Analytical Mechanics This page intentionally left blank
Download 10.87 Mb. Pdf ko'rish
|
x
H Fig. 11.5 Indeed (see Fig. 11.5 and recall Remark 8.12), we have dσ = ds dh, with dh |∇ x H | = δE, so that A(E + δE) = A(E) + δE γ E ds |∇ x H | + O((δE) 2 ), where x = (p, q) and by our hypotheses |∇ x H | =
/ 0 on γ E . Hence d A dE = lim δE →0 A(E + δE) − A(E) δE = γ E ds |∇ x H | = / 0. From equation (11.62) we have that dJ/dE = / 0, and therefore the existence of the inverse function E = K(J ) = A − 1 (2πJ ) follows. Substituting it into (11.59) we obtain the generating function of the canonical transformation to the action- angle variables. The latter is F (q, J ) = W (q, K(J )) (Example 11.3). Thus we have proved the following. T heorem 11.3 Every Hamiltonian system (11.46) with one degree of freedom and with motions of librations or rotations is completely canonically integrable. As a consequence of (11.52), (11.54) and (11.61), the period of the motion has the following simple expression: T =
d A dE = 2π dJ dE . (11.63)
Example 11.10 Consider the harmonic oscillator with Hamiltonian H = p
2m + 1 2 mω 2 q 2 . 11.3 Analytic mechanics: Hamilton–Jacobi theory and integrability 437
2mE 2E
2
Fig. 11.6 In the phase plane, the cycles γ E of the equation p 2 2m + 1 2 mω 2 q 2 = E
enclose the area (2π/ω)E (Fig. 11.6), and hence it follows from (11.61) that the action variable is J = E/ω, i.e. K(J ) = ωJ ; we have rederived equation (11.56). The generating function is (see Example 11.2) F (q, J ) = W (q, ωJ ) = J arcsin ωm 2J
+ mωJ
2 q 1 − mω 2J q 2 , and hence χ =
∂F ∂J = arcsin mω 2J q , from which we obtain the relations q = 2J
sin χ, p =
√ 2mωJ
1 − mω 2J q 2 = √ 2mωJ cos χ, which coincide with (11.55). Figure 11.6 shows the geometric meaning of the variable χ. The example of the harmonic oscillator illustrates well the statement made at the beginning of this section: the transformation from the variables (p, q) to variables of the kind (f (R E ), ψ) is not in general canonical. 438 Analytic mechanics: Hamilton–Jacobi theory and integrability 11.3 Let us compute R E and ψ as functions of (J, χ) in dimensionless variables (therefore setting m = 1). We find that πR 2 E = 2πJ , and hence R E = √ 2J .
Having chosen the point ( √ 2ωJ , 0) on γ E to be the origin of the arcs, for χ = 0, we have s(J, χ) = χ 0 ∂p ∂χ 2 + ∂q ∂χ 2 1 / 2 dχ = χ 0 2ωJ sin 2 χ +
2J ω cos 2 χ dχ .
In particular, the length λ E of γ E is only a function of J . Finally, we find ψ(J, χ) = 2π s(J, χ)
λ E (J ) . We now compute the Poisson bracket: {ψ, f(R E
} (J,χ)
= ∂ψ ∂χ dR E dJ f (R E ) = 2π λ E (J ) f (R
E ) √ 2J 2ωJ sin
2 χ +
2J ω cos 2 χ 1/2 . It follows that if ω = 1 (hence if γ E is not a circle) then {ψ, f(R E ) } = 1, inde- pendent of the choice of f (R E ), and the variables (f (R E ), ψ) are not canonical. If instead ω = 1 we have that {ψ, f(R
E ) } = f (R E )/R
E . Therefore, choosing f (R E
1 2 R 2 E = J we naturally obtain the same canonical variables (J, χ). Example 11.11 The Hamiltonian of a simple pendulum is (see Section 3.3) H(p, ϑ) = p 2 2ml 2 − mgl cos ϑ. Setting e = E/mgl, if |e| < 1 the motion is oscillatory and the action is equal to J = 2
2m 2 gl 3 ϑ m 0 √ e + cos ϑ dϑ, where ϑ m = arccos( −e). Setting k 2 = (e + 1)/2 and sin ϑ/2 = k sin ψ, we find J = ml gl 8 π π/ 2 0 k 2 cos 2 ψ 1 − k
2 sin
2 ψ dψ = ml gl 8 π [(k 2 − 1)K(k) + E(k)], (11.64) where K(k) and E(k) are the complete elliptic integrals of the first and second kind, respectively.
11.4 Analytic mechanics: Hamilton–Jacobi theory and integrability 439 If e > 1 the motions are rotations, and the action is equal to J = 1 π 2m 2 gl 3 π 0 √ e + cos ϑ dϑ. Setting k 2 = 2/(e + 1) and ψ = ϑ/2, we find J = 2 π 2m 2 gl 3 π/ 2 0 e + 1
− 2 sin 2 ψ dψ = 4 π m 2 gl 3 k 2 E(k). (11.65) The function K(J ) can be found by inverting the function J (E); J depends on E and e. Writing the formula for the period T =
2π ω = 2π dJ dE and computing dJ dE = dJ de 1 mgl , we easily find the formulae (3.18) and (3.21) of Chapter 3. We take into account in the calculations the relations dE(k)
dk = E(k) − K(k) k , dK(k) dk = 1 k E(k) 1 − k
2 − K(k)
(see Whittaker and Watson 1927, p. 521). Introducing action-angle variables for systems with more degrees of freedom requires some preliminary ideas. These are discussed in the following sections. 11.4
Integrability by quadratures. Liouville’s theorem Integrating a system of 2l ordinary differential equations of first order requires more than just knowledge of the l first integrals. However, if the system of equations is canonical, the fact that the flow preserves the symplectic structure of the phase space has among its consequences that it is enough to know l independent integrals in order to solve the Hamilton–Jacobi equations, thus leading to integration of the equations of motion. It is necessary, however, for the l first integrals to be in involution (Definition 10.18). This concept is clarified in the following. T heorem 11.4 (Liouville) Consider an autonomous Hamiltonian system with Hamiltonian H(p, q) having l degrees of freedom. Assume that the system admits 440 Analytic mechanics: Hamilton–Jacobi theory and integrability 11.4 l first integrals f 1 (p, q), . . . , f l (p, q) which are independent (hence such that for every (p, q) the gradients ∇ (p,q) f i are l linearly independent vectors) and that they are in involution. Consider the level set M a = {x = (p, q) ∈ R 2l |f
(p, q) = a i , i = 1, . . . , l }, (11.66)
where a ∈ R
l is fixed. If M a is not empty: (1) M a is a regular submanifold of dimension l, invariant with respect to the Hamiltonian flow S t and the phase flows g t 1 , . . . , g t l associated with f 1 , . . . , f l ;
t 1 , . . . , g t l commute with each other. In addition, if det
∂f i ∂p j = / 0, (11.67) then locally there exists a function S = S(f , q, t) such that l i
p i dq i − H(p, q) dt (p,q)∈M a
(11.68) The function is a complete integral of the Hamilton–Jacobi equation (11.5) corresponding to H. The system is therefore integrable by quadratures. Before giving the proof, we list some remarks. Remark 11.9 The system is autonomous, and hence we can include the Hamiltonian H among the l integrals of the motion considered in Theorem 11.4. In all cases, both H and all of the first integrals f i are constant not only along their own flow g t i , but also along the flow generated by other integrals. This is due to the mutual involution condition. In addition H is always constant on each connected component of the manifold M a . Indeed, even if f i = / H for every i = 1, . . . , l it is always possible to connect any pair of points belonging to the same connected component of M a through
successive applications (in any order) of the flows g t i . This intuitive concept will be rigorously proven in Lemma 11.2 below. Since H |
a is constant, equation (11.68) takes locally the form l i
p i dq i | M a = dW (a, q), (11.69) where
W (a, q) = S(a, q, t) + E(a)t, (11.70)
and E(a) is the value taken by H on M a . 11.4 Analytic mechanics: Hamilton–Jacobi theory and integrability 441 An l-dimensional submanifold of the phase space satisfying condition (11.69) is called Lagrangian. The significance of this property will be made clear when we construct the action-angle variables for systems with several degrees of freedom (see Section 11.6). Remark 11.10 The condition (11.67) is not restrictive, because the condition that the first integ- rals f
1 , . . . , f l are independent ensures that there exist l canonical coordinates x i 1 , . . . , x i l such that det
∂(f 1 , . . . , f l ) ∂(x i 1 , . . . , x i l ) = / 0.
(11.71) We saw (see Example 10.8) that the exchange of canonical coordinates is a completely canonical transformation. Hence if (i 1 , . . . , i l ) =
/ (1, . . . , l), exchanging some of the coordinates q i k
−p i k , we can always write (11.71) in the form (11.67). Remark 11.11 In general the condition (11.67) cannot be globally satisfied on all of M a : consider for example what happens in the case of the harmonic oscillator. Proof of Theorem 11.4 The properties (1) and (2) are an immediate consequence of the linear inde- pendence of the integrals, and of the fact that they are in involution (see Theorem 10.18). The condition (11.67) and the implicit function theorem guarantee the local existence of l regular functions ˆ p 1 (f , q), . . . , ˆ p l (f , q) such that p i = ˆ p i (f , q) and
f i (ˆ p(f , q), q) = a i , (11.72) for all i = 1, . . . , l. By Remark 11.9, equation (11.68) is equivalent to l i
ˆ p i (f , q)dq i = dW (f , q), (11.73) with f = a and d acting on q only. The existence of a function W satisfying (11.73) is guaranteed if for every j, k = 1, . . . , l we have ∂ ˆ
p k ∂q j = ∂ ˆ p j ∂q k , (11.74) and hence if the matrix B = (∂ ˆ p j /∂q k ) is symmetric. On the other hand, by differentiating with respect to q k the second of equations (11.72), we find l j =1 ∂f i ∂p j ∂ ˆ
p j ∂q k + ∂f i ∂q k = 0. (11.75)
442 Analytic mechanics: Hamilton–Jacobi theory and integrability 11.4 It follows that, setting A = (∂f i /∂p
j ) and C = (∂f i /∂q
k ) we have B = −A
1 C, (11.76) and equation (11.74) becomes −A − 1 C =
−C T (A T ) − 1 , and hence CA T − AC
T = 0. In componentwise form, this reads l k
∂f i ∂q k ∂f j ∂p k − ∂f i ∂p k ∂f j ∂q k = {f i , f j } = 0,
(11.77) confirming the validity of (11.73). The function W is therefore defined on M a as W (f , q) = q q 0 l i =1 ˆ p i (f ,
ξ) dξ i , (11.78) computed along an arbitrary path belonging to M a joining q 0 and q. The extension of W to non-constant values of f is possible because of the arbitrariness of a. Consider now ˆ H(f , q) = H(ˆ p(f , q), q). (11.79)
For fixed f = a, from (11.74) it follows that ∂ ˆ
H ∂q i = ∂H ∂q i + l j =1 ∂H ∂p j ∂ ˆ p j ∂q i = ∂H ∂q i + l j =1 ∂ ˆ p i ∂q j ∂H ∂p j = − ˙p i + l i =1 ∂ ˆ p i ∂q j ˙ q j = 0, by the first of equations (11.72). Thus ˆ H is independent of q, and the Hamiltonian can be expressed through the integrals f 1 , . . . , f l : H(ˆ p(f , q), q) = ˆ H(f ).
(11.80) It follows that setting S(f , q, t) = W (f , q) − ˆ
H(f )t, (11.81)
it can immediately be verified that (11.68) is satisfied and S is a solution of the Hamilton–Jacobi equation. Indeed, by hypothesis det
∂ 2 W ∂q i ∂f j = det (∂p i ∂f
) = / 0.
11.4 Analytic mechanics: Hamilton–Jacobi theory and integrability 443 From equation (11.73) it follows that p i = ∂W ∂q i = ∂S ∂q i , while equation (11.81) implies that H + ∂S/∂t = 0. In addition, S is a complete integral, because it depends on the l arbitrary constants a 1 , . . . , a l (the fixed values of f 1 , . . . , f l ) and H is independent of t. Example 11.12 Consider a system of l non-interacting harmonic oscillators, with Hamiltonian H(p, q) = l i =1 p 2 i 2m i + 1 2 m i ω 2 i q 2 i . (11.82) Evidently H(p, q) = l i =1 f i (p i , q i ), (11.83) where f i (p i , q i ) =
p 2 i 2m i + 1 2 m i ω 2 i q 2 i (11.84)
is the energy of the ith oscillator. The functions f i are integrals of the motion, independent and in involution. The level manifold M a is compact, connected, and diffeomorphic to an l-dimensional torus. The condition (11.67) is satisfied (as long as p i = 0 for every i = 1, . . . , l). Note that this property is not globally satisfied on M a , see Remark 11.11; however the condition (11.71) is certainly globally satisfied. The function S is then given by S(f , q, t) = l i
q i (q i ) 0 ± 2m i f i − m
2 i ω 2 i ξ 2 i dξ i − t
l i =1 f i . (11.85) Note that since the condition (11.67) is not globally satisfied on M a , S is not a single-valued function. Remark 11.12 Liouville’s theorem ensures that the integrals f 1 , . . . , f l can play the role of new canonical coordinates, together with the variables β i = ∂W ∂f i , i = 1, . . . , l. (11.86) The function W (f , q) is thus the generating function of a completely canonical transformation of the variables (p, q) into (f , β). Therefore it satisfies l i
(p i dq i + β
i df i ) = dW (f , q). (11.87)
444 Analytic mechanics: Hamilton–Jacobi theory and integrability 11.4 Note that on M a we have df = 0 and equation (11.87) reduces to (11.69). The new Hamiltonian takes the form ˆ H = ˆ H(f ) = H(ˆ p(f , q), q), (11.88) and hence Hamilton’s equations become ˙f = 0, ˙ β = ∇ f ˆ H(f ). (11.89) From this it follows that f is constant (as was known) and β(t) = β(0) + ∇ f ˆ H(f (0))t. (11.90)
Remark 11.13 We saw that every time we can solve the Hamilton–Jacobi equation (11.9) and compute the Hamilton characteristic function (as is the case, for example, when we can apply the method of separation of variables) then we determine l first integrals, independent and in involution. These are precisely the new canonical coordinates α 1 , . . . , α l . The theorem of Liouville gives the converse: the knowledge of l integrals, independent and in involution, yields Hamilton’s characteristic function. Note finally that in the separable cases equation (11.73) is simplified in a similar way to the Hamilton–Jacobi equation, as each function ˆ p i depends on f and only on the corresponding q i .
Consider a point particle of mass m in free motion on an (l − 1) dimensional ellipsoid embedded in R l , described by the implicit equation l i =1 x 2 i a i = 1, (11.91) where 0 < a 1
2
l , and
√ a i is the length of the ith semi-axis of the ellipsoid. We introduce a convenient parametrisation of the ellipsoid (due to Jacobi, see Arnol’d et al. (1983), p. 126–9) via the equation f (x, λ) = l i
x 2 i a i − λ = 1. (11.92)
This associates to any generic point x = (x 1 , . . . , x l ) ∈ R l , l real numbers λ 1
l (the l roots of equation (11.92)) which evidently alternate with the a i
1 < a 1 ≤ λ 2 < a 2 ≤ . . . ≤ λ l < a l . To show this, it is enough to note that for every fixed non-zero point x, f as a function of λ has l vertical asymptotes in λ = a
i , and for λ = a i one has ∂f /∂λ > 0. 11.4 Analytic mechanics: Hamilton–Jacobi theory and integrability 445 If x belongs to the ellipsoid (11.91), necessarily λ 1 = 0 and the variables λ 2
l yield a system of orthogonal coordinates on the ellipsoid. It is not difficult to show that for every i = 1, . . . , l we have x 2 i = l j =1 (a i − λ
j ) l j =1,j=i
(a i − a j ) , (11.93) from which it follows that 1 l
=1 ˙ x i 2 = 1 4 l i =2 M i ˙λ 2 i , (11.94) where M i = j =i (λ j − λ i ) l j =1 (a j − λ
i ) , i = 2, . . . , l. (11.95)
The variables µ i canonically conjugate to the λ i are
µ i = mM i ˙ λ i 4 , i = 2, . . . , l, (11.96) and the Hamiltonian of our problem is given by H(µ 2 , . . . , µ l , λ
2 , . . . , λ l ) =
2 m l i =2 µ 2 i M i (λ 2 , . . . , λ l ) . (11.97)
A set of independent first integrals is constructed by means of a remarkable formula due to Jacobi: l i
λ n i j =i (λ i − λ
j ) = 0, if n < l
− 1, 1, if n = l − 1. (11.98)
We leave its verification as an exercise. From equation (11.98), and substituting the definition (11.95) of M i into (11.97), we find that the following identity holds: l i =1 l −1 n =0 F n λ n i j =i (λ i − λ
j ) = 2 m l i =1 µ 2 i l k =1 (λ i − a
k ) j =i (λ i − λ j ) . (11.99)
1 An easy proof is provided by the computation of the residues of (11.92) considered as a rational function of λ.
446 Analytic mechanics: Hamilton–Jacobi theory and integrability 11.5 In this formula F l − 1 = H, while for the moment, F 0 , F 1 , . . . , F l −
are arbitrary. However, if we set l −
n =0 F n λ n i = 2 m µ 2 i l k =1 (λ i − a k ), i = 2, . . . l, (11.100)
from this system of equations we find F 0 , F 1 , . . . , F l −
as functions of λ and µ. These, together with F l −
, yield a set of l independent integrals of the motion which can be seen to be in involution. 11.5 Invariant l-dimensional tori. The theorem of Arnol’d Liouville’s theorem implies that if an autonomous Hamiltonian system with l degrees of freedom has l integrals that are independent and in involution, then the Hamilton–Jacobi equation has a complete integral and the equations of motion are integrable by quadratures. In this section we intend to study the geometry of invariant manifolds of integrable systems with several degrees of freedom. In particular, we prove the following theorem, which clarifies in which cases it is possible to give a global parametric representation of the manifold M a using l angular coordinates (in which case M a is diffeomorphic to a torus T l ). T heorem 11.5 (Arnol’d) Let H(p, q) be a given autonomous Hamiltonian system with l degrees of freedom and which has l first integrals of the motion f 1
l (p, q) that are independent and in involution. If the level manifold M a
l-dimensional torus. Remark 11.14 Sometimes M a has several connected components. In this case, Theorem 11.5 applies separately to each connected component. Remark 11.15 There exist Hamiltonian systems such that the level manifold M a is not com- pact and/or not connected. These systems satisfy the hypotheses of Liouville’s theorem, but not of the theorem of Arnol’d above. An important example is the case of linearised equations of a system with two degrees of freedom in a neighbourhood of a saddle point of the potential energy: H(p, q) = 1 2 [(p 2 1 + ω 2 1 q 2 1 ) + (p 2 2 − ω 2 2 q 2 2 )]. Setting f 1 = (p
2 1 + ω 2 1 q 2 1 /2), f 2 = (p
2 2 − ω 2 2 q 2 2 /2), M a is the Cartesian product of an ellipse (corresponding to the curve f 1 = a 1 in the (p 1 , q
1 ) plane) with two branches of the hyperbola (corresponding to f 2 = a 2 in the (p 2 , q
2 ) plane), and hence it is neither compact nor connected.
11.5 Analytic mechanics: Hamilton–Jacobi theory and integrability 447 Theorem 11.5 is a non-trivial extension of a very simple property, which we observed when l = 1 (the manifold M a reduces to the phase curve γ E ). The
proof of Theorem 11.5 can be omitted at a first reading. It is possible to skip directly to the following section, after reading the statement of Proposition 5.1 and the subsequent remarks. We devote this section to the proof of the theorem of Arnol’d and to its consequences. We have already remarked (see Remark 10.30) that the integrals f 1 , . . . , f l induce l Hamiltonian phase flows g t 1
t l that leave M a invariant. The idea of the proof of Theorem 11.5 is to use these flows to construct an atlas of the manifold M a , and then to prove that this atlas is compatible with the definition of the l-dimensional torus (see Examples 1.38 and 1.39). Choose t = (t 1 , . . . , t l ) ∈ R l , and consider the composition g t of the flows g t i : g t = g t 1 1 ◦ · · · ◦ g t l l . (11.101) Since {f i , f j } = 0, the flows commute, and g t does not depend on the order in which the individual flows are applied, but only on t. We hence define an l-parameter family of transformations from M a to itself, i.e. a map g : R l ×M a → M a , defined by g(t, x) = g t (x), satisfying the group conditions required by Definition 1.33. We then say that g t is an l-parameter group of transformations of M a , and that R l acts on M a through g t , and hence that g t defines an action of R l on M a . L emma 11.1 Let x 0 be any point of M a . The map g x 0
l → M
a , g
x 0 (t) = g t (x 0 ) is a local diffeomorphism (Section 1.7), and hence there exist an open neighbourhood U of t = 0 in R l and an open neighbourhood V of x 0 in M
a such
that g x 0 (U ) = V , and g x 0 restricted to U is a diffeomorphism (Fig. 11.7). Proof
Since the integrals f 1 , . . . , f l are independent, for every x ∈ M a
I∇ x f i (x)
∈ T x M a are linearly independent and are a basis of T x M
. Integrating along the directions of these vectors, it is possible to parametrise every point y ∈ V of a neighbourhood of x 0 ∈ M
a through t: y = y(t 1
l ) = g
t (x 0 ), (11.102)
where t = (t 1 , . . . , t l ) belongs to a neighbourhood U of 0 (note that x 0 = g
0 (x 0 )). The invertibility of the transformation is a consequence of the independence of the first integrals, which ensures that the determinant of the Jacobian matrix of the parametrisation (11.102) is non-zero. Indeed if, for example, M a is paramet- risable through the variables (q 1 , . . . , q l ) (hence p = ˆ p(a, q)) in the neighbourhood V of x
0 = (p
0 , q
0 ) = (ˆ
p(a, q 0 ), q 0 ), equation (11.102) can be written as q = q(t 1
l ) = g
t (ˆ p(a, q 0 ), q
0 ). (11.103) 448 Analytic mechanics: Hamilton–Jacobi theory and integrability 11.5
Download 10.87 Mb. Do'stlaringiz bilan baham: |
ma'muriyatiga murojaat qiling