4(xo, x, Po, P) = 0.

Transcription

1 128 MATHEMATICS: F. D. MURNAGHAN PRoe. N. A. S. THE PRINCIPLE OF MA UPERTUIS By FRANCIS D. MURNAGHAN DEPARTMENT OF MATHEMATICS, JOHNS HOPKINS UNIVERSITY Communicated December For a dynamical system whose kinetic energy is T and whose potential energy is V, Hamilton's principle tells us that the first variation of the time integral of the Lagrangian function L = T- V is zero along a trajectory of the system. The kinetic energy T is a quadratic function (not necessarily homogeneous) in the velocities and for a "natural" dynamical system L does not involve explicitly the time variable t. In this latter case the Principle of Maupertuis states that the Hamiltonian integral may be replaced by another which is homogeneous of the first degree in the derivatives of the coordinates with respect to the independent variable, but which involves the energy constant. It is the purpose of this note to show the real basis of this Principle of Maupertuis and to give explicitly the formulation of the integrand of Maupertuis for an arbitrary Lagrangian function which does not involve explicitly one of the coordinates. Denoting the coordinates of a dynamical system with n degrees of freedom by (x1,x2,..., x") the time t may conveniently be denoted by x and time derivatives will be indicated by a superposed dot. Let L be any function of the time x, the coordinates x = (xl, x2,..., x") and their time derivatives (x1, x2, 2. n). On introducing arbitrarily a new independent variable r by means of the equation x = x0(t) and denoting derivatives with respect to r by a prime, the integrand in the Hamiltonian integral becomes F(xO, x, x ' x) = L (x, x,) F is homogeneous of degree unity in the quantities (xo', x') so that the (n + 1) derivatives af af Pa P P7 = (r = 1, 2,..., n) (which constitute the momentum-energy covariant tensor) are homogeneous of degree zero in the n + 1 derivatives (xo', x') and hence involve these derivatives only through their n ratios. Eliminating these n ratios we obtain one or more relations of the type 4(xo, x, Po, P) = 0. (A)

2 VOL. 17, 1931 MATHEMATICS: F. P. MURNAGHAN 129 The homogeneous character of F enables us to write the Hamiltonian integral f/ldt =JFdr in the form J d~ xof + )d f (podx pdo) + where the summation symbol a runs over the values 1 to n. The variational principle may be then stated in the 2n + 2 dimensional (stateenergy) space in which the co6rdinates are (xo, x, po, p) but it is a constrained problem, as these coordinates must satisfy the relation (or relations) (A). The Euler-Lagrange equations for the problem so stated give (when there is only one relation) the usual canonical equations of Hamilton dx ao dxr N dpo dp,r p(b) dtr po dr apby dr bxo; dr = ocr( where,u is an undetermined multiplier. When there are s relations (A) these must be replaced by the equations dx af dx aa dpo 6a dpr _ a =_P(P( dr a po dt a')pr dr x dt t where the summation label a runs from 1 to s and there are now s undetermined multipliers (A, Au,...,,uA). In the dynamical problem it is assumed that the equations Pr = 6F = 6L (r = 1, 2,.,n) may be solved for the n quantities xt in terms of the n quantities Pr and on substituting these in the equation Po = a = L(xO, x, x) - XaP we obtain the single relation (p(xo, x, Po. P) = Po + H(xO, x, p) = 0 where H is the usual Hamiltonian function. On using this particular form for so it is dlear that the canonical equations (B) take the familiar form

3 130 MA THEMA TICS: F. D. MURNAGHAN PROC. N. A. S. dx ah H -_OH -=dr = dpo - dpr = A ' 1 ~~PrI rof~~xo =x Let us now consider, conversely, a scalar function qp(xo, x, Po, p) of the coordinates (x, x) and a covariant vector (Po, p) which has the value zero: lo(x0, x, Po,p) = 0. (C) The n + 1 partial derivatives ( constitute a contravariant vector and we write apo k' afr kr (D) apo k I bpr k where k is an undetermined factor of proportionality. Assuming that the determinant 2 (i, j = 0, 1,..., n) does not vanish identically these equations may be solved for (Po, p) in terms of the X/k and on substituting these in (C) the expression for k in terms of (xo, x, XO, X) follows. When this is done (po, p) appear as homogeneous functions of degree zero in the (XO, X). Denoting the scalar product Xapa (where the summation label runs over the values 0, 1, 2,..., n) by S: S = XaP we have ds = Xadpa + padx. On differentiating (C) and using (D) we have 0= dxa + a dpa= dxa + dpa so that ds = padda- k - a dxa. As the variables x and X are independent this gives as as as af as afo = axo Po; bxr ax = Pr; axo -x = -k ao' -'o 6 -x = -k as (X E) On comparing these equations with the -equations (B) we see that if we set X0 = _dx_ r = k dx MdT / dt we have

4 VOL. 17, 1931 MA THEMA TICS: F. D. MURNAGHAN 131 d (S\ a as d (as\ Os d\r xo'/ xo' dr\ax' ax' so that S is the homogeneous Lagrangian function corresponding to the Hamiltonian relation so(xo, xy po, p) = 0. In particular, if so has the form po + H(xO, x, p) we have ap = 1 so that k = XO and the equations -)Po ah Xr X0 bjpr Pr~~~~~x serve to determine (pi,.., Pr) in terms of the ratios X-. The remaining xo component po follows from po + H(xO, x, p) = 0. When one of the co6rdinates (xo, x) is missing from sp(xo, x, po, p) (i.e., is ignorable) the equations (B) show that the corresponding momentum is constant and the two equations of the set (B) which correspond to this coordinate and momentum may be omitted when solving for the remaining n x's and n p's. The Principle of Maupertuis is merely the application of this observation to the case where the ignorable coordinate is the time x and the Lagrangian function has the form L = T - V where T (in the usual formulation) = 1/2 g'ay;p and V is a function of (xl,..., Then =po + H(x, p) where H(x, p) = 1/2 gpapppp + V. On setting - = - we obtain grape = - so that Pr = g,ra k. Direct ajpr k k k substitution in p = 0 gives ex"x PO + '/2 g k-2 + = 0 so V that k = ~~~~~~2( + V) The homogeneous Lagrangian integrand is accordingly S=XPa= k =V -2(p+V). dx' Writing Xr = - and po = - h this gives the usual statement of the d,r Principle of Maupertuis. It is sufficiently evident from what we have written that the force of the principle is this: When a problem in the calculus of variations (one independent variable, n dependent variables), has a non-homogeneous integrand the problem can always be transformed

5 132 MA THEMA TICS: A. D. MICHAL PROC. N. A. S. so as to have a homogeneous integrand but at the expense of increasing the number of dependent variables by one (from n to n + 1). However when there is an ignorable coordinate the problem can be transformed into one with a homogeneous integrand with only n dependent variables, this integrand involving the constant momentum which corresponds to the ignorable coordinate. Since the parameter can be chosen arbitrarily when the integrand is homogeneous this enables us to take one of the coordinates themselves as the independent variable thus reducing the problem to one with n - 1 (instead of n) dependent variables. NOTES ON SCALAR EXTENSIONS OF TENSORS AND PROP- ERTIES OF LOCAL COORDINA TES By A. D. MICHAL DEPARTMENT OF MATHEMATICS, CALIFORNIA INSTITUTE OF TECHNOLOGY. Communicated December 22, The differential geometry of the n-dimensional group manifold of an n-parameter continuous group of transformations was, I believe, first studied by Cartan' and Schouten' in This group manifold possesses two teleparallelisms2 (distant parallelisms). There results a unique symmetric affine connection 1'a,, ry = 1/2 r (a + f).i~r 0. (1) Einstein's2 convention for Greek and Latin indices as well as his summation convention will be employed throughout our paper. A few exceptions will occur (such as those of formula (5) and (10) below) but this need not cause any confusion. Now the ennuple of contravariant vectors it' (the associated ennuple of covariant vectors is denoted by ita) is not arbitrary but satisfies the differential equations ita jt ir Ct kv (2) where ckj are the structural constants of the group. In the September, 1928, issue of these PROCEEDINGS,3 I published a paper entitled "The Group Manifold of Finite Continuous Point and Functional Transformation Groups." In this paper I developed the subject in such a manner as to bring into the foreground the various fundamental invariants in an arbitrary co6rdinate system. This was achieved with the