Analytical Mechanics This page intentionally left blank
Download 10.87 Mb. Pdf ko'rish
|
x x 1
A O A = g s 2
t 1 (x), B = g t 1
s 2 (x) Fig. 10.3 while the other terms of degree 2 are zero. We now seek an explicit expression for it. D
1 , v
2 is the vector field w, denoted by w = [v 1 , v 2 ], with components w i
N j =1 (v 1 ) j ∂(v
2 ) i ∂x j − (v 2 ) j ∂(v 1 ) i ∂x j = L v 1 (v 2 ) i − L
v 2 (v 1 ) i . (10.141)
For the fields of Example 10.26 the commutator is [v 1 , v 2 ] = (0,
−1). Remark 10.28 For any function f we find: L v 1 L v 2 f − L v 2 L v 1 f = (v T 1 ∇ x )v T 2 ∇ x f − (v
T 2 ∇ x )v T 1 ∇ x f = N i,j =1 (v 1 ) j ∂ ∂x j (v 2 ) i ∂f ∂x i − (v
2 ) j ∂ ∂x j (v 1 ) i ∂f ∂x i = N i,j =1 (v 1 ) j (v 2 ) i ∂ 2 f ∂x i ∂x j + (v 1 ) j ∂(v 2 ) i ∂x j ∂f ∂x i − (v 2 ) j ∂(v
1 ) i ∂x j ∂f ∂x i − (v 2 ) j (v 1 ) i ∂ 2 f ∂x i ∂x j = w · ∇f, (10.142)
10.6 Analytical mechanics: canonical formalism 377 since the terms containing the second derivatives cancel each other. Hence we obtain the important result [L v 1 , L
v 2 ] = L w = L
[v 1 , v 2 ] , (10.143)
so that the commutator of the Lie derivatives Lv 1 , Lv 2 is the Lie derivative L w associated with the commutator [v 1 , v
2 ]. P roposition 10.6 Let ( ∆ f )(t, s, x) be defined as in (10.140). Then ∂ 2 ( ∆ f )
∂t∂s (0, 0, x) = (L [v 1
v 2 ] f )(x). (10.144)
Proof From equation (10.138) it follows that (L v
f )(g s 2 (x)) = ∂ ∂t t =0 f (g t 1 g s 2 (x)). Therefore, applying (10.138) to the function (L v 1 f ) we find (L v 2 (L v 1 f ))(x) = ∂ 2
t =s=0
f (g t 1 g s 2 (x)). Interchanging the order
of g t 1 and
g s 2 and subtracting we reach
the conclusion. For the fields of Example 10.26, we have L [v 1 , v 2 ] f =
−∂f/∂x 2 . To define the degree of non-commutativity δ we can now apply the operator L w to the functions x i and set δ = sup i =1,...,2l |L w x i |. Remark 10.29 Two flows commute if and only if their commutator is zero (see Arnol’d 1978a, section 39e). We now seek a characterisation of a pair of Hamiltonian flows which commute. This analysis has important consequences, which will be considered in the next chapter. To this end, we define D f = {f, ·},
(10.145) the first-order differential operator which to each function g : R 2l → R, g = g(x) associates its Poisson bracket with f : R 2l → R: D f g = {f, g}. If v
f denotes a Hamiltonian vector field associated with f , i.e. v f
I∇ x f, 378 Analytical mechanics: canonical formalism 10.6 we have
L v f = ( I∇ x f ) T ∇ x = −(∇ x f )
T I∇ x = −D f . (10.146)
We can now prove our main result. T heorem 10.18 Consider two Hamiltonian fields v i = I∇ x f i , i = 1, 2. The respective Hamiltonian flows g t i
{f 1 , f 2 } =
constant. Proof
Compute the commutator w of the two Hamiltonian fields. Following the definition (10.141) and applying equation (10.145), we obtain w i
∇ x {f 1 , f
2 }) i +l , 1 ≤ i ≤ l, (10.147)
w i = −(∇ x {f 1 , f
2 }) i −l , l + 1 ≤ i ≤ 2l, (10.148)
and hence w =
−I∇ x {f 1 , f
2 }. (10.149) From (10.149) and Remark 10.29 the conclusion follows immediately. D efinition 10.18 Two regular functions f 1 , f
2 : R
2l → R such that {f 1
2 } = 0
(10.150) are said to be in involution. Remark 10.30 From Theorem 10.18 it follows in particular that pairs of Hamiltonians in invol- ution generate commuting flows. If {f 1 , f 2 } = 0 then not only w = 0 but also L v 1 f 2 = L v 2 f 1 = 0.
Note that in Example 10.26 the two non-commuting flows have Hamiltonians that satisfy {f 1
2 } = x
1 , which is non-constant. For two uniform rectilinear motions in orthogonal directions in R 2 , generated by f 1 = x
1 , f
2 = x
2 , we have {f 1
2 } = 0 and hence commutativity. To conclude, we summarise the main properties of the Poisson brackets. T heorem 10.19 The Poisson brackets satisfy the following properties: (1) they define a bilinear skew-symmetric form on functions, i.e. (f, g) → {f, g}
is linear in both arguments and {f, g} = −{g, f}; (2) Leibniz’s rule: {f 1 f 2 , g } = f 1 {f 2 , g
} + f 2 {f 1 , g
}; (10.151)
(3) the Jacobi identity: {f, {g, h}} + {g, {h, f}} + {h, {f, g}} = 0; (10.152)
10.6 Analytical mechanics: canonical formalism 379 (4) if D
f and D
g are operators associated with f and g by equation (10.145) then [D f
g ] = D
{f,g} ; (10.153) (5) non-degeneration: if a point x 0 ∈ R 2l is not a critical point of f , there exists a function g such that {f, g}(x
0 ) =
/ 0. Proof
Properties (1) and (2) have a trivial verification, left to the reader. The Jacobi identity can be shown without any lengthy calculation, by noting that if we take the expansion of (10.152) we find a sum of terms each containing a second derivative of one of the functions f , g and h multiplied by the first derivatives of the other two. If we prove that in the expansion of (10.152) no second derivative of any of the three functions appear, than we prove that all these terms cancel out, and their sum is then equal to zero. On the other hand, if we consider for example the terms containing the second derivatives of h, we note that they necessarily come from the first two terms of (10.152). However, {f, {g, h}} + {g, {h, f}} = {f, {g, h}} − {g, {f, h}} = D f D
h − D
g D f h = [D
f , D
g ]h = [L
v f , L v g ]. (10.154) Remark 10.28 ensures that the commutator of two Lie derivatives is again a Lie derivative, and hence it does not contain second derivatives. The Jacobi identity is then proved. The property (4) is an immediate consequence of the Jacobi identity. Indeed, from (10.152) and (10.154) it follows that for any h we have D {f,g}
h = {{f, g}, h} = {f, {g, h}} + {g, {h, f}} = [D f , D
g ]h.
Finally, the non-degeneration property is an obvious consequence of the non- degeneration of the symplectic product. The Jacobi identity yields another interesting result. C orollary 10.6 If f and g are two first integrals then the same holds for {f, g}. Proof
If in equation (10.152) we set h = H, it follows from {f, H} = {g, H} = 0 that {{f, g}, H} = 0. Example 10.28 Recall the formula {L i , L j } = ijk L k of Example 10.25. Then Corollary 10.6 guarantees that if two components of the angular momentum of a point are constant, then the third component must also be constant.
380 Analytical mechanics: canonical formalism 10.7 10.7
Symplectic rectification The canonical transformations are a powerful tool, allowing the construction of new canonical variables with the aim of writing the Hamiltonian in some desired form. A significant example of such an application is given by the following theorem. T heorem 10.20 (Symplectic rectification) Let f(x) be a function C 1 in R
2l and x
0 a point which is not critical for f , so that ∇ x
0 ) =
/ 0. Then there exists a completely canonical transformation X = X(x), defined in a neighbourhood of x 0
f (X) = f (x(X)) = X i for some i. To understand the meaning of this theorem, and the technique of its proof, we start by analysing a simple non-trivial example. Example 10.29 rectification of the harmonic oscillator Consider the Hamiltonian H = 1 2 p 2 + ω 2 /2q
2 and the flow that it generates, given by ˙ p = −ω 2 q, ˙ q = p.
Endowed with the initial conditions p(0) = η, q(0) = ξ this gives p = η cos ωt − ωξ sin ωt, q = η ω sin ωt + ξ cos ωt. (10.155)
The retrograde flow (with Hamiltonian − 1 2 η 2 − ω 2 /2ξ 2 ) with initial conditions η(0) = p, ξ(0) = q is η = p cos ωt + ωq sin ωt, ξ = −
ω sin ωt + q cos ωt. (10.156) Our goal is to find a completely canonical transformation (in an open set excluding the origin), such that the new coordinate P is given by P =
1 2 p 2 + ω 2 2 q 2 . (10.157) Since the transformation is completely canonical, P is also the new Hamiltonian, so that the Hamilton equations yield the solution P = constant, Q = t
− t 0 . (10.158) After imposing equation (10.157), the problem is reduced to making a correct choice for Q = Q(p, q). Its value along the motion must coincide with time (up to translations). Hence in the plane (ξ, η) we take a regular curve ϕ(ξ, η) = 0, with {ϕ, − 1
η 2 − 1 2 ω 2 ξ 2 } = / 0 (so that the curve cannot be a trajectory of (10.156)). Fix a point (q, p) such that the trajectory (10.156) intersects the curve and denote
10.7 Analytical mechanics: canonical formalism 381 by t(p, q) the time of impact. If we denote the functions (10.156) by ξ(p, q, t), η(p, q, t), then the function t(p, q) is defined implicitly by ϕ[ξ(p, q, t), η(q, p, t)] = 0. (10.159) We now complete the transformation (10.157), that is independent of time, with Q = t(p, q) (10.160)
and check that the variables P, Q are canonical. Let us compute the Poisson bracket
t(p, q), 1 2 (p 2 + ω 2 q 2 ) = ∂t ∂q p − ∂t ∂p ω 2 q. (10.161) Equation (10.159) yields for the derivatives of t: ∂t ∂p = − ∂ϕ ∂ξ ∂ξ ∂p + ∂ϕ ∂η ∂η ∂p 1 D , ∂t ∂q = − ∂ϕ ∂ξ ∂ξ ∂q + ∂ϕ ∂η ∂η ∂q 1 D , where D =
∂ϕ ∂ξ ∂ξ ∂t + ∂ϕ ∂η ∂η ∂t = {ϕ(ξ, η), −H(η, ξ)} = / 0. Now it is easy to note that ∂t ∂q p − ∂t ∂p ω 2 q = − 1 D ∂ϕ ∂ξ p ∂ξ ∂q − ω 2 q ∂ξ ∂p + ∂ϕ ∂η p ∂η ∂q − ω
2 q ∂η ∂p = − 1 D ∂ϕ ∂ξ {ξ(p, q, t), H(p, q)} + ∂ϕ ∂η
Since the transformation (p, q) (η, ξ) is canonical and the Poisson brackets are preserved, −{ξ(p, q, t), H(p, q)} = {ξ, −H(η, ξ)} = ∂ξ ∂t
−{η(p, q, t), H(p, q)} = {η, −H(η, ξ)} = ∂η ∂t . We finally find t(p, q), 1 2 (p 2 + ω 2 q 2 ) = 1,
(10.162) which shows that the transformation is completely canonical. The time t(p, q) is also equal to the time, on the direct flow, necessary to reach the point (p, q) starting from the curve ϕ(ξ, η) = 0. Note that the curve ϕ(ξ, η) = 0 in this 382 Analytical mechanics: canonical formalism 10.7 procedure is arbitrary. Therefore there exist infinitely many transformations of the kind sought. For example, for ϕ(ξ, η) = ξ we find Q = (1/ω)arccot(p/ωq), for ϕ(ξ, η) = η we find Q = −(1/ω)arccot(ωq/p), for ϕ(ξ, η) = ωξ + η we find Q = (1/ω)arccot[(p − ωq)/(p + ωq)], and finally for ϕ(ξ, η) = ωξ − η we have Q = −(1/ω)arccot[(p + ωq)/(p − ωq)]. Each of these formulae, together with (10.157), gives a symplectic of the harmonic oscillator. Verify that in all cases, knowledge of integrals (10.158) leads, through the transformation, to the known integral of the harmonic motion. Proof of Theorem 10.14 By hypothesis, in a neighbourhood of x 0 at least one of the first derivatives of f is different from zero. We can assume without loss of generality that ∂f /∂p 1 = / 0. As in the example, we consider the direct flow ˙ p = −∇
f , ˙q = ∇ p f with initial conditions p = η, q = ξ and the inverse flow with Hamiltonian −f(η, ξ) and initial conditions η = p, ξ = q. The condition ∂f /∂p 1 =
∂ξ 1 /∂t = −∂f/∂η 1 = / 0 in the retrograde flow. Hence it is possible to express the function t(p, q) explicitly from the equation ξ 1
(10.163) As in the example, the function t(p, q) can also be defined in alternative ways, but for simplicity we consider only (10.163), keeping in mind that for every different choice of t(p, q) we obtain a different transformation. From (10.163) we deduce ∂t
i = − ∂ξ 1 ∂p i ∂ξ 1 ∂t , ∂t ∂q i = − ∂ξ 1 ∂q i ∂ξ 1 ∂t , with ∂ξ 1 ∂t = − ∂f ( η, ξ)
∂η 1 , and hence {t, f}
(p,q) = 1 ∂f /∂η 1 {ξ 1 , f
} (ξ,η)
= 1 (we used the invariance of {ξ 1
} passing from (p, q) to (η, ξ) and in addition {ξ 1 , −f} = ∂ξ
1 /∂t). Hence also on the direct flow the function t takes the values of time t. Therefore, if l = 1, the transformation P = f (p, q), Q = t(p, q) is the one we were seeking, and the theorem is proved. If l > 1, we set P 1
Q 1 = t(p, q) (10.164) and complete the transformation (independent of time) by defining P i
i = η
i (p, q, t(p, q)), Q i
i +l = ξ i (p, q, t(p, q)), i = 2, . . . , l. (10.165)
10.7 Analytical mechanics: canonical formalism 383 To verify that the transformation is canonical we must compute all the fundamental Poisson brackets. We start with {P i , P j }, with i, j = / 1: {P i , P j } = l k =1 ∂η i ∂q k + ∂η i ∂t ∂t ∂q k ∂η j ∂p k + ∂η j ∂t ∂t ∂p k − l k =1 ∂η j ∂q k + ∂η j ∂t ∂t ∂q k ∂η i ∂p k + ∂η i ∂t ∂t ∂p k = {η i , η j } − 1 ∂ξ 1 /∂t ∂η i ∂t {ξ 1 , η j } − ∂η j ∂t {ξ 1 , η i } .
Since (η, ξ) are canonical variables, then {η i , η j } = 0 for every i, j and {ξ 1 , η
i } = 0
for i > 1. Therefore {P i , P j } = 0 for i, j = / 1. The expression for {Q i , Q j } for i, j = / 1 is analogous, with ξ i and ξ
j in place of η i , η
j ; the conclusion is the same. The evaluation of {P i , f } and {Q
i , f
} with i > 1 is made easy by the fact that these are the derivatives of the functions η Download 10.87 Mb. Do'stlaringiz bilan baham: |
ma'muriyatiga murojaat qiling