Geometric integration methods that preserve Lyapunov functions (Q2490363)

Some projection Runge-Kutta methods that preserve Lyapunov type functions of ordinary differential systems are proposed. Assuming that the differential system \( y'(t)= f(y(t)), (1) \) has a smooth scalar Lyapunov function \( V(y)\) which is non increasing along the solutions of (1) in a certain region of the \(y\)-space and \( \psi_h\) is a Runge-Kutta method with order \(p\) defined by the Butcher array \((A=(a_{ij}),b=(b_j))\) with non negative weights \(b_j\), the authors propose an algorithm \( \widetilde{\psi}_h = P \cdot \psi_h\) which projects orthogonally the numerical solution \( \psi_h (y_n)\) into the manifold \( V(y)= V_{n+1}\) where \(V_{n+1}\) is some approximation to the Lyapunov function at \( y(t_{n+1})\). It is proved that this projection retains the order of the original method and the equilibrium points of the system coicide with the fixed points of the numerical method. Further some symmetric projection methods are also proposed. Finally, the paper presents some numerical experiments testing the behaviour of Euler, Heun and two-stage Gauss methods for some nonlinear planar problems, concluding that for these problems the preservation of some Lyapunov functions improves the qualitative phase space produced by the numerical methods over the standard approach.