Analytical Mechanics This page intentionally left blank
Download 10.87 Mb. Pdf ko'rish
|
p q 0
1
Natural motion Perturbed motion Fig. 9.4 Sketch of the synchronous perturbations. T heorem 9.4 A necessary and sufficient condition that δA = 0, to first order in the class of perturbations (9.44), is that q ∗ k , p ∗ k are solutions of the Hamilton equations (8.24). Proof We immediately find that δA = t 1 t 0 ( ζ · ˙q ∗ + p ∗ · ˙η + ζ · ˙η − δH) dt, (9.45) and since we are only interested in the first-order variation we can neglect the term ζ · ˙η and write δH (
q H) ∗ · η + (∇ p H) ∗ · ζ,
(9.46) where (
·) ∗ denotes the values taken along the natural motion. Integrating by parts the term containing ˙ η, we arrive at the expression δA = t
t 0 [ ζ · ( ˙q − ∇ p H) ∗ − η · ( ˙p + ∇ q H)
] dt. (9.47)
From this we can deduce the equivalence δA 0 ⇔ ( ˙q − ∇ p H) ∗ = ( ˙
p + ∇ q H) ∗ = 0, (9.48) if we proceed as in the final part of the proof of Theorem 9.1. Remark 9.5 The previous theorem is still valid if we restrict to the class of perturbations (9.44), imposing the limitations ζ k (t 0 ) = ζ k (t 1 ) = 0 (what needs to be modified in the proof is not essential).
316 Analytical mechanics: variational principles 9.5 9.5
Principle of the stationary action Besides Hamilton’s principle, there exist several other variational principles. 1 We will discuss only one more, the most famous, which has special interest for its geometric implications. This principle is called the principle of stationary action, or Maupertuis’ principle. It is valid for systems with a time-independent Hamiltonian. It is convenient to refer to the space (p, q, t) and to parametrise not only p and q but also t, thus considering the curves in R 2 +1
given by the equations p = p(u), q = q(u), t = t(u). To obtain a parametrisation of the natural motion, it is enough to consider a function t = t(u), u 0 ≤ u ≤ u 1 , in
C 2 [u 0 , u
1 ] with
t (u) = / 0 in [u 0 , u
1 ], and consequently define the functions q k (u) = q
∗ k (t(u)), p k = p ∗ k (t(u)). We find the curve of equations p = p(u), q = q(u), t = t(u), (9.49) along which we introduce the perturbations p = p(u) + ζ (u), q = q(u) + η (u), t = t(u) + τ(u), (9.50) in such a way that the new functions p(u), q(u), t(u) are also C 2 , and satisfy η(u 0 ) = η(u 1 ) = 0. The relevant novelty is that perturbations now include a variation in the temporal scale. Therefore they are called asynchronous perturbations (Fig. 9.5). We note that, in analogy with the case discussed in the previous section, only a subset of the curves (9.50) is associated with possible motions. However, every stationarity result obtained in this wider class applies to the subfamily of possible motions.
In what follows we select a particular subclass of perturbations, satisfying H(p(u), q(u)) = H(p(u), q(u)). (9.51) The asynchronous perturbations subject to the condition (9.51) are called isoen- ergetic. It will soon be clear that the need to introduce asynchronous variations is due to the constraint imposed on the energy. The functional we want to study is A =
t (u 1 ) t (u 0 ) p · ˙q dt, (9.52)
which is also called the action. The integrand must be understood in the Hamilto- nian formalism. This functional is obviously linked to the time average of the kinetic energy (see (4.34)). 1 See, for example, Levi-Civita and Amaldi (1927), Whittaker (1936) and Agostinelli and Pignedoli (1989). 9.5 Analytical mechanics: variational principles 317
1
q 0
t 0
1 Natural motion Perturbed motion Fig. 9.5 Sketch of the asynchronous variations. T heorem 9.5 (Stationary action of Maupertuis’ principle) If the Hamiltonian does not depend explicitly on time, the functional (9.52) along the natural motion is stationary with respect to the class of isoenergetic asynchronous perturbations. Proof We make the change of variables t = t(u) in (9.52) and we write A( ζ, η, τ) = u 1
0 (p +
ζ(u)) · d du (q + η(u)) du. (9.53) Neglecting the higher-order term ζ · d du η and integrating by parts where necessary, we arrive at the expression for the variation δA: δA u
u 0 ζn · d du q − η · d du p du.
(9.54) Using the Hamilton equations we find immediately that, to first order, ζ · d
q − η ·
d du p δH dt du . (9.55)
Since by hypothesis δH = 0, the proof is finished. 318 Analytical mechanics: variational principles 9.6 Remark 9.6 The functional (9.52) contains only information on the geometric-material struc- ture of the system. The dynamic information comes into play because of the isoenergetic constraint. Before examining the geometric consequences of this principle, we consider a few simple examples. Example 9.8: motion of a free point in the absence of forces Isoenergetic motions are in this case the uniform motions with the same mag- nitude of velocity as the natural motion. It is clear then that it is impossible to perturb the trajectory without perturbing the temporal scale. The functional A can be written as A = 1
mv t (u 1 ) t (u 0 ) v d t = 1 2 mvs, (9.56) where s is the length of the path travelled. The geometric interpretation of the principle of the stationary action is then that the natural motion is the motion which makes the length of the travelled path stationary with respect to any other path with the same velocity connecting the same start and end points. Example 9.9 motion of a point on an equipotential surface Let P be constrained on the surface V = constant. In this case as well the isoenergetic motions are the uniform motions with the same magnitude of velocity as the natural motion, and the conclusion is the same as in the previous case: the trajectory is a geodesic of the surface (Proposition 2.2). In fact, the minimality property of the path stressed by the previous examples holds in general, as long as the manifold of configurations is endowed with the appropriate metric. We shall develop this concept in the next section. 9.6
The Jacobi metric Consider a holonomic system with fixed, smooth constraints, not subject to any force directly applied to it. For such a system the kinetic energy is constant: T = constant. (9.57) Recall that T = 1 2 h,k =1 a hk ˙ q h ˙ q k is a positive definite quadratic form. We interpret (a hk ) as the metric tensor of the manifold of configurations of the 9.6 Analytical mechanics: variational principles 319 system, as we did in Theorem 4.3: (ds) 2 = h,k = 1
a hk dq h dq k . (9.58)
With this metric, the velocity of the representative point in the space is such that
| ˙q| 2 = 2T, (9.59) and hence | ˙q| = constant. We can apply the principle of stationary action and conclude that the natural motion is an extremal for the length of the path travelled on the Riemannian manifold
V endowed with the metric (9.58) (this is indeed the meaning of the action). On the other hand, note that in this case Theorem 9.2 refers to the extremal for the functional t 1 t 0 √ 2T dt, the Euler equations coinciding, for T = constant, with the Lagrange equations for L = T (see Problem 9.5). Turning to the general case, when there is a conservative force field with potential energy V (q), it is still possible to obtain an analogous result, as long as the chosen metric incorporates the function V (q) in a suitable way, at the same time preserving the information encoded in T . More precisely, we write (ds) 2
− V ) h, k
=1 a hk dq h dq k , (9.60) so that | ˙q| = 2T (9.61) and consequently the action coincides directly with the length of the arc of the trajectory travelled by the point in the space of configurations. The metric (9.60) is called the Jacobi metric and it is defined in the regions V E
{V (q) ≤ E}. For a fixed energy E, the manifold V E
nian manifold with boundary (∂ V E = {V (q) = E}), and from the Maupertuis principle it follows that the natural motion travels along the geodesics of this manifold. Note that the metric (9.60) is singular on ∂ V E
The following examples make reference to systems with two degrees of freedom. In the space (q 1 , q
2 ) we look for the trajectories of the form q 1 = f (q
2 ). Hence
the functional we have to study is of the form l(f ) =
b a [E − V (f, q 2 )] 1/2 a 11 df dq 2 2 + 2a
12 df dq 2 + a
22 1/2
dq 2 , 320 Analytical mechanics: variational principles 9.6 with a
ij functions of f and q 2 . After elimination of time, we can neglect the perturbations of this variable. Example 9.10 Verify that the trajectory of a central motion with potential energy V (r) is a geodesic with respect to the Jacobi metric (9.60). We seek the extremals of the functional ϕ 2 ϕ 1 [(ρ 2 + ρ
2 )(E
− V (ρ))] 1/2
d ϕ, (9.62) from which we obtain the Euler equation ρρ − 2ρ 2 − ρ
2 ρ 2 + ρ 2 + ρ V 2(E
− V ) = 0
(9.63) for the trajectory r = ρ(ϕ). We want to check that by integrating (9.63) we find the trajectory of the motion under consideration. Indeed, it is enough to note that, setting u = 1/ρ and ˆ V (u) = V (ρ), equation (9.63) becomes u + u
u 2 + u 2 = ˆ V (u) 2( ˆ
V − E)
, (9.64)
admitting the first integral u 2 + u 2 = k(E − ˆ V ),
(9.65) with k constant (it suffices to multiply the two sides of (9.64) by 2u ). This is simply the energy integral, after identifying k with 2m/L 2 z (see (5.27)). Indeed, by substituting (9.65) into (9.64) we find u + u = −
L 2 z ˆ V (u),
(9.66) and hence we obtain (5.26). This proves that the solution of the variational problem, i.e. the integra- tion of equation (9.64), is equivalent to the classical solution of the dynamical problem. Example 9.11: motion of a point mass in a one-dimensional field Choose the x 3 -axis in the direction of the field, and let V (x 3 ) be the potential energy, with V (0) = 0. We study the motion in the (x 1 , x 3 ) plane, with the initial conditions x 1 (0) = x 3 (0) = 0, ˙ x 1
01 , ˙
x 3 (0) = v 03 .
9.6 Analytical mechanics: variational principles 321 Since x
1 (t) = v
01 t and ˙
x 3 = ± −2/m (V (x 3 ) + v 2 03 ) 1/2 , we find by separation of variables in the latter t =
± x 3 0 − 2 m V (ζ) + v 2 03
d ζ. (9.67) The equation of the trajectory is then x 1 = ±v 01 x 3 0 − 2 m V (ζ) + v 2 03 −1/2 d ζ, (9.68) where the sign must be changed in correspondence to the possible singularities of the integrand. We now solve the problem using the variational technique considered in this section, by finding the extremal of the length of the arc of the trajectory with respect to the metric (9.60). Hence we find the function x 1 = ξ(x
3 ) which is an extremal of the functional (ξ) =
ζ 0 0 [ −2V (x
3 ) + m(v
2 01 + v 2 03 )] 1/2 [m(1 + ξ
2 (x 3 ))] 1/2
d x 3 , (9.69)
where ξ(0) = 0 and ξ(ζ 0 ) must coincide with the value taken by (9.68) for x 3 = ζ
0 . Since the integrand in (9.69) does not depend on ξ(x 3 ), the Euler equation admits the first integral [ −2V (x 3 ) + m(v
2 01 + v 2 03 )] 1/2 ξ (1 + ξ 2 ) 1/2 = c, i.e.
ξ = c[ −2V (x
3 ) + m(v
2 01 + v 2 03 ) − c 2 ] −1/2 . (9.70) To find the desired value of ξ( ζ 0 ) we take c 2 = mv 2 01 , and hence c = ± √ mv 01 , and the integral of equation (9.70) then coincides with (9.68). Example 9.12 Consider a rod AB constrained in the (x, z) plane, and with the point A sliding on the x 1 -axis, without any directly applied force. The rod has length 2 and mass m. We seek the equation of the trajectory in the Lagrangian coordinate space. Choose the coordinates ϕ, ξ = x/ as in Fig. 9.6. We compute the kinetic energy T = 1
m 2 ˙ ξ 2 − 2 ˙ξ ˙ϕ sin ϕ + 4 3 ˙ ϕ 2 (9.71) 322 Analytical mechanics: variational principles 9.6
Fig. 9.6
and deduce that the correct metric to use in solving the problem is d s 2 = d ξ 2 − 2 sin ϕ d ξ d ϕ + 4 3 d ϕ 2 . By seeking the trajectories in the form ξ = ξ(ϕ), we must find the extremal of the functional (ξ) =
ϕ 2 ϕ 1 ξ 2 − 2 sin ϕξ + 4 3 1/2 d ϕ. (9.72) The Euler equation admits the first integral ∂ ∂ξ
2 − 2 sin ϕξ + 4 3
= k, from which ξ = sin ϕ ± (1 − k
2 ) −1/2 4 3 − sin 2 ϕ 1/2 , with
|k| < 1. (9.73)
This equation leads to an elliptic integral. Classically we can solve the problem by writing the conservation of the kinetic energy: ˙
2 − 2 ˙ξ ˙ϕ sin ϕ + 4 3
ϕ 2 = c 0 (9.74)
and of the first component of the momentum: ˙ ξ − ˙ϕ sin ϕ = c 1 . (9.75) Note that we must have c 0 ≥ c
2 1 . Solving the system (9.74), (9.75) with respect to ˙ ξ, ˙
ϕ and eliminating time, we find (9.73), with k = c 1 /
c 0 . 9.7 Analytical mechanics: variational principles 323 9.7
Problems 1. Determine all plane curves of equation y = y(x) passing through the origin, and through the point with coordinates (π/2, 1) that are extremals for the functional π/ 2
[(y ) 2 − y 2 ] d x. 2. Consider all plane curves y = y(x) passing through two fixed points A and B. Show that the area of the surface of rotation obtained by rotating the graph of the curve around the x-axis is given by S(y, y ) = 2π x B
A y 1 + (y ) 2 d x. Show that the area is stationary if y = a cosh(x − b)/a, and hence a catenary. The constants of integration a and b are determined by requiring that the curve passes through the points A and B (depending on the relative position of the points, the solution may or may not be unique, or may not exist. Discuss all possible cases). (Hint : since the integrand is independent of x, use Remark 9.2(c).) 3. Determine the extremals of the following functionals, for fixed values of q(t
0 ), q(t
1 ): (a) t 1 t 0 (t ˙
q + ˙ q 2 ) d t; (b) t 1 t 0 (q 2 + ˙
q 2 − 2qt) d t; (c) t 1 t 0 ( ˙
q + t 2 ˙ q 2 ) d t. 4. Let h = R 2 → R be of the form h(x, x ) = (x − x ) 2 /2 + u(x), where u : R → R is of class C ∞ . Given any finite sequence of real numbers (x j , . . . , x k ),
j , . . . , x k ) =
k −1 i =j h(x
i , x
i +1 ). A (k − j)-tuple is minimal for h if h(x
j , . . . , x k )
j , . . . , x k )
j , . . . , x k ) such that x j = x
j , x
k = x
k . Prove that if (x j , . . . , x k ) is
minimal then it satisfies the following condition to be stationary: x i +1 − 2x
i + x
i −1 = u (x i ) for all j < i < k. Determine all the stationary n-tuples for the case u ≡ constant and u = ax. Which are the minimal ones? 5. Deduce from the principle of stationary action that the orbit of a point particle in a central force field of potential energy V (r) = 1 2 kr 2 , k > 0, is an ellipse with centre at the origin. 6. Within special relativity theory, the Lagrangian of a point particle with mass m (at rest) and in the absence of forces, is L( ˙ q) =
−mc 2 1 − (| ˙q| 2 /c 2 ), where c is the speed of light. 324 Analytical mechanics: variational principles 9.9 Determine the kinetic momentum p, the Hamiltonian H and show than, for any speed much smaller than the speed of light, H ∼ mc
2 + |p| 2 /2m. Write the Euler equations for the relativistic action functional S = L( ˙q)
d t and show that, in the case | ˙q|
c, they reduce to the equation m¨ q = 0.
9.8 Additional remarks and bibliographical notes Although the first studies in the calculus of variations date back to the seventeenth century, it was only in 1736 that Euler proved Theorem 9.1, which is still today considered the fundamental result in this field. The proof we gave is due to Lagrange, who obtained it in 1756. He also introduced the principle of stationary action, without, of course, using the Hamiltonian formalism. The problem of the additional conditions that a solution of the Euler equation must satisfy, in order to effectively provide the maximum or minimum of the functional, was successfully considered by Legendre, who gave an additional necessary condition. It was only in 1837 that Jacobi succeeded in strengthening the condition of Legendre to make it a sufficient condition, when he discovered the existence of conjugate points at which the minimisation problem loses uniqueness. A detailed discussion of these beautiful results goes beyond the scope of this work; for an elementary and pleasant introduction, we recommend Fox (1987, chapters 1–3). We simply note that if (1) the Euler equation is satisfied, (2) the interval of integration [t 1 , t
2 ] is sufficiently small, (3) the × matrix ∂ 2 F /∂ ˙
q i ∂ ˙ q j is either positive definite or negative definite, then there is a maximum or a minimum according to whether ∂ 2 F/∂ ˙ q i ∂ ˙ q j is negative or positive definite. This is enough to show that the Hamiltonian action (9.35) is minimised along the natural motion (for sufficiently short time intervals). 9.9 Additional solved problems Problem 1 Let S be a surface given as the graph z = f (x, y), with f ∈ C 2
2 ). Find the periodic function ρ(ϕ) > 0 such that the area of the portion of S projected in the region bounded by the curve r = ρ(ϕ) on the plane (x, y), with prescribed length , is an extremal. Solution
The length of the curve is = 2π 0 ρ 2 (ϕ) + ρ 2 (ϕ) d ϕ.
9.9 Analytical mechanics: variational principles 325 The area we are considering is A(ρ) = 2π 0 ⎧ ⎪ ⎨ ⎪ ⎩ ρ (ϕ) 0 EG − F 2 d r ⎫ ⎪ ⎬ ⎪ ⎭ d ϕ, where E, F, G are obtained by the following parametrisation of S: x = r cos ϕ, y = r sin ϕ, z = f (r cos ϕ, r sin ϕ). One verifies that EG − F
2 = r
2 (1 + (
∇f) 2 ), and hence the functional for which we seek an extremal is 2π 0 ⎧ ⎪ ⎨ ⎪ ⎩ ρ (ϕ) 0 r(1 + ( ∇f) 2 ) 1/2 d r − λ ρ 2 + ρ 2 ⎫ ⎪ ⎬ ⎪ ⎭ d ϕ. Note that if S is a plane, then ( ∇f) 2 is constant and A(ρ) is simply the area enclosed by the curve r = ρ(ϕ), and hence 1 2 2π 0 ρ 2 (ϕ)
d ϕ, divided by 1/(1 + ( ∇f)
2 ) 1/2 . In this case the problem is equivalent to that of Example 9.4. More generally, we must solve the Euler equation d d
λρ ρ 2 + ρ 2 + ρ(1 + ( ∇f) 2 ) 1/2 r =ρ(ϕ) − λρ ρ 2 + ρ
2 = 0
requiring that the solution is periodic, with period 2π, and determining λ using the constraint on length. As an example, in the case of a surface of rotation z = f (r) we find ( ∇f)
2 = f
2 and the above equation becomes λρ ρρ
2 − ρ
2 (ρ 2 + ρ 2 ) 3/2 + ρ(1 + f 2 (ρ))
1/2 = 0,
which admits the solution ρ = R 0 , with R 0 = /2π, as long as R 0 is inside the domain of definition of f ; indeed, it is enough to choose λ = (1 + f 2 (R 0 )) 1/2 /R 0 . In the case of the sphere f (r) = √ R 2 − r, to find the circular solution we need R 0
Problem 2 On the surface of rotation x = ρ(z) cos ϕ, y = ρ(z) sin ϕ, z = z
326 Analytical mechanics: variational principles 9.9 consider the family of elicoidal curves defined by ϕ = f (z), f (z
1 ) = 0,
f (z 2 ) = 2π, with f ∈ C
2 increasing, and the interval (z 1 , z
2 ) inside the domain of definition of ρ(z). Find f so that the length of the curve is stationary. Solution
The length of the curve is given by the functional (f ) =
z 2 z 1 [(1 + ρ
2 ) + ρ
2 f 2 ] 1/2
d z. Since the integrand does not depend on f , we can immediately write a first integral of the Euler equation: ρ 2 f = c[(1 + ρ 2 ) + ρ 2 f 2 ] 1/2
, c > 0,
(9.76) from which we find f : f = c
1 + ρ 2 ρ 2 − c
2 1/2
. The constant c has to be determined by imposing, if possible, c z
z 1 1 ρ 1 + ρ
2 ρ 2 − c 2 1/2 d z = 2π.
(9.77) If the surface has a vertex in z = z ∗ (i.e. ρ(z ∗ ) = 0) and z ∗ lies in [z 1 , z
2 ], then equation (9.76) is incompatible with f > 0, because it forces c = 0. In this case the problem does not admit a solution. Even when this is not the case, equation (9.77) is not always solvable. Take for example the cone ρ = zα, with opening angle α, for which f =
c zα 1 + α 2 z 2 α 2 − c 2 1/2
= γ z (1 + α 2 ) 1/2 α 1 z 2 sin 2 α − γ 2 1/2
, with γ = c(sin α)/α. Setting z sin α = ζ and
ζ = −γ 1 + t 2 1 − t 2 , the integral can easily be computed and yields f (z) = 2 α (1 + α 2 ) 1/2 arctan
z 1 sin α + γ z 1 sin α − γ 1/2
− arctan
z sin α + γ z sin α
− γ 1/2
. (9.78)
9.9 Analytical mechanics: variational principles 327 It is easy to see that the difference of the arctangents is positive for z > z 1 , and
it is always less than π/4. Hence the condition f (z 2 ) = 2π cannot be satisfied if α 2 > 1/15. If on the other hand there exists a solution, it is unique, as the right-hand side of (9.78) is an increasing function of γ in the interval (0, z 1 sin α). It is not at all intuitive that there may be cases when no solution exists. In fact a solution always exists, but when it is singular it cannot be found as a solution of the Euler equation. Indeed, in the class considered, a path that follows meridians (f = 0) and parallels (f singular) may be the most economical (in terms of length).
Problem 3 A point particle travels along the smooth curve z = −f(x) ≤ 0, 0 < x < a, in the vertical plane (x, z). The curve joins the two points (x = 0, z = 0), (x = a, z = 0). Initially the point is at (0, 0) with zero velocity, and its motion is periodic. Find the curve, in the family of curves of class C 2 (a, b) with fixed length > a, for which the period is an extremal. Solution Without the constraint on length, the curve would be a cycloid. The period is twice the travelling time along the curve between the points (0,0) and (a, 0). Conservation of energy implies that ˙s = 2gf (x). Since d s = 1 + f 2 d x we have
d t = ((1 + f 2 )/2gf )
1/2 d x. The period is then T (f ) = 2 a 0 1 + f 2 2gf 1/2 d x. We need to find the extremals of the functional T (f ) − λ
a 0 (1 + f 2 ) 1/2 d x, where
λ is the Lagrange multiplier. The corresponding Euler equation has first integral given by the Legendre transform of F (f, f ) = (1 + f 2 )
(2/gf ) − λ , i.e. f ∂F
− F = λ − 2 gf 1 1 + f
2 . Introducing the integration constant c we can write f 2 = c 2 λ − 2 gf 2 − 1, and separate variables: f (x)
0 d f c 2 λ − 2 gf 2 − 1 −1/2
= x, 0 < x <
a 2
328 Analytical mechanics: variational principles 9.9 (the branch a/2 < x < a is symmetric, and equal values of f correspond to opposite values of f ). Set λ − 2/gf = ζ, i.e. gf/2 = (λ − ζ) −2 and
d f =
−4/g(λ − ζ) −3 d ζ. This puts the indefinite integral in the form − 4 g (λ − ζ) −3 (c 2 ζ 2 − 1) −1/2 d ζ. The transformation (c 2 ζ 2 − 1)
1/2 = (cζ
− 1)t, i.e. ζ =
− 1 c 1 + t 2 1 − t 2 , d ζ =
− 4 c t d t (1 − t
2 ) 2 carries the integral to a rational form − 8 cg λ +
1 c 1 + t 2 1 − t 2 −3 d t 1 − t 2 . For f ↓ 0 we have ζ → −∞ and t → 1. The upper extremum can be deduced from t = ((cζ + 1)/(cζ − 1)) 1/2
with ζ expressed through f (x). We then obtain an implicit expression for f (x), where the constant c must be determined through the condition f (a/2) = 0, i.e. c 2 λ − 2/gf(a/2) 2 = 1. As usual the multiplier λ is found by imposing the length constraint. Problem 4 Prove that if for a functional ϕ(q) =
t 1 t 0 F (q(t), ˙ q(t), t) d t, with q ∈ C 2 ([t 0 , t
1 ], R), q(t 0 ) = q
0 , q(t
1 ) = q
1 , the Euler equation (9.12) becomes an identity, then ϕ does not depend on the integration path q but only on (t 0 , q 0 ) and (t
1 , q
1 ). Solution Writing explicitly equation (9.12) we find ∂ 2 F ∂ ˙
q 2 ¨ q + ∂ 2 F ∂q∂ ˙
q ˙ q + ∂ 2 F ∂ ˙ q∂t
− ∂F ∂q = 0. If this equation is an identity, i.e. it is satisfied by any q, necessarily the coefficient of ¨ q must be identically zero, because there is no other way to eliminate ¨ q. Hence F must be of the form F = a(t, q) ˙ q + b(t, q). Substituting in the equation, we find ∂a/∂t = ∂b/∂q, and hence the 1-form F d t is exact: F d t = a(t, q) d q + b(t, q) d t =
d f (t, q). From this it follows that t 1
0 F (q, ˙
q, t) d t = f (t 1 , q
1 ) − f(t 0 , q
0 ).
9.9 Analytical mechanics: variational principles 329 Problem 5 Consider the variational problem for the functional (9.9) in the function class (9.8). Find the necessary and sufficient condition for its solutions to also be solutions of the variational problem for the problem ψ(q) =
t 1 t 0 G[F (q, ˙q, t)] d t
∈ C 2 , G = / 0. Solution
Setting F(q, ˙q, t) = G[F (q, ˙q, t)], we immediately find d d
∇ ˙ q F − ∇ q F = ∇ ˙ q F d 2 G d F 2 d F d t . Hence the required condition is that the function F (q, ˙q, t) is a first integral of the Euler equation for the functional (9.9). 10 ANALYTICAL MECHANICS: CANONICAL FORMALISM 10.1 Symplectic structure of the Hamiltonian phase space Consider the real 2l × 2l matrix I = 0
1 0 (10.1) (with 1 and 0 we henceforth denote the identity and the null matrix, with the obvious dimensions, e.g. l × l in (10.1)). Note that I is orthogonal and skew-symmetric, i.e. I −1
I T = −I (10.2)
and that I 2 = −1. As observed in Chapter 8, setting x = (p, q), the Hamilton equations can be written in the form ˙x =
I∇ x H(x, t). (10.3) Example 10.1 Let S be a real symmetric constant 2l × 2l matrix. A linear Hamiltonian system with constant coefficients is a system of 2l ordinary differential equations of the form (10.3), where H(x) = 1
x T Sx. (10.4) The Hamiltonian is then a quadratic form in x and (10.3) takes the form ˙x = J Sx.
The solution of this system of differential equations with the initial condition x(0) = X is given by x(t) = e tB X, (10.5) where we set B = J S.
The matrices with this structure deserve special attention. 332 Analytical mechanics: canonical formalism 10.1 D
symplectic) if B T I + IB = 0. (10.6)
T heorem 10.1 The following conditions are equivalent: (1) the matrix B is Hamiltonian; (2) B =
IS, with S a symmetric matrix; (3)
IB is a symmetric matrix. In addition, if B and C are two Hamiltonian matrices, B T , βB (with β ∈ R), B ± C and [B, C] = BC − CB are Hamiltonian matrices. Proof From the definition of a Hamiltonian matrix it follows that IB = −B T
T , and hence (1) and (3) are equivalent. The equivalence of (2) and (3) is immediate, as S = −IB.
The first three statements of the second part of the theorem are obvious (for the first, note that B T =
and C = IR (with S and R symmetric matrices) we have [B, C] = I(SIR − RIS) and (S
T = −RIS + SIR. It follows that the matrix [B, C] is Hamiltonian. Remark 10.1 Writing B as a 2l × 2l block matrix B = a
c d , where a, b, c, d are l × l matrices, (10.6) becomes B T
c + c T −a T − d
a + d T b − b T , and hence B is Hamiltonian if and only if b and c are symmetric matrices and a T + d = 0. If l = 1, B is Hamiltonian if and only if it has null trace. Remark 10.2 From Theorem 10.1 it follows that the Hamiltonian matrices form a group (with
10.1 Analytical mechanics: canonical formalism 333 respect to matrix sum) called sp(l, R). If we identify the vector space of real 2l ×2l matrices with R 4l 2
l(2l + 1) (indeed, from what was previously discussed we may choose l(l + 1)/2 elements of the matrices b and c and, for example, l 2 elements of the matrix a). In addition, since the Lie product (or commutator ) [ , ] preserves the group of Hamiltonian matrices, sp(l, R) has a Lie algebra structure (see Arnol’d 1978a). D efinition 10.2 A real 2l × 2l matrix A is called symplectic if A T IA = I. (10.7) T heorem 10.2 Symplectic 2l × 2l matrices form a group under matrix mul- tiplication, denoted by Sp(l, R). The transpose of a symplectic matrix is symplectic. Proof Evidently the 2l × 2l identity matrix is symplectic, and if A satisfies (10.7) then it is necessarily non-singular, since from (10.7) it follows that (det(A)) 2 = 1. (10.8) In addition, it can be easily seen that A −1
−IA T I, (10.9) so that
(A −1 ) T IA −1 = (A T ) −1 I(−IA
T I) = (A
T ) −1 A T I = I, i.e. A −1 is symplectic. If C is another symplectic matrix, we immediately have that (AC)
T IAC = C
T A T IAC = C T IC = I. In addition, A T = −IA −1 I, from which it follows that A IA T = AA −1 I = I. Example 10.2 The group of symplectic 2 × 2 matrices with real coefficients, Sp(1, R), coincides with the group SL(2, R) of matrices with determinant 1. Indeed, if A = α
γ δ , the symplecticity condition becomes A T IA = 0 −αδ + βγ −βγ + αδ 0 = I. 334 Analytical mechanics: canonical formalism 10.1 Hence A is symplectic if and only if det(A) = αδ − βγ = 1. It follows that every symplectic 2 × 2 matrix defines a linear transformation preserving area and orientation. The orthogonal unit matrices (with determinant equal to 1) are a subgroup of SL(2, R), and hence also of Sp(1, R). Remark 10.3 Let A be a symplectic 2l × 2l matrix. We write it as an l × l block matrix: A = a
c d . (10.10) The condition that the matrix is symplectic then becomes A T
a T c + c T a −a T d + c
T b −b T c + d
T a −b T d + d
T b = 0 −1 1 0 , (10.11) and hence A is symplectic only if a T c and b T d are l
× l symmetric matrices and a
T d − c T b = 1. The symplecticity condition is therefore more restrictive in dimension l > 1 than in dimension l = 1, when it becomes simply det(A) = 1. It is not difficult to prove (see Problem 1) that symplectic matrices have determinant equal to 1 for every l (we have already seen that det(A) = ±1, see (10.8)). Remark 10.4 Symplectic matrices have a particularly simple inverse: from (10.9) and (10.10) it follows immediately that A −1 = d T −b T −c T a T . (10.12)
Remark 10.5 If we identify the vector space of the 2l ×2l matrices with R 4l 2 , the group Sp(l, R) defines a regular submanifold of R 4l 2
immediately in view of the conditions expressed in Remark 10.3; indeed, starting from the dimension of the ambient space, 4l 2 , we subtract 2(l(l − 1))/2, since the matrices a T c and b
T d must be symmetric, and l 2 since a
T d − c T b = 1.)
P roposition 10.1 The tangent space to Sp(l, R) at 1 is the space of Hamiltonian matrices: T 1 Sp(l, R) = sp(l, R). (10.13)
10.1 Analytical mechanics: canonical formalism 335 Proof
Let A(t) be a curve in Sp(l, R) passing through 1 when t = 0, and hence such that
A(t) T IA(t) = I (10.14) for every t and A(0) = 1. By differentiating (10.14) with respect to t we find ˙ A T IA + A
T I ˙
A = 0, from which, setting B = ˙ A(0) ∈ T
1 Sp(l, R)
B T I + IB = 0, and hence B ∈ sp(l, R). Conversely, to every Hamiltonian matrix there corresponds a curve in Sp(l, R), as shown in the following. P roposition 10.2 Let B be a Hamiltonian matrix. The matrix A(t) = e tB is symplectic for every t ∈ R. Proof
We must show that A(t) satisfies (10.7) for every t, i.e. (e tB ) T Ie tB = I. It follows immediately from the definition e tB = ∞ n =0 t n n! B n that (e tB ) T = e
tB T , and (e tB ) −1 = e −tB
. Hence the condition for the matrix to be symplectic becomes e tB
I = Ie −tB
. But
e tB T I = ∞ n =0 t n n! B T n−1 B T I = ∞ n =0 t n n! B T n−1
( −IB).
Iterating, we find e tB T I = I
∞ n =0 t n n! ( −1)
n B n = Ie −tB . D efinition 10.3 The symplectic product on a real vector space V of dimension 2l is a skew-symmetric, non-degenerate bilinear form ω : V × V → R. The 336 Analytical mechanics: canonical formalism 10.1 space V endowed with a symplectic product has a symplectic structure and V is a symplectic space. We recall that a bilinear skew-symmetric form is non-degenerate if and only if ω(v 1
2 ) = 0 for every v 2 ∈ V implies v 1 = 0. We note also that only vector spaces of even dimension admit a symplectic structure. Indeed, all bilinear skew-symmetric forms are necessarily degenerate in a space of odd dimension. Consider the canonical basis e 1 , . . . , e 2l in R
2l . The symplectic product ω has a matrix representation W obtained by setting W ij = ω(e i , e j ). Evidently the representative matrix W is skew-symmetric and the non-degeneracy condition is equivalent to det(W ) = / 0. Moreover, for every x, y ∈ R 2l
ω(x, y) = 2l i,j =1 W ij x i y j = x
T W y.
(10.15) By choosing the matrix W = I we obtain the so-called standard symplectic product (henceforth simply referred to as symplectic product unless there is a possibility of confusion) and correspondingly the standard symplectic structure. Remark 10.6 The standard symplectic product has an interesting geometric characterisation. Given two vectors x, y we have x T
1 y l +1 − . . . − x l y
+ x l +1 y 1 + . . . + x 2l y l = (x l +1 y 1 − x 1 y l +1 ) + . . . + (x 2l y
− x l y 2l ), corresponding to the sum of the (oriented) areas of the projection of the par- allelogram with sides x, y on the l planes (x 1 , x l +1 ), . . . , (x l , x
2l ). Hence, if p is the vector constructed with the first l components of x, and q is the one constructed with the remaining components, we have x = (p, q), and analogously if y = (p , q ), we have ω(x, y) = x T Iy = (q
1 p 1 − p 1 q 1 ) + . . . + (q l p
− p l q l ). (10.16) Note that in R 2 the symplectic product of two vectors coincides with the unique non-zero scalar component of their vector product. D efinition 10.4 Suppose we are given a symplectic product in R 2l . A symplectic basis is a basis of R 2l with respect to which the symplectic product takes the standard form (10.16), and hence it has as representative matrix the matrix I. Given a symplectic product ω,
symplectic basis
e 1 , . . . , e 2l = e p 1 , . . . , e p l , e q 1 , . . . , e q l satisfies ω(e q i , e q j ) = ω(e p i , e p j ) = 0, (10.17)
10.1 Analytical mechanics: canonical formalism 337 for every i, j = 1, . . . , l and ω(e q i , e p j ) = δ ij . (10.18) Remark 10.7 It follows that the choice of standard symplectic structure for R 2l coincides with the choice of the canonical basis of R 2l as symplectic basis. Using a technique similar to the Gram–Schmidt orthonormalisation for the basis in an inner product space, it is not difficult to prove the following theorem. T heorem 10.3 In any space endowed with a symplectic product it is possible to construct a symplectic basis. As for inner product spaces, it is possible to choose as the first vector of the basis any non-zero vector. Pursuing the analogy between an inner and a symplectic product, we can define a class of transformations that preserve the symplectic product, taking as a model the orthogonal transformations, which preserve the inner product. D efinition 10.5 Given two symplectic spaces V 1 , ω
1 and V
2 , ω
2 , a linear map S : V 1
2 is symplectic if ω 2 (S(v), S(w)) = ω 1 (v, w) for every v, w ∈ V 1
moreover S is an isomorphism, we say that S is a symplectic isomorphism. Remark 10.8 From Theorem 10.3 it follows, as an obvious corollary, that all symplectic spaces of the same dimension are symplectically isomorphic. A ‘canonical’ isomorphism can be obtained by choosing a symplectic basis in each space, and setting a correspondence between the basis elements with the same index. In particular, all symplectic spaces of dimension 2l are symplectically isomorphic to R 2l with its standard structure. T heorem 10.4 Let R 2l be considered with its standard structure. A linear map S : R 2l
2l is symplectic if and only if its representative matrix is symplectic. Proof This is a simple check: given x, y ∈ R 2l we have ω(Sx, Sy) = (Sx) T ISy = x T S T ISy, which is equal to ω(x, y) = x T Iy for every x, y if and only if S T IS = I. We conclude this section with the definition and characterisation of Hamiltonian vector fields (or symplectic gradient vector fields). These are useful in view of the fact that the Hamilton equations can be written in the form (10.3). 338 Analytical mechanics: canonical formalism 10.1 D
2l is Hamiltonian if there exists a function f (x, t) in C 2 such that X(x, t) = I∇ x
(10.19) In this case f is called the Hamiltonian corresponding to the field X and the field X is called the symplectic gradient of f . If X is Hamiltonian, the system of differential equations ˙x = X(x, t) (10.20)
is called Hamiltonian. The system of Example 10.1 is Hamiltonian. Remark 10.9 A Hamiltonian vector field determines the corresponding Hamiltonian f up to an arbitrary function h(t) depending only on time t. This arbitrariness can be removed by requiring that the Hamiltonian associated with the field X = 0 be zero. Remark 10.10 In R 2
Iv can be obtained by rotating v by π/2 in the positive direction. It is easy to check that, in R 2l ,
in a Hamiltonian field, for every fixed t, the Hamiltonian is constant along the lines of the field (Fig. 10.1). If the field is independent of time the Hamilto- nian is constant along its integral curves, i.e. along the Hamiltonian flow (recall equation (8.26)). It is essential to characterise Hamiltonian vector fields. This is our next aim. T heorem 10.5 A necessary and sufficient condition for a vector field X(x, t) in R 2l to be Hamiltonian is that the Jacobian matrix ∇ x X(x, t) is Hamiltonian for every (x, t). Proof
The condition is necessary. Indeed, if f is the Hamiltonian corresponding to X we have that ∂X i
j = l k =1 I ik ∂ 2 f ∂x k ∂x j , and hence the matrix ∇ x X can be written as the product of the matrix I and
the Hessian matrix of f , which is evidently symmetric. The condition is also sufficient: if ∇ x
setting Y(x, t) = IX(x, t), by (3) of Theorem 10.1, we have that ∂Y i
j = ∂Y j ∂x i . |
ma'muriyatiga murojaat qiling