Analytical Mechanics This page intentionally left blank
Download 10.87 Mb. Pdf ko'rish
|
r G q – r G q a S S T Fig. 12.1 acceleration a S of the origin of the reference system considered. Evidently the system is quasi-integrable, with h 0 (p, q) = |p|
2 2 − K |q|
, f (p, q, t) = − K
G (t)
| + K
r G (t) · q |r G (t) | 3 · Example 12.4: the spin–orbit problem Consider a satellite S in orbit around a planet P . Suppose that the satellite is a rigid body with the form of a tri-axial homogeneous ellipsoid. The three axes A 1
2 > A
3 of the ellipsoid coincide with the principal axes of inertia. Since the ellipsoid is homogeneous, the corresponding principal moments of inertia are I 1 < I 2 < I 3 , and hence the maximum momentum I 3 is associated with the shortest axis of the ellipsoid. Suppose also that the orbit of the satellite is a fixed Keplerian ellipse with P at one of the foci. We denote by e the eccentricity of the orbit. We also assume that the axis of rotation of the satellite coincides with the x 3 -axis and is directed orthogonally to the plane of the orbit. Since the orientation of the satellite is completely determined by the angle between the major axis of the ellipsoid and the direction of the pericentre of the orbit, the problem has only one degree of freedom. We also neglect dissipative forces which may be acting on the system and all perturbations due to other bodies (which 12.1 Analytical mechanics: canonical perturbation theory 491
1
w Fig. 12.2 may for example be responsible for changes in the orbital parameters, cf. Laskar and Robutel (1993)). The orientation of the satellite varies only under the effect of the torque of the gravitational attraction of P on the ellipsoid S. If α is the angle between the x 1 -axis of the ellipsoid and the direction of the pericentre of the orbit, ϕ is the polar angle, a is the semi-major axis and r is the instantaneous orbital radius (Fig. 11.2) the equation of the motion can be written as (cf. Goldreich and Peale 1966, Danby 1988, section 14.3) ¨ α + 3 2 I 2 − I
1 I 3 a r(t)
3 sin(2α
− 2ϕ(t)) = 0. Note that if the ellipsoid is a surface of revolution, then I 1
I 2 and the equation is trivially integrable. In addition, since r and ϕ are periodic functions of t (with period equal to the period T of revolution of S around P ), by choosing the unit of time appropriately we can assume that r and ϕ are 2π-periodic functions. Finally, setting x = 2α, ε = 3
I 2 − I 1 I 3 , and expanding (a/r(t)) 3 sin(x
− 2ϕ(t)) in Fourier series we find ¨ x + ε m ∈Z m=0 ˆ s m (e) sin(x − mt) = 0. This equation corresponds to a quasi-integrable Hamiltonian system (depending on time) with one degree of freedom: H(p, x, t, ε) = p 2 2 − ε
m ∈Z m=0 ˆ s m (e) cos(x − mt),
492 Analytical mechanics: canonical perturbation theory 12.1 which will be the object of a more detailed study in the next section (ε is a small parameter because, in the majority of cases arising in celestial mechanics, ε ≈ 10 −3 –10
−4 ). The computations of the coefficients ˆ s m (e) is somewhat laborious (see Cayley 1861). They can be expressed as a power series in e and at the lowest order they are proportional to e |m−2| . As an example, we have ˆ s −2 (e) = e 4 24 + 7e 6 240
+ O(e
8 ), ˆ s −1 (e) = e 3 48 + 11e
5 768
+ O(e
7 ), ˆ s 1 (e) = − e 2 + e 3 16 − 5 384 e 5 + O(e
7 ), ˆ s 2 (e) = 1 − 5e 2 2 + 13e 4 16 − 35e 6 288 + O(e
8 ), ˆ s 3 (e) = 7e 2 − 123e 3 16 + 489e
5 128
+ O(e
7 ), ˆ s 4 (e) = 17e 2 2 − 115e
4 6 + 601e 6 48 + O(e
8 ), ˆ s 5 (e) = 845 48 e 3 − 32525 768 e 5 + O(e
7 ). In the Earth–Moon system (cf. Celletti 1990) the orbital eccentricity is e = 0.0549, while ε = 7 × 10
−4 . If we neglect the terms which give a contribution to the Hamiltonian of less than 10 −6 we find H(p, x, t) = p 2 2 − ε −
e 2 cos(x − t) + 1 − 5 2 e 2 cos(x − 2t) + 7e 2 cos(x − 3t)
+ 17 2 e 2 cos(x − 4t) + 845
48 e 3 cos(x − 5t) .
If ε = 0 the system (12.4) is integrable and Hamilton’s equations ˙ J i = 0,
˙ χ i = ∂H 0 ∂J i (J) (12.5) are trivially integrable: the actions are first integrals of the motion for the system, i.e. J i
i (0) for every i = 1, . . . , l, while each angle has a period 2π/ω i , where
ω i = ω i (J(0)) =
∂H 0 ∂J i (J(0))
is the frequency of the angular motion, depending on the initial conditions for the action variables. All motions are therefore bounded and quasi-periodic and the system admits as many independent first integrals as the number of
12.1 Analytical mechanics: canonical perturbation theory 493 degrees of freedom. The phase space is foliated into invariant tori of dimension l (cf. Remark 11.19) for the Hamiltonian flow and each torus is identified by the constant values of the actions J. When ε = / 0 the motion equations change; in particular for the action variables we have ˙
i = −ε ∂F ∂χ i (J, χ),
i = 1, . . . , l, (12.6)
and they are no longer constants of the motion. From the regularity of F there follows the possibility of estimating the time difference of the action from its initial value: |J i (t) − J
i (0)
| ≤ ∂F ∂χ i εt,
(12.7) where
· indicates the maximum norm on a compact subset K of R l to which J(0) belongs, and on T l for the angles. The estimate (12.7), while significant for times t of order O(1), may yield little information for longer times. This is shown by the following trivial example. Example 12.5 Let l = 1 and H(J, χ, ε) = J + ε cos χ. In this case, Hamilton’s equations are ˙ J = ε sin χ, ˙ χ = 1,
and hence J (t) = J (0) + ε[cos χ(0) − cos(χ(0) + t)], χ(t) = χ(0) + t. It follows that |J(t) − J(0)| ≤ 2ε for all times t and not only for times t = O(1) as predicted by (12.7). This drawback of (12.7) can be attributed to the fact that in deriving this inequality we did not take into account the sign variations in ∂F/∂χ i . These
variations can yield some compensations which extend the validity of the estimate. The perturbation ∂F/∂χ i is not generally constant (except when its arguments are constant), and it does not have a constant sign. Indeed, the function ∂F/∂χ i is periodic but has zero mean, and therefore it cannot have a constant sign unless it is identically zero. The perturbation method for Hamiltonian systems of type (12.4) consists of solving the following problem. Problem
Find a completely canonical transformation which eliminates the dependence of the Hamiltonian on the angular variables, to first order in ε. Then iterate this 494 Analytical mechanics: canonical perturbation theory 12.1 procedure until the dependence on χ to all orders in ε, or at least to a prescribed order, is eliminated. Hence we seek the generating function W (J , χ, ε) of a canonical transformation from the action-angle variables (J, χ) corresponding to the integrable system with Hamiltonian H 0 to new variables (J , χ ), with respect to which the Hamiltonian (12.4) has an expression H (J , χ , ε) that is independent of the angular variables, at least in the terms up to order O(ε 2
H (J , χ , ε) = H 0 (J ) + εH 1 (J ) + ε
2 F (J ,
χ , ε). (12.8)
Here F is a remainder depending on ε, but that may fail to tend to 0 when ε → 0 (however we assume it to be bounded together with its first derivatives). When ε = 0 the starting Hamiltonian is independent of the angle variables. Hence the transformation sought is ε-near the identity and we can try to expand the generating function W into a power series in ε whose zero-order term is the generating function of the identity transformation. We therefore write W (J , χ, ε) = J · χ + εW (1) (J ,
χ) + O(ε 2 ), (12.9) with W
(1) (J ,
χ) unknown. The transformation generated by (12.9) is J i = J i + ε ∂W (1)
∂χ i (J , χ) + O(ε 2 ), i = 1, . . . , l, χ i = χ i + ε ∂W (1)
∂J i (J , χ) + O(ε 2 ), i = 1, . . . , l. (12.10)
Substituting the first of equations (12.10) into (12.4) and requiring that the transformed Hamiltonian has the form (12.8), we find the equation H 0
∇ χ W (1) ) + εF (J , χ) + O(ε 2 ) = H 0 (J ) + εH 1 (J ) +
O(ε 2 ), (12.11) where the functions H 0 , H
1 are to be determined. Expanding H 0 to first order and equating the corresponding powers of ε we find for the term of zero order in ε:
H 0 (J ) = H 0 (J ).
(12.12) This ensures—as was obvious from the previous considerations—that to zero order in ε the new Hamiltonian coincides with the starting one (expressed in the new action variables). At the first order in ε we find the equation ω(J ) · ∇ χ W
(J , χ) + F (J , χ) = H 1 (J ),
(12.13) for the unknowns W (1) (J ,
χ) and H 1 (J ), where ω(J ) = ∇ J H 0 is the vector of frequencies of the new Hamiltonian. For fixed actions J , equation (12.13) is a
12.1 Analytical mechanics: canonical perturbation theory 495 linear partial differential equation of first order on the torus T l whose solution will be studied in Sections 12.3 and 12.4. We shall see that the iteration to higher order terms of the perturbation method always leads to solving equations of the type (12.13). For this reason, the latter is called the fundamental equation of classical perturbation theory. If equation (12.13) admits a solution, i.e. if there exist two functions H 1 (J ) and W (1)
(J , χ) (the second 2π-periodic with respect to χ) which satisfy (12.13), the equations of motion for the new action variables are ˙ J i = − ∂H ∂χ i (J , ε) = O(ε
2 ), where i = 1, . . . , l. Therefore, for all times t in the interval [0, 1/ε] we have |J (t) − J (0)| = O(ε). The new action variables are approximately (up to O(ε) terms) constant over a time interval of length 1/ε. One arrives at the same conclusion for the action variables J, exploiting the fact that the transformation (12.10) is near the identity. Indeed,
J(t) − J(0) = (J(t) − J (t)) + (J (t) − J (0)) + (J (0) − J(0)), and given that the first and last terms are also O(ε) (uniformly with respect to time t) we have |J(t) − J(0)| = O(ε), for every t ∈ [0, 1/ε]. Remark 12.2 Equation (12.11) is simply the Hamilton–Jacobi equation approximated up to terms of order ε 2 for the Hamiltonian (12.4). Indeed, the Hamilton–Jacobi equation for the Hamiltonian (12.4) can be written as H( ∇ χ W, χ, ε) = H 0 ( ∇ χ W ) + εF ( ∇ χ
χ) = H (J , ε), (12.14)
and equation (12.11) is then obtained by substituting the expansion (12.9) into equation (12.14) and neglecting all terms of order O(ε 2
Before starting a more detailed study of equation (12.13) when l ≥ 2, we
consider the case l = 1. If the system has only one degree of freedom, then as we saw (cf. Section 11.3), it is completely canonically integrable, as long as the motions are periodic (hence outside the separatrix curves in phase space). Therefore the following theorem should not come as a surprise to the reader. 496 Analytical mechanics: canonical perturbation theory 12.1 T
/ 0, equation (12.13) has solution H 1 (J ) = 1 2π 2π 0 F (J , χ) dχ, (12.15) W (1) (J , χ) = 1 ω(J ) χ 0 [H 1 (J )
− F (J , x)] dx. (12.16)
This solution is unique, if we require that the mean value of W (1)
on S 1 be zero, and hence that 1 2π 2π 0 W (1) (J , χ) dχ = 0. (12.17) Proof
Expression (12.15) is the only possible choice for H 1 (J ), because the χ-average of ω(J )∂W (1)
/∂χ vanishes due to the periodicity of W (1)
. Therefore H 1 (J ) must be the mean of F (J , χ) with respect to χ. After this, it is immediate to check that (12.16) actually satisfies (12.13). The uniqueness of the solution follows in a similar way. Let W (1)
, H 1 be a second solution of (12.13). Then ω(J ) ∂ ∂χ (W (1)
− W (1)
)(J , χ) = H 1 (J ) − H 1 (J ). (12.18) However
2π 0 ∂ ∂χ (W (1) − W (1)
)(J , χ) dχ = W
(1) (J , 2π)
− W (1)
(J , 2π) − (W
(1) (J , 0)
− W (1)
(J , 0)) = 0, by the periodicity of W (1) and W
(1) . Hence integrating both sides of equation (12.18) we find that H 1 (J ) = H 1 (J ). Therefore ω(J ) ∂
(W (1)
− W (1)
)(J , χ) = 0, from which it follows that W (1) (J , χ) = W (1) (J , χ) + g(J ). If we impose that W (1)
has zero average, then necessarily g ≡ 0.
Example 12.6 Consider the following quasi-integrable system with one degree of freedom (dimensionless variables): H(J, χ, ε) = J 2 + εJ
3 sin
2 χ.
12.1 Analytical mechanics: canonical perturbation theory 497 The generating function W (J , χ, ε) = J χ + εJ 2 8 sin 2χ
transforms the Hamiltonian H to H (J , ε) = J 2 +
2 J 3 + O(ε
2 ). The frequency of the motions corresponding to H is ω (J , ε) = 2J + 3 2 εJ 2 . In the case of one degree of freedom, it is possible to formally solve the Hamilton–Jacobi equation (12.14) to all orders in ε (neglecting the question of the convergence of the series), assuming that the frequency of the motions is not zero. Canonical perturbation theory thus yields (at least formally) the complete integrability of these systems. T heorem 12.2 If l = 1 and ω(J ) = / 0, the Hamilton–Jacobi equation (12.14) admits a formal solution: H (J , ε) = ∞ n =0 ε n H n (J ), (12.19) W (J , χ, ε) = J χ + ∞ n
ε n W (n) (J , χ).
(12.20) The solution is unique if we require that W (n) has zero average with respect to χ for every n ≥ 1.
Proof Substituting (12.19) and (12.20) into equation (12.14) we have H 0
∞ n =1 ε n ∂W (n) ∂χ + εF J + ∞ n =1 ε n ∂W (n)
∂χ , χ
= ∞ k =0 ε k H k (J ), and expanding H 0 in Taylor series around J we find H 0 J + ∞ n =1 ε n ∂W (n) ∂χ = H 0 (J ) + ω(J ) ∞ n
ε n ∂W (n) ∂χ + 1 2 d 2 H 0 dJ 2 ∞ n =2 ε n n 1 +n 2 =n ∂W (n 1 ) ∂χ ∂W (n 2 ) ∂χ + · · · + 1 k! d k H 0 dJ k ∞ n =k ε n n 1 +n 2 +···+n k =n ∂W (n 1 ) ∂χ ∂W (n 2 ) ∂χ · · ·
∂W (n k ) ∂χ + · · · , (12.21)
498 Analytical mechanics: canonical perturbation theory 12.1 where ω = dH 0 /dJ . Similarly, expanding F we find F J +
∞ n =1 ε n ∂W (n) ∂χ , χ = F (J , χ) + ∂F ∂J ∞ n =1 ε n ∂W (n) ∂χ + 1 2 ∂ 2 F ∂J 2 ∞ n =2 ε n n 1 +n 2 =n ∂W (n 1 ) ∂χ ∂W (n 2 ) ∂χ + · · ·
+ 1 k! ∂ k F ∂J k ∞ n =k ε n Download 10.87 Mb. Do'stlaringiz bilan baham: |
ma'muriyatiga murojaat qiling