Comments
Description
Transcript
Some extensions
22.3 SOME EXTENSIONS z b ρ −b (a) (b) a (c) Figure 22.4 Possible soap films between two parallel circular rings. surface area between z and z + dz is 1/2 , dS = 2πρ (dz)2 + (dρ)2 so the total surface area is given by b S = 2π ρ(1 + ρ )1/2 dz. 2 (22.11) −b Since the integrand does not contain z explicitly, we can use (22.8) to obtain an equation for ρ that minimises S, i.e. ρ(1 + ρ )1/2 − ρρ (1 + ρ )−1/2 = k, 2 2 2 where k is a constant. Multiplying through by (1 + ρ 2 )1/2 , rearranging to find an explicit expression for ρ and integrating we find cosh−1 ρ z = + c. k k where c is the constant of integration. Using the boundary conditions ρ(±b) = a, we require c = 0 and k such that a/k = cosh b/k (if b/a is too large, no such k can be found). Thus the curve that minimises the surface area is ρ/k = cosh(z/k), and in profile the soap film is a catenary (see section 22.4) with the minimum distance from the axis equal to k. 22.3 Some extensions It is quite possible to relax many of the restrictions we have imposed so far. For example, we can allow end-points that are constrained to lie on given curves rather than being fixed, or we can consider problems with several dependent and/or independent variables or higher-order derivatives of the dependent variable. Each of these extensions is now discussed. 781 CALCULUS OF VARIATIONS 22.3.1 Several dependent variables Here we have F = F(y1 , y1 , y2 , y2 , . . . , yn , yn , x) where each yi = yi (x). The analysis in this case proceeds as before, leading to n separate but simultaneous equations for the yi (x), ∂F d ∂F = , i = 1, 2, . . . , n. (22.12) ∂yi dx ∂yi 22.3.2 Several independent variables With n independent variables, we need to extremise multiple integrals of the form ∂y ∂y ∂y I= · · · F y, , ,..., , x1 , x2 , . . . , xn dx1 dx2 · · · dxn . ∂x1 ∂x2 ∂xn Using the same kind of analysis as before, we find that the extremising function y = y(x1 , x2 , . . . , xn ) must satisfy n ∂F ∂ ∂F = , (22.13) ∂y ∂xi ∂yxi i=1 where yxi stands for ∂y/∂xi . 22.3.3 Higher-order derivatives If in (22.1) F = F(y, y , y , . . . , y (n) , x) then using the same method as before and performing repeated integration by parts, it can be shown that the required extremising function y(x) satisfies n ∂F ∂F d ∂F ∂F d2 n d − + − · · · + (−1) = 0, (22.14) ∂y dx ∂y dx2 ∂y dxn ∂y (n) provided that y = y = · · · = y (n−1) = 0 at both end-points. If y, or any of its derivatives, is not zero at the end-points then a corresponding contribution or contributions will appear on the RHS of (22.14). 22.3.4 Variable end-points We now discuss the very important generalisation to variable end-points. Suppose, as before, we wish to find the function y(x) that extremises the integral b I= F(y, y , x) dx, a but this time we demand only that the lower end-point is fixed, while we allow y(b) to be arbitrary. Repeating the analysis of section 22.1, we find from (22.4) 782 22.3 SOME EXTENSIONS ∆y y(x) + η(x) y(x) ∆x h(x, y) = 0 b Figure 22.5 Variation of the end-point b along the curve h(x, y) = 0. that we require η ∂F ∂y b + a a b ∂F d − ∂y dx ∂F ∂y η(x) dx = 0. (22.15) Obviously the EL equation (22.5) must still hold for the second term on the LHS to vanish. Also, since the lower end-point is fixed, i.e. η(a) = 0, the first term on the LHS automatically vanishes at the lower limit. However, in order that it also vanishes at the upper limit, we require in addition that ∂F = 0. (22.16) ∂y x=b Clearly if both end-points may vary then ∂F/∂y must vanish at both ends. An interesting and more general case is where the lower end-point is again fixed at x = a, but the upper end-point is free to lie anywhere on the curve h(x, y) = 0. Now in this case, the variation in the value of I due to the arbitrary variation (22.2) is given to first order by b ∂F d ∂F ∂F b − η + η dx + F(b)∆x, (22.17) δI = ∂y a ∂y dx ∂y a where ∆x is the displacement in the x-direction of the upper end-point, as indicated in figure 22.5, and F(b) is the value of F at x = b. In order for (22.17) to be valid, we of course require the displacement ∆x to be small. From the figure we see that ∆y = η(b) + y (b)∆x. Since the upper end-point must lie on h(x, y) = 0 we also require that, at x = b, ∂h ∂h ∆x + ∆y = 0, ∂x ∂y which on substituting our expression for ∆y and rearranging becomes ∂h ∂h ∂h + y η = 0. ∆x + ∂x ∂y ∂y 783 (22.18) CALCULUS OF VARIATIONS x = x0 A x B y Figure 22.6 A frictionless wire along which a small bead slides. We seek the shape of the wire that allows the bead to travel from the origin O to the line x = x0 in the least possible time. Now, from (22.17) the condition δI = 0 requires, besides the EL equation, that at x = b, the other two contributions cancel, i.e. F∆x + ∂F η = 0. ∂y (22.19) Eliminating ∆x and η between (22.18) and (22.19) leads to the condition that at the end-point ∂F ∂h ∂F ∂h − = 0. (22.20) F − y ∂y ∂y ∂y ∂x In the special case where the end-point is free to lie anywhere on the vertical line x = b, we have ∂h/∂x = 1 and ∂h/∂y = 0. Substituting these values into (22.20), we recover the end-point condition given in (22.16). A frictionless wire in a vertical plane connects two points A and B, A being higher than B. Let the position of A be fixed at the origin of an xy-coordinate system, but allow B to lie anywhere on the vertical line x = x0 (see figure 22.6). Find the shape of the wire such that a bead placed on it at A will slide under gravity to B in the shortest possible time. This is a variant of the famous brachistochrone (shortest time) problem, which is often used to illustrate the calculus of variations. Conservation of energy tells us that the particle speed is given by ds v= = 2gy, dt where s is the path length along the wire and g is the acceleration due to gravity. Since the element of path length is ds = (1 + y 2 )1/2 dx, the total time taken to travel to the line x = x0 is given by x=x0 x0 1 ds 1 + y 2 = √ dx. t= v y 2g 0 x=0 Because the does not contain x explicitly, we can use (22.8) with the specific integrand √ form F = 1 + y 2 / y to find a first integral; on simplification this yields 1/2 2 y(1 + y ) = k, 784