Analytical Mechanics This page intentionally left blank
Download 10.87 Mb. Pdf ko'rish
|
L
i h r Fig. 11.12 Table 11.1 Orbital elements of the planets of the Solar System. Here , h, i and λ are expressed in degrees, a is expressed in astronomical units (1 A.U. = 1.5 ×10
8 km),
is the perihelion longitude: = g +h,
λ = + l is the average longitude (at a fixed time) Planet h
e λ a Mercury 77.4561
48.3309 7.0050
0.205632 252.2509
0.387104 Venus
131.5637 76.6799
3.3947 0.006772
181.9798 0.723307
Earth 102.9373
0.016709 1000.4664 1.000012 Mars
336.0602 49.5581
1.8497 0.093401
355.4333 1.523711
Jupiter 374.3313
100.4644 1.3033
0.048495 34.3515
5.210216 Saturn
93.0568 113.6655
2.4889 0.055509
50.0775 9.538070
Uranus 173.0052
74.0060 0.7732
0.046296 314.0550
19.183302 Neptune
48.1237 131.7841
1.7700 0.008989
304.3487 30.055144 Pluto 224.6148
110.4065 17.1323
0.250877 218.8874
39.537580 This difficulty can be resolved by introducing a new set of action-angle variables ( Λ
1 , Z
2 , λ, ζ
1 , ζ
2 ) ∈ R 3 ×T 3 : Λ = L, Z 1 = L−G, Z 2 = G−H, λ = h+g +l, ζ 1 = −g −h, ζ 2 = −h (λ is called the mean longitude, −ζ 1 is the perihelion longitude). Hence considering the pairs (Z 1 , ζ 1 ) and (Z
2 , ζ
2 ) as polar coordinates we find ξ 1
2Z 1 cos ζ 1 , η
1 = 2Z 1 sin ζ
1 , ξ
2 = 2Z 2 cos ζ
2 , η
2 = 2Z 2 sin ζ
2 . (11.174) 11.9 Analytic mechanics: Hamilton–Jacobi theory and integrability 471 The canonical variables ( Λ , ξ
1 , ξ
2 , λ, η
1 , η
2 ) ∈ R + × R
2 × T
1 × R
2 are called Poin- car´ e variables and are also well defined in the case of a circular orbit (Z 1 = 0) or
a horizontal orbit (Z 2 = 0). In these new variables the Hamiltonian of Kepler’s problem is H = −mk
2 /2 Λ 2 ; therefore Λ , ξ
1 , ξ
2 , η
1 , η
2 are constants of the motion. The advantage of the Poincar´ e variables over the Delaunay ones is that the former are then suitable for both the complete resonance of the Kepler problem and for the study of the planets of the Solar System. The relation between the Poincar´ e and the original variables momentum and position (p, q) is more complicated and will not be discussed here (the interested reader can refer to Poincar´ e (1905,
chapter III) or Laskar (1989a)). Note however that Λ is proportional to √ a, ξ 2 1 + η 2 1 √ Λ e(1 +
O(e 2 )), ξ 2 2 + η 2 2 √ Λ i(1 + O(i 2 ) + O(e 2 )). In applications one often uses the orbital elements as (non-canonical) coordinates. 11.9
Wave interpretation of mechanics In this section we intend to illustrate how through the Hamilton–Jacobi equation, we can associate a wave front to a Hamiltonian system. What follows is a seem- ingly abstract analysis of classical mechanics, which however comes surprisingly close to the fundamental concepts of quantum mechanics. Consider an autonomous system with Hamiltonian H(p, q) having l ≥ 2 degrees of freedom, and assume that the Hamilton principal function is known. This function can be written in the form S(q,
α, t) = W (q, α) − E(α)t, (11.175)
up to an inessential additive constant. The constants α 1 , . . . , α l are determined by the initial conditions. For t = 0 we have S = W
0 = W (q(0), α). (11.176)
For every t > 0 the equation S(q,
α, t) = W 0 (11.177) defines a regular (l − 1)-dimensional manifold M(t) in the space of configurations R l
W (q, α) = W
0 + E(
α)t. (11.178)
At time t = 0 equation (11.178) selects a family of initial states, determined by the pairs of vectors (q(0), α), such that W (q(0), α) = W 0 . As t varies, M (t) moves within the family Σ of manifolds W (q, α) = constant, according to (11.178). Hence the dynamics of the system becomes associated with the propagation of a ‘front’ M (t). There exist interesting relations between the motion of the system and the motion of M (t) in Σ .
472 Analytic mechanics: Hamilton–Jacobi theory and integrability 11.9 P
1 2 l i,j =1 a ij ˙ q i ˙ q j is the kinetic energy of the system and if R l is endowed with the metric ds 2 T = l i,j =1 a ij dq i dq j , (11.179) then the trajectories of the system in the configuration space are orthogonal to each manifold in the family Σ .
For a fixed time t 0 and a point q 0 ∈ M(t
0 ), for every vector δq ∈ T q
M (t 0 ) we have by (11.178), ∇ q W (q 0 , α) · δq = 0 (11.180)
to first order in |δq|. We know that (q 0 ,
0 through the relation p 0 =
q W (q
0 , α) and that in addition, by definition, p 0i = l j =1 a ij (q 0 ) ˙
q j . Equation (11.180) can be interpreted as ( ˙q, δq) T = 0,
where we denote by (x, y) T = l i,j
=1 a ij x i y j the scalar product induced by the metric (11.179). We can now deduce information on the velocity of the points of M (t). Consider the family of the trajectories of the system issuing from the points of M (0) and the family Σ of the manifolds M (t) (Fig. 11.13). For every fixed trajectory γ we define the vector q γ (t) = γ ∩ M(t). P roposition 11.5 According to the metric (11.179) we have | ˙q γ | T = |E| 2(E − V (q
γ )) . (11.181) Proof
From the identity W (q
γ (t),
α) = W 0 + Et (11.182) we deduce ∇ q
γ (t),
α) · ˙q γ = E, (11.183) which can be interpreted as ( ˙q, ˙q γ )
= E, where ˙q is defined through the vector p =
∇ q W (q γ (t),
α). By construction ˙q γ and ˙q are proportional at every instant, and therefore ( ˙q, ˙q γ ) T = | ˙q| T | ˙q
γ | T = |E|.
Since | ˙q|
2 T = 2T = 2(E − V ), we can deduce equation (11.181). In the case of a single free point particle, the space of configurations coincides with the physical space, the manifold M (t) is a surface, and the metric (11.179) can be identified with the Euclidean metric. 11.9 Analytic mechanics: Hamilton–Jacobi theory and integrability 473
(t) q(0) M(t) M(0) ⌺ Fig. 11.13 Example 11.17 For a free point particle we have ˙q = c 0 , a constant, p = mc 0 = α, and hence W = α ·q. In this case, W is the generating function of the identity transformation. The surfaces W = constant are the planes orthogonal to α (Fig. 11.14) and the ‘front’ M (t) behaves as the phase of a plane wave: S = α · q − Et. (11.184) This simple example leads us to reinterpret the propagation of the front M (t) in the context of a field theory analogous to the theory describing the propagation of light in a non-uniform optical medium. We start from the observation that a plane wave with velocity c 0 can be represented in the form φ = φ
0 e i(k·q−ωt) , (11.185)
with φ 0 a constant, k a vector parallel to c 0 and
ω = kc 0 . (11.186) The absolute value k is the wave number which defines the wavelength λ = 2π/k. Equation (11.185), with ω given by (11.186), is a solution of the wave
474 Analytic mechanics: Hamilton–Jacobi theory and integrability 11.9
Fig. 11.14 equation ∆ φ − 1 c 2 0 ∂ 2 φ ∂t 2 = 0,
(11.187) describing the propagation of the electromagnetic field in a medium with refractive index n
= c c 0 , (11.188) where c is the speed of light in the void. The wave phase can also be written in the form
k 0 (n 0 e · q − ct) = 2π n 0 λ 0 e · q − νt , (11.189) where e = k/n, k 0 = kc
0 /c, λ
0 = 2π/k
0 , ν = ω/2π. Up to here, the analogy with Example 11.17 is evident. We now consider the modifications which must be introduced in order to describe the motion of a more general mechanical system by means of this optical model. For simplicity we shall deal only with the case of a single free point particle, subject to a field with potential energy V (q). We know then
11.9 Analytic mechanics: Hamilton–Jacobi theory and integrability 475 that at every point of the associated moving surface M (t) we can define the propagation velocity u =
E 2m(E
− V ) , (11.190) and we try to reproduce this behaviour in the case of a wave of the form φ = φ
0 e A (q)+ik 0 (L(q)−ct) , (11.191)
by imposing the validity of an equation of the kind (11.187). Note that in contrast to the case of a plane wave, we now have a variable amplitude φ 0 e
(q) and that
the function L(q), called eikonal, replaces the linear function n 0 e · q. If we insert the function (11.191) into the modified equation ∆ φ
1 u 2 (q) ∂ 2 φ ∂t 2 = 0, (11.192)
with u(q) given by (11.190), separating real and imaginary parts we find |∇L|
2 = n
2 (q) +
λ 0 2π 2 [ ∆ A + |∇A|
2 ], (11.193) ∆ L + 2
∇L · ∇A = 0, (11.194)
where n(q) = c/u(q). If λ
0 tends to zero in (11.193) (geometrical optic limit ) we find for L(q) the equation of geometrical optics: |∇L|
2 = n
2 (q),
(11.195) which is structurally a Hamilton–Jacobi equation. The analogy with the equation for the function W can be made closer by noting that the front with constant phase:
2π 1 λ 0 L(q)
− νt = constant (11.196)
varies in the family Σ of surfaces L(q) = constant. We can impose proportionality between the phase (11.196) and the function S = W
− Et. This yields: (a) a proportionality relation between |E| and ν: |E| = hν
(11.197) (h is the Planck constant ), 476 Analytic mechanics: Hamilton–Jacobi theory and integrability 11.9 (b) the proportionality W = h/λ 0 L, so that equation (11.193) coincides with the Hamilton–Jacobi equation for W if n =
λ 0 h 2m(E − V ),
(11.198) in agreement with the definition n = c/u = c 2m(E − V ) E = c hν 2m(E − V ), c/ν = λ 0 . The hypothesis that allowed us to develop this analogy is that we assumed the validity of the geometrical optics approximation, corresponding to neglecting the terms in λ 2 0 in equation (11.193). We can also introduce the wave number of the equivalent field: k = 2π
= 1 2m(E − V ), = h 2π . (11.199) We can easily check that if we rewrite equation (11.191) in the form φ = ψ(q)e − iωt
(11.200) ψ must satisfy the equation ∆ ψ + k
2 ψ = 0,
(11.201) and hence ∆ ψ +
2 2 m(E − V )ψ = 0, (11.202)
called the Schr¨ odinger equation of wave mechanics. We note that this can also be derived from the Schr¨ odinger equation for quantum mechanics: 1 2m ∆ φ − 1 2 V φ = − i ∂φ
∂t , (11.203) by substituting into it the expression (11.200). If we now return to the parallel idea of considering a wave with phase S/ , and hence φ = φ 0 e iS/ h – , equation (11.203) yields the equation 1 2m |∇S| 2 + V + ∂S ∂t = i 2m ∆ S, (11.204)
which reduces to the Hamilton–Jacobi equation if we consider the limit h → 0
(classical limit ). It is interesting to note that h → 0 is equivalent to ν → ∞ (by (11.197)), and hence to λ 0 → 0: the classical limit of the Schr¨odinger equation is equivalent to the limit of geometrical optics in the context of wave theory. 11.10 Analytic mechanics: Hamilton–Jacobi theory and integrability 477 11.10
Problems 1. Solve the Hamilton–Jacobi equation for the motion in space of a point particle of mass m subject to weight. 2. Separate variables in the Hamilton–Jacobi equation for the motion of a point particle of mass m subject to the action of weight and constrained to move on a surface of rotation around the z-axis. 3. A point particle of unit mass moves without any external forces on a surface whose first fundamental form is (ds) 2 = (U (u) + V (v))((du) 2 + (dv)
2 ), where (u, v) ∈ R 2 (Liouville surface). (a) Write down the Hamiltonian of the system. (b) Write down the Hamilton–Jacobi equation and separate variables. (c) Solve the Hamilton–Jacobi equation in the case that U (u) = u 2 , V (v) = v, where u > 0, v > 0. (Answer: (a) H = (p 2 u
2 v )/[2(U (u) + V (v))]; (b) setting S = W 1 (u) + W
2 (v)
− Et, we have (W 1 (u))
2 − 2EU(u) = α = (W 1 (v))
2 − 2EV (v); (c) W 1 (u) =
1/2[u α + 2Eu
2 + (α/
√ 2E)arcsinh( 2E/αu)], and W 2 (v) = 1/3E(α + 2Ev) 3/2 .) 4. Use the Hamilton–Jacobi method to solve Problems 23, 24 and 28 of Section 1.13. 5. Write down the Hamiltonian, solve the Hamilton–Jacobi equation and find the action variables for the systems described in Problems 11 and 15 of Section 3.7 and Problems 8, 9 and 12 of Section 4.12. 6. Consider a point particle of unit mass freely moving on the surface of a tri-axial ellipsoid: x 2
2 + y 2 b 2 + z 2 c 2 = 1, where a < b < c. Prove that in the variables (u, v) ∈ [b, c] × [a, b] defined by x = a
− a)(v − a) (b − a)(c − a) , y =
b (u − b)(v − b) (c − b)(a − b) , z =
c (u − c)(v − c) (a − c)(b − c) , the Hamiltonian is given by H(u, v, p u , p v ) =
1 2 p 2 u (u − v)A(u) + p 2 v (v − u)A(v) ,
478 Analytic mechanics: Hamilton–Jacobi theory and integrability 11.10 where
A(λ) = 1 4 λ (a − λ)(b − λ)(c − λ) . Write down the Hamilton–Jacobi equation and separate variables. 7. Two point particles of equal mass move along a line interacting through a force field with potential energy V (x 1 −x
), where x 1 and x 2 are the coordinates of the two points. (a) Write down the Hamiltonian. (b) Write down the Hamilton–Jacobi equation and separate variables (hint: introduce as new coordinates x = (x 1 + x
2 )/ √ 2 and y = (x 1 − x 2 )/ √ 2). (c) Setting V (x 1 −x
) = V 0 e (x 1 −x 2 )/d
, where V 0 and d are two prescribed positive constants, compute the solution of the Hamilton–Jacobi equation. 8. Write down and solve the Hamilton–Jacobi equation for a point particle of mass m moving in space under the action of a central field V (r) = k/r 2 + hr 2 , where k and h are two positive constants. Find the action variables and express the energy as a function of them. Compute the frequencies of the motions. Find the resonance conditions. Do periodic orbits exist? 9. Consider the Hamiltonian (see Problem 9 of Section 3.7) H(p, q) = p 2
+ V 0 q d 2n , where m, V 0 and d are prescribed positive constants and n is an integer greater than or equal to 2. Let E be the fixed value of energy. Prove that if J indicates the action variable, then E = πnJ
dB n 2n/(n+1) 1 2m n/ (n+1) V 1/(n+1) 0 , where B n = 1 0 √ 1 − uu 1/2n
− 1 du. Prove that the period T of the motion is equal to T = d
2m E E V 0 1/2n n + 1 n 2 B n . 10. Given a system of Hamiltonian (see Problem 14 of Section 3.7) H(p, q) = p 2
+ V 0 (e − 2q/d
− 2e − q/d ), 11.10 Analytic mechanics: Hamilton–Jacobi theory and integrability 479 where V
0 , d are prescribed positive constants, compute the action variable J and check that the energy expressed in terms of the action variable is E =
−V 0 1 − J d √ 2mV
0 2 . Compute the period of the motion. 11. Given a system of Hamiltonian (see Problem 13 of Section 3.7) H(p, q) = p 2 2m − V 0 cosh
2 (q/d)
, where V
0 , d are prescribed positive constants, compute the action variable cor- responding to librations, and verify that the energy expressed in terms of the action variable is E = −V
1 − J d √ 2mV 0 . Compute the period of the motions. 12. Given a system with Hamiltonian H(p, q) = p 2
+ V 0 |q|, where V 0 is a prescribed positive constant, find the transformation to action-angle variables and determine the frequency of the motion. 13. Given a system with Hamiltonian H(p, q) = p 2 2m + V (q),
where V is periodic with period 2π and V (q) =
−V 0 q, if −π ≤ q ≤ 0, +V 0
if 0 ≤ q ≤ π,
with V 0 a prescribed positive constant (see Problem 12 of Section 3.7), find the transformation to action-angle variables and determine the frequency of the motion for librations and for rotations. 14. Consider the Hamiltonian system (see Problem 10 of Section 3.7) H(p, q) = p 2
+ V 0 tan 2 (q/d),
where V 0 , d are two fixed positive constants. Compute the action variable J and prove that the energy E expressed in terms of J is given by E =
1 2m J d 2 (J + 2d 2mV 0 ). 480 Analytic mechanics: Hamilton–Jacobi theory and integrability 11.11 Verify that the period of the motion is T = 2πmd
2 J + 2d
√ 2mV
0 , and compute the angle variable. 15. A point particle of mass m moves with velocity v along a segment and is reflected elastically at the endpoints of the segment. Prove that the action J for the system is J = pl/π, where p = mv and l is the length of the segment. Prove that the energy is E = 1/2m (πJ /l) 2 .
sum of three functions related to the three cycles in Fig. 11.11. Then compute the angle variables, showing that one of them coincides with the mean anomaly. 11.11 Additional remarks and bibliographical notes In this chapter we discussed the Hamilton–Jacobi method for solving the equations of motion and a notion of integrability 5 for the Hamilton’s equations, correspond- ing to the existence of bounded and quasi-periodic orbits, and hence of a foliation of the phase space in invariant tori. The presentation of the Hamilton–Jacobi equations and of the method of separation of variables for its solution follows the text of Landau and Lifschitz (1976). The book by Levi-Civita and Amaldi (1927) contains a detailed discussion of the theory presented in the first two sections, and of the examples we chose to illustrate it. General references for this chapter are the book of Whittaker (1936), which contains many of the classical results obtained by the Italian school 6 at the beginning of the twentieth century on the classification of the cases when the Hamilton–Jacobi equation is solvable by separation of variables, and the treatise of Agostinelli and Pignedoli (1989). For a more complete treatment of the notion of a completely canonically integrable system, and of action-angle variables, we recommend the article of Nekhoroshev (1972), and the review of Gallavotti (1984). The latter is also an excellent basis for a more advanced study of the topics discussed in the next chapter. We partially followed it in our proof of the theorem of Liouville given in Section 11.4. The lecture notes of Giorgilli (1990) have been particularly useful in preparing Section 11.5, in particular for the proof of Proposition 11.1. A more detailed discussion of the action-angle variables for the problem of Kepler, and their relation with the orbital elements, can be found in the first 5 There indeed exist various other notions of integrability, just as there exist methods different from the Hamilton–Jacobi one for solving the equations of motion. The reader interested in a more detailed treatment of these themes can start by reading the review article of Kozlov (1983). 6 Morera, Levi-Civita, Burgatti, Dall’Acqua and many more. 11.12 Analytic mechanics: Hamilton–Jacobi theory and integrability 481 chapter of Beletski (1986). In the second chapter of this very pleasant and informal introduction to celestial mechanics one can find a study of the problem of two centres of force (see Example 11.6), with the completion of its solution via separation of variables in the Hamilton–Jacobi equation, and the explicit computation of the trajectory of a polar satellite using elliptic functions. In the following third chapter, parabolic coordinates are used for the study of the pressure of solar radiation upon an artificial satellite orbiting around the Earth, and there is a complete classification of the trajectories corresponding to the planar case. We stress that for the sake of brevity we did not illustrate the mechanics of rigid systems as an example of completely canonically integrable systems. The book of Gallavotti (1980) contains this illustration fully. In addition, we did not develop very extensively the theme, touched upon in Section 11.9, of the relations existing between the Hamilton–Jacobi equation, geo- metrical optics and the semiclassical approximation in quantum mechanics. The text of Arnol’d (1978a), already referred to, also contains these topics (sections 46 and 47 and appendices 11 and 12). The complex relation between the notion of action variable and the so-called ‘old quantum’ of Bohr and Sommerfeld, described in Graffi (1993), is also very interesting. An excellent reading on this is the treatise of Born (1927), which can also be read as an introduc- tion to the study of the canonical theory of perturbations, studied in the next chapter. 11.12
Additional solved problems Problem 1 A point particle of unit mass moves along a plane curve of equation y = P (x), where P is a polynomial of degree n ≥ 1, and it is subject to a conservative force field with potential energy V (x) = ax 2 + bx + c, a = / 0. Use the method of Hamilton–Jacobi to determine the travelling time t = ˆ t(x, x(0), ˙ x(0)) of a solution. Solution The Lagrangian of the particle is L(x, ˙ x) =
1 2 ˙ x 2 (1 + P (x) 2 ) − V (x), from which it follows that p = ∂L/∂ ˙ x = ˙
x(1 + P (x) 2 ) and H(p, x) = p 2 2(1 + P (x) 2 ) + V (x). (11.205)
The Hamilton–Jacobi equation for Hamilton’s characteristic function W = W (x, E) is 1 2
∂x 2 (1 + P (x) 2 ) −1 + V (x) = E, (11.206)
482 Analytic mechanics: Hamilton–Jacobi theory and integrability 11.12 and hence W (x, E) = ± x x (0)
(1 + P (ξ) 2 ) 2(E − V (ξ)) dξ, (11.207) where the sign is determined by the sign of ˙ x(0). Differentiating (11.207) with respect to E and recalling Remark 11.5, we find t = ∂W
= ± x x (0)
1 + P (ξ) 2 2[E − V (ξ)] dξ.
(11.208) The integrand is of the form Q(ξ)/ αξ 2
k j =0 q j ξ j (in
our case q 0 = 1, k = 2n − 2, α = −2a, β = −2b, γ = 2(E − c)). Its primitives are of the form k −1
=0 q j ξ j αξ 2 + βξ + γ + q k dξ
2 + βξ + γ
, (11.209)
where the k coefficients q 0 , q 1 , . . . , q k can be determined starting from the known coefficients q 0 , . . . , q k , α, β, γ and multiplying the relation k j
q j ξ j αξ 2 + βξ + γ = d dξ k −1 j =0 q j ξ j αξ 2 + βξ + γ + q k αξ 2 + βξ + γ
(11.210) by the square root and identifying the polynomials obtained on the two sides of the identity. Recall that, setting ∆ = β 2 −4αγ, we have, up to additive constants, dξ αξ
+ βξ + γ = 1 √ α ln |2 αξ 2 + βξ + γ + 2αξ + β | (if α > 0) = ⎧ ⎪ ⎪ ⎪ ⎪ ⎪ ⎨ ⎪ ⎪ ⎪ ⎪ ⎪ ⎩ 1 √ α arcsinh 2αξ + β
√ − ∆ , if ∆ < 0, α > 0, 1 √ α ln |2αξ + β|, if ∆ = 0, α > 0, − 1 √ −α arcsin
2αξ + β √ ∆ , if ∆ > 0, α < 0. (11.211)
From (11.209)–(11.211) it follows that it is possible to compute explicitly the integral (11.208) and hence obtain t = ˆ t(x, x(0), ˙ x(0)).
Problem 2 Consider the following canonical transformation of R + × R
3 : Q 1 = −e t (1 + p
1 ) √ q 1 , Q 2 = arcsin q 2 p 2 2 + q 2 2 , P 1 = e −t (1 − p 1 ) √ q 1 , P 2 = p 2 2 + q 2 2 2 . (11.212) 11.12 Analytic mechanics: Hamilton–Jacobi theory and integrability 483 How does the Hamiltonian H(p 1 , p
2 , q
1 , q
2 ) = q
1 (p 2 1 − 1) + (p 2 2
2 2 )/2 transform? Use the result obtained to completely solve Hamilton’s equations associated with H. For the associated system determine whether the hypotheses of the theorem of Liouville and of the theorem of Arnol’d are satisfied. Solution
The transformation (11.212) admits the generating function F 2 (q 1 , q 2 , P
1 , P
2 , t) =
− 2(e t P 1 ) √ q 1 + 1 2 (e t P 1 ) 2 + q 1 + q 2 2 2P 2 − q
2 2 + P 2 arcsin
q 2 √ 2P 2 . The Hamiltonian H is transformed into K(P 1 , P 2 , Q
1 , Q
2 , t) = H + ∂F 2 /∂t:
K = P 1 Q 1 + P
2 + ∂F 2 ∂t = P 1 Q 1 + P 2 − 2e t P 1 √ q 1 + P 2 1 e 2t = 2P 1 Q 1 + P 2 . (11.213) Hamilton’s equations associated with K can be solved immediately: P 1
1 (0)e
−2t , Q 1 (t) = Q
1 (0)e
2t , P 2 (t) = P
2 (0),
Q 2 (t) = Q 2 (0) + t.
(11.214) Then from the inverse of (11.122): q 1
e 2t 4 (P 1 − e −2t Q 1 ) 2 , p 1 = − P 1 + e −2t
Q 1 P 1 − e
−2t Q 1 , q 2 = 2P 2 sin Q 2 , p 2 = 2P 2 cos Q 2 , (11.215) we deduce the solution of Hamilton’s equations associated with H by substituting (11.214) into (11.215) and using the relations P 1
− p 1 (0)) q 1 (0), P 2 (0) = (p 2 2 (0) + q 2 2 (0))/2, Q 1 (0) = −(1 + p 1 (0)) q 1 (0), Q 2 (0) = arcsin q 2 (0) p 2 2 (0) + q 2 2 (0) . The two functions f 1 (p 1 , p
2 , q
1 , q
2 ) =
−q 1 (1 − p 2 1 ), f 2 (p 1 , p 2 , q
1 , q
2 ) =
p 2 2 + q 2 2 2 are first integrals for H = f 1 + f
2 , in involution and independent except in three planes π 1 , π 2 , π
3 of equations q 1 = 0, p
1 = 1; q
1 = 0, p
1 = −1; q 2 = p
2 = 0,
respectively. The hypotheses of the theorem of Liouville are therefore satisfied 484 Analytic mechanics: Hamilton–Jacobi theory and integrability 11.12 on R
4 \ (π
1 ∪ π
2 ∪ π
3 ) while those of the theorem of Arnol’d are not, because the level sets of f 1 are not compact. Problem 3 Consider the system described by the Hamiltonian H : R 2
1 , p
2 , q
1 , q
2 ) =
p 2 1 q 2 1 − 1 (1 + p
2 2 + q 2 2 ). (11.216) (i) Write down Hamilton’s equations and determine all constant solutions. (ii) Linearise the equations around p 1 = p 2 = q
1 = q
2 = 0 and solve the linearised equations. (iii) Determine two first integrals of the motion, independent and in involution, and express the Hamiltonian through these first integrals. (iv) Write down the Hamilton–Jacobi equation associated with H and solve it by separation of variables. (v) Construct when it is possible the action-angle variables, write the Hamilto- nian as a function of the actions only and determine the frequencies. For what initial conditions are the motions periodic? Solution Hamilton’s equations are ˙ p
= 2q 1 p 2 1 (q 2 1 − 1) 2 (1 + p 2 2 + q 2 2 ), ˙ q 1 = 2p 1 q 2 1 − 1 (1 + p
2 2 + q 2 2 ), ˙ p 2 = − 2q 2 p 2 1 q 2 1 − 1
, ˙ q 2 = 2p 2 p 2 1 q 2 1 − 1
, (11.217)
from which we see immediately that the constant solutions are given by p 1 = 0, for any (p 2 , q 1 , q
2 ). The equations linearised around the origin are ˙ p
= 0, ˙ p 2 = 0,
˙ q 1 = −2p
1 , ˙ q 2 = 0. Denoting by (P 1 , P 2 , Q
1 , Q
2 ) the initial conditions, the corresponding solution is clearly p
(t) ≡ P
1 , p 2 (t)
≡ P 2 , q 1 (t) = Q 1 − 2P
1 t, q 2 (t)
≡ Q 2 . Since the Hamiltonian is of the form H(p 1 , p 2 , q
1 , q
2 ) = f
1 (p 1 , q 1 )f 2 (p 2 , q 2 ), with f 1 (p 1 , q
1 ) = p
2 1 /(q 2 1 − 1), f 2 (p 2 , q 2 ) = 1 + p 2 2 + q 2 2 , we deduce immediately that f 1 and f
2 are two first integrals that are independent and in involution. Indeed, the involution is guaranteed by the fact that f 1 and f 2 are functions of distinct pairs of canonically conjugate variables. Moreover {f 1 , H } = {f
1 , f
1 f 2 } = {f 1 , f 2 }f 1 + 11.12 Analytic mechanics: Hamilton–Jacobi theory and integrability 485 {f
, f 1 }f 2 = 0 and similarly {f 2
} = 0. The independence holds on the whole of R
2 × (R \ {−1, 1}) × R except where ∇f 1
2p 1 q 2 1 − 1 , 0, − 2q 1 p 2 1 (q 2 1 − 1)
2 , 0
= (0, 0, 0, 0), or ∇f 2 = (0, 2p
2 , 0, 2q
2 ) = (0, 0, 0, 0), and hence {p 1 = 0 } and {p
2 = q
2 = 0
}, respectively. Since H is independent of time, if E denotes the energy, the Hamilton–Jacobi equation for the characteristic function W (q 1 , q 2 , α
1 , α
2 ) can be written as 1 q
1 − 1
∂W ∂q 1 2 1 +
∂W ∂q 2 2 + q
2 2 = E and can be solved by separation of variables: W (q, α) = W
1 (q 1 , α 1 ) + W 2 (q 2 , α 2 ), with ∂W 1 ∂q 1 2 = α 1 (q 2 1 − 1), ∂W 2 ∂q 2 2 = α 2 − 1 − q
2 2 , α 2 ≥ 1 E = α 1 α 2 , from which it follows that, up to additive constants, W 1 (q 1 , α
1 ) =
± ⎧ ⎪ ⎪ ⎪ ⎪ ⎨ ⎪ ⎪ ⎪ ⎪ ⎩ q 1 2 α 1 (q 2 1 − 1) − 1 2 √ α 1 ln √ α 1 q 1 + α 1 (q 2 1 − 1) , if α 1 > 0,
0, if α
1 = 0,
q 1 2 −α 1 (q 2 1 − 1) + 1 2 √ −α 1 arcsinq 1 , if α 1 < 0 W 2 (q 2 , α 2 ) =
± q 2 2 α 2 − 1 − q 2 2 + 1 2 (α 2 − 1)arcsin q 2 √ α 2 − 1 . Following the steps outlined in Section 11.1 it is then possible to compute explicitly the Hamiltonian flow associated with (11.216). For a completely canon- ical transformation to action-angle variables to exist, the level set M α 1
2 = {(p 1 , p
2 , q
1 , q
2 ) | f 1 (p 1 , q 1 ) = α 1 , f
2 (p 2 , q 2 ) = α 2 } must be compact and connec- ted. This is the case only if (α 1 , α 2 ) ∈ (−∞, 0) × (1, +∞). In this case, the two equations p 2 1 − α
1 q 2 1 = −α 1 , p 2 2 + q 2 2 = α 2 − 1
486 Analytic mechanics: Hamilton–Jacobi theory and integrability 11.12 determine two ellipses in the planes (p 1 , q
1 ), (p
2 , q
2 ) and therefore M α 1
2 is evidently diffeomorphic to a two-dimensional torus T 2 . From equation (11.61) it follows that J 1 = 1 2π π √ −α 1 = √ −α 1 2 , J 2 = 1 2π π(α 2 − 1) = α 2 − 1 2 , from which α 1 = −4J 2 1 , α 2 = 2J
2 +1, E =
−4J 2 1 (2J 2 +1) = K(J 1 , J
2 ). Substituting the latter into W 1 and W 2 we find the generating function of the transformation to action-angle variables. The frequencies are ω 1 =
1 (2J
2 + 1), ω
2 = −8J 2 1 ; therefore the motions are periodic only if ω 1 /ω 2 is rational, i.e. for initial conditions such that (2J 2 (0) +
1)/J 1 (0) ∈ Q (or if J 1 (0) = 0). 12 ANALYTICAL MECHANICS: CANONICAL PERTURBATION THEORY 12.1 Introduction to canonical perturbation theory The so-called ‘perturbation methods’ for studying differential equations are tradi- tionally of great importance for their applications to celestial mechanics (indeed, this field of study initially motivated their development). In spite of the efforts of generations of celebrated mathematicians (Lagrange, Laplace, Weierstrass and above all Poincar´ e, who can be considered the father of the modern theory) until recently the majority of techniques used did not have a rigorous mathematical justification. Proving the convergence (or divergence) of the perturbation series is not just an abstract goal, of secondary interest for physicists. On the contrary, it stems from the need to understand in depth the domains of applicability to physical problems, and the limitations, of perturbation techniques. On the other hand, the number of problems that can be treated with these tech- niques justifies a more detailed analysis, even if most of the modern developments go beyond the scope of the present text. The central question we want to consider is the study of a system whose Hamiltonian is a ‘small’ perturbation of the Hamiltonian of a completely canon- ically integrable system. According to Poincar´ e (1893) this is to be considered the ‘fundamental problem of classical mechanics’. In what follows we assume systematically that the Hamiltonian functions we consider are sufficiently regular. D efinition 12.1 A Hamiltonian system is called quasi-integrable if its Hamilton function is of the form h(p, q, ε) = h 0 (p, q) + εf (p, q), (12.1) where (p, q) ∈ R 2l
≤ |ε| 1) and h
0 is the Hamiltonian of a completely canonically integrable system. Remark 12.1 The previous definition is not completely satisfactory, unless we make more precise in what sense ε is a small parameter. As (p, q) varies in a compact subset K of R 2l (in which we want to study the motions and where h 0 does not
have singularities) there exists a constant M K > 0 such that max (p,q)∈K
|f(p, q)| ≤ M K max (p,q)∈K |h 0 (p, q) |. (12.2) 488 Analytical mechanics: canonical perturbation theory 12.1 The requirement that the perturbation be small can be expressed through the condition |εM
K | 1. (12.3) (Naturally, we suppose that neither h 0 nor f contain terms that are independent of q, p.) Since the system with Hamiltonian h 0 is completely canonically integrable, there exists a completely canonical variable transformation from (p, q) to action- angle variables (J, χ) with respect to which the Hamiltonian h 0 is expressed through a function H 0 that depends only on the action variables. After this transformation, the Hamiltonian (12.1) in the new coordinates is H(J,
χ, ε) = H 0 (J) + εF (J, χ), (12.4)
where H 0 and F are the functions h 0 and f expressed in the new variables, respectively. The action variables J are defined on some open subset of R l , while the angle variables are by their nature variables on a torus T l of dimension l. In other words, the function F is periodic separately in each of its variables χ 1 , . . . , χ l with fixed periodicity, for example equal to 2π. We also assume that the functions H, H 0 and F are regular (of class C ∞ or analytic when needed) in each argument. Example 12.1 The so-called Fermi–Pasta–Ulam model (Fermi et al. 1954) consists of a chain of l + 2 equal particles linked by non-linear springs. The two particles at the extremes of the chain are fixed. If (p, q) = (p 1 , . . . , p l , q
1 , . . . , q l ) are the kinetic moments and the coordinates of the l moving particles, setting q 0 = q l +1 = 0, the Hamiltonian of the model is H(p, q, ε) = h 0 (p, q) + εf (q) where h 0 (p, q) = l i =1 p 2 i 2m + k 2 l i =0 (q i +1 − q i ) 2 , corresponds to the integrable part, and f (q) = λ
l i =0 (q i +1 − q i ) r , where r = 3 or r = 4, defines the non-linearity of the springs of the chain, and λ is a constant. Fermi, Pasta and Ulam introduced this model to study numerically the ergodic hypothesis and the equipartition theorem of statistical mechanics (cf. Section 15.2). Example 12.2 Consider a system of l identical particles, each performing a uniform rotation on a fixed circle. Let q i be the angular coordinate identifying the ith particle and 12.1 Analytical mechanics: canonical perturbation theory 489 we examine a weak perturbation with potential energy V (q i , q i +1 ) = −εV 0 cos(q i +1 − q i ), V 0 > 0,
where 0 ≤ |ε|
1 measures the intensity of the coupling. If we assume that the last particle is coupled to the penultimate and the first, setting q 0 = q
l , the
Hamiltonian of the system is H(p, q, ε) = h 0 (p) + εf (q), with h 0 (p) = l i =1 p 2 i 2 , f (q) = −V 0 l i =1 cos(q i − q i −1 ). The action variables for the unperturbed system coincide with the kinetic moments p and the angle variables correspond to the angles q ∈ T l
system can also be considered as a classic model for the study of the so-called ‘spin systems’ of statistical mechanics. In the following two examples the perturbation is periodic in time. Example 12.3 The Hamiltonian of the restricted three-body problem is quasi-integrable. Consider the effect of the attraction of Jupiter on the revolution around the Sun of a minor planet of the Solar System (the Earth, or Mercury) or of an asteroid. As a first approximation we can consider the orbit of Jupiter as circular and fixed (hence neglecting the effect of the attraction of the minor body on Jupiter). The resulting problem has three degrees of freedom, but the Hamiltonian depends on time t periodically (the period is equal to the period of revolution of Jupiter around the Sun). The motion of the minor body is then described with respect to a reference system with origin at the Sun, and axes moving with respect to an inertial reference system (Fig. 12.1). Note that this system is however not inertial, because the Sun has an acceleration (due to the attraction of Jupiter): a S = Kε r G |r G | 3 , where K is a constant proportional to the mass of the Sun, ε is the ratio M G /M S between the mass of Jupiter and of the Sun (so that ε ≈ 10
−3 ), and r
G is the
position vector of Jupiter. The Hamiltonian of the system is then equal to H(p, q, t, ε) = |p| 2
− K |q| − Kε |q − r G (t)
| + Kε
r G (t) · q |r G (t) | 3 , where the mass of the minor body is equal to 1, its position vector is denoted by q ∈ R
3 and p denotes the relative conjugate kinetic momentum. The last term in H is the (generalised) potential energy of the inertia force responsible for the
490 Analytical mechanics: canonical perturbation theory 12.1
Download 10.87 Mb. Do'stlaringiz bilan baham: |
ma'muriyatiga murojaat qiling