Analytical Mechanics This page intentionally left blank
Download 10.87 Mb. Pdf ko'rish
|
n 1 +n 2 +···+n k =n ∂W (n 1 ) ∂χ ∂W (n 2 ) ∂χ · · · ∂W (n k ) ∂χ + · · · .
(12.22) Therefore to order k ≥ 2 we must solve the equation ω(J )
∂W (k)
∂χ (J , χ) + F (k) (J , χ) = H k (J ),
(12.23) where the functions F (k) are given by F (k)
= k i =2 1 i! d i H 0 dJ i n 1 +···+n i =k ∂W (n 1 ) ∂χ · · ·
∂W (n i ) ∂χ + k −1 i =1 1 i! ∂ i F ∂J i n
1 +···+n
i =k−1
∂W (n 1 ) ∂χ · · · ∂W (n i ) ∂χ (12.24) and thus contain W (n)
only with n < k. Equation (12.23) is exactly of the type (12.13). It then follows from Theorem 12.1 that H k
1 2π 2π 0 F (k) (J , χ) dχ, (12.25)
W (k)
(J , χ) = 1 ω(J ) χ 0 [H k (J )
− F (k)
(J , x)] dx. (12.26)
The uniqueness of the solution follows from the uniqueness of the Taylor series expansion in ε of H and W , and from Theorem 12.1. Remark 12.3 It is not difficult to prove the uniform convergence of the series expansions (12.19) and (12.20) under the assumption that H 0 and F are analytic functions of all their arguments, but the proof goes beyond the scope of this introduction as it requires some knowledge of the theory of analytic functions of one or more complex variables. Example 12.7 Consider the following quasi-integrable system with one degree of freedom: H(J, χ, ε) = cos J + ε J 2
sin 2 χ, and solve the Hamilton–Jacobi equation up to terms of order O(ε
3 ).
12.2 Analytical mechanics: canonical perturbation theory 499 Substituting the expansions (12.19) and (12.20) into cos ∂W ∂χ + ε 2 ∂W ∂χ 2 sin 2 χ = H (J , ε) and neglecting terms of order higher than 3 in ε we find cos
J + ε ∂W (1) ∂χ + ε
2 ∂W (2) ∂χ + ε 2 J + ε
∂W (1)
∂χ 2 sin 2 χ = H
0 + εH
1 + ε
2 H 2 . Then it follows that cos J − sin J ε ∂W (1)
∂χ + ε
2 ∂W (2) ∂χ − 1 2 ε 2 cos J ∂W (1) ∂χ 2 + ε 2 J 2 + 2εJ
∂W (1)
∂χ sin
2 χ = H
0 + εH
1 + ε
2 H 2 . Equating the terms corresponding to the same power of ε in the expansion and solving the resulting equations by using equations (12.25), (12.26), we find H 0 (J ) = cos J , H 1 (J ) = 1 4 J 2 , H 2 (J ) = − 1 64 (cos J )J 4 sin 2 J and W (1)
(J , χ) = − J 2 8 sin J
sin(2χ). We leave the computation of W (2) as an exercise for the reader. 12.2 Time periodic perturbations of one-dimensional uniform motions Consider a point particle of unit mass moving along a line under the action of a weak force field depending periodically on the position x of the particle and on time t. For simplicity of exposition, we systematically use dimensionless variables. The Lagrangian and the Hamiltonian of the system can then be written as follows: L(x, ˙ x, t, ε) = 1 2 ˙ x 2 − εV (x, t), H(p, x, t, ε) = p 2 2 + εV (x, t), (12.27) where ε is a small parameter, 0 ≤ ε 1, and V is the (generalised) potential of the applied force (which we assume to be non-constant and of class C ∞ ). 500 Analytical mechanics: canonical perturbation theory 12.2 The periodicity assumption implies that V (x + 2π, t) = V (x, t + 2π) = V (x, t), for appropriately normalised units of space and time. The periodicity in space of the force field acting on the particle yields that the x-coordinates of the particle that differ by an integer multiple of 2π are identifiable; the phase space of the system is therefore a cylinder (x, ˙ x) ∈ S
1 × R.
An example of a field that satisfies these assumptions is the case discussed in Example 2.4. If ε = 0, the velocity ˙ x = ω of the particle is constant. Since ε is small, it is reasonable to expect that there exists an invertible coordinate transformation, depending on time, which transforms the equation of motion ¨ x + εV
x (x, t) = 0, (12.28) where V
x = ∂V /∂x, into ¨ ξ = 0,
(12.29) and for which the velocity ˙ ξ = ω is a conserved quantity. Such a transformation certainly exists—if ω = / 0 and ε is sufficiently small—when V does not depend on time (it is easy to conclude this in view of the complete integrability of the associated Hamiltonian system; ξ is the angle variable corresponding to rotations). We therefore seek a transformation of the type x = ξ + u(ξ, t; ε), (12.30)
with u ∈ C
∞ , where (ξ, t) ∈ T 2
to ξ and t), transforming equation (12.28) into (12.29). Since for ε = 0 the two equations coincide, we must impose that u(ξ, t; 0) = 0, and hence u(ξ, t, ε) = O(ε). If we also require that ∂u ∂ξ
(12.31) the local invertibility of the transformation is guaranteed. Condition (12.31) is satisfied as long as ε is chosen sufficiently small. Differentiating equation (12.30) once with respect to t, and recalling that ˙ ξ = ω is constant, we find ˙ x = ω + (D ω u)(ξ, t; ε), (12.32) where D
ω denotes the linear partial differential operator of first order: (D ω
∂u ∂ξ (ξ, t; ε) + ∂u ∂t (ξ, t; ε). (12.33) Differentiating (12.32) again with respect to time, we find ¨ x = D
2 ω u = ω 2 ∂ 2 u ∂ξ 2 + 2ω ∂ 2 u ∂ξ∂t
+ ∂ 2 u ∂t 2 . (12.34)
12.2 Analytical mechanics: canonical perturbation theory 501 The equation of motion (12.28) then becomes (D 2 ω u)(ξ, t; ε) + εV x (ξ + u(ξ, t; ε), t) = 0. (12.35) Equation (12.35) is a partial differential equation of second order, which is non- linear because of the term V x (ξ + u, t). We try to solve equation (12.35) starting from the remark that this equation is identically satisfied by u ≡ 0 if ε = 0. Then we expand the function u in power series of ε: u(ξ, t; ε) = ∞ n
ε n u (n) (ξ, t).
(12.36) Each function u (n) (ξ, t) is periodic with period 2π, both in the space coordinate ξ and in time t. If we substitute the expansion (12.36) into the term V x (ξ + u, t) of equation (12.35) we find V x (ξ + u, t) = V x (ξ, t) + ∞ m =1 1 m! ∂ m ∂x m V x (ξ, t)(u(ξ, t; ε)) m = V
x (ξ, t) +
∞ m =1 1 m! ∂ m ∂x m V x (ξ, t) ∞ n =1 ε n u (n) (ξ, t)
m = V
x (ξ, t) +
∞ m =1 1 m! ∂ m ∂x m V x (ξ, t) × n 1 ,...,n m ε n 1 +···+n m u (n 1 ) (ξ, t) . . . u (n m ) (ξ, t). (12.37)
Hence, reordering the second sum in increasing powers n 1 + · · · + n m = n of ε we have V x (ξ + u, t) = V
x (ξ, t) +
∞ n =1 ε n n m =1 1 m! ∂ m ∂x m V x (ξ, t)
n 1 +···+n m =n u (n 1 ) (ξ, t) . . . u (n m ) (ξ, t).
(12.38) Substituting (12.38) and (12.36) into equation (12.35) we find ∞ n
ε n (D 2 ω u (n) )(ξ, t) + εV x (ξ, t)
+ ∞ n =2 ε n n −1 m =1 1 m! ∂ m ∂x m V x (ξ, t) n 1 +···+n m =n−1 u (n 1 ) (ξ, t) . . . u (n m
(ξ, t) = 0. (12.39)
502 Analytical mechanics: canonical perturbation theory 12.3 Therefore, equation (12.35) has a solution if the following infinite system of linear equations admits a solution: D 2 ω u (1) (ξ, t) + V x (ξ, t) = 0, D 2 ω u (2)
(ξ, t) + V xx (ξ, t)u (1) (ξ, t) = 0, D 2
u (3)
(ξ, t) + V xx (ξ, t)u (2) (ξ, t) +
1 2 V xxx (ξ, t)(u
(1) (ξ, t))
2 = 0,
. . . , D 2 ω u (n) (ξ, t) + P n (ξ, t) = 0, . . . , (12.40)
where V xx = ∂ 2 V /∂x
2 , V
xxx = ∂
3 V /∂x
3 and P
n is a function depending only on V , on its derivatives (up to order n) and on the functions u (1)
, u (2)
, . . . , u (n−1)
. We must hence study the linear equation (D 2
u)(ξ, t) = v(ξ, t), (12.41)
where v is a known function, periodic in x and in t. Equation (12.41) is a partial differential equation with constant coefficients on the torus T 2 , analogous to equation (12.39). Evidently, the existence of a solution of equation (12.41) is a consequence of the invertibility of the linear operator D ω which we will now discuss. 12.3 The equation D ω
We remark first of all that the eigenvalues λ and the eigenvectors u λ of the linear operator D ω = ω ∂ ∂ξ + ∂ ∂t (12.42) are given by D ω
λ = λu
λ , (12.43) and take the form u λ = e i(mξ+nt)
, λ = i(mω + n), (12.44) where (m, n) ∈ Z 2 . Hence if ω is an irrational number, the eigenvalue λ = 0 corresponds to the choice m = n = 0, and hence has multiplicity one. If, on the other hand, ω is rational, ω = j/k, the eigenvalue λ = 0 corresponds to the choice of (m, n) ∈ Z
2 such that mj + nk = 0 and therefore has infinite 12.3 Analytical mechanics: canonical perturbation theory 503 multiplicity. We shall now see that if ω ∈ R\Q, it is sufficient to impose the zero mean condition ˆ v
= 1 (2π) 2 2π 0 dξ 2π 0 v(ξ, t)dt = 0 (12.45)
to ensure the existence of a formal solution u of the equation D ω u = v. (12.46)
This means that it is possible to determine the coefficients of the Fourier series expansion of u, neglecting the question of its convergence. If, on the other hand, ω is rational, then it is necessary to impose infinitely many conditions, corresponding to the vanishing of all the coefficients ˆ v m,n
of the Fourier expansion of v with m, n such that mj + nk = 0. We will not study further the latter case, which would lead to the study of the so-called resonant normal forms. P roposition 12.1 If ω ∈ R\Q and if v(ξ, t) has zero mean (hence if it satisfies the condition (12.45)) there exists a formal solution u of the equation (12.46). The solution is unique if we impose that the mean of u be zero: ˆ u 0,0
= 0. Proof
Expanding both v and u in Fourier series, and substituting these series into (12.46), we find (m,n)∈Z 2
u m,n
i(mω + n)e i(mξ+nt)
= (m,n)∈Z
2 ˆ v m,n e i(mξ+nt) . Hence, by the uniqueness of Fourier expansions, it follows that for every (m, n) ∈ Z
2 we have
ˆ u m,n i(mω + n) = ˆ v m,n (yielding ˆ v 0,0 = 0), and therefore ˆ u m,n = ˆ v m,n
i(mω + n) if (m, n) = / (0, 0), while ˆ u 0,0 is undetermined. It follows that if ω is irrational and v has zero mean, there exists a unique formal solution u(ξ, t) = (m,n)∈Z 2
ˆ v m,n i(mω + n) e i(mξ+nt) (12.47) of D
ω u = v, and similarly, a unique formal solution w(ξ, t) = (m,n)∈Z
2 \{(0,0)}
ˆ v m,n −(mω + n) 2 e i(mξ+nt) (12.48)
504 Analytical mechanics: canonical perturbation theory 12.3 of D
2 ω w = v. We now discuss the convergence of the series (12.47) and (12.48). We start by remarking that the most serious difficulty is the need to control the denominators |mω + n| which can become arbitrarily small even if ω is irrational. T heorem 12.3 (Dirichlet) Let ω be irrational; there exist infinitely many distinct pairs (m, n) ∈ Z
2 \{(0, 0)}, m > 0, such that |mω + n| < 1/|m|. Proof Let M be a fixed integer, and consider the sequence {|mω + n|, m = 0, . . . , M}, where n
∈ Z is prescribed for every m in such a way that |mω + n| ≤ 1 (such a choice is always possible). The points of the sequence then belong to the interval [0,1] and are necessarily distinct, as ω is irrational. If we consider the decomposition of [0, 1] = M −1 j =0 j M , j + 1 M into M intervals, it follows that at least two points in the sequence |mω+n| must belong to the same subinterval [j/M, (j + 1)/M ] (indeed, there are M + 1 points and only M intervals). Denote these two points by m ω + n and m ω + n , and note that it is not restrictive to assume that 0 ≤ m − m ≤ M. Therefore |(m − m )ω + n − n | ≤ 1 M
1 m − m , and we have found one pair satisfying the claim. The existence of infinitely many such pairs (m, n) follows from a simple proof by contradiction. Suppose that (m 1 , n 1 ), (m 2 , n
2 ), . . . , (m k , n
k ) are all the solutions of |mω + n| < 1/|m|. Then if M is an integer such that |m j
j | >
1 M , j = 1, . . . , k, there exists a pair (m, n) ∈ Z 2
|mω + n| ≤ 1 M < 1 |m| , which is a contradiction. From this theorem it also follows that inf
(m,n)∈Z 2 \{(0,0)} |mω + n| = 0, and therefore 1/ |mω + n| is not bounded from above. This fact yields serious difficulties in the proof of the convergence of the series (12.47) and (12.48). This problem is called the problem of small divisors and was already known to Poincar´
e and the astronomers of the nineteenth century. 12.3 Analytical mechanics: canonical perturbation theory 505 We shall now see that it is possible to make some hypotheses on ω—verified by almost any ω with respect to the Lebesgue measure—under which it is possible to prove the convergence of the series (12.47) and (12.48). D efinition 12.2 We say that an irrational number ω satisfies a diophantine condition (with constant γ > 0 and exponent µ ≥ 1), and we denote it by ω ∈ C
γ,µ , if for every (m, n) ∈ Z 2
|mω + n| ≥ γ(|m| + |n|) −µ . (12.49) Remark 12.4 The need for the condition µ ≥ 1 in the previous definition is an immediate consequence of the theorem of Dirichlet, which guarantees that there do not exist diophantine irrationals with exponent µ < 1. P roposition 12.2 Let µ > 1 be fixed. The Lebesgue measure | · | of the set C γ,µ
∩ (0, 1) satisfies the inequality |C γ,µ ∩ (0, 1)| ≥ 1 − 4ζ(µ)γ, (12.50)
where ζ(µ) = ∞ k =1 k −µ is the Riemann zeta function computed at µ. Proof
Let R γ,µ
= (0, 1) \(C
γ,µ ∩ (0, 1)) be the complement in (0, 1) of C γ,µ . From the definition of C γ,µ
it immediately follows that R γ,µ = j,k
x ∈ (0, 1)||kx − j| < γ (k + j)
µ = j,k x ∈ (0, 1)| x − j k
γ k(k + j)
µ , with the conditions k ≥ 1, 0 ≤ j ≤ k. This yields |R γ,µ | ≤ ∞ k =1 k j =0 2γ k(k + j) µ ≤ ∞ k =1 k j =0 2γk −1−µ ≤ 4γ
∞ k =1 1 k µ · Evidently ζ(µ) = ∞ k
1/k µ
∞, since µ > 1. Equation (12.50) follows by observing that |C γ,µ
∩ (0, 1)| = 1 − |R γ,µ
|. From the previous result it follows immediately that γ> 0
γ,µ ∩ (0, 1) = 1, (12.51) and therefore for almost every ω ∈ (0, 1) there exists a constant γ > 0 such that ω ∈ C γ,µ . Note that if γ < γ, then C γ,µ ⊆ C
γ ,µ .
506 Analytical mechanics: canonical perturbation theory 12.3 Remark 12.5 It is not difficult to prove that if ω is an algebraic number of degree d ≥ 2, i.e. if ω ∈ R \ Q is a zero of an irreducible polynomial with rational coefficients and Download 10.87 Mb. Do'stlaringiz bilan baham: |
ma'muriyatiga murojaat qiling