1

SIMPLE METHOD FOR PERFORMANCE EVALUATION
OF MULTISTAGE ROCKETS
Mauro Pontani
*
and Paolo Teofilatto


Multistage rockets are commonly employed to place spacecraft and satellites in
their operational orbits. Performance evaluation of multistage rockets is aimed at
defining the maximum payload mass at orbit injection, for specified structural,
propulsive, and aerodynamic data of the launch vehicle. This work proposes a
simple method for a fast performance evaluation of multistage rockets. The
technique at hand is based on three steps: (i) the flight path angle at each stage
separation is guessed, (ii) the spacecraft velocity is maximized at the first and
second stage separation, and (iii) for the last stage the thrust direction is obtained
through the particle swarm optimization technique, in conjunction with the use
of the Euler-Lagrange equations and the Pontryagin minimum principle. The
coast duration at the second stage separation is optimized as well. The method at
hand is intended to generate a near optimal trajectory through a simple, easy-to-
implement approach, with the final intent of obtaining a rapid evaluation of the
performance of a specified multistage launch vehicle, with reference to different
operational orbits.
INTRODUCTION
Multistage rockets are commonly employed to place spacecraft and satellites in their opera-
tional orbits. Performance evaluation of multistage rockets is basically aimed at defining the
maximum payload mass that can be inserted in the desired orbit. Usually, if the rocket character-
istics are specified, this evaluation is obtained from optimizing the rocket trajectory, i.e. by de-
termining the optimal control law that leads to maximizing the final mass at orbit injection. In
general, the numerical solution of aerospace trajectory optimization problems is not trivial and
has been pursued with different approaches in the past. Indirect methods, such as the gradient-
restoration algorithm
1,2
and the shooting method
3
, or direct techniques, such as direct colloca-
tion
4,5
, direct transcription
6,7
, and differential inclusion
8,9
, are to name a few. However, only a
relatively small number of publications are concerned with trajectory optimization of multistage
launch vehicles
1,2,10,11
.
This work is concerned with a novel approach that is intended to allow a fast performance
evaluation of multistage rockets with given characteristics, under some simplifying assumptions.
The technique described in this work is applied to a three-stage rocket, whose three-dimensional

*
PhD in Aerospace Engineering, Research Assistant, Scuola di Ingegneria Aerospaziale, University of Rome “La
Sapienza”, via Salaria 851/881, 00138 Rome, Italy

Professor of Flight Mechanics, Scuola di Ingegneria Aerospaziale, University of Rome “La Sapienza”, via Salaria
851/881, 00138 Rome, Italy
(Preprint) IAA-AAS-DyCoSS1-11-05
2
trajectory is composed of the following thrust phases and coast arcs:
(a) first stage propulsion
(b) second stage propulsion
(c) coast arc (after the second stage separation)
(d) third stage thrust phase
In particular, the methodology presented and applied in this paper is based on three steps:
(1) the flight path angle at each stage separation is determined through an iterative process;
(2) for each stage, the native MATLAB routine fmincon is employed for finding the thrust
direction that maximizes the velocity at the stage separation, at which the flight path
angle is constrained to the value found at step (1);
(3) for the third stage, the existence and duration of a coast arc and the optimal thrust
direction are determined through a heuristic technique, i.e. the particle swarm algorithm.
The Euler-Lagrange equations and the Pontryagin minimum principle are employed to
express the control as a function of the adjoint variables conjugate to the dynamics
equations.
The method that is being presented requires a reduced deal of effort in programming,
debugging, and testing the algorithmic codes, as existing routines are used, in conjunction with
analytical developments and a simple implementation of swarming algorithm. Nevertheless, the
trajectories yielded through the technique at hand resemble (qualitatively) the ones found by
Miele
2
and those of existing rockets as the Scout
12
. Hence, the methodology treated in this paper
is intended to: (i) yield a reasonable solution for performance evaluation of multistage rockets and
(ii) represent a technique for generating a suitable first-attempt guess trajectory to be employed
by more refined algorithms tailored to optimizing the overall trajectory.
PROBLEM DEFINITION
This research addresses the problem of performance evaluation of multistage rockets through
the determination of a near optimal ascending trajectory, which terminates at injection of the
payload in the desired operational orbit. This is a circular orbit of radius
d
R , inclination
d
i , and
right ascension of the ascending node (RAAN)
d
Ω . The three-stage launch vehicle is modeled as
a point mass, in the context of a three-degree-of-freedom problem.
Rocket characteristics
The three-stage rocket that is being considered is the Muralm missile. It has specified
structural, propulsive, and aerodynamic characteristics.
For the sake of simplicity, the mass distribution of the launch vehicle can be described in
terms of masses of subrockets: subrocket 1 is the entire rocket, including all the three stages,
subrocket 2 is the launch vehicle after the first stage separation, subrocket 3 is the launch vehicle
after separation of the first two stages, and therefore is represented by the third stage only. Let
( )
0
i
m denote the initial mass of subrocket i. This mass
( )
0
i
m is composed of a structural mass
( ) i
S
m ,
a propellant mass
( ) i
P
m , and a payload mass
( ) i
U
m (
( ) ( ) ( ) ( )
0
i i i i
S P U
m m m m = + + ). For the first two
subrockets
( ) i
U
m ( 1, 2 i = ) coincides with the initial mass of the subsequent subrocket (i.e.
( ) ( ) 1
0
i i
U
m m
+
≡ ). With regard to the third subrocket
( ) 3
U
m is the (actual) payload mass, which is to be
maximized. The mass distribution for the Muralm three-stage rocket is the following:
3

( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
1 1 1 1
0
2 2 2 2
0
3608 kg 306 kg 2480 kg 822 kg
822 kg 46 kg 370 kg 406 kg
S P U
S P U
m m m m
m m m m
= = = =
= = = =
(1)
The initial mass of the third stage is specified and this implies that minimizing the third stage
propellant expenditure implies maximizing the payload mass that can be injected in the desired
orbit.
The propulsive characteristics of the launch vehicle can be described in terms of thrust
magnitude T and specific impulse
SP
I . Both these quantities are assumed as time-independent for
all of the stages. If the superscript “
( ) j
” ( ) 1, 2, 3 j = refers to stage j, the following propulsive data
characterize the Muralm rocket:

( ) ( ) ( )
( ) ( ) ( )
1 2 3
1 2 3
235180 N 23472 N 7500 N
290 sec 291 sec 320 sec
SP SP SP
T T T
I I I
= = =
= = =

(2)
For each subrocket i it is then straightforward to derive the thrust acceleration as a function of the
time t:

( )
( ) ( )
( )
( )
( )
0
0 0
1
i
i i i
i
ign SP
n T
m
n t t g I
| |
=
|
\ ¹ − −

(3)
where
( )
0
i
n is the thrust acceleration at ignition of stage i (which occurs at the time
( ) i
ign
t ) and
0
g is
the gravitational acceleration at sea level. Lastly, due to the specified mass distribution, the
burnout times
( ) i
BO
t ∆ for the first two subrockets ( 1, 2 i = ) are given by

( )
( )
( )
( )
( )
0
0 0
i i
i
SP P
BO
i i
g I m
t
n m
∆ = (4)
The aerodynamics of the Muralm rocket has been modeled through the Missile DATCOM
database. The aerodynamic force is assumed to be composed of two terms, lift L and drag D,
( )
( )
( )
( ) 2 2
1 1
, and ,
2 2
i i
L A D A
L c M S v D c M S v α ρ α ρ = = (5)
where the lift and drag coefficients
L
c and
D
c depend on M and α , i.e. Mach number and angle
of attack, respectively.
( ) i
S is the reference surface area for subrocket i,
A
ρ is the (local)
atmospheric density, and v is the relative velocity (cf. Appendix 1). For the Muralm launch
vehicle
( )
( )
2
0.636 m 1, 2, 3
j
S j = = .
Equations of motion
The equations of motion that govern the rocket dynamics can be conveniently written in terms
of its radius r, (geographical) longitude ξ , latitude φ , flight path angle γ , velocity v, and head-
ing angle ζ . These variables refer to the relative motion in an Earth-centered rotating frame
(ECEF) and are described in more detail in Appendix 1. They form the state vector
4
[ ]
( )
T
R
r v ξ φ γ ζ x ≜ of the launch vehicle (in rotating coordinates). Omitting the su-
perscript “
( ) i
”, for each subrocket the equations of motion are
sin r v γ = ɺ
(6)

cos cos
cos
v
r
γ ζ
ξ
φ
=
ɺ

(7)

cos sin v
r
γ ζ
φ =
ɺ
(8)
( )
2
2
sin
cos cos 2 cos cos cos sin sin sin
T E E
E
r T v L
m v r r v mv v
α µ ω
γ γ φ ω ζ φ γ φ γ ζ
(
| |
= + − + + + +
( |
\ ¹
¸ ¸
ɺ
(9)
( )
2
2
cos
sin cos cos sin sin cos sin
T E
E
T D
v r
m v r m
α µ
γ ω φ φ γ φ γ ζ = − − + − ɺ (10)

2
tan cos cos 2 cos tan sin sin cos cos 2 sin
cos
E
E E
r v
r v
ω
ζ φ γ ζ ω φ γ ζ φ φ ζ ω φ
γ
= − + − −
ɺ
(11)
where
( )
3 2
398604.3 km sec
E
µ = and
( )
5 -1
7.292115 10 sec
E
ω

= ⋅ are the Earth gravitational
parameter and rotation rate, respectively, L and D are given by Eq. (5), and the time-varying mass
is
( ) ( ) ( )
( )
( )
( ) 0 0
i i i i
ign SP
m m T t t g I = − − . As the thrust vector is assumed to be coplanar with the posi-
tion vector r and the velocity vector v, the angle
T
α suffices to define its direction, which is taken
counterclockwise from v. The state
R
x is continuous across stage separations, which occur at
time
1
t (first stage separation) and
2
t (second stage separation).
Since the Muralm is assumed to be air-launched from a Tornado fighter, the initial conditions
for Eqs. (6)-(11) are the following:
( ) ( ) ( ) ( ) ( ) ( )
0 0 0
0 0 0 0 0 0
2
E L L L
r R h v v
π
ξ µ φ φ γ γ ζ χ = + = = = = = − (12)
where
0
h ,
0
γ , and
0
v are the rocket altitude, flight path angle and (relative) velocity at release
from the carrier aircraft.
L
φ ,
L
χ ,
L
µ represent respectively the latitude, azimuth, and (geographi-
cal) longitude at release. ( ) 6378.165 km
E
R = is the Earth radius. For the problem at hand the
following data are assumed:

0 0 0
6.7 km 52.6 deg 5.2 deg 35.0 deg 0.280 km sec
L L
h v µ φ γ = = − = = = (13)
The azimuth at release
L
χ is determined in relation with the desired orbit inclination (cf. next
section).
5
METHOD OF SOLUTION
This work employs two distinct approaches for determining the trajectory of the first two
stages and that of the last stage. The orbital plane selection is assumed to occur during the pow-
ered motion of the first two subrockets. Moreover, for the upper stage aerodynamic forces are
negligible, and therefore they are not considered. These circumstances imply that the trajectory of
the last stage is planar.
Trajectory of the first two stages
The determination of the trajectory of the first two stages is based on maximizing the final ve-
locity at each stage separation, at which the flight path angle is constrained to a specified value.
The control is represented by the thrust direction, defined by
T
α , and by the aerodynamic angle
of attack, α . However, as the thrust direction is always aligned with the rocket longitudinal axis,
these two angles coincide:
T
α α ≡ .
While ( )
1
t γ is set to 30 deg, the flight path angle at the second stage separation, ( )
2
t γ , is ad-
justed iteratively up to obtaining an apogee altitude in the range [ ] ,
d d
h h h − ∆ , where
d
h is the
altitude of the desired circular orbit, and h ∆ is set to 10 km. The first-attempt value for ( )
2
t γ is
chosen to be 25 deg.
The orbit inclination is selected through a simultaneous iterative process on the initial azimuth
at rocket release. This process is relatively straightforward. In fact, in the range [0, 2] π the orbit
inclination increases as
L
χ decreases (for 2
L
χ π ≤ ). Convergence is declared when inclination
is in the range [ ] ,
d d
i i i i − ∆ + ∆ , where i ∆ is set to 3 deg . At each iteration of the process (for
given values of ( )
1
t γ , ( )
2
t γ , and
L
χ ), the MATLAB routine fmincon determines the angle ( ) t α
that maximizes v at each stage separation.
Finally, the correct selection of the release time
L
t leads to obtaining the desired right ascen-
sion of the ascending node (RAAN),
d
Ω . In fact,
L
t is directly related to the Greenwich sidereal
time at release
gL
θ ,
( )
gL g E L
t t θ θ ω = + −
(14)
where
g
θ is the (known) Greenwich absolute longitude at a reference time t . The RAAN is
given by the relations reported in Appendix 2 and, once the rocket final state (at orbit injection) is
known, specifying the RAAN leads to obtaining
gL
θ and then
L
t :
( )
2

gL g
L
E
k
t t k
θ θ π
ω
− +
= + ∈Z (15)
At separation of the second stage, the state is denoted with ( )
2 R
t x . Through the relations re-
ported in Appendix 2 ( )
2 R
t x is employed to derive the inertial state, denoted with ( )
2 I
t x and
including the following components: radius r, absolute longitude
I
ξ , latitude φ , inertial flight
path angle
I
γ , inertial velocity
I
v , and inertial heading angle
I
ζ . Then ( )
2 I
t x is employed to
6
derive the orbital elements at
2
t , i.e. the orbit semi-major axis (SMA)
2
a , eccentricity
2
e , incli-
nation i, RAAN Ω, argument of perigee
2
ω , and true anomaly ( )
2
f t (cf. Appendix 2).
Trajectory of the last stage
As remarked previously, the last stage trajectory is assumed to be planar. This circumstance
implies that the orbit inclination and RAAN do not change during the last stage propulsion.
The third stage trajectory is assumed to be composed of two phases: a coast arc and a thrust
phase. It is relatively straightforward to prove that during the coast arc the true anomaly variation,
denoted with f ∆ , suffices to describe the rocket dynamics. In fact, if
3
t represents the ignition
time of the third state, then ( ) ( )
3 3 2
f f t f t f = + ∆ ≜ . The orbital elements
2
a ,
2
e , and
2
ω do not
vary along a coast arc. As a result the radius, inertial velocity and inertial flight path angle at
3
t
are simply given by
( )
( )
2
2 2
3
2 3
1
1 cos
a e
r t
e f

=
+
(16)
( )
2 3
3
2 3
sin
arctan
1 cos
I
e f
t
e f
γ
(
=
(
+
¸ ¸

(17)
( )
( )
2
3 2 2 3
2
2 2
1 2 cos
1
E
I
v t e e f
a e
µ
= + +


(18)
In the powered phase, the third stage motion can be conveniently described through the use of the
equations of motion for r,
I
v , and
I
γ (also referred to as the state equations hence forward):

2 2
cos sin
sin sin cos
I E I I E
I I I I I I
I I I
v T T
r v v
m v r m v r r v
α µ α µ
γ γ γ γ
| |
= = − = + −
|
\ ¹
ɺ ɺ ɺ
(19)
The thrust direction is now specified through the thrust angle
I
α , taken counterclockwise from
the direction of the inertial velocity vector
I
v . The operational orbit is assumed to a be circular
orbit of radius
d
R . This means that the final conditions at orbit injection (which occurs at
f
t ) are

( ) ( ) ( )
0
E
f d I f I f
d
r t R v t t
R
µ
γ = = =
(20)
The trajectory for the last stage is obtained by minimizing fuel consumption, which is equivalent
to minimizing the thrust duration. Hence, the objective function for the third stage trajectory op-
timization is

3 f
J t t = −
(21)
7
In summary, the optimization problem for the upper stage is the following: determine the optimal
control law ( )
I
t α and the optimal true anomaly
3
f such that J is minimized. The ignition time
3
t
is then calculated through the Kepler’s law:
( ) ( ) ( ) ( ) { }
3
2
3 2 3 2 2 3 2 2
sin sin
E
a
t t E t E t e E t e E t
µ
= + − − ( + (
¸ ¸ ¸ ¸
(22)
where ( )
k
E t is the eccentric anomaly associated with ( ) ( ) 2, 3
k
f t k = .
Letting [ ]
1 2 3
T T
I I
x x x r v γ = (
¸ ¸
x ≜ , to obtain the necessary conditions for an optimal
solution a Hamiltonian H and a function of the boundary conditions Φ are introduced as

2
1 2 3 2 3 3 3 2 2
2 1 2 1 1 2
cos sin
sin sin cos
I E I E
x T T
H x x x x
m x x m x x x x
α µ α µ
λ λ λ
( ( | |
+ − + + −
( | (
¸ ¸ \ ¹ ¸ ¸

(23)

( )
( )
( )
2
2 2
2
3 1 10 2 20 2 2 3
2
2 3 2 2
2 3
3 30 4 1 5 2 6 3
2 3
1
1 2 cos
1 cos 1
sin
arctan
1 cos
E
f
E
f d f f
d
a e
t t x x e e f
e f a e
e f
x x R x x
e f R
µ
υ υ
µ
υ υ υ υ
( (

( ( Φ − + − + − + +
+ ( − (
¸ ¸ ¸ ¸
( (
( + − + − + − +
( (
¸ ¸
+
( ¸ ¸ ¸ ¸


(24)
where ( )
0 3 k k
x x t = and
( ) ( ) 1, 2, 3
kf k f
x x t k = = ; [ ]
( ) 1 2 3
T
λ λ λ ≜ λ and
[ ]
( ) 1 2 3 4 5 6
T
υ υ υ υ υ υ ≜ υ represent respectively the adjoint variable conjugate to the
dynamics equations (19) and to the boundary conditions (16)-(18) and (20). The necessary condi-
tions for optimality
13
yield the following adjoint equations for the costate [ ]
( ) 1 2 3
T
λ λ λ ≜ λ :
( )
3
1 2 3 3 2 3 3 2 3
1 2 1
cos 1 1
cos 2 sin 2
E E
x
x x x
x x x
λ λ µ λ µ λ
| |
= − +
|
\ ¹
ɺ

(25)

2 1 3 3 3 2 2 2
1 1 2 2
1
sin cos sin
E P
I
P
T
x x
x x x m x
µ
λ λ λ α
( | |
= − − + −
( |
\ ¹ ¸ ¸
ɺ

(26)

3 2
3 2 1 3 2 3 3 2 2
1 1 1 2
cos
cos sin
E
E
x x
x x x
x x x x
µ
λ λ µ λ λ
| |
= − + + −
|
\ ¹
ɺ

(27)
in conjunction with the respective boundary conditions:
( )
0 3
and 1, 2,3
k k kf k
k λ υ λ υ
+
= − = =
(28)
In the presence of initial conditions depending on a parameter ( )
3
f a pair of additional necessary
conditions must hold
14
:
8

2
2
3 3
0 and 0
f f
∂Φ ∂ Φ
= ≥
∂ ∂

(29)
The first equation yields a relation that expresses
30
λ as a function of
10
λ ,
20
λ , and
3
f :

( )
( )
2
2 2
2 2
3 2 2 3 2 2 3
30 20 10 2
2 3 2 3
2 3
1
sin 1 2 cos 1 2 cos
cos cos
1 cos
a e
f e e f e e f
e f e f
e f
λ λ λ

+ + + +
= −
+ +
+
(30)
In addition, the optimal control
*
I
α can be expressed as a function of the costates through the
Pontryagin minimum principle:

1 1
2 2
2 2
* * 2 * 2 3 3 3
2 2 2
2 2 2
argmin sin and cos
I
I I I
H
x x x
α
λ λ λ
α α λ α λ λ
− −
( (
| | | |
( ( = ⇒ = − + = − +
| |
( (
\ ¹ \ ¹
¸ ¸ ¸ ¸

(31)
Lastly, as the final time is unspecified, the following transversality condition must hold:

( )
( )
( )
( )
( )
( )
2
3
2
3
0
2
3 3
2
0 3 0
0 1 0
1
f
f f
f f
f SP
n
H t
t x
n t t g I
λ
λ
(
∂Φ
( + = ⇒ + − =
(
¸ ¸

− − (
¸ ¸
(32)
The necessary conditions for optimality allow translating the optimal control problem into a two-
point boundary-value problem involving Eqs. (25)-(32), with unknowns represented by the initial
values of λ ,
3
f , and
f
t .
Equality constraints reduce the search space where feasible solutions can be located. However,
what will be demonstrated is that for the problem at hand the transversality condition can be ne-
glected by the swarming algorithm. To do this, one has first to recognize the special structure of
the costate equations (25)-(27), which are homogeneous in λ . This circumstance implies that if
an optimization algorithm is capable of finding some initial value of λ such that
( ) ( )
*
1 1
0 0 , k
λ
λ λ = ( ) ( )
*
2 2
0 0 , and k
λ
λ λ = ( ) ( )
*
3 3
0 0 k
λ
λ λ = ( ) 0 k
λ
> (where the superscript “*” de-
notes the actual optimal value of a variable), then the same proportionality holds between λ and
*
λ at any t, due to homogeneity of Eqs. (25)-(27). Moreover, the control
I
α can be written as a
function of λ through Eq. (31) and one can recognize that
I
α coincides with
*
I
α :

( )
1 1
2 2
2 2
* *
2
2 * * 3 3 3 3
2 2
2 2 2 2
sin sin
I I
k k
k
x x x x
λ λ
λ
λ λ λ λ
α λ λ α
− −
( (
| | | |
( ( = − + = − + ≡
| |
( (
\ ¹ \ ¹
¸ ¸ ¸ ¸

(33)

( )
1 1
2 2
2 2
*
2
2 * * * 3 3
2 2 2 2
2 2
cos cos
I I
k
k k
x x
λ
λ λ
λ λ
α λ λ λ λ α
− −
( (
| | | |
( ( = − + = − + ≡
| |
( (
\ ¹ \ ¹
¸ ¸ ¸ ¸

(34)
9
This circumstance implies that if λ is proportional to
*
λ then the final conditions are fulfilled at
the minimum final time
*
f
t . In contrast, the transversality condition is violated, because the value
of
( )
*
f
H t , due to Eq. (32), turns out to be

( )
( )
( )
( )
( )
( )
2
3
2
3 * 0
2
3 3 * *
2
0 3 0
1
1
f
f f
f
f SP
n
H t k
x
n t t g I
λ
λ
λ
(
( = − + = − ≠ −
(
¸ ¸
− − (
¸ ¸
(35)
Therefore, provided that the proportionality condition holds, the optimal control
*
I
α can be de-
termined without considering the transversality condition, which in fact is ignorable in this con-
text. Thus, this condition is discarded, in order to reduce the number of equality constraints con-
sidered by the swarming algorithm, with the intent of improving its performance.
In short, the following parameter set can be employed in the solution process:
{ }
10 20 3
, , ,
f
f t λ λ .
The remaining parameter
30
λ can be easily obtained by means of the relation (30). Specifically,
the swarming technique is based on the following points:
(a) the control variable is expressed as a function of the costate through Eq. (31)
(b) the value of
30
λ is calculated by means of Eq. (30), after picking the unknown values of the
remaining Lagrange multipliers at the initial time ( )
10 20
and λ λ , and the true anomaly
3
f
(c) the state equations (19) are integrated numerically up to the final time
f
t , together with the
adjoint equations (25)-(27). The respective initial conditions are known once the parameters
3
f (for the state equations) and { }
10 20 30
, , λ λ λ (for the adjoint equations) are specified
(d) the inequality in Eq. (29) (not expanded for the sake of brevity) and the conditions at injec-
tion (20) are evaluated
In summary, the problem reduces to the determination of four unknown parameters,
{ }
10 20 3
, , ,
f
f t λ λ , that lead the dynamical system to satisfying three conditions (Eqs. (20)).
Outline of the particle swarm optimization method
The particle swarm optimization (PSO) method is a heuristic technique aimed at finding the
optimal values of a set of unknown parameters, for a generic dynamical system. In this research
the optimization problem involves a continuous time-dependent control variable and is translated
into a parameter optimization problem through the necessary conditions for optimality, which
allow expressing the control variable as a function of the adjoint variables conjugate to the dy-
namics equations.
Unconstrained parameter optimization problems can be stated as follows: determine the opti-
mal values of the n unknown parameters { }
1
, ,
n
χ χ … such that the objective function J is mini-
mized. The PSO technique is a population-based method, where the population is represented by
a swarm of N particles. Each particle i ( ) 1, , i N = … is associated with a position vector ( ) i χ and
with a velocity vector ( ) i w . The position vector includes the values of the n unknown parameters
of the problem, ( ) ( ) ( )
1
T
n
i i i χ χ (
¸ ¸
χ ≜ … , whereas the velocity vector, whose components are
10
denoted with ( ) ( ) 1, ,
k
w i k n = … , determines the position update. Both the position and the ve-
locity components are constrained to suitable ranges:
( ) and 1, ,
k k k k k k
a b d w d k n χ ≤ ≤ − ≤ ≤ = … (36)
where
k k k
d b a − ≜ . Each particle represents a possible solution to the problem, and corresponds
to a specific value of the objective function. The expressions for position and velocity update de-
termine the swarm evolution toward the location of the globally optimal position, which corre-
sponds to the globally optimal solution to the problem of interest. The initial population is ran-
domly generated by introducing N particles, whose positions and velocities are (stochastically)
uniformly distributed in the respective search spaces, defined by Eq. (36). The algorithm termi-
nates when the maximum number of iterations
IT
N is reached. The position vector of the best
particle is expected to contain the optimal values of the unknown parameters, which correspond
to the global minimum of J. A fairly large number of iterations are set to ensure that the final so-
lution is stable enough to be considered optimal. The central idea underlying the method is con-
tained in the formula for velocity updating. This formula includes three terms with stochastic
weights: the first term is the so called inertial component and for each particle is proportional to
its velocity in the preceding iteration; the second component is termed the cognitive component,
directed toward the personal best position, i.e. the best position experienced by the particle; and
finally the third term is the social component, directed toward the global best position, i.e. the
best position yet located by any particle in the swarm. A detailed description of the algorithm
(with also the same settings adopted in this work) is contained in Reference 15.
Constrained optimization problems involve equalities and/or inequalities, regarding (directly
or indirectly) the unknown parameters. Equality constraints narrow considerably the search space
where feasible solutions can be located. This is due to the fact that (non-redundant) equality con-
straints actually reduce the degree of freedom of the problem according to their number. In fact,
m equality constraints reduce the degree of freedom by m. Therefore, in the presence of n un-
known parameters, at most m n = equality constraints are admissible ( ) m n ≤ : ( ) 0
r
l = χ
( ) 1, , r m = … . The most popular approach for dealing with these constraints consists in penalizing
them by adjoining additional terms to the objective function:
( )
2
1
m
r r
r
J J l η
=
= +

χ
ɶ

(37)
This approach is employed also in this research, with ( ) 1 1, 2, 3
r
r η = = . The equality constraints
are represented by the relations (20). As the necessary conditions for optimality are used to ex-
press the control variable, the PSO assumes as objective the expression (37) with 0 J ≡ . Inequal-
ity constraints are less problematic because they reduce the search space of feasible solutions
without decreasing the degree of freedom of the problem. For each particle the simplest way of
treating inequality constraints consists in assigning a fictitious infinite value to the objective func-
tion if the particle violates at least one of them. In this work a single inequality constraint exists,
i.e. the second relation reported in Eq. (29). Further options exist for dealing with constraints
(e.g., methods based on preserving feasibility of solutions, methods that distinguish between fea-
sible and infeasible solutions, cf. References 16 and 17). However, the simple strategy described
previously for equality and inequality constraints will be shown to be quite effective for the prob-
lem considered in this paper.
11
ASCENDING TRAJECTORIES
The ascending trajectories are determined by employing canonical units: the distance unit
(DU) is the Earth radius, whereas the time unit (TU) is such that
3 2
1 DU TU
E
µ = . Thus,
1 DU 6378.165 km = and 1 TU 806.8 sec = .
With regard to the last stage trajectory, the swarming optimizer is employed with the follow-
ing settings: 50 N = (number of particles) and 500
IT
N = (number of iterations). The optimal
values of the unknown parameters are sought in the following ranges: ( )
0
1 1 1, 2,3
k
k λ − ≤ ≤ = ,
0 f π ≤ ∆ ≤ , and 0.01 TU 0.2 TU
f
t ≤ ≤ . It is worth noticing that the constraint reduction allows
arbitrarily defining the search space for the initial values of the Lagrange multipliers. This means
that they can be sought in the interval
0
1 1
k
λ − ≤ ≤ by the PSO algorithm, and only a posteriori
their correct values (fulfilling also the transversality condition (32)) can be recovered.
Three test cases have been considered:

Case 1: 280 km and 5.2 deg
Case 2: 280 km and 45 deg
Case 3: 280 km and 90 deg
d d
d d
d d
h i
h i
h i
= =
= =
= =


The main results (azimuth at release
L
χ , flight path angle at the second stage separation
( )
2
t γ , orbit semi-major axis
f
a , eccentricity
f
e , inclination
f
i , coast duration
CO
t ∆ , and final
mass at injection
f
m ) are reported in the following. Figures 1, 3, and 5 portray some state com-
ponents (radius, velocity and flight path angle) and the ground tracks, whereas Figures 2, 4, and 6
illustrate the corresponding thrust pointing angle time histories for each of the three subrockets.


( )
2
7
Case 1: 0.0 deg 20.0 deg 175.2 sec
280.004 km 7.260 10 5.2 deg 140.4 kg
L CO
f E f f f
t t
a R e i m
χ γ

= = ∆ =
= + = ⋅ = =

(38)

( )
2
7
Case 2: 40.0 deg 20.5 deg 171.3 sec
280.002 km 4.377 10 45.7 deg 133.7 kg
L CO
f E f f f
t t
a R e i m
χ γ

= = ∆ =
= + = ⋅ = =

(39)

( )
2
7
Case 3: 4.0 deg 21.0 deg 158.2 sec
279.994 km 4.403 10 87.7 deg 120.6 kg
L CO
f E f f f
t t
a R e i m
χ γ

= − = ∆ =
= + = ⋅ = =

(40)
From inspecting the results summarized in Eqs. (38)-(40) it is apparent that the final conditions at
injection are fulfilled in a satisfactory way. Moreover, the final mass decreases as the orbit incli-
nation increases. This is a predictable result and is related to the launch site, which is located in
the proximity of the Earth equator. This circumstance implies that as the operational orbit inclina-
tion increases, an increasing amount of fuel is employed to correct the initial inertial velocity,
which has a nonnegligible component directed toward East.
12

Figure 1. Case 1: altitude, inertial velocity and flight path angle, ground track


Figure 2. Case 1: control for the first two stages (a) and for the last stage (b)
13

Figure 3. Case 2: altitude, inertial velocity and flight path angle, ground track


Figure 4. Case 2: control for the first two stages (a) and for the last stage (b)
14

Figure 5. Case 3: altitude, inertial velocity and flight path angle, ground track


Figure 6. Case 3: control for the first two stages (a) and for the last stage (b)
15
CONCLUSION
The generation of an optimal trajectory for multistage launch vehicles is a challenging prob-
lem, treated with different approaches in the past. This work proposes and successfully applies a
simple technique for generating near optimal three-dimensional ascending trajectories for multi-
stage rockets, for the purpose of performance evaluation. Only existing routines and a simple im-
plementation of swarming algorithm are employed, in conjunction with the analytical necessary
conditions for optimality, applied to the upper stage trajectory. A particular configuration of the
launch vehicle is assumed, taking into account realistic propulsive, structural, and aerodynamic
data.
This study proves how the desired orbit right ascension of the ascending node and inclination
can be obtained by properly selecting the launch time and the azimuth angle at release, respec-
tively. The remaining orbit elements of the operational orbit are attained by choosing a suitable
control law for the thrust direction time history, and by selecting the optimal coast duration at the
second stage separation.
The solution found in this work appears suitable also for being employed as a guess for more
refined optimization algorithms.
APPENDIX 1. REFERENCE FRAMES
The rocket motion is conveniently described in rotating, i.e. non-inertial, reference frames.
The Earth-centered Earth-fixed (ECEF) reference frame represents a reference system that is rig-
idly attached to the Earth. Under the assumption that the planet rotates with the constant angular
rate
E
ω this circumstance implies that also the reference frame rotates with angular rate
E
ω with
respect to an inertial Earth-centered (ECEI) reference frame, denoted with ( )
1 2 3
ˆ ˆ ˆ , , c c c . The origin
of both the inertial and the rotating coordinate system is the center of the planet, O;
1
ˆ c is the ver-
nal axis. The unit vector
3
ˆ
ˆ c k ≡ is aligned with the planet axis of rotation and is positive north-
ward, and therefore
( )
ˆ

E E
k ω ≜ ω represents the (vector) rotation rate of the ECEF-frame with
respect to the ECEI-frame. The unit vector ˆ ι intersects the Greenwich meridian at all times. The
ECEF-frame is associated with
( )
ˆ ˆ
ˆ, , j k ι , which form a right-handed, time-dependent sequence of
unit vectors. As the reference Greenwich meridian rotates with rotation rate
E
ω , its angular posi-
tion (with respect to the ECEI-frame) is identified by its absolute longitude (usually termed
Greenwich sidereal time) ( ) ( ) ( )
g g E
t t t t θ θ ω = + − , where t denotes a generic time instant.
The position vector of the orbiting spacecraft in the ECEF-frame is denoted with r, whereas
the subscript “I ” corresponds to a quantity in the ECEI-frame. As the ECEF-frame and the ECEI-
frame have a common origin, it is evident that
I
≡ r r . In contrast, the inertial velocity
I
v is re-
lated to the (relative) velocity v through the following expression:

I E
= + × v v ω r (41)
This means that, unlike the position vector, the velocity vector in rotating coordinates, v , does
not coincide with the inertial velocity vector,
I
v .
An additional useful reference frame is represented by the right-handed Cartesian coordinate
system
( )
ˆ ˆ
ˆ, , r E N , where ˆ r is the unit vector aligned with r,
ˆ
E is directed along the local East
16
direction, and
ˆ
N is aligned with the local North direction. It is straightforward to obtain
( )
ˆ ˆ
ˆ, , r E N from
( )
ˆ ˆ
ˆ, , j k ι through two elementary rotations: the first (counterclockwise by the geo-
graphical longitude angle ξ ) about axis 3, the second (clockwise by the latitude angle φ ) about
axis 2.
A third useful rotating frame is
( )
ˆ ˆ
ˆ r, ,h θ , where
ˆ
θ is aligned with the projection of the velocity
v onto the horizontal plane
( )
ˆ ˆ
E, N . This frame is obtained from
( )
ˆ ˆ
ˆ, , r E N through a counter-
clockwise rotation about axis 1 by the heading angle ζ (cf. Fig. 7). As a result,
( )
ˆ ˆ
ˆ r, ,h θ can be
written as a function of
( )
ˆ ˆ
ˆ, , j k ι :
ˆ ˆ ˆ ˆ
ˆ ˆ
T T
A
r h j k θ ι
( (
=
¸ ¸ ¸ ¸
R , where

cos cos cos sin sin
sin sin cos cos sin sin sin sin cos cos sin cos
cos sin cos sin sin cos sin sin sin cos cos cos
A
φ ξ φ ξ φ
ζ φ ξ ζ ξ ζ φ ξ ζ ξ ζ φ
ζ φ ξ ζ ξ ζ φ ξ ζ ξ ζ φ
(
(
= − − − −
(
( − + − −
¸ ¸
R
(42)
Finally, the reference frame
( )
ˆ
ˆ ˆ n,v,h (with ˆ v aligned with v) is obtained from
( )
ˆ ˆ
ˆ r, ,h θ through
a clockwise rotation about axis 3 by the flight path angle γ (Fig. 7). From inspecting Fig. 7 it is
apparent that
[ ]
ˆ ˆ
ˆ sin cos cos cos sin
T
v r E N γ γ ζ γ ζ
(
=
¸ ¸
v (43)
The overall aerodynamic force A is conveniently written in the
( )
ˆ
ˆ ˆ n,v,h -frame as the sum of two
terms, i.e. lift L and drag D:
ˆ ˆ Ln Dv = − A (44)


Figure 7.
( )
ˆ ˆ
ˆ r, E, N -frame,
( )
ˆ
ˆ ˆ n,v,h -frame, and related angles
17
APPENDIX 2. STATE VARIABLES AND ORBITAL ELEMENTS
The derivation of the orbital elements ( ) , , , , , a e i f ω Ω from the relative state variables
( ) , , , , , r v ξ φ γ ζ follows two steps:
(1) derivation of the inertial state variables ( ) , , , , ,
I I I I I I
r v ξ φ γ ζ from the relative state vari-
ables ( ) , , , , , r v ξ φ γ ζ
(2) derivation of the orbital elements ( ) , , , , , a e i f ω Ω from the inertial state vari-
ables ( ) , , , , ,
I I I I I I
r v ξ φ γ ζ
With regard to step (1), the fact that
I
≡ r r and the definition of the reference
( )
ˆ ˆ
ˆ, , j k ι implies
that

I I g I
r r ξ ξ θ φ φ = = + =
(45)
The velocity vectors
I
v and v have the following expressions in the rotating frame
( )
ˆ ˆ
ˆ, , r E N :

ˆ ˆ
sin sin
ˆ ˆ
cos cos and cos cos
ˆ ˆ cos sin cos sin
T T
I
I I I I
I I
r r
v E v E
N N
γ γ
γ ζ γ ζ
γ ζ γ ζ
( (
( (
( (
( (
= =
( (
( (
( (
( (
¸ ¸ ¸ ¸
¸ ¸ ¸ ¸
v v
(46)
Due to Eq. (46) and to the fact that
ˆ
cos
E E
r E ω φ × = ω r , Eq. (41) yields three relations:
sin sin
I I
v v γ γ = (47)
cos cos cos cos cos
I I I E
v v r γ ζ γ ζ ω φ = + (48)
cos sin cos sin
I I I
v v γ ζ γ ζ = (49)
which lead to the following expressions of
I
γ ,
I
v , and
I
ζ as functions of γ , v , and ζ :
( )
2
2
cos 2 cos cos cos
I E E
v v r v r ω φ ω φ γ ζ = + + (50)

sin
arcsin
I
I
v
v
γ
γ
(
=
(
¸ ¸

(51)

cos cos cos cos sin
cos and sin
cos cos
E
I I
I I I I
v r v
v v
γ ζ ω φ γ ζ
ζ ζ
γ γ
+
= =
(52)
With regard to step (2), the in-plane orbital elements a, e, and f can be easily calculated from r,
I
γ , and
I
v . In fact, the conservation of energy yields a:
18

2
2
E
E I
r
a
rv
µ
µ
=


(53)
Then, the angular momentum magnitude equals cos
I I
rv γ and can also be written in terms of or-
bital elements as
( )
2
1
E
a e µ − . Hence, the orbit eccentricity is given by

( )
2
cos
1
I I
E
rv
e
a
γ
µ
= − (54)
The true anomaly f can be found by considering the polar equation of the ellipse,

( ) ( )
2 2
1 1
cos
1 cos
a e a e r
r f
e f re
− − −
= → =
+
(55)
in conjunction with the radial component of velocity:

( )
( )
2
2
1
sin
sin sin sin
1
a e
v
e f v f
e a e
µ γ
γ
µ

= → =

(56)
The remaining orbital elements (i, Ω, and ω ) can be obtained by considering the rotation matrix
that relates
( )
ˆ ˆ
ˆ r, ,h θ to ( )
1 2 3
ˆ ˆ ˆ , , c c c , written in terms of the angles i, Ω, and
( )
f θ ω + ≜ :
[ ]
1 2 3
ˆ ˆ
ˆ ˆ ˆ ˆ
T
T
B
r h c c c θ
(
=
¸ ¸
R , where
18
:

cos cos sin cos sin cos sin sin cos cos sin sin
sin cos cos cos sin sin sin cos cos cos cos sin
sin sin sin cos cos
B
i i i
i i i
i i i
θ θ θ θ θ
θ θ θ θ θ
Ω− Ω Ω+ Ω (
(
= − Ω− Ω − Ω+ Ω
(
( Ω − Ω
¸ ¸
R
(57)
( )
ˆ ˆ
ˆ r, ,h θ and ( )
1 2 3
ˆ ˆ ˆ , , c c c are also related through a rotation matrix
A
ɶ
R formally identical to
A
R
(cf. Eq. (42)), with
I
ξ and
I
ζ in place of ξ and ζ , respectively. The fact that
A B

ɶ
R R leads to
the following relations:
( ) arccos cos cos
I
i ζ φ = (58)
cos sin cos sin sin cos sin sin sin cos
sin cos
sin sin
I I I I I I I I
i i
ζ φ ξ ζ ξ ζ φ ξ ζ ξ − + +
Ω= Ω = (59)

sin cos sin
sin cos
sin sin
I
i i
ζ φ φ
θ θ = = (60)
and finally to obtaining f ω θ = − . Due to Eqs. (14), (15), and (59) the proper choice of the re-
lease time allows setting
I
ξ to the value that corresponds to the desired RAAN
d
Ω .
19
REFERENCES
[1] A. Miele, "Multiple-Subarc Gradient-Restoration Algorithm, Part 1: Algorithm Structure." Journal of Optimiza-
tion Theory and Applications, Vol. 116, No. 1, pp. 1-17, 2003
[2] A. Miele, "Multiple-Subarc Gradient-Restoration Algorithm, Part 1: Application to a Multistage Launch Vehicle
Design." Journal of Optimization Theory and Applications, Vol. 116, No. 1, pp. 19-39, 2003
[3] K. R. Brown, E. F. Harrold, and G. W. Johnson, "Rapid Optimization of Multiple-Burn Rocket Flights." NASA
CR-1430, 1969
[4] C. R. Hargraves and S. W. Paris, "Direct Trajectory Optimization Using Nonlinear Programming and Colloca-
tion." Journal of Guidance, Control and Dynamics, Vol. 10, No. 4, pp. 338-342, 1987
[5] A. L. Herman and B. A. Conway, "Direct Optimization Using Collocation Based on High-Order Gauss-Lobatto
Quadrature Rules," Journal of Guidance, Control and Dynamics, Vol. 19, No. 3, pp. 592-599, 1996
[6] P. J. Enright and B. A. Conway, "Discrete Approximations to Optimal Trajectories Using Direct Transcription
and Nonlinear Programming," Journal of Guidance, Control and Dynamics, Vol. 15, No. 4, pp. 994-1002, 1992
[7] J. T. Betts, "Optimal Interplanetary Orbit Transfers by Direct Transcription," Journal of the Astronautical Sci-
ences, Vol. 42, No. 3, pp. 247-326, 1994
[8] H. Seywald, "Trajectory Optimization Based on Differential Inclusion," Journal of Guidance, Control and Dy-
namics, Vol. 17, No. 3, pp. 480-487, 1994
[9] V. Coverstone-Carroll and S. N. Williams, "Optimal Low Thrust Trajectories Using Differential Inclusion Con-
cepts," Journal of the Astronautical Sciences, Vol. 42, No. 4, pp. 379-393, 1994
[10] P. F. Gath and A. J. Calise, "Optimization of Launch Vehicle Ascent Trajectories with Path Constraints and Coast
Arcs." Journal of Guidance, Control and Dynamics, Vol. 24, No. 2, pp. 296-304, 2001
[11] P. Lu and B. Pan, "Trajectory optimization and guidance for an advanced launch system." 30
th
Aerospace Sci-
ences Meeting and Exhibit, Reno, NV, 1992; paper AIAA 1992-732
[12] M. Sirinian, "La Traiettoria di lancio del San Marco II." Atti del Centro Ricerche Aerospaziali, Università di
Roma, Scuola di Ingegneria Aerospaziale, 1967
[13] A. E. Bryson and Y. C. Ho, Applied Optimal Control, Hemisphere, New York, NY, 1975, pp. 71-89
[14] D. Hull, Optimal Control Theory for Applications, Springer, New York, NY, 2003, pp. 247-257
[15] M. Pontani and B. A. Conway, "Particle Swarm Optimization Applied to Space Trajectories." Journal of Guid-
ance, Control, and Dynamics, Vol. 33, No. 5, 2010, pp. 1429-1441
[16] X. Hu, Y. Shi, and R. Eberhart, "Recent Advances in Particle Swarm." Proceedings of the IEEE Congress on
Evolutionary Computation (CEC 2004), Portland, OR, 2004, pp. 90-97
[17] Z. Michalewicz and M. Schoenauer, "Evolutionary Algorithms for Constrained Parameter Optimization Prob-
lems." Evolutionary Computation, Vol. 4, No. 1, 1996, pp. 1-32
[18] J. E. Prussing and B. A. Conway, Orbital Mechanics, Oxford University Press, New York, NY, 1993, pp. 46-54

which terminates at injection of the payload in the desired operational orbit. It has specified structural. at which the flight path angle is constrained to the value found at step (1). The Euler-Lagrange equations and the Pontryagin minimum principle are employed to express the control as a function of the adjoint variables conjugate to the dynamics equations. PROBLEM DEFINITION This research addresses the problem of performance evaluation of multistage rockets through the determination of a near optimal ascending trajectory. With regard to the third subrocket mU ) is the (actual) payload mass. and aerodynamic characteristics. in the context of a three-degree-of-freedom problem. The method that is being presented requires a reduced deal of effort in programming. including all the three stages. the trajectories yielded through the technique at hand resemble (qualitatively) the ones found by Miele2 and those of existing rockets as the Scout12. as existing routines are used. Rocket characteristics The three-stage rocket that is being considered is the Muralm missile. the native MATLAB routine fmincon is employed for finding the thrust direction that maximizes the velocity at the stage separation.e. This mass m0 ) is composed of a structural mass mS ) . and therefore is represented by the third stage only. ( ( ( mU ) ≡ m0 ) ).trajectory is composed of the following thrust phases and coast arcs: (a) first stage propulsion (b) second stage propulsion (c) coast arc (after the second stage separation) (d) third stage thrust phase In particular. the methodology presented and applied in this paper is based on three steps: (1) the flight path angle at each stage separation is determined through an iterative process. inclination id . For the first two ( subrockets mUi ) ( i = 1. which is to be maximized. The mass distribution for the Muralm three-stage rocket is the following: i i +1 3 2 .e. and testing the algorithmic codes. the particle swarm algorithm. For the sake of simplicity. Hence. 2 ) coincides with the initial mass of the subsequent subrocket (i. The three-stage launch vehicle is modeled as a point mass. the mass distribution of the launch vehicle can be described in terms of masses of subrockets: subrocket 1 is the entire rocket. subrocket 3 is the launch vehicle after separation of the first two stages. the existence and duration of a coast arc and the optimal thrust direction are determined through a heuristic technique. in conjunction with analytical developments and a simple implementation of swarming algorithm. ( ( ( ( ( ( a propellant mass mPi ) . debugging. (2) for each stage. Let (i (i (i m0 ) denote the initial mass of subrocket i. This is a circular orbit of radius Rd . and a payload mass mUi ) ( m0i ) = mSi ) + mPi ) + mUi ) ). Nevertheless. subrocket 2 is the launch vehicle after the first stage separation. (3) for the third stage. i. propulsive. the methodology treated in this paper is intended to: (i) yield a reasonable solution for performance evaluation of multistage rockets and (ii) represent a technique for generating a suitable first-attempt guess trajectory to be employed by more refined algorithms tailored to optimizing the overall trajectory. and right ascension of the ascending node (RAAN) Ω d .

S ( ) is the reference surface area for subrocket i.( m0 ) = 3608 kg 1 ( mS ) = 306 kg 1 ( mP ) = 2480 kg 1 ( mU ) = 822 kg 1 m0 = 822 kg ( 2) mS = 46 kg ( 2) mP = 370 kg ( 2) mU = 406 kg ( 2) (1) The initial mass of the third stage is specified and this implies that minimizing the third stage propellant expenditure implies maximizing the payload mass that can be injected in the desired orbit. Both these quantities are assumed as time-independent for all of the stages. 2. flight path angle γ . (geographical) longitude ξ . due to the specified mass distribution. 2 ) are given by () ∆t BO = i ( ( g 0 I SP) mP ) i i ( ( n0 ) m0 ) i i (4) The aerodynamics of the Muralm rocket has been modeled through the Missile DATCOM database. latitude φ . M ) S ( ) ρ A v 2 2 and 1 i D = cD ( α . i Equations of motion The equations of motion that govern the rocket dynamics can be conveniently written in terms of its radius r. If the superscript “ ( ) ” ( j = 1. respectively. and v is the relative velocity (cf. the following propulsive data characterize the Muralm rocket: j T ( ) = 235180 N 1 T ( ) = 23472 N 2 T ( ) = 7500 N 3 ( I SP) = 290 sec 1 ( I SP) = 291 sec 2 ( I SP) = 320 sec 3 (2) For each subrocket i it is then straightforward to derive the thrust acceleration as a function of the time t: ( n0 ) T  =   ( (i)  m i 1 − n0i ) t − tign i ( ) (g I( ) ) i 0 SP (3) ( () where n0 ) is the thrust acceleration at ignition of stage i (which occurs at the time tign ) and g 0 is i i the gravitational acceleration at sea level. velocity v.e. The propulsive characteristics of the launch vehicle can be described in terms of thrust magnitude T and specific impulse I SP . ρ A is the (local) atmospheric density. and heading angle ζ . the (i) burnout times ∆t BO for the first two subrockets ( i = 1.3) . For the Muralm launch j vehicle S ( ) = 0. These variables refer to the relative motion in an Earth-centered rotating frame (ECEF) and are described in more detail in Appendix 1. Lastly.636 m 2 ( j = 1. They form the state vector 3 . lift L and drag D. M ) S ( ) ρ A v 2 2 (5) where the lift and drag coefficients cL and cD depend on M and α . Mach number and angle of attack. i. The aerodynamic force is assumed to be composed of two terms. 2. 1 i L = cL ( α . Appendix 1).3) refers to stage j.

φL . and (geographi- h0 = 6. χ L . (5). As the thrust vector is assumed to be coplanar with the posii 0 SP tion vector r and the velocity vector v.292115 ⋅ 10−5 sec-1 ) are the Earth gravitational parameter and rotation rate. γ 0 .165 km ) is the Earth radius.3 km3 sec 2 ) and ωE ( = 7. µ L represent respectively the latitude. Since the Muralm is assumed to be air-launched from a Tornado fighter. next section). and v0 are the rocket altitude. the angle α T suffices to define its direction. flight path angle and (relative) velocity at release from the carrier aircraft.xR ≜ [ r ξ φ γ i perscript “ ( ) ”. which is taken counterclockwise from v. the initial conditions for Eqs. respectively.6 deg φL = 5.280 km sec (13) The azimuth at release χ L is determined in relation with the desired orbit inclination (cf.7 km µ L = −52.2 deg γ 0 = 35. Omitting the suɺ r = v sin γ ɺ ξ= v cos γ cos ζ r cos φ v cos γ sin ζ r (6) (7) ɺ φ= (8) γɺ =   ω2r T sin α T  v µ E  L +  − 2  cos γ + + cos φ  2ωE cos ζ + E ( cos φ cos γ + sin φ sin γ sin ζ )  m v mv v r r v   ɺ v= v r T cos α T µ E D 2 − 2 sin γ − + ωE r cos φ ( cos φ sin γ − sin φ cos γ sin ζ ) m v r m 2 ωE r sin φ cos φ cos ζ − 2ωE sin φ v cos γ (9) (10) ζɺ = − tan φ cos γ cos ζ + 2ω E cos φ tan γ sin ζ − (11) where µ E ( = 398604. for each subrocket the equations of motion are ( v ζ] T ) of the launch vehicle (in rotating coordinates). which occur at time t1 (first stage separation) and t2 (second stage separation). (6)-(11) are the following: r ( 0 ) = RE + h0 ξ ( 0) = µL φ ( 0 ) = φL γ (0) = γ 0 v ( 0 ) = v0 ζ (0) = π 2 − χL (12) cal) longitude at release. For the problem at hand the following data are assumed: where h0 . and the time-varying mass ( () is m = m0 ) − T ( ) t − tign i i i ( ) ( g I ( ) ) . L and D are given by Eq. RE ( = 6378. azimuth. The state x R is continuous across stage separations.0 deg v0 = 0. 4 .

is in the range [id − ∆i. inertial flight path angle γ I . absolute longitude ξ I . The orbit inclination is selected through a simultaneous iterative process on the initial azimuth at rocket release. and therefore they are not considered. However. γ ( t2 ) . for the upper stage aerodynamic forces are negligible. In fact. γ ( t2 ) . and inertial heading angle ζ I . Convergence is declared when inclination given values of γ ( t1 ) . the MATLAB routine fmincon determines the angle α ( t ) that maximizes v at each stage separation. Moreover. These circumstances imply that the trajectory of the last stage is planar. tL is directly related to the Greenwich sidereal time at release θ gL . At each iteration of the process (for Finally. the flight path angle at the second stage separation. θ gL = θ g + ωE ( t L − t ) (14) where θ g is the (known) Greenwich absolute longitude at a reference time t . π 2] the orbit inclination increases as χ L decreases (for χ L ≤ π 2 ). is ad- altitude of the desired circular orbit. latitude φ . once the rocket final state (at orbit injection) is known. In fact. This process is relatively straightforward. and χ L ). The orbital plane selection is assumed to occur during the powered motion of the first two subrockets. The control is represented by the thrust direction. hd ] . id + ∆i ] . the correct selection of the release time tL leads to obtaining the desired right ascension of the ascending node (RAAN). inertial velocity vI . Then x I ( t2 ) is employed to At separation of the second stage. denoted with x I ( t2 ) and including the following components: radius r. where ∆i is set to 3 deg . The first-attempt value for γ ( t2 ) is chosen to be 25 deg. at which the flight path angle is constrained to a specified value. and ∆h is set to 10 km . specifying the RAAN leads to obtaining θ gL and then tL : tL = t + θ gL − θ g + 2kπ ωE (k ∈ Z) (15) ported in Appendix 2 x R ( t2 ) is employed to derive the inertial state. defined by α T . justed iteratively up to obtaining an apogee altitude in the range [ hd − ∆h. these two angles coincide: α T ≡ α . as the thrust direction is always aligned with the rocket longitudinal axis. α . Ω d . and by the aerodynamic angle of attack.METHOD OF SOLUTION This work employs two distinct approaches for determining the trajectory of the first two stages and that of the last stage. the state is denoted with x R ( t2 ) . where hd is the While γ ( t1 ) is set to 30 deg. Through the relations re- 5 . Trajectory of the first two stages The determination of the trajectory of the first two stages is based on maximizing the final velocity at each stage separation. in the range [0. The RAAN is given by the relations reported in Appendix 2 and.

which is equivalent to minimizing the thrust duration. Appendix 2). the objective function for the third stage trajectory optimization is J = t f − t3 (21) 6 . The orbital elements a2 . taken counterclockwise from the direction of the inertial velocity vector v I . This means that the final conditions at orbit injection (which occurs at t f ) are r ( t f ) = Rd vI ( t f ) = µE Rd γ I (t f ) = 0 (20) The trajectory for the last stage is obtained by minimizing fuel consumption. In fact. i. then f 3 ≜ f ( t3 ) = f ( t2 ) + ∆f . As a result the radius. inclination i. inertial velocity and inertial flight path angle at t3 are simply given by r ( t3 ) = 2 a2 (1 − e2 ) 1 + e2 cos f 3 (16) γ I ( t3 ) = arctan  vI ( t3 ) =  e2 sin f3   1 + e2 cos f 3  2 1 + e2 + 2e2 cos f3 (17) 2 a2 (1 − e2 ) µE (18) In the powered phase. Hence.derive the orbital elements at t2 . RAAN Ω .e. eccentricity e2 . This circumstance implies that the orbit inclination and RAAN do not change during the last stage propulsion. argument of perigee ω2 . denoted with ∆f . and ω2 do not vary along a coast arc. It is relatively straightforward to prove that during the coast arc the true anomaly variation. the third stage motion can be conveniently described through the use of the equations of motion for r. and true anomaly f ( t2 ) (cf. The operational orbit is assumed to a be circular orbit of radius Rd . vI . and γ I (also referred to as the state equations hence forward): ɺ r = vI sin γ I ɺ vI = T cos α I µ E − 2 sin γ I m vI r γɺI = µ T sin α I  vI +  − 2E m vI  r r vI   cos γ I  (19) The thrust direction is now specified through the thrust angle α I . e2 . the orbit semi-major axis (SMA) a2 . the last stage trajectory is assumed to be planar. suffices to describe the rocket dynamics. The third stage trajectory is assumed to be composed of two phases: a coast arc and a thrust phase. Trajectory of the last stage As remarked previously. if t3 represents the ignition time of the third state.

In summary. to obtain the necessary conditions for an optimal   solution a Hamiltonian H and a function of the boundary conditions Φ are introduced as x2 T T Letting x ≜ [ x1  T sin α I  x2   T cos α I µ E  µ  H ≜ λ 1x2 sin x3 + λ 2  − 2 sin x3  + λ 3  +  − 2 E  cos x3  x1  m x2   x1 x1 x2   m x2  2    a2 (1 − e2 )  µE 2  + υ2  x20 − 1 + e2 + 2e2 cos f3  Φ ≜ ( t f − t3 ) + υ1  x10 − 2 1 + e2 cos f 3    a2 (1 − e2 )        µE  e sin f3  + υ3  x30 − arctan 2  + υ6 x3 f  + υ 4  x1 f − Rd  + υ5  x2 f −   1 + e2 cos f3  Rd      (23) (24) where υ ≜ [υ1 υ2 υ3 υ4 υ5 υ6 ] ( xk 0 = xk ( t3 ) and T ) xkf = xk ( t f ) ( k = 1. The necessary conditions for optimality13 yield the following adjoint equations for the costate λ ≜ [ λ1 λ 2 λ 3] ɺ λ1 = ( x2 λ 3 cos x3 ) ( ): cos x3  1 1  −  2 µ E λ 2 sin x3 + 2 µ E λ 3  2 x1  x2  x13    1 µ  T + 2 E 2  − P 2 sin α I   x1 x1 x2  mP x2  (25) ɺ λ 2 = −λ1 sin x3 − λ 3 cos x3  (26) ɺ λ 3 = − x2 λ1 cos x3 + µ E λ 2 x cos x3 µ  + λ 3 sin x3  2 − 2 E  2 x1  x1 x1 x2  (27) in conjunction with the respective boundary conditions: λk 0 = −υk and λkf = υk + 3 ( k = 1. 2. λ ≜ [ λ1 λ 2 λ 3] ( T ) T and represent respectively the adjoint variable conjugate to the dynamics equations (19) and to the boundary conditions (16)-(18) and (20).2. The ignition time t3 is then calculated through the Kepler’s law: t3 = t2 + µE 3 a2 {E ( t ) − E ( t ) − e sin  E ( t ) + e sin  E ( t )}     3 2 2 3 2 2 (22) where E ( tk ) is the eccentric anomaly associated with f ( tk ) ( k = 2.3) .3) .3) (28) In the presence of initial conditions depending on a parameter ( f 3 ) a pair of additional necessary conditions must hold14: 7 . x3 ] = r vI γ I  . the optimization problem for the upper stage is the following: determine the optimal control law α I ( t ) and the optimal true anomaly f 3 such that J is minimized.

f 3 . and λ 3 ( 0 ) = kλ λ 3* ( 0 ) ( kλ > 0 ) (where the superscript “*” denotes the actual optimal value of a variable). what will be demonstrated is that for the problem at hand the transversality condition can be neglected by the swarming algorithm. and t f . λ 2 ( 0 ) = kλ λ 2* ( 0 ) . (25)-(32). This circumstance implies that if an optimization algorithm is capable of finding some initial value of λ such that λ1 ( 0 ) = kλ λ1* ( 0 ) . the optimal control α I* can be expressed as a function of the costates through the Pontryagin minimum principle: 2  λ 3  λ 3  * * 2 α I = arg min H ⇒ sin α I = −   + λ 2  αI x2  x2     − 1 2  λ  2  * 2 and cos α I = −λ 2  3  + λ 2   x2     − 1 2 (31) Lastly. as the final time is unspecified. the control α I can be written as a function of λ through Eq. Equality constraints reduce the search space where feasible solutions can be located. λ 20 . one has first to recognize the special structure of the costate equations (25)-(27). then the same proportionality holds between λ and λ* at any t. with unknowns represented by the initial values of λ . which are homogeneous in λ . (25)-(27). and f 3 : λ 30 = λ 20 2 sin f 3 1 + e2 + 2e2 cos f 3 e2 + cos f3 − λ 10 2 2 a2 (1 − e2 ) 1 + e2 + 2e2 cos f 3 e2 + cos f 3 (1 + e2 cos f 3 ) 2 (30) In addition. (31) and one can recognize that α I coincides with α I* : 2  λ 3  λ 3  2   + λ 2  sin α I = − x2  x2     − 1 2 2  kλ λ  kλ λ 3*  * 2  =−  + ( kλ λ 2 )  x2  x2     * 3 − 1 2 ≡ sin α I* (33)  λ  2  2 cos α I = −λ 2  3  + λ 2   x2     − 1 2 * 2   * k λ  * 2 = − kλ λ 2   λ 3  + ( kλ λ 2 )   x2     − 1 2 ≡ cos α I* (34) 8 . However. the following transversality condition must hold: ∂Φ =0 H (t f ) + ∂t f ⇒ ( 1 − n0 ) ( t f − t3 ) 3 ( n0 ) 3 (g I( ) ) 3 0 SP  λ 3f  2     + λ 2 f  − 1 = 0  x2 f    2 (32) The necessary conditions for optimality allow translating the optimal control problem into a twopoint boundary-value problem involving Eqs. Moreover. To do this. due to homogeneity of Eqs.∂Φ =0 ∂f3 and ∂ 2Φ ≥0 ∂f32 (29) The first equation yields a relation that expresses λ 30 as a function of λ 10 .

f3 . together with the adjoint equations (25)-(27). f3 . in order to reduce the number of equality constraints considered by the swarming algorithm. In this research the optimization problem involves a continuous time-dependent control variable and is translated into a parameter optimization problem through the necessary conditions for optimality. whereas the velocity vector. (31) (b) the value of λ 30 is calculated by means of Eq. the transversality condition is violated. for a generic dynamical system. which allow expressing the control variable as a function of the adjoint variables conjugate to the dynamics equations. (c) the state equations (19) are integrated numerically up to the final time t f . λ 20.of H ( t * ) . with the intent of improving its performance. The respective initial conditions are known once the parameters f 3 (for the state equations) and {λ10. The remaining parameter λ 30 can be easily obtained by means of the relation (30). where the population is represented by a swarm of N particles. The position vector includes the values of the n unknown parameters of the problem. λ 20. turns out to be f H (t * f This circumstance implies that if λ is proportional to λ* then the final conditions are fulfilled at the minimum final time t * . Thus.…. N ) is associated with a position vector χ ( i ) and T with a velocity vector w ( i ) . t f } . this condition is discarded. The PSO technique is a population-based method. after picking the unknown values of the remaining Lagrange multipliers at the initial time ( λ10 and λ 20 ) . Specifically. t f } . due to Eq. the optimal control α I* can be determined without considering the transversality condition. (30). provided that the proportionality condition holds. λ 30} (for the adjoint equations) are specified (d) the inequality in Eq. Unconstrained parameter optimization problems can be stated as follows: determine the optimal values of the n unknown parameters {χ1 . (32). (20)). the swarming technique is based on the following points: (a) the control variable is expressed as a function of the costate through Eq. χ ( i ) ≜  χ1 ( i ) … χ n ( i )  . the following parameter set can be employed in the solution process: {λ10. that lead the dynamical system to satisfying three conditions (Eqs. because the value f )=− ( * 1 − n0 ) ( t * − t3 ) f 3 ( n0 ) 3 (g I( ) ) 3 0 SP  λ 3f   x2 f   2  +  λ 2 f  = − kλ ≠ −1     2 (35) Therefore. and the true anomaly f 3 In short. (29) (not expanded for the sake of brevity) and the conditions at injection (20) are evaluated In summary.…. In contrast. χ n } such that the objective function J is minimized. {λ10. Outline of the particle swarm optimization method The particle swarm optimization (PSO) method is a heuristic technique aimed at finding the optimal values of a set of unknown parameters. Each particle i ( i = 1. whose components are   9 . the problem reduces to the determination of four unknown parameters. λ 20. which in fact is ignorable in this context.

(29). and finally the third term is the social component. Equality constraints narrow considerably the search space where feasible solutions can be located. the PSO assumes as objective the expression (37) with J ≡ 0 . A fairly large number of iterations are set to ensure that the final solution is stable enough to be considered optimal. The most popular approach for dealing with these constraints consists in penalizing ɶ J = J + ∑η r lr2 ( χ ) r =1 m them by adjoining additional terms to the objective function: (37) This approach is employed also in this research. In fact. directed toward the global best position. determines the position update. the best position yet located by any particle in the swarm. the best position experienced by the particle. methods that distinguish between feasible and infeasible solutions. Both the position and the velocity components are constrained to suitable ranges: ak ≤ χ k ≤ bk and − d k ≤ wk ≤ d k ( k = 1.3) . The expressions for position and velocity update determine the swarm evolution toward the location of the globally optimal position. Further options exist for dealing with constraints (e. the second relation reported in Eq.e. For each particle the simplest way of treating inequality constraints consists in assigning a fictitious infinite value to the objective function if the particle violates at least one of them.. The position vector of the best particle is expected to contain the optimal values of the unknown parameters. Each particle represents a possible solution to the problem. directed toward the personal best position. n ) .…. i. which correspond to the global minimum of J. A detailed description of the algorithm (with also the same settings adopted in this work) is contained in Reference 15.e. 2. This is due to the fact that (non-redundant) equality constraints actually reduce the degree of freedom of the problem according to their number. defined by Eq. However. in the presence of n unknown parameters. As the necessary conditions for optimality are used to express the control variable. Inequality constraints are less problematic because they reduce the search space of feasible solutions without decreasing the degree of freedom of the problem. The equality constraints are represented by the relations (20). In this work a single inequality constraint exists. The initial population is randomly generated by introducing N particles. regarding (directly or indirectly) the unknown parameters. (36).denoted with wk ( i ) ( k = 1. Therefore. This formula includes three terms with stochastic weights: the first term is the so called inertial component and for each particle is proportional to its velocity in the preceding iteration. with η r = 1 ( r = 1. The central idea underlying the method is contained in the formula for velocity updating. whose positions and velocities are (stochastically) uniformly distributed in the respective search spaces. and corresponds to a specific value of the objective function. m ) .….…. the simple strategy described previously for equality and inequality constraints will be shown to be quite effective for the problem considered in this paper. i. methods based on preserving feasibility of solutions. n ) (36) where d k ≜ bk − ak . which corresponds to the globally optimal solution to the problem of interest. 10 . References 16 and 17). i. Constrained optimization problems involve equalities and/or inequalities. m equality constraints reduce the degree of freedom by m. The algorithm terminates when the maximum number of iterations N IT is reached. the second component is termed the cognitive component.g. at most m = n equality constraints are admissible ( m ≤ n ) : lr ( χ ) = 0 ( r = 1. cf.e.

11 .377 ⋅ 10−7 ∆tCO = 171. which has a nonnegligible component directed toward East. the swarming optimizer is employed with the following settings: N = 50 (number of particles) and N IT = 500 (number of iterations). It is worth noticing that the constraint reduction allows arbitrarily defining the search space for the initial values of the Lagrange multipliers.8 sec . inclination i f . Case 1: The main results (azimuth at release χ L .3) . flight path angle at the second stage separation χ L = 0. and 0.0 deg e f = 7.2 TU .004 km Case 2: χ L = 40. 3.2 sec i f = 5. Moreover. velocity and flight path angle) and the ground tracks. and 5 portray some state components (radius. 2. Figures 1.7 deg m f = 133.2 deg id = 45 deg id = 90 deg γ ( t2 ) .2 deg m f = 140. 4. orbit semi-major axis a f .4 kg (38) a f = RE + 280. The optimal values of the unknown parameters are sought in the following ranges: −1 ≤ λk 0 ≤ 1 ( k = 1. an increasing amount of fuel is employed to correct the initial inertial velocity.165 km and 1 TU = 806. 1 DU = 6378. eccentricity e f .0 deg γ ( t2 ) = 20. This circumstance implies that as the operational orbit inclination increases. the final mass decreases as the orbit inclination increases.002 km Case 3: χ L = −4.7 deg m f = 120.ASCENDING TRAJECTORIES The ascending trajectories are determined by employing canonical units: the distance unit (DU) is the Earth radius.01 TU ≤ t f ≤ 0.0 deg γ ( t2 ) = 20. Three test cases have been considered: Case 1: Case 2: Case 3: hd = 280 km hd = 280 km hd = 280 km and and and id = 5.6 kg (40) a f = RE + 279. whereas Figures 2. and only a posteriori their correct values (fulfilling also the transversality condition (32)) can be recovered. This is a predictable result and is related to the launch site. (38)-(40) it is apparent that the final conditions at injection are fulfilled in a satisfactory way. With regard to the last stage trajectory.0 deg e f = 4. and final mass at injection m f ) are reported in the following.260 ⋅ 10−7 ∆tCO = 175. coast duration ∆tCO . 0 ≤ ∆f ≤ π . which is located in the proximity of the Earth equator.5 deg e f = 4.0 deg γ ( t2 ) = 21.403 ⋅ 10−7 ∆tCO = 158.994 km From inspecting the results summarized in Eqs. and 6 illustrate the corresponding thrust pointing angle time histories for each of the three subrockets.2 sec i f = 87. Thus.3 sec i f = 45. This means that they can be sought in the interval −1 ≤ λk 0 ≤ 1 by the PSO algorithm.7 kg (39) a f = RE + 280. whereas the time unit (TU) is such that µ E = 1 DU 3 TU 2 .

Figure 1. Case 1: altitude. Case 1: control for the first two stages (a) and for the last stage (b) 12 . ground track Figure 2. inertial velocity and flight path angle.

Case 2: control for the first two stages (a) and for the last stage (b) 13 .Figure 3. inertial velocity and flight path angle. Case 2: altitude. ground track Figure 4.

ground track Figure 6. Case 3: control for the first two stages (a) and for the last stage (b) 14 . Case 3: altitude. inertial velocity and flight path angle.Figure 5.

A particular configuration of the launch vehicle is assumed. The Earth-centered Earth-fixed (ECEF) reference frame represents a reference system that is rigidly attached to the Earth. An additional useful reference frame is represented by the right-handed Cartesian coordinate ˆ ˆ. The remaining orbit elements of the operational orbit are attained by choosing a suitable control law for the thrust direction time history. The unit vector ιˆ intersects the Greenwich meridian at all times.e. i. which form a right-handed. unlike the position vector. it is evident that r ≡ rI . ˆ ˆ ˆ system r E . This work proposes and successfully applies a simple technique for generating near optimal three-dimensional ascending trajectories for multistage rockets. where r is the unit vector aligned with r. Only existing routines and a simple implementation of swarming algorithm are employed. and by selecting the optimal coast duration at the second stage separation. ˆ. c3 ) . the velocity vector in rotating coordinates. v . non-inertial. REFERENCE FRAMES ˆ ward. where t denotes a generic time instant. This study proves how the desired orbit right ascension of the ascending node and inclination can be obtained by properly selecting the launch time and the azimuth angle at release. applied to the upper stage trajectory. the inertial velocity v I is related to the (relative) velocity v through the following expression: v I = v + ωE × r (41) This means that. The solution found in this work appears suitable also for being employed as a guess for more refined optimization algorithms. for the purpose of performance evaluation. As the reference Greenwich meridian rotates with rotation rate ωE . In contrast. v I . The position vector of the orbiting spacecraft in the ECEF-frame is denoted with r. The unit vector c ≡ k is aligned with the planet axis of rotation and is positive north3 The rocket motion is conveniently described in rotating. its angular position (with respect to the ECEI-frame) is identified by its absolute longitude (usually termed Greenwich sidereal time) θ g ( t ) = θ g ( t ) + ω E ( t − t ) . respectively. taking into account realistic propulsive. in conjunction with the analytical necessary conditions for optimality. does not coincide with the inertial velocity vector. whereas the subscript “I ” corresponds to a quantity in the ECEI-frame. O.CONCLUSION The generation of an optimal trajectory for multistage launch vehicles is a challenging problem. c2 . structural. As the ECEF-frame and the ECEIframe have a common origin. N . reference frames. and aerodynamic data. and therefore ωE k ( ≜ ωE ) represents the (vector) rotation rate of the ECEF-frame with ˆ ˆ ˆ respect to an inertial Earth-centered (ECEI) reference frame. APPENDIX 1. treated with different approaches in the past. E is directed along the local East ( ) 15 . denoted with ( c1 . Under the assumption that the planet rotates with the constant angular rate ωE this circumstance implies that also the reference frame rotates with angular rate ωE with respect to the ECEI-frame. time-dependent sequence of ( ) unit vectors. c1 is the verˆ ˆ nal axis. k . The origin ˆ of both the inertial and the rotating coordinate system is the center of the planet. The j ˆ ECEF-frame is associated with ιˆ.

v. E. n.v.θ .h -frame as the sum of two terms. and related angles ( ) ( ) 16 . N from ιˆ.e.h can be ˆ ˆ ˆ written as a function of ιˆ. ( ) ( ) ˆ ˆ ˆ ˆ A third useful rotating frame is r . N ) through a counter- ˆ ˆ ˆ clockwise rotation about axis 1 by the heading angle ζ (cf. lift L and drag D: ˆ ˆ A = Ln − Dv ( ) (44) ˆ ˆ ˆ ˆ ˆ ˆ Figure 7.h through a clockwise rotation about axis 3 by the flight path angle γ (Fig.h -frame. r .θ . where θ is aligned with the projection of the velocity v onto the horizontal plane ( ) ˆ ˆ ˆ ˆ ( E. ˆ.θ .h .h (with v aligned with v) is obtained from r . ˆ. k :  r θ h  = R A ιˆ j ˆ    cos φ cos ξ   − sin ζ sin φ cos ξ − cos ζ sin ξ RA =   − cos ζ sin φ cos ξ + sin ζ sin ξ  ( ) ( ) T ˆ k  . where j ˆ  sin φ  sin ζ cos φ   cos ζ cos φ   T cos φ sin ξ − sin ζ sin φ sin ξ − cos ζ cos ξ − cos ζ sin φ sin ξ − sin ζ cos ξ (42) ˆ ˆ ˆ ˆ ˆ ˆ ˆ Finally.ˆ direction. 7 it is apparent that v = v [sin γ ( ) ( ) cos γ cos ζ ˆ ˆ cos γ sin ζ ]  r E  ˆ N  T (43) ˆ ˆ ˆ The overall aerodynamic force A is conveniently written in the n. 7). i.v. E. ˆ ˆ r E .N ) . and N is aligned with the local North direction. the second (clockwise by the latitude angle φ ) about axis 2. N -frame. It is straightforward to obtain ˆ. the reference frame n. Fig. r. From inspecting Fig. This frame is obtained from ( rˆ. k through two elementary rotations: the first (counterclockwise by the geoj ˆ graphical longitude angle ξ ) about axis 3. 7). As a result.

Ω. f ) from the inertial state vari- With regard to step (1). φI . v. γ I . ξ I . ˆ. φ . k implies j ˆ that rI = r ( ) ξI = ξ + θg φI = φ (45) ˆ. Eq. STATE VARIABLES AND ORBITAL ELEMENTS The derivation of the orbital elements ( r . i. (46) and to the fact that ωE × r = ωE r cos φ E . ξ I . γ . ω . f ) from the relative state variables follows two steps: (1) derivation of the inertial state variables ( rI . ζ ) ( a. γ . γ I . the conservation of energy yields a: 17 . ξ . γ I . and f can be easily calculated from r.φ . ˆ ˆ The velocity vectors v I and v have the following expressions in the rotating frame r E . (41) yields three relations: vI sin γ I = v sin γ vI cos γ I cos ζ I = v cos γ cos ζ + ωE r cos φ vI cos γ I sin ζ I = v cos γ sin ζ which lead to the following expressions of γ I . and ζ I as functions of γ . ζ ) (2) derivation of the orbital elements ables ( rI . the in-plane orbital elements a. vI . ζ I ) from the relative state variables ( r . and vI . e. vI . ω . the fact that r ≡ rI and the definition of the reference ιˆ. and ζ : vI = v 2 + (ωE r cos φ ) + 2vω E r cos φ cos γ cos ζ 2 (47) (48) (49) (50) γ I = arcsin  v cos γ cos ζ + ωE r cos φ vI cos γ I  v sin γ    vI  and sin ζ I = v cos γ sin ζ vI cos γ I (51) cos ζ I = (52) With regard to step (2). ξ . vI . i. In fact. φI . v . e. N : ( )  sin γ I   cos γ cos ζ  v I = vI  I I   cos γ I sin ζ I    T ˆ r  ˆ E N  ˆ   and  sin γ  v = v  cos γ cos ζ     cos γ sin ζ    T ˆ r   ˆ E N  ˆ   (46) ˆ Due to Eq. Ω. ζ I ) ( a. v. e.APPENDIX 2.

θ .a= 2µ E − rvI2 µE r (53) Then. (15). Hence. Ω .h ) (57) ɶ ˆ ˆ ˆ and ( c1 . c ) . Due to Eqs. with ξ I and ζ I in place of ξ and ζ . respectively. (42)). c2 . (14). written in terms of the angles i. 18 . the angular momentum magnitude equals rvI cos γ I and can also be written in terms of orbital elements as µ E a (1 − e2 ) . Eq.θˆ . c3 ) are also related through a rotation matrix R A formally identical to R A ɶ (cf. the orbit eccentricity is given by e= ( rv cos γ I ) 1− I µE a 2 (54) The true anomaly f can be found by considering the polar equation of the ellipse. where18: T cos θ sin Ω + sin θ cos i cos Ω sin θ sin i   cos θ cos Ω − sin θ cos i sin Ω  − sin θ cos Ω − cosθ cos i sin Ω − sin θ sin Ω + cos θ cos i cos Ω cos θ sin i  RB =    sin i sin Ω − sin i cos Ω cos i    ˆ ˆ ( r . The fact that R A ≡ R B leads to the following relations: i = arccos ( cos ζ I cos φ ) sin Ω = − cos ζ I sin φ cos ξ I + sin ζ I sin ξ I sin i sin θ = sin φ sin i cos Ω = cos ζ I sin φ sin ξ I + sin ζ I cos ξ I sin i sin ζ I cos φ sin i (58) (59) cos θ = (60) and finally to obtaining ω = θ − f . and (59) the proper choice of the release time allows setting ξ I to the value that corresponds to the desired RAAN Ω d .h to ( c . r= a (1 − e 2 ) 1 + e cos f → cos f = a (1 − e 2 ) − r re (55) in conjunction with the radial component of velocity: a (1 − e 2 ) µ e sin f = v sin γ → sin f = v sin γ e a (1 − e 2 ) µ (56) The remaining orbital elements (i. Ω . and θ ( ≜ ω + f ) : ( ) 1 2 3  r θ h  = R B [ c1 c2 ˆ ˆ ˆ ˆ ˆ   T ˆ c3 ] . and ω ) can be obtained by considering the rotation matrix ˆ ˆ ˆ ˆ ˆ ˆ that relates r . c .

" Journal of Guidance. Control. 15. Reno. 1975. 42. No. 2. "Optimal Interplanetary Orbit Transfers by Direct Transcription. 3. A. 2003 [2] A. 1996 [6] P. Seywald. Shi. "Trajectory Optimization Based on Differential Inclusion." Journal of the Astronautical Sciences. pp. New York. Herman and B." Atti del Centro Ricerche Aerospaziali. R. "Direct Trajectory Optimization Using Nonlinear Programming and Collocation." Journal of Guidance. NV. pp. pp. Control and Dynamics. 1993. Bryson and Y. Optimal Control Theory for Applications. J. "Particle Swarm Optimization Applied to Space Trajectories. W. "Multiple-Subarc Gradient-Restoration Algorithm. 1994 [9] V. Applied Optimal Control. Conway. Enright and B. "La Traiettoria di lancio del San Marco II. Springer. Università di Roma. "Recent Advances in Particle Swarm. No." Journal of Optimization Theory and Applications. Miele. Vol. Vol. Hull. 338-342. pp. Oxford University Press. 19-39. Coverstone-Carroll and S. pp. Conway. 1987 [5] A. No. Pontani and B. Control and Dynamics. 4. 480-487. paper AIAA 1992-732 [12] M. 2003. Pan. 24. NY. Prussing and B. No. Control and Dynamics." Journal of Guidance. Vol. No. F. No. 592-599. pp. pp. Johnson. 1994 [10] P. T. Ho. A." Proceedings of the IEEE Congress on Evolutionary Computation (CEC 2004). Gath and A. Paris. 46-54 19 . No. 1-17. Orbital Mechanics. Brown. Scuola di Ingegneria Aerospaziale. Vol. 1996. Part 1: Application to a Multistage Launch Vehicle Design. A. R. 33. Williams. 2001 [11] P. pp. 10. 1992. Eberhart.REFERENCES [1] A. pp. Sirinian. "Discrete Approximations to Optimal Trajectories Using Direct Transcription and Nonlinear Programming." NASA CR-1430. 71-89 [14] D." Journal of Guidance. "Optimization of Launch Vehicle Ascent Trajectories with Path Constraints and Coast Arcs. Conway. New York. Hemisphere. pp. Vol. 17. "Direct Optimization Using Collocation Based on High-Order Gauss-Lobatto Quadrature Rules. 247-257 [15] M. Miele. No. L. N. pp. 1994 [8] H. No. 2004." Journal of the Astronautical Sciences." Journal of Optimization Theory and Applications. and Dynamics. 1969 [4] C. "Evolutionary Algorithms for Constrained Parameter Optimization Problems. 2003 [3] K. 1. No. 247-326. Betts. 1992 [7] J. 2010. pp. and G. pp. 116. Lu and B. F. Control and Dynamics. 4. and R. "Optimal Low Thrust Trajectories Using Differential Inclusion Concepts. J. NY." Journal of Guidance. Calise. 5. New York. Y. Harrold. Michalewicz and M. 994-1002. A. Vol. "Multiple-Subarc Gradient-Restoration Algorithm. 4. 379-393. Hargraves and S. W. 1. 116. 1429-1441 [16] X. Vol. C. 3. Schoenauer. E. 1." Journal of Guidance. 3. Vol. 1967 [13] A. 90-97 [17] Z. 42. E. Vol. 4. pp. 296-304." Evolutionary Computation. "Trajectory optimization and guidance for an advanced launch system. 19. E. 1-32 [18] J." 30th Aerospace Sciences Meeting and Exhibit. "Rapid Optimization of Multiple-Burn Rocket Flights. Vol. Conway. pp. Vol. No. Control and Dynamics. Hu. NY. Part 1: Algorithm Structure. OR. Portland.