attempt it must be admitted it was rather a
teaser.
Narrow and much encumbered, made more difficult by a violent
current, such is the pass of Fafa.
We took as guide the son of the chief of the village, who was later
to pay us a visit at Say. Thanks to him and with the help of his men
we crossed the first rapids without too much difficulty; but, alas! the
rope which was used to transmit to the rudder the movements of the
helm broke just as we emerged from them. Had this happened thirty
seconds sooner the Davoust could not have answered to her helm,
and would have been flung upon the rocks. The damage repaired,
we steered once more into the current, wending our way cautiously
amongst the numerous islands, skirting the course of the reef, our
good star bringing us safely into a quiet reach extending as far as
Wataguna, where we again came to flints lining the bed of the
stream.
In the evening we reached Karu, the Aube having struck once by
the way, but without sustaining much damage; still all these shocks
did not add to her waterproof qualities, and as she shipped more and
more water our anxiety and fatigue became greater and greater. We
had constantly to empty the hold, which did not conduce to the
repose of the passengers, who were often woke up by the noise we
made with our buckets.
 FAFA.

Karu is a pretty little village with thatched huts, amongst which

were many of the barns of a bee-hive shape used for storing millet
alluded to by Barth. We had noticed a good number during the last
few days. The inhabitants of this village are Rimaïbes or serfs of the
Fulahs and Bellates or slaves of the Tuaregs.
The chief of the latter told us how glad he was to see some white
men before he died. He added that he would like to give us some
sheep, but he understood that we never ate anything except the
flesh of black animals, and he had none of that colour.
I said that the colour of the wool did not trouble us at all; all we
cared for was the quality of the flesh, and he went and fetched us a
fine ram. It was the marabouts, who, to add to the probability of their
report that we were sorcerers, had made this assertion about black
animals. There is a custom in the Sudan that animals given as
presents should be as white as possible, as a sign of peace between
donor and receiver. We were now told that Bokar Wandieïdu, chief of
the Logomaten, had assembled a column of troops and was about to
attack us.
 At Karu the mountains were
pointed out to us which line the
famous rapid of Labezenga, which we
expected to reach the next day. A
guide was given to us who was said to
be wonderfully clever, but we saw no
particular sign of his intelligence.
It was on March 14 that we first
saw the terrible Labezenga rapid, and
KARU WITH MILLET I am very sure that we shall none of
GRANARIES. us ever forget it.
Our guide began the day by
performing a number of mummeries, the aim of which appears to
have been to make various evil genii propitious to us. From a leather
bag he took out a lot of flat and shaly flints which had been picked up
in the rapid. He wrapped each one of these flints in a separate piece
of cloth, spat upon them, and arranged them here and there all over
the boat.
The current rapidly swept us into a part of the river pretty free
from obstruction, and every now and then I tried to distract our
guide’s attention from his spells and to get him to give me a little
information, but he merely replied without looking at me that there
was no danger, and that he would stop us at the right time.
 THE LABEZENGA RAPIDS.

Often from behind some little jutting out point which intercepted
our view I heard a peculiar noise, a sort of dull but vague roar. The
rate of the current too increased rapidly, and we rushed along at a
rate of five miles an hour at the least. We listened eagerly, but all of a
sudden we saw that the stream was barred from side to side, a
distance of something like a thousand yards, by a positive wall of
rocks against which the water was dashing up in foam.
Our idiot of a guide looked up at last and saw the danger. He
motioned to us to steer for the bank, but rushing along as we were
with the tremendous current, to attempt to do so would have been
merely to drift helplessly on to the line of rocks, so we continued to
dash on with a speed which almost made me giddy, and presently, to
my intense relief, I saw a place on the right where there was less
foam. Yes, it was the pass, it was the gate of safety, we must make
for it, but was there any hope of our reaching it?
Our coolies bent to their oars and rowed so hard that they were in
danger of breaking them, whilst the sweat poured down their shining
black skins. I had just time to hoist the signal “Do as we do!” which
most fortunately Baudry and the captain of the Dantec understood.
They were just behind us. Now up with the oars and trust to our luck!
The speed increases yet more, the stream sweeps the boat towards
the pass, where it flings itself into the lower reach: we feel ourselves
falling, we shudder, we realize the fatal attraction drawing us in the
direction of the whirlpool; then like an arrow we shoot safely through
the opening. All is well with us at least. Our next anxiety is for our
comrades; we look behind, and a cry of terror bursts from our lips.
The Dantec, which is the next to attempt the pass, has stopped
suddenly; her mast is swept asunder, and has been flung across the
bow by the violence of the shock. All the men were thrown at the
same moment to the bottom of the boat, for the unlucky barge, which
had tried to pass about three feet on one side of the place where we
had got safely through, had struck against a rock which was hidden
by the whirling foam. She received a tremendous blow, but
fortunately did not sink.
But where was the Aube? That was our care now. She was
approaching rapidly, borne on by the current, but the whole pass was
blocked before her. She would crash into the Dantec, and both
vessels must inevitably be wrecked.
But no! Clouds of spray dash up over bow and stern alike; Baudry
has flung out the anchor and the grappling-iron: oh that they may
grip properly!
Thank God! They have. The Aube stops short some three
hundred yards at least from the Dantec at the brink of the rapid.
But what in the world is up now? The Aube is tilted at an angle of
some 45 degrees! The force of the current is such that it has taken
her in the rear and forced her into this extraordinary position, whilst
the grappling-chains and those of the anchor are strained to the
uttermost, producing the terrifying result described.
I now moored the Davoust to the bank, for we must try to save our
other boats.
 With regard to the Dantec it was a simple affair enough, for she is
a wonderful little craft, answering readily to the helm, and so buoyant
that we got off with no worse damage than the bursting asunder of a
couple of planks of her bottom. I sent Digui to help the men on board
of her, and she got safely through.
The rescue of the Aube was a more difficult matter, especially as
her rudder had got broken in the struggle. The anchor was raised all
right, but when it came to the grappling-iron we could not make it
budge; it had probably got jammed between two rocks, and all our
efforts to move it were in vain, indeed they only seemed to fix it more
firmly.
Driven on by the wind and whirled round by the strong eddies of
the current, the unfortunate barge began to describe semicircles
round her own grappling-iron. Of course when we once cut the chain
there would be no time to steer her, and we must therefore manage
to divide it exactly at the moment when she was opposite to the
opening she had to pass through. One second too soon or too late
and she would be lost.
I had climbed to the top of a little ridge, and with fast beating heart
I watched Baudry making his dispositions for the manœuvre he had
to attempt. A Tuareg chose this moment of awful suspense to tap me
on the shoulder and greet me with the formal salutation, Salam
radicum mahindia, and you can imagine how much notice I took of
him.
Without being at all put out by my silence, however, he went on—
“I see that you are in trouble. I have watched all that has been
going on from my camp behind the hills, and ever since early
morning I have felt sure that you were all lost. But God has saved
you and your people. I have forbidden my tribe to come and bother
you, for you know that we always beg of every one. Well, I am going
now, but if you have need of us, Tuaregs and negroes alike are
ready to help you, you have only to send me a messenger. Our
Amenokal has ordered us to meet your wishes.”
 As he finished his speech, I saw Digui deal a great blow to the
chain of the grappling-iron. The Aube fell into the rapid, but she
could not avoid the rock on which the Dantec had struck already.
She strikes, and the whole of her starboard side is completely
immersed. Is she staved in? No, her speed is such that she rushes
on as if nothing had happened. She is saved. A moment later she is
moored beside the Davoust.
“Not so much as a hole in her, Baudry!” I cried.
“No, I don’t think there is,” he replied, “but we had a narrow
escape.” We overhauled her, and there was not a leak anywhere. In
fact, Baudry declared that her planks were really more watertight
than ever.
Then my Tuareg, who had not gone away after all, but whom I
had completely forgotten, spoke to me again: “Enhi!” he said, which
means simply “look!” but his great wild black eyes shone with
pleasure from out of his veil as if some piece of good luck had
happened to himself.
Now are these Tuaregs brutes? are they men who can only be
swayed by interested motives? What nonsense to say they are!
Where did the interested motives come in here? Would it not have
been better for him if our boats had all been sucked down in the
rapids? We ourselves and all our goods would then have been his
lawful prey.
May Providence only grant that I never find any of my fellow-
countrymen worse than the Tuaregs.
You may be sure the brave fellow got his parcel of goods and
many other things as well. With his long swinging step he went off to
his people again, shouting to us by way of adieu, “Ikfak iallah el
Kheir” (“may God give thee all good things!”)
This was, however, but the first of the Labezenga rapids, and that
the easiest. We had scarcely gone a hundred yards further when we
came to a regular cataract some two feet high, barring our passage.
On one side rose lofty heights, on the left the stream was broken into
several arms by islands. In fact, there did not seem to be any
opening on either side, and we were all but in despair of getting
through this time.
Baudry spent the whole afternoon with our guide from Karu,
seeking a practicable pass, but everywhere the scene before him
was most forbidding, one cataract succeeding another and
alternating with boiling whirlpools, whilst the current rushed on at a
rate of seven or eight miles at the least. The river simply seems to
writhe in its course, and here and there it dashes backwards and
forwards from one side to the other of its bed as if in a state of
frenzy. There must be a difference of something like seven feet in the
height of the water.
The least impracticable place seemed to be on the left of our
anchorage between two islands, but I never should have believed
that any boat could pass through even that. We had, however, to
make the venture, and any delay would only render it more difficult,
for the water was falling rapidly.
On the morning of Sunday the 15th
Father Hacquart celebrated mass and
we then prepared for the passage.
The crew of our two big barges was
not strong enough to navigate both at
once, so we decided to send each
vessel separately past the dangerous
THE ‘AUBE’ IN THE RAPIDS. spots, supplementing one crew from
the other, and later we always
adopted this plan, which worked well on emergencies.
Digui was the only one of our captains who could manage such
tours de force, for really there is no other word for the work he had to
perform. Idris, the quarter-master of the Aube, rather loses his head
amongst the rapids, and is absolutely no good as a leader. Of course
all that can be done is to give a general indication of the course to be
pursued, and when the manœuvre has once begun everything must
be left to the intelligence of the pilot, and Digui alone of all my men
was really worthy to be trusted at the helm.
 We fortified ourselves with a good cup of coffee, feeling that it
might be our last, and the Davoust started, Baudry following us in a
canoe.
The scene before us was very much what it had been the day
before—a narrow pass, a diabolical current producing an impression
of unfathomable depth, which made our hearts sink and our breath
come in gasps. On either side the water whirled and surged and
roared unceasingly as it dashed over the huge rocks. Suddenly there
was a tremendous shock, and the boat seemed to slide away from
under our feet. It was the Davoust’s turn to-day. A hidden rock had
battered a hole in her bow in my cabin. Through the gap, some 20
inches big, the water came in in floods, and in less than ten seconds
it was a couple of feet deep.
But it was written in the book of fate that we were to go down to
the sea in the Davoust, and in spite of all our misfortunes, in spite of
everything being against us, in spite of reason, in spite of logic,
something always turned up to save us even at what seemed the
very last moment. The expected miracle always happened, and it is
no exaggeration to say that we experienced dozens of such
miracles.
We were going at such a rate when we struck the rock that for
one instant the barge remained as it were suspended on it, but the
next it was over it and in deep water again.
It so happened, as good luck would have it, that my servant
Mamé was in my cabin when the boat struck, and the water rushed
in at his very feet.
For the brave fellow to tear off his burnous, roll it into a ball and
shove it into the gap in the planks was the work of a few seconds;
that is to say, of just the time during which the rock held us fixed,
preventing us from settling down. We were saved once more. The
miracle had been performed. Only do not fail to notice what a
combination of circumstances was required to bring about the result:
the immense speed with which we were going making us actually
mount the rock, with the presence of Mamé in my cabin all ready to
stop up the hole!
 The Dantec passed through with us without difficulty, and it was
now the turn of the Aube. Digui attempted a manœuvre with her of
positively extraordinary audacity. Knowing all too well that the rock
which had been nearly fatal to us could not possibly be evaded, he
simply flung the boat upon the grass-covered bank, and she climbed
up, driven on by the great speed of the current. Then he let her slide
down again backwards, or, to use the strictly nautical term, to fall
astern.
For all this, however, we every one of us had to pay toll in one
way or another at this infernal Labezenga. The Aube grated on the
point of a hidden rock just as she was about to join us again in quiet
water.
It was now two o’clock in the afternoon, and we had been eight
hours getting over a little more than half a mile in a straight line. We
were famished with hunger, and our craving for food became almost
unbearable. I constituted myself cook, and drawing upon our
reserves of tinned meats and preserved vegetables, which we all felt
we were justified in doing under the circumstances, I seized what
came first, and tumbled everything helter-skelter into a saucepan.
We all devoured the result, which I called tripes à la Labezenga,
without in the least knowing what we were eating. I will give the
recipe to all who wish to emulate Vatel: tripes à la mode de Caen,
truffles, esculent boletus, haricots verts, with plenty of pepper and
spice, served hot. In N. Lat. 14° 57′ 30″, after just escaping from
drowning or from death in the jaws of a crocodile, nothing could be
more delicious, but somehow I have never ventured to try my olla
podrida again in France.
After a little rest, which was indeed well earned, Baudry went with
Digui to the village of Labezenga to try and get guides. He came
back in a state of terror at what he had seen.
For more than a month we had to lead a life such as I have just
described. What I have said will give an idea of all we went through. I
don’t want to dwell too much on our sufferings now that they are
over. Once embarked on such an enterprise as this there is nothing
for it but to go straight ahead, and by degrees one gets accustomed