AOE 5244 - E.M.

Cli

Goddard Problem

Posed by R.H. Goddard, A

Method of Reaching Extreme Altitudes, Smithsonian Inst. Misc. Coll. 71, 1919, reprinted by Am. Rocket Soc., 1946.
8

AOE 5244 - E.M. Cli

Three state variables

r : distance from Earths center v : radial velocity m : rockets mass

One control variable : mass

ow-rate
8

AOE 5244 - E.M. Cli

The dynamical model

r = v D(r, v) c 2 v = m m r m = with [0, max ].
8

AOE 5244 - E.M. Cli

The initial conditions are:

r(0) = Re Earths radius v(0) = 0 start from rest m(0) = M0 initial mass

The (only) specied

end-condition is 1(x(tf )) = m(tf ) Mf
8

AOE 5244 - E.M. Cli

where Mf < M0 is the mass with all fuel expended

The cost functional is

g(r(tf ), v(tf ), m(tf )) = r(tf ) , that is, we maximize the nal altitude.
8

AOE 5244 - E.M. Cli

Nondimensional Form

Its convenient to
nondimenionalize. We select M : M0 unit mass L : Re unit length T :
8

3 Re / unit time

AOE 5244 - E.M. Cli

Note that this leads to /Re : unit speed

2 M0/Re

: unit force

The scaled dynamical model

r = v D(, v ) r c 1 v = 2 m m r m =

AOE 5244 - E.M. Cli

with [0, max ], where, for example, r = r/Re

In the following we drop the

and note that all quantities have been non-dimensionalized.
8

AOE 5244 - E.M. Cli

Applying the M.P.

We form the variational

AOE 5244 - E.M. Cli

10

Hamiltonian H = r v + v c m m m D(r, v) 1 r2
a

The adjoint dierential

8

AOE 5244 - E.M. Cli

11

equations are m r r3 v D v = r + m v v [c D(r, v)] . m = m2 The terminal transversality

8

r = v

1D

AOE 5244 - E.M. Cli

12

conditions imply H(tf ) = 0 r (tf ) = 0 v (tf ) = 0 m(tf ) = 1 Note that H is constant along an extremal path.
8

AOE 5244 - E.M. Cli

13

Applying the M. P. min H

We are to minimize the

variational Hamiltonian H subject to the bounds.

Observe that H can be written

8

AOE 5244 - E.M. Cli

14

as H = v c m + terms independent of m

Use the symbol S for the terms in square brackets

For the mass ow-rate we nd

8

AOE 5244 - E.M. Cli

15

three possibilities 0 = max singular

if S > 0 S<0 S=0

The singular case arises only if S() vanishes on an arc of nite width.
8

AOE 5244 - E.M. Cli

16

Since m is positive we can

multiply S by m without changing the conclusions. Hence we re-dene S = [v c m m]

AOE 5244 - E.M. Cli

17

Singular Control

For S to stay constant at zero we require S = 0 and this leads

to S = v c m m m m

Substituting the appropriate

8

AOE 5244 - E.M. Cli

18

state/adjoint dierential equations this simplies to v D S= D+c r c m v Note that the terms have cancelled out.

We could take a second

time-derivative and insist that
8

AOE 5244 - E.M. Cli

19

S = 0. Equivalently, observe that the three conditions H = 0 S = 0 S = 0 are three linear homogeneous equations in the adjoint
8

AOE 5244 - E.M. Cli

20

variables r , v , m. Since the adjoints can not all vanish simultaneously, this implies that the determinant must be zero. That is, v D+c D v c D+ m r2 =0

AOE 5244 - E.M. Cli

21

Since this involves only state

variables its somewhat simpler than S. Setting the time-derivative of this expression to zero will lead to an expression for the singular control . It is still somewhat messy.
8

AOE 5244 - E.M. Cli

22

In general, the control will

appear rst in an even time-derivative of the switching function. If the control appears rst in the 2q-th time-derivative of S, we say the singular arc is of order q. The Goddard problem has a rst order singular arc.
8