Analytical Mechanics This page intentionally left blank
Download 10.87 Mb. Pdf ko'rish
|
of degree d, then ω is diophantine with exponent µ = d − 1 (Liouville’s theorem). It is possible to prove that in fact all algebraic numbers are diophantine for every exponent µ > 1, independent of their degree (by a theorem of Roth, cf. Schmidt 1980). If ω satisfies a diophantine condition, and v is of class C ∞ or analytic, the series (12.47) and (12.48) converge uniformly and define a function of class C ∞ and analytic, respectively. T heorem 12.4 Let ω ∈ C γ,µ and v
∈ C ∞ . Then the series (12.47) and (12.48) converge uniformly. Proof
Consider the series (12.47) (a similar argument applies to the series (12.48)): (m,n)∈Z
2 \{(0,0)}
ˆ v m,n i(mω + n) e i(mξ+nt) ≤ (m,n)∈Z
2 \{(0,0)}
|ˆv m,n
| |mω + n|
≤ (m,n)∈Z
2 \{(0,0)}
|ˆv m,n
| γ ( |m| + |n|) µ . Since v ∈ C
∞ , for r > µ + 2 we have the inequality (cf. Appendix 7) (m,n)∈Z 2
|ˆv m,n
| γ ( |m| + |n|) µ ≤ M γ (m,n)∈Z 2 \{(0,0)}
1 ( |m| + |n|) r −µ
∞. To prove that (m,n)∈Z 2 \{(0,0)} 1 ( |m| + |n|) r −µ
∞ compare the series with the integral
∞ 1 ∞ 1 dx dy
(x + y) r −µ . We can now conclude our discussion of the example from which we started. Indeed, we have reduced the solution of equation (12.35) to the system of linear equations (12.40). Each of the equations in (12.40) has the form (12.41) and the previous theorem guarantees that if ω satisfies a diophantine condition with constant γ and exponent µ, and if V (x, t) is of class C ∞
system (12.40) admits a solution of class C ∞ (or analytic), so that the functions u (1) , . . . , u (k)
, . . . that are solutions of (12.40) exist and are functions of class C ∞ (or analytic) of (ξ, t). In fact, from the first equation of the system (12.40): D 2 ω u (1) (ξ, t) + V x (ξ, t) = 0, (12.52) 12.4 Analytical mechanics: canonical perturbation theory 507 and expanding u (1) and V in Fourier series: u (1)
(ξ, t) = (m,n)∈Z
2 \{(0,0)}
ˆ u (1) m,n e i(mξ+nt) , V (ξ, t) = (m,n)∈Z 2
ˆ V m,n e i(mξ+nt)
, (12.53)
we find −(mω + n) 2 ˆ
(1) m,n
+ im ˆ V m,n = 0, for every (m, n) ∈ Z 2
u (1)
(ξ, t) = (m,n)∈Z
2 \{(0,0)}
im ˆ V m,n (mω + n) 2 e i(mξ+nt) . (12.54) The regularity of u (1)
follows from the regularity of V and from the assumption that ω satisfies a diophantine condition. Since u (1)
is C ∞ , we can substitute this into the second equation of the system (12.40):
D 2 ω u (2)
(ξ, t) + V xx (ξ, t)u (1) (ξ, t) = 0. (12.55) One can check that V xx (ξ, t)u
(1) (ξ, t) has zero mean, and thus we can compute u (2)
, which is then of class C ∞ , and so on. We are still left with the more difficult problem of the convergence of the series (12.36). What we have seen so far only guarantees that each term in the series is well defined. The convergence of (12.36) under our assumptions (regularity of V and ω satisfying a diophantine condition) is guaranteed by the following theorem, whose proof is beyond the scope of this introduction (cf. Salomon and Zehnder 1989). T heorem 12.5 Let ω ∈ C γ,µ and suppose that V is analytic. Then there exists a unique solution u(ξ, t; ε) of (12.35) that is analytic in (ξ, t; ε). Moreover there exists a constant ε 0 > 0 such that the series expansion (12.36) of u(ξ, t; ε) converges uniformly with respect to (ξ, t) for all ε such that |ε| < ε
0 . The constant ε 0 of the previous theorem depends only on V and on ω. If ω = ( √
− 1)/2 and V = − cos ξ − cos(ξ − t), ε 0 has a value of approximately 0.03. The computation of ε 0 —and its physical significance—have been discussed, e.g. in Escande (1985). 12.4
Discussion of the fundamental equation of canonical perturbation theory. Theorem of Poincar´ e on the non-existence of first integrals of the motion We consider again the fundamental equation of canonical perturbation theory (12.13), and we show how the discussion of equation (12.46) extends to the more general case.
508 Analytical mechanics: canonical perturbation theory 12.4 Since the mean on the torus T l of the term ω(J ) · ∇ χ W (1) is equal to zero because of the periodicity of W , a necessary condition (which clearly is not sufficient) for (12.13) to have a solution is 1 (2π)
l T l (H 1 (J ) − F (J , χ)) dχ 1 . . . dχ l = 0,
(12.56) which allows the determination of H 1 as the mean of the perturbation: H 1 (J ) = 1 (2π)
l T l F (J , χ) dχ
1 . . . dχ
l = F
0 (J ),
(12.57) as we have already seen when l = 1 (cf. (12.15)). Fixing the values of the actions J , the linear operator D ω = ω · ∇
χ (12.58)
has constant coefficients. Its eigenvalues λ and eigenfunctions u λ (χ) are of the form λ = im
· ω, u λ = e im·χ
, (12.59)
where m ∈ Z
l and
ω = ω(J ) is the vector of frequencies. D efinition 12.3 The frequencies ω ∈ R l are called non-resonant if for every m ∈ Z
l , m =
/ 0, m · ω = / 0. (12.60)
Otherwise (hence if there exists m ∈ Z
l , m =
/ 0, such that m · ω = 0) the frequencies ω are said to be resonant. Example 12.8 The vector (1, √ 2,
3) ∈ R
3 is non-resonant, while 1, √ 2, 1/
√ 2 is resonant (for example consider m = (0, 1, −2)).
Remark 12.6 We could naturally examine the various possible kinds of resonance, and consider the associated modules of resonance (cf. Definition 11.7). This would lead us to the study of resonant normal forms, which goes beyond the scope of this introduction. If ω is non-resonant, the eigenvalue λ = 0 of D ω corresponds to the choice m = 0 and has multiplicity one. The fundamental equation of the canonical theory of perturbations is therefore formally solvable (neglecting the question of the convergence of the series arising when considering the Fourier expansions of W (1) and F ). T heorem 12.6 If ω is non-resonant, there exists a formal solution W (1) of equation (12.13). The solution is unique if we require that the mean of W (1) on the torus T l is zero: ˆ W (1)
0 = 0.
12.4 Analytical mechanics: canonical perturbation theory 509 Proof
Expanding both F and W (1)
in Fourier series (see Appendix 7): F (J ,
χ) = m∈Z
l ˆ F m (J )e
im·χ , W (1) (J ,
χ) = m∈Z
l ˆ W (1) m (J )e im·χ , (12.61) and substituting these expansions into (12.13) we find im · ω(J ) ˆ W (1)
m (J ) + ˆ
F m (J ) = 0, (12.62) for every m ∈ Z l
ˆ W (1) m (J ) =
ˆ F m (J ) −im · ω(J ) . (12.63)
The non-resonance hypothesis (12.60) guarantees that the denominators in (12.63) never vanish. When the Hamiltonian H 0 is linear in the action variables (harmonic oscillators) H 0 (J) = J · ω = l k =1 ω k J k , (12.64) the non-resonance condition is a hypothesis on the unperturbed system, and not on the values of the action variables, as the frequencies do not depend on the actions. However, in general the frequencies ω depend on the action variables, and hence contrary to the case of (12.64), the function ω(J ) is not constant, and the non-resonance condition will only hold on a subset of the phase space. D efinition 12.4 A Hamiltonian integrable system H 0 (J) is non-degenerate (in an open subset A ⊂ R
l ) if there exists a constant c > 0 such that for every J ∈ A, det
∂ 2 H 0 ∂J i ∂J k (J) ≥ c. (12.65)
If a system is non-degenerate, by the local invertibility theorem the map ω : A → R l ,
→ ω(J) = ∇ J H 0 (J),
is a local diffeomorphism. In this case, the hypothesis of non-resonance (12.60) selects some values of the action variables, and disregards others. Since the set of 510 Analytical mechanics: canonical perturbation theory 12.4 vectors of R l orthogonal to vectors of Z l is dense in R l , the resonance condition ω · m = 0 is satisfied for any m ∈ Z l , m = / 0 in a dense subset Ω r of R l : Ω r = m∈Z l m=0 {ω ∈ R l |ω · m = 0}. However, since the frequencies ω are in continuous one-to-one correspondence with the action variables, the resonance condition is satisfied by values of the actions J which belong to a dense subset A r of A:
A r = {J ∈ A|ω(J) ∈ Ω r } = m∈Z
l m=0
{J ∈ A|ω(J) · m = 0}. We shall see shortly (cf. Theorem 12.7) that the density of A r in A makes it impossible to define the canonical transformation generated by J · χ + εW
(1) as a regular transformation on an open susbset of the phase space, and it precludes the existence of analytic first integrals of the motion, independent of the Hamiltonian, in quasi-integrable systems (cf. Theorem 12.8). This was proved by Poincar´ e in 1893. D efinition 12.5 A function F : A×T l → R, F = F (J, χ) has a generic Fourier series expansion if for every J ∈ A and every m ∈ Z l there exists m ∈ Z l parallel to m such that ˆ F m (J) = / 0.
T heorem 12.7 (Poincar´e) If the integrable part of the Hamiltonian (12.4) is non-degenerate in an open set A and the perturbation F has a generic Fourier series expansion, the fundamental equation of perturbation theory (12.13) does not admit a solution W (1)
(J , χ) which is regular as the action variables vary in the open set A. Proof
The proof is by contradiction. Suppose that the fundamental equation of perturb- ation theory (12.13) admits a solution W (1) regular with respect to the actions. The non-degeneracy of the Hamiltonian H 0 guarantees the invertibility of the relation between the actions J and the frequencies ω, as well as the continuity of both transformations (from actions to frequencies and vice versa). The set Ω r of resonant frequencies is dense in every open subset of R l . It follows that the set A r of the J resonant actions, to which there corresponds a resonant frequency ω(J ), is dense in A. Therefore, for every J ∈ A there exists an action J ∈ A, arbitrarily close to J , and a vector m ∈ Z, m = / 0, such that m · ω(J) = 0 for m = m and for all vectors m = m parallel to it. From (12.62) it then follows that necessarily F m (J) = 0 and by continuity also that F m (J) = 0, and hence F m
a generic Fourier series expansion. The density of the set A r of the actions corresponding to resonant values of the frequencies has significant consequences for the problem of the existence of analytic first integrals, independent of the Hamiltonian. 12.4 Analytical mechanics: canonical perturbation theory 511 Consider the Hamiltonian quasi-integrable system (12.4) and seek a solution for the equation of the first integrals {I, H} = 0 (12.66) in the form of a power series in ε: I(J, χ, ε) =
∞ n =0 ε n I (n) (J,
χ). (12.67)
Substituting equation (12.67) into (12.66), taking into account the form (12.4) of H and equating terms of the same order in ε, we obtain an infinite system of equations for the (unknown) coefficients of the expansion (12.67) of the first integral sought: {I (0)
, H 0 } = 0, {I (1)
, H 0 } = {F, I (0) }, . . . . . . {I (n)
, H 0 } = {F, I (n−1) }. (12.68) We remark first of all that the Poisson bracket with H 0 is an operator of the form {·, H
0 } = ω(J) · ∇ χ ,
and hence it coincides with the operator D ω (12.58). Each equation of the infinite system (12.68) therefore has the form of the fundamental equation of canonical perturbation theory (12.13). We start by proving that the first of equations (12.68) implies that I (0)
does not depend on the angles χ. P roposition 12.3 If the Hamiltonian H 0 (J) is non-degenerate and I (0) is a
first integral that is regular for the Hamiltonian flow associated with H 0 , i.e. a regular solution of the equation {H 0 , I (0)
} = 0, (12.70)
then I (0)
does not depend on the angles χ, and hence I (0) = I
(0) (J).
Proof Assume that I (0) (J,
χ) is a solution of (12.70). Substituting the equation into the Fourier series expansion of I (0) :
(0) (J,
χ) = m∈Z
l ˆ I (0) m (J)e im·χ , we find i m∈Z
l (m · ω(J))ˆI (0) m (J)e im·χ = 0,
512 Analytical mechanics: canonical perturbation theory 12.4 and hence it follows that for every m ∈ Z l we have ˆ I (0) m (J)
≡ 0 or m · ω(J) ≡ 0. Differentiating the latter relation with respect to the actions, we find l i
m i ∂ω i ∂J k = 0, for every k = 1, . . . , l, which, when m = / 0, is satisfied only if det
∂ω i ∂J k = det
∂ 2 H 0 ∂J i ∂J k = 0, contradicting the hypothesis of non-degeneracy (12.65). It follows that the only non-zero Fourier coefficient is the one corresponding to m = 0 and the solutions of the first of equations (12.68) are necessarily of the form I (0) = I (0)
(J). We now use induction, and assume that we have solved equations (12.68) for I (1)
, . . . , I (n−1)
. Consider then the equation {I (n) , H 0 } = {F, I (n−1) }. (12.71) Indicating by F (n)
the term, known by the inductive hypothesis, which appears on the right-hand side, by expanding in Fourier series both I (n) and F
(n) we find the relation im · ω(J)ˆI
(n) m (J) = ˆ F (n)
m (J),
(12.72) which must hold for every m ∈ Z l
There are therefore two problems to be solved in order to prove the existence of a solution of (12.72). (a) We must prove that ˆ F (n) 0 (J)
≡ 0, and hence that {F, I (n−1)
} has zero mean value. This is immediate for n = 1 (since ˆ I (0)
is independent of χ and F is periodic in χ) but it is non-trivial for n ≥ 2. (b) We again need a non-resonance condition for ω(J) (unless ˆ F (n)
m (J) vanishes when m · ω(J) = 0) to guarantee at least the existence of a formal solution (still neglecting the problem of the convergence of the series). While the first problem can be solved generally by a more in-depth study of the series (cf. Cherry 1924a,b; Whittaker 1936, chapter 16; Diana et al., 1975), the second is at the heart of the non-existence theorem of Poincar´ e (Poincar´ e 1892, sections 81–3). 12.4 Analytical mechanics: canonical perturbation theory 513 D
there exists a non-constant analytic function g of one variable such that I = g(H). Otherwise, I is independent of H. T heorem 12.8 (Poincar´e) If H(J, χ, ε) is a Hamiltonian quasi-integrable system satisfying the same hypotheses as Theorem 12.6 (non-degeneracy and genericity), there does not exist an analytic first integral of the motion I(J, χ, ε) (for which the expansion (12.67) is therefore well defined and convergent if ε is sufficiently small, uniformly with respect to J ∈ A and χ ∈ T l ) which is independent of H. The proof of the theorem of Poincar´ e uses the following. L emma 12.1 An analytic first integral I, such that I (0) is independent of H 0 ,
independent of H, one can associate with it an analytic first integral ˜ I with ˜
I (0)
independent of H 0 . Proof If I depends on H, I (0) necessarily depends on H 0 . Indeed since I = g(H) = g(H 0
0 ) + εg (H 0 )F +
· · · . Comparing with (12.67) we find I (0) = g(H
0 ), proving the first part of the proposition. Now let I 0 be an analytic first integral that is independent of H and consider the power series expansion in ε: I 0 = I (0)
0 + εI
(1) 0 + ε 2 I (2) 0 + · · · . (12.73) We want to prove that if I (0) 0
0 , starting from I 0 one can
construct another first integral I, analytic and independent of H, for which I Download 10.87 Mb. Do'stlaringiz bilan baham: |
ma'muriyatiga murojaat qiling