You are on page 1of 10

Rend. Sem. Mat. Univ. Poi. Torino Voi. 54, 3 (1996) Geom. Struc. for Phys.

Theories, I

S. Benenti GEOMETRICAL ASPECTS OF THE DYNAMICS OF NON-HOLONOMIC SYSTEMS*

Abstract. Dynamics of non-holonomic mechanical systems is interpreted as a submanifold of TT*Q where Q is the configuration manifold. Integrability of dynamics is discussed for linear and non-linear constraints. The case of constrained geodesics of a Riemannian manifold studied by Synge is also considered. Locai coordinate representations are used. An example of an ideal non-linear non-holonomic constraint is proposed.

1. First order equations A first order (differential) equation on a manifold M is a submanifold D of the tangent bundle TM. A first order equation is said to be integrable if for each v E D there exists a differentiable curve 7: / • — * • M such that 7(0) = v and y(t) G D for each t G / , where y:I —*• TM is the tangent curve to'7 and I is an open real interval containing 0. Such a curve is called an integrai curve of D based on v è D. It is possible to extend this definition to the case in which D is a submanifold with boundary or a subset of TM. If D is a non-integrable first order equation, then the integrable part of D is the maximal subset of D which is integrable according to the definition above. If (xA) are locai coordinates of M, then we denote by (xA, xB) the corresponding fibered coordinates on TM. A first order equation D is locally described by a system of equations Da(xA,xB) =0 (A,B = 1 , . . . , m; m = dim(M); a — 1 , . . . , /; / = dim(D)). An integrai curve has a locai representation xA = yA(t) such that for each t D"(7A(t),jB(t))=0
""This paper was delivered at the Université de Savoie in 1987 and published in the proceedings of the "Journées Relativistes 1987", Chambery, May 14-16, 1987. As this volume is not easily accessible and some of the contributors to the present collection refer to this paper, it was thought to be useful to reprint it here.

204 S. Equations (1) follow from the d'Alembert-Lagrange principle. Hence.I' — * M are integrai curves based on v £ A then they coincide in the intersection / 0 / ' of the intervals of definition.. For each v € D there exists an integrai curve of D based on v. 2. but it is not unique. EXAMPLES. In this case the uniqueness property holds (Cauchy theorem): if 7: / — * •M and Y'. The image D = X(M) of X is an integrable first order equation. We cali (1) and (2) the Lagrangian representation and the Hamiltonian representation of the dynamics D respectively.Pk) the corresponding fibered coordinates on TQ. a differentiable vectorfield on M..e.e. a regular distrìbution on M. i. qh. and H:T*Q —>• M is the Hamiltonian function locally represented by a function of the coordinates (q\pj). Benenti where DyA is the derivative of the real function yA..Pj. i. M is the Lagrangian function locally represented by a function of the coordinates (q1. (1) A subbundle D of TM.. T*Q and TT*Q respectively.e. The dynamics of the mechanical system is the submanifold D of TT*Q locally defined by equations or by equations where L:TQ —>.pj) and (ql.TM -»• M. D is integrable with the uniqueness . run from 1 to n = dim(Q). is an integrable first order equation on M. (ql. Dynamics of holonomic systems A fundamental example of first order equation is given by the dynamics of holonomic mechanical systems. In the following discussion Latin indices ij.^). The manifolds TQ and T*Q represent the velocity space and the phase space of the mechanical system. We denote by (q\qi).k. h. Let Q be the configuration manifold of a holonomic mechanical system with n degrees of freedom. Equations (2) follow from equations (1) under the regularity condition The Hamiltonian representation D shows that the dynamics D is the image of a vector field X on the phase space T*Q. Lagrangian coordinates of the mechanical system). (2) Let X: M —• TM be a differentiable section of the tangent bundle rM'.. Let (q1) be locai coordinates on Q (i.

The distribution K can be represented by locai equations (1) K?V'=0 (a=l. /. dq> where x dpj ~ Wi' ixdO= '.e. The terms (5) Ri = Xa A7 represent the possible reaction forces of the constraints.f = «. api fìT (4) Pi + « .. In most of the applications K is a subbundle of TQ. A point of TT*Q belongs to D if and only if its coordinates satisfy equations (3) or (4) with some values of the parameters (Aa). 3. It follows from the D'Alembert-Lagrange principle that the dynamics D is the subset of TT*Q locally defined by equations (3) or by equations » . The vector field is globally defined by equation where 9 = pi dql is the fundamental 1-form of T*Q (the Lìouville form). Dynamics with non-holonomic linear constraints We assume that further constraints are imposed on the holonomic system. where (Aa) are the Lagrange multipliers. « f -0.= Aa A. The possible kinematical states of the system are represented by vectors v € TQ which belong to a subset K of T. a regular distribution on Q.. i.. If K is not completely integrable. then the constraints are called nonholonomic linear constraints.~ W -dH. The locai expression of the vector field X is = A A ~—r 4. oq° A? «• = 0. ./)..Geometrìcal aspects of the dynamics of non-holonomic systems property..À j — A — T J 205 ... where A'f are functions on the domain of the coordinates (ql) forming a matrix of maximal rank: (2) rank(A7) =..

The equivalence of the representations (1) and (6) implies that (8) Fla K? = 0. This proves 1.. The distribution K can be represented by parametric equations of the kind (6) 9'=X«. a straightforward calculation shows that the Jacobian matrix of the functions at the left sides of (6) and (9). oql . PROPOSITION 4.u«) + \a K?(xi). has maximal rank. The image by the tangent fìbration TT*Q:TT*Q -*T*Q of the first order equation D considered in Section 3 is the subset C C T*Q locally defined by equations (1) Ca = KfH*=0 which are obtained by combining the first and the third set of equations (4.§2) is equivalt to (2) det(Hij) ^0. The system of equations (6) and (9) gives the parametric representation of the dynamics D.§2) and (2) the subset D C TT*Q defined by equations (3) is a submanifold of dimension 2n. . Benenti The Lagrangian representation (3) of the dynamics D is suitable for proving that D is a submanifold. The integrability will be discussed in the next section. The 2n parameters (x\ \a. Under the regularity conditions (3. pi = gi{xi. .206 S. OH H% = T . dpi etc. The parameters (x — q . . n ) J where the functions F%a form a matrix of maximal rank: (7) 7 l a rank(FÌ) =' „ .ua) can be interpreted as coordinates on D and equations (6) and (9) as representing a locai immersion of D into TT*Q. Coordinates (uja) are known in the classical literatue as "pseudo-velocities".§3).uja). . The integrability theorem and the elimination of the Lagrangian multipliers For the sake of simplicity we shall use the following notation: „ OH Hi = • F 7 ./.u ) can be interpreted as coordinates on K and equations (6) as representing a locai immersion of K into TQ. < f *= i ^ > » (« = / + l . . The substitution of (6) into the first two sets of the Lagrange equations (3) yields equations of the kind (9) Pi = /*(*>>«). The Hamiltonian representation (4) is suitable for discussing the integrability of D. Indeed. with respect to the variables (x\Xa. The regularity condition (3.

REMARE . i. Hence. C = A(A').•} denotes the canonical Poisson bracket of functions on the cotangent bundle T*Q. This is the case of the ordinary holonomic mechanical systems. Let 7 : / —• T*Q be an integrai curve of D. Hence. (ii) The explicit form (6)-(7) of the Lagrangian multipliers has been derived by Eden [7] by a different method. then the integrable part of the first order differential equation D defined by equations (4. PROPOSITION 1. i.§3)./. the intersection D fi TC coincides with the integrable part of D. Cb} = Kf (Hij Hj .1 .§3). obtained from (1) by formai derivation. the subset C is a submanifold of T*Q of dimension 2n .Geometrìcal aspects of the dynamics of non-holonomic systems 207 The regularity eonditions (2. for each point p G C there exists one and only one element of D belonging to the intersection D(~)TPC. We obtain (6) where \a^GahL\ j Lb = {H.§3) can be interpreted as the locai definition of a fibre bundle isomorphism A: THQ —• T*Q. Hence. we can solve equations (4) with respect to the multipliers (A a ). The first set of the Lagrange equations (3.§3).Kfj H* H>. \(Gab) = (Gab)~\ Gab = HiiK?I<l In the first equation {•. The image y(I) is contained in C.Hj W) . Under the condition (5) det (Hij Kf K)) £ 0.Hij Hj) + Kf K] Hij \b = 0. This proves: 2.e. the intersection DC\TC is characterized by equations (4) Kfj H{ Hj + Kf (H) Hj .§3) and (2) imply that the / functions at the left side of (1) are independent. The first set of the Hamilton equations can be interpreted as the locai definition of A .TC —*• C. (2) and (5) are saisfied.13) is the image of a vector field X in the submanifold C C T*Q defined by equations (1). (i) If the quadratic form defined by the matrix (H**) is positivedefinite. Hence. then the regularity condition (5) is a consequence of condition (2. The submanifold TC is defined by equations (1) and equations (3) q> (Kfj H* + Kf H)) + j>j Kf H* = *. Since the submanifold D is defined by equations (4. Hence. It follows that the integrable part of D is contained in the intersection DC\TC. If the regularity eonditions (2. the image of a vector field X on C. This means that DC\TC is the image of a section X:C —> TC of the tangent fibration TQ'. the inage y(I) of the tangent curve 7: / —• TT*Q is contained in TC.. of the Legendre transformation.e..

Hence. 2. The image Y (K) of the vector field Y. (11) Then the vector field X is intrinsically defined by equation (12) where 0 is the Liouville 1-form.. i. REMARK iYdO= -dH + fi. which is a submanifold of TK. The Hamilton-Jacobi method for integrating first order equations can be applied only to Hamiltonian vector fields on cotangent bundles. the solutions of the following system of differential equations PROPOSITION T = H\ Xi= -Hi + GahLaKl whose initial conditions satisfy equations (1). is locally represented by equations involving the coordinates (q\ ua. then we can find a locai integrai F :: J — • IR of 0| J and a locai extension F of this REMARK . to vector fields Z such that iz d0 is an exact form.. This vector field is the geometrie representation of the gibbs-Appel equatios or the Maggi-Volterra equations. 5.e. It follows from Remarks 3 and 4 that if we know a submanifold J C C such that (i) X is tangent to J. but within the Lagrangian formalism. equations (1) are invariant relations of the differential system (10). According to the classical terminology. Le. the vector field X can be interpreted as the restriction to C of the vector field It on T*Q whose components are (9) This proves 3. Any other extension can be chosen for finding the integrai curves of D.] 4. ql. To the vector field À' defined above we can add any arbitrary (smooth) vector field vanishing on the submanifold C. have been obtained by Synge [13] (see also Agostinelli [1]) for quadratic Hamiltonians. REMARK REMARK 3. The integrai curves of the first order equation D are the integrai curves of the vector field X based on the points of C.ùa) of TK.208 S. [Let us introduce on T*Q thè vertical 1-form <P = GabLaKfdqi. Pi + Hi = GabLaKl A?<f = 0. as we shall see in the next section. Benenti The intersection D fi TC is defined by equations (8) «•"-#•=(). Analogous results. By the Legendre transformation h-l\T*Q -+ TQ the vector field X:C -»• TC is transformed into a vector field Y: K — • TK over the subbundle K C TQ. (ii) the pull-back of <f> to J is closed.

6. Let U be the domain of the coordinates (q1). (6) det (Hij) ^ 0.^ = <». If the constraints are linear. Dynamics vvith non-Iinear non-holonomic constraints When the kinematical constraints are represented by a submanifold K of TQ defined by locai equations (1) K"(q'. If the regularity conditions rank(At) = /. then equations (2) and (3) reduce to equations (3) and (4) of §3 respectively. are satified. .e. . it can be shown that PROPOSITION 4.Geometrical aspects of the dynamics of non-holonomic systems 209 function on T*Q. (iì) the subset C = TT*Q(D) CT*Q locally defined by equations (7) Ca = 0. We mention here the recent articles by Vershik [14] and Weber [15] on geometrical approaches to non-linear constraints which generalize this assumption. i. we can apply the Hamilton-Jacobi method to the vector field Z for finding the integrai curves of X\J.Bf Hl = {H. . then: (i) the subset D C TT*Q defined by equations (2) is a submanifold of dimension 2n.dKa A „ a = ' where (A a ) are the Lagrange multipliers. . where the symbol |* denotes the substitution ql = Hl. OH =A " 'l?- A =0 r" n 0 n (3) ^ ^ = 0 n ' « + 5? = -!?•'• A ^ .H] Hj) . Ca]. By a procedure analogous to that of §3 and §4.q') =0 (a = l . 0 . Equations (2) follow from the Gauss principle (see Prange [11]). The Hamiltonian vector field Z generated by H' = H — ~F is such that Z\J = X\J.. Gab =AfAbjHij (5) La = Aai (Hij Hj . Hence. it is assumed that the dynamics is the first order equation D C TT*Q locally defined by equations to\ dL n • dL Ì dK (2) or by equations . . det (Gab) # 0. Then in the open subset T*U C T*Q we can define the following functions: (4) Aa = dq % B? = dK dq " i Ca = Kal. • dH «-a? . the motions of the non-holonomic system lying on J.oX « .

[For homogeneous (non-linear) constraints the regularity conditions (6)2 are not fulfilled for q% = 0. however. Fufaev-Neimark [9] and Pironnau [10]). when the mechanical system is at rest (see the example below).] It seems that no mechanical system with ideal non-linear non-holonomic constraints is known other than that of Appell (see Fufaev and Neimark [9]). but its construction seems to be rather complicated. Caratheodory [2] pointed out that if the constraints are homogeneous then no work is done by the reaction forces (see also Saletan and Cromer [12]).210 S. which arises when one of the bars is orthogonal to the skates. To guarantee that the two "skates" r\ and r2 remain parallel we constrain four points (Ai. We propose here a simple example.C2. when certain masses ànd distances become negligible. Remarks analogous to those of §4 hold for non-linear constraints.b.d) respectively. The equations (1) can be chosen to be homogeneous in the coordinates (ql).c. Benenti is a submanifold ofdimension 2n — l..I>i) of r\ and corresponding four points (A2ÌB2.D2) of r2 to slide without friction along four rigid bars (a. i.. suffers of some defects (for criticisms and discussions we refer to Delassus [6]. whose centers of masses P{ and P'2 can move along the skates slightly from the midpoints P\ and P2 respectively. These four bars can pivot without friction around a common point P which moves freely in the piane. A non-holonomic constraint is called homogeneous if v E K implies rv E K for each real number r. The corresponding singular points of D should be analyzed more closely: in general the integrability is preserved but not the uniqueness. and we disregard the masses of ali the components of the device]. The use of four bars instead of three avoids a certain singularity in the construction. then we have constructed a system of two material points Pi and P2 which move in a piane and are constrained to have parallel velocities. by a (single) scalar EXAMPLE. leaving to further investigation the question as to wheter it is realistic or not and whether it confirms the theory or not. Le. Castoldi [3]. rods ri and r2 move on a piane in such a way that the rods and the velocities v\ and v2 of the midpoints Pi and P2 remain parallel. It is represented by equation v\ x v2 — 0.5i. Then the constraints are said to be ideal. (Hi) the integrable pan of D is the image of a vector field X on C. (iv) the vector field X is the restriction of a vector field X onT*U with components (8) T = H\ Xi = -Hi + Gab LaAl where (Gab) ìs the inverse matrix of(Gab). in order that their velocities and those of the skates remain parallel. This example.Ci. At each conflguration the two skates ri and r2 are in a symmetrical position with respect to the point P. This constraint can be produced by installing a sharp wheel or a sharp biade (as in an ice skate) at the center of each rod.e. TWO identical . This constraint is nonlinear and homogenous. Castoldi proposed a different example. If we consider [on each skate an heavy small body. Hertz pointed out that non-linear constraints can be realized as a limit of linear constraints. In fact.

U. Athenes 1934.Geometrical aspects of the dynamics of non-holonomic systems 211 homogeneous quadratic equation in the components of the velocities. Norm. Paris 256 (1963). Rend. Die allgemeinen Integrationsmethoden der analytischen Mechanik. A variational principle for nonholonomic systems. 564-583. Proc.J. this singularity is first of ali due to the construction. A. CASTOLDI L. Rend. 221-228. C. Physics 38 (1970). 211-214. Soc. [11] [12] . Am. Hermann.. Translations of Math.. U. the regularity condition (6) 2 is not satisfied for v\ = v2 = 0. Dynamics of non-holonomic systems. CATTANEO C .M. EDEN R.S. Torino 1982. Sulle equazioni di moto di un sistema soggetto a un vincolo anolonomo mobile.H. CARATHEODORY C . Proceedings of the IUTAMISIMM Symposium on "Modem Developments in Analytical Mechanics". Ec. J. 351-370. 1 (1956).R. FUFAEV N. DELASSUS E^Legons sur la dynamique des système matériels. 22 (1963).I. PRANGE G.A.M.I. 33 (1912). If we leave the two points at rest in a configuration. Ili 3 (1947). London 205(1951).. Paris (1913). s. 2089-2091. Sur les liaisons non holonomes non linéaires.S. REFERENCES [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] AGOSTINELLI C . CROMER A. Acta Academiae Scientiarum Taurinensis (1983).. Monographs 33 (1972). IV-2 (1933). Math. Unfortunately. Lincei 34 (1963). CATTANEO-GASPARINI I.. 305-369. FERRARESE G. Boll. Derivée covariante "liée" dans une Vn-i riemanienne à structure presque-produit. d. NEIMARK J. Sur les liaisons et les mouvements.J.. then we do not know the behaviour of the system without specifying the initial directions of the skates. Acc.I.M. But this information must be "a priori" ignored because of our assumption of disregarding [ali the remaining parts] of the device.A. PIRONNAU Y. In fact. SALETAN E. 1-9. The Hamiltonian dynamics of non-holonomic systems. Ann. Wissensch. Actes Congr.. Sulla struttura locale delle equaizoni dinamiche di un sistema anolonomo.. 396-402. Roy.. di Matem. Naz. / "movimenti astratti" di Appell e un nuovo esempio di vincoli anolonomi non lineari nelle velocità. Interbalcanian Math. Encycl. 892-897. Boll. Nuova forma sintetica delle equazioni del moto di un sistema anolonomo ed esistenza di un integrale lineare nelle velocità lagrangiane...

. Math.M. [15] WEBER R. Lect. Notes in Math. Analysis 81 (1968). Ann. Springer 1984. Dipartimento di Matematica. 278-301. Mech.L. W. Classical and non-classical dynamics with constraints. 1108. SYNGE VERSHIK [13] [14] Sergio BENENTI. Rat.. Arch. 99 (1928). Hamiltonian systems with constraints and their meaning in mechanics. 309-335. 738-751. .212 S. Via Carlo Alberto 10. Geodesics in non-holonomic geometry.1-10123 Torino. Italy. Benenti J. A. Università di Torino..