You are on page 1of 7

NON-CANONICAL PERTURBATIONS IN SYMPLECTIC INTEGRATION

SEPPO MIKKOLA
Tuorla Observatory, University of Turku, 21500 Piikkiö, Finland

(Received: 18 July 1997; accepted: 20 November 1997)

Abstract. The inclusion of non-canonical perturbations in symplectic integration schemes has been
discussed. A rigorous derivation of an analog for the Wisdom–Holman (1991) method, such that
velocity dependent forces can be included, has been outlined. This is done both by using the δ-
function formalism and also by means of formal Hamiltonization. Application to the relativistic
corrections in Solar System integrations is discussed as an example. Numerical experiments confirm
the usefulness of the method.
Key words: symplectic integration, relativistic terms, drag forces

1. Introduction

The methods of symplectic integration have often been suggested as ideal for the
study of the long term behavior of dynamical systems. Recent works include the pub-
lications by Kinoshita, Yoshida and Nakai (1991) and Wisdom and Holman (1991)
(hereafter WH). Mikkola (1997) used the same principle as the above studies, but
with time transformation.
Recently also generalizations to dissipative systems have been proposed by Mal-
hotra (1994) and Cordeiro, Gomes and Martins (1997) (CGM hereafter) who consider
mainly dissipative systems. These authors offer their methods as generalizations and
analogs to the celebrated Wisdom–Holman algorithm. The analogies are, however,
in technical details rather than in principle, and those methods can produce secu-
lar dissipation effects if applied to conservative systems, such as planetary motions
with relativistic corrections. Saha and Tremaine (1992) suggest the use of a potential
mimicking the effect of relativistic terms, but this approach is not always applicable.
More recently Saha and Tremaine (1996) used the relativistic Hamiltonian to obtain
rigorously symplectic results for Solar System orbits with relativistic corrections.
In this paper the inclusion of non-canonical perturbations, such as the relativistic
terms (in the usual non-canonical formulation) and drag forces, are considered in the
framework of the δ-function formalism introduced by Wisdom and Holman, as well
as by using the device of formal Hamiltonization.

2. Treatment of Non-Canonical Perturbations

In this section, first the proper analog of the WH-method for any type of perturbation
is considered, then the same results are derived from a formal Hamiltonian.

Celestial Mechanics and Dynamical Astronomy 68: 249–255, 1998.


© 1998 Kluwer Academic Publishers. Printed in the Netherlands.
250 SEPPO MIKKOLA

2.1. analog of wh-method

Consider the perturbed two-body Hamiltonian

H = K + R(r), (1)

where K the Keplerian part and R is a perturbation. As known the basic ideal in the
WH-method was to replace this by the expression

H = K + 1(h, t)R(r), (2)


P
where h is the stepsize and 1(h, t) = h +∞ k=−∞ δ(t − (k + 2 )h) is the so called
1

‘delta-function fence’. The two Hamiltonians differ only by short period terms and
therefore produce similar long term behavior.
The solution for the motion defined by the Hamiltonian (2) results in the so called
generalized leapfrog method:
To move the coordinates and momenta over a timestep of magnitude h, first
advance the system for time h/2 using the Hamiltonian K, then update the
momenta by adding to them the ‘perturbation’ of magnitude −h∂R/∂r and
finally advance the system again for the half timestep h/2 using only K.
If we have a system with non-canonical perturbations, it is not possible to maintain
the Hamiltonian formalism as such. However, if we write the differential equation
following from the above Hamiltonian

r̈ = F + 1(h, t)f(r), (3)

where F is the two-body acceleration and f is the perturbation, we can immediately


generalize this to

r̈ = F + 1(h, t)f(r, v, t), (4)

where v = ṙ is the velocity. The solution for this differential equation is:
(1) The two-body motion governed by the equations

ṙ = v; v̇ = F, (5)

when the system moves in between the point of action of the delta-function fence.
(2) At the point of action of a delta-function the solution must be obtained from

v̇ = f(r, v, t) (6)
r = constant; t = constant,

integrated over a time interval of length = h. This result is easily obtained, e.g.
by replacing the δ-function by its approximation, a function with value = 1/ε over
NON-CANONICAL PERTURBATIONS 251

a time interval of length = ε, and letting ε → 0: When the δ-function acts, the
differential equations become
h
v̇ = f(r, v, t) (7)
ε
ṙ = v, (8)
in which the time advances over an interval of length ε. After taking a new time
variable tδ = εt/ h we may write
 
dv ε
= f r, v, t + tδ (9)
dtδ h
dr ε
= v, (10)
dtδ h
and the tδ -interval = h. In the limit this then reduces back to (6) when we use again
dot as notation for the derivative.
Note, especially, that the result is not the same thing as using a velocity jump
computed with the most recent values of the variables r, v, t (as CGM (1997) seem to
imply), but the variation of v must be obtained by solving the differential Equation (6)
where r and t are considered constants.

2.2. use of formal hamiltonian

Let us introduce the Hamiltonian


H = Pt + Pr · v + Pv · F + Pv · f(r, v, t) (11)
where t, r and v are considered as coordinates and Pt , Pv and Pr their conjugate
momenta. This formal Hamiltonian gives the equations of motion
∂H
v̇ = = F + f, (12)
∂Pv
∂H
ṙ = = v, (13)
∂Pr
∂H
t˙ = = 1, (14)
∂Pt
which are obviously correct.
Now we divide the Hamiltonian into two parts
H0 = Pt + Pr · v + Pv · F, (15)
H1 = Pv · f(r, v, t), (16)
and apply the generalized leapfrog principle to get the ‘unperturbed step equations’:
∂H0
v̇ = = F, (17)
∂Pv
252 SEPPO MIKKOLA

∂H0
ṙ = = v, (18)
∂Pr
∂H0
t˙ = = 1, (19)
∂Pt
and from the perturbation part we obtain
∂H1
v̇ = = f, (20)
∂Pv
∂H1
ṙ = = 0, (21)
∂Pr
∂H1
t˙ = = 0, (22)
∂Pt
which are the same as above, while the rest of the equations (for the formal momenta)
are not needed in practice.
More generally, by defining a formal Hamiltonian

H = P · F(x), (23)

it is possible to apply the above to any flow

ẋ = f(x). (24)

If it is possible to write

F(x) = F(x) + f(x) (25)

one may divide the Hamiltonian into two parts

H0 = P · F (26)
H1 = P · f, (27)

and subsequently apply the generalized leapfrog into this system (assuming both
parts individually integrable).
However, often the perturbation part is not easily integrable analytically. If it is
small, often an approximate solution (numerical) is sufficient. The formal Hamilto-
nian formulation guides us in selecting the method. If the ‘unperturbed’ part of the
derivatives, and possibly a major part of the perturbation also, arises from a conser-
vative Hamiltonian system, then a symplectic method is advisable. A simple choice
is the (symplectic) implicit midpoint (e.g. Liao, 1997) method [which may be defined
for any differential equation y 0 = f (y) as yn+1 = yn + hf ( 21 (yn + yn+1 ))].
This method can be efficient despite its implicit character if the perturbation is
sufficiently weak. Another reason for this choice is that it gives a time-reversible
NON-CANONICAL PERTURBATIONS 253

result if the system itself is time-reversible. Also, time-reversibility alone is normal-


ly sufficient for good energy conservation in long term integrations (Funato, Hut,
McMillan, and Makino 1996, Hut, Makino and McMillan 1995).

2.3. relativistic terms in the solar system

An important special case of application is the computation of the motions of the


bodies in the Solar System with inclusion of the relativistic corrections. Although
Hamiltonian formulations exist (e.g. Saha and Tremaine (1996)), the equations are
often used in the non-canonical formulation. For any planet we must thus solve a
problem of the form
1
r̈ = −mr/r 3 + f0 (r, t) + f1 (r, v), (28)
c2
where f0 is the Hamiltonian perturbation, c is the speed of light and f1 is the expression
for the first Post-Newtonian contribution (e.g. Soffel, 1989). If we apply the implicit
midpoint method in connection with the generalized leapfrog, we obtain by (20) for
the velocity jumps the equation
h
δv = hf0 (r, t) + f1 (r, v + 21 δv)), (29)
c2
which must be solved for δv. In Solar System applications the relativistic contribution
is largest for Mercury and thus we tested this for a ‘Mercury only’ problem. Due to
the weakness of the relativistic perturbation it is sufficient to evaluate the right-hand
side only twice. In our 1 Myr test computation, with a stepsize of 6 days, we found
that the implicit midpoint method (Equation (29)), which is ‘rigorous’ in the sense
of this paper) gives a tiny secular error of

ȧ/a = −1.3 × 10−16 /year.

The origin of this small error remains unknown, but is likely to be the result of
a systematic roundoff. The phenomenon is qualitatively similar to what Wisdom,
Holman and Touma (1996) obtained in accuracy tests of their N-body method (with
Hamiltonian perturbations only). They, however, used quadruple precision in some
key parts of their code and thus obtained even smaller numerical errors.
On the other hand the method of CGM (which corresponds to neglecting the δv
in the right hand side of (29)), gives a small but clear secular error in semi-major axis
ȧ/a = −1.3 × 10−13 /year, which is three orders of magnitude larger than what is
obtained using the new formulation.

2.4. a drag force

A simple drag force −εv in a two-body orbit allows the comparison of the method
of CGM against the new formulation with the implicit midpoint method (in this case
254 SEPPO MIKKOLA

explicit solution is possible) and the analytical solution of Equation (20) [which in
this case reads v̇ = −εv]. Thus we have the three possibilities for the computation
of the velocity jump:

δv = −hεv0 , (30)
δv = −hεv0 /(1 + hε/2), (31)
δv = −(1 − exp(−hε))v0 , (32)

With a value of ε = 10−5 , initially circular orbit (mass = 1, G = 1, initial semi-


major axis = 1) and length of integration = 1000 initial periods, with 30 integration
steps per period, we find the results:
The semi-major axis shrank by 12%, the CGM method (30) produced in the semi-
major axis a secular error which reached the value of 1.3×10−9 (relative error) at the
end of the computation, while the new method(s) (31) and (32) gave almost identical
results in which the error fluctuated in the much smaller range of ±2 × 10−13 .
The ‘accurate’ solution, with which the computations were compared, was
obtained by using ten times smaller timestep in the integration. Although such a
calculation is not error-free, the inaccuracies are anyway expected to be much small-
er than those in the test computations, thus giving a reliable estimate for the precision
of the test calculations.

3. Conclusion

The generalized leapfrog, known as the Wisdom–Holman method in planetary system


computations, can be rigorously generalized to include velocity dependent forces.
However, it is not clear that the way suggested in this paper is the only possible or
the best.
The formal Hamiltonization is a useful tool for handling new kind of effects to
be implemented into symplectic codes.
Test calculations show that the relativistic corrections do not form any special
obstacle for symplectic integration of planetary systems even if used in the usual non-
canonical formulation. Also drag forces can be accurately treated with the present
formalism.

References

Cordeiro, R. R., Gomes, R. S. and Vieira Martins, R.: 1997, ‘A Mapping for Non-Conservative
Systems’, Celest. Mech. Dynam. Astron. 65, 407–419.
Funato, Y., Hut, P., McMillan, S. and Makino, J.: 1996, ‘Time Symmetrized Kustaanheimo–Stiefel
Regularization’ Astron. J. 112, 1697–1708.
Hut, P., Makino, J. and McMillan, S.: 1995, ‘Building a Better Leapfrog’, Ap. J. 443, L93–L96.
Kinoshita, H., Yoshida, H. and Nakai. H.: 1991, ‘Symplectic Integrators and their Application in
Dynamical Astronomy’, Celest. Mech. Dynam. Astron. 50, 59–71.
NON-CANONICAL PERTURBATIONS 255

Liao, X.: 1997, ‘Symplectic integrator for general near-integrable Hamiltonian system’,
Celest. Mech. Dynam. Astron. 66, 243–253.
Malhotra, R.: 1994, ‘A Mapping Method for the Gravitational Few-Body Problem with Dissipation’,
Celest. Mech. Dynam. Astron., 60, 373–385.
Mikkola, S.: 1997, ‘Practical Symplectic Methods with Time Transformation for the Few-Body
Problem’, Celest. Mech. Dynam. Astron. (in press).
Saha, P. and Tremaine, S. D.: 1992, ‘Symplectic Integrators for Solar system Dynamics’, Astron.
J. 104, 1633–1640.
Saha, P. and Tremaine, S. D.: 1996, ‘Long-Term Planetary Integration with Individual Timesteps’,
A. J. 108, 1962–1969.
Soffel, M. H.: 1989, Relativity in Astrometry, Celestial Mechanics and Geodesy, Springer-Berlin,
p. 141.
Wisdom, J. and Holman, M.: 1991, ‘Symplectic Maps for the N-Body Problem’, Astron. J. 102,
1520.
Wisdom, J., Holman, M. and Touma, J.: 1996, ‘Symplectic Correctors’, Proc. Integration Methods in
Classical Mechanics Meeting, Waterloo, October 14–18, 1993, Fields Institute Communications
10, p. 217.

You might also like