Applications
5.4 Explicit Lie-Poisson Integration
In this section, we introduce a new explicit Lie-Poisson algorithm. Take a Hamiltonian system on the cotangent bundle of some Lie group. Assume that the Hamiltonian is left-invariant under the group action of G on T*G.
We saw from section 1.5 that the co-adjoint orbits on
S*,
when the phase space has been left reduced, is given by the negative Lie-Poisson bracket, i.e., if p EG*,
then the co-adjoint orbit through p EG*
will be (Q,,w,) where (3, = {Ad;-,(p)lg E G) and< .,
. >is the natural pairing between the Lie algebra and its dual and [,]is the Lie algebra bracket. From section 3.5, we know that the equation of
motion for some p E
E*
and Hamiltonian, H ECw(E*)
are given byThe minus arises if the Lie-Poisson system was derived by identification with right invariant vector fields and the positive arises in the left-invariant case. Therefore, for a left invariant Hamiltonian, the equations of motion for p E
E*
is given byIn order t o integrate tlzis system, we have resorted t o the construction of symplectic integrators which are based on the reduced Hamilton-Jacobi equation of Chapter 4. While this approach is applicable t o a very broad class of Lie-Poisson systems, it is only easily realizable in the subcase of regular quadratic Lie algebras, i.e., algebras endowed with an Ad-invariant, bilinear, non-degenerate form. As we have seen, tlzis property enables us t o construct a con-slngu!ar identity transfarmati~n =:I
G*
for the momentum preserving generating fullctions of the first kind on T*G. This was accom- plished by the application of the strict generating pair theory of Ge[13].Armed with such an identity transformation generator, we can then form a kth-order perturbative solution to the Hamilton-Jacobi equation which is implicitly Casimir preserving and Poisson, i.e., the mapping on
G*
is sym- plectic with respect to the KAMS symplectic form on co-adjoint orbits.We claim that it is possible to build an explicit Lie-Poisson integrator for precisely these Lie algebras without reverting to a Ruth[l4] type of approx- imation to the Hamiltonian. If
6
is regular quadratic, then it is possible t oform a diffeomorphism between co-adjoint and adjoint orbits. For brevity, we will present a result proven in [I] for a left-invariant Poisson system.
Proposition 5.4.1 Let H : T*G i R be a left-invariant Hamiltonian, i.e.,
N
o T * L , =N
for all g E G. O n (3, we defined the symplectic KAKS 2-form with the negative sign having been selected. If there ezists a bi- invariant 2-form (., .) o nG,
then the adjoint orbit throught
EG
given by (3( = {Ad,tlg E G) has symplectic 2-formfor ( anzd q i n
6.
As we will be using the Cartan-Killing form in our examples, let us denote (., .) by I<(., .). I< is non-degenerate, so for every p E
G*,
there exists a uniqueb
E6
such that<
p , t>=
Ii'(fi,t) for allt.
We now show that as a consequence of K being Ad-invariant, the adjoint action is skew-symmetric with respect to it.Consider
I<(adtC, 7 ) = -lt=oI~(Adexpt[C, 7) = d dt
With this information, we return to the original equation of motion on
G*
Since is the derivative of an element of a linear space, it can be also considered an element of that space. Thus, there exists some uniquefi
in6
such that
< b,
'I>= ~ ( j i ,
rl).Also, setting
t
= E O , we deduce-
<
p, adeq>= li(F,
adtq) =-
K(ad@, 7 ) . (5.21)Therefore by the Ad-invariance of K , the system reduces t o
We can now approximate this flow by discretizing its equation of motion.
If we choose as our initial position some element of the Lie algebra, po,then for time-step, T , the time advanced element p1 will be given by
where h = is the first derivative of the Hamiltonian evaluated at the starting point. Each iteration may then be executed by replacing po by p~
and iterating through time.
The trick which makes this algorithm pratical without any neccessity t o separate the Hamiltonian into simpler pairings, is found in a formula which we have already encountered in the last chapter and proven in [24]. It can be shown that for
t
in the neigborhood of the origin inG ,
where, just as in the final sections of the last chapter, ad< is simply a linear operator of a linear representation of (7 and thus the right-hand side of equation[1.23] can be interpreted as a power series in the operator.
The above time-stepping obviously preserves the Casimirs by definition.
However, we sliould check t o make sure that the KAKS symplectic 2-form on the adjoint leaves is preserved.
Proposition 5.4.2 Ad,-1 : O5 -i C3t preserves the symplectic 2-form, wo, i.e., (Ad,-l)*wo = wo.
To prove this, consider
By definition this equals
~O(A~,-~$)(TGA~,-~E‘B($),
T$Ads-l ?B($)). (5.26) NDW, in genera! (f )*Y ( 3 ) = (T f)-lY( f (z))? so T$.Ad,-lt g ( $ )
= (Ad,)*JG(Ad,-I$).So, equation(5.24) equals
where
4
= Ady-I$. We next use the fact that (Ad,)*& = ( A d , - ~ t ) ~ , t o obtainw0(4)((Ady-1t>8(4), (Adg-l?)~(d))). (5.28) By the definition of the KAIIS bracket on adjoint orbits via the Cartan- Killing form, we find that equation(5.27) becomes
- I{(+,
[& ~ 1 ) ~
(5.29) by Ad-invariance of the Cartan-Icilling form. The transformed 2-form (5.24) now equalsWO(@>(&(+>, ??G(?It))- (5.30) Therefore, the Ad mapping preserves the KAKS 2-form.
This algorithm will be implemented in the next section for the case of the rigid body dynamics and compared to the results obtained via the implicit Hamilton- Jacobi integrator.