Sufficient conditions for penalty formulation methods in analytical dynamics (Q684903): Difference between revisions

Constraints of analytical mechanics can be formulated with penalty methods using an asymptotically growing potential field. For nonlinear multi-degree-of-freedom systems, relatively few convergence results are known. This paper gives for a certain class of penalty methods sufficient conditions for convergence, extending a theorem of Rubin-Ungar. The paper emphasizes also the necessity of a rank condition of the Jacobian; without this the procedure can proceed along some incorrect trajectory (bifurcation, quasiperiodic response, chaotic behaviour) not existing in the true solution, but indistinguishable from the correct one. Main assumptions: ``natural'' systems are considered subject to holonomic and scleronomic constraints; the potential is continuously differentiable in a closed bounded region \(D \subset \mathbb{R}^ n\); the mass matrix is symmetrical and positive definite and twice continuously differentiable in \(D\); the constraints are twice continuously differentiable in \(D\). Further assumptions for the constraints, for the rank of the Jacobian and the \(\varepsilon\)-penalty matrix are made. These latter are constructed in such a way that their smallest eigenvalues become unbounded as \(\varepsilon \to 0\). The so-called inertial penalty method is used (the holonomic constraints are satisfied only approximately, if \(\varepsilon \to 0)\). It is proved: coordinates and velocities tend to the solution of the constrained problem, if \(\varepsilon \to 0\). The family \(q_ \varepsilon\) is equicontinuous and uniformly bounded. A numerical example (in connection a slider-crank mechanism) shows the situation (incorrect limit cycle of chaotic behaviour) if Jacobian loses its full rank.