Approximating Runge-Kutta matrices by triangular matrices (Q1359972): Difference between revisions

Given a stiff initial value problem \[ y'(t)= f(t,y(t)), \qquad y(t_0)= y_0,\tag{1} \] \(y,f\) vectors with \(d\) components, \(t_0\leq t\leq t_e\), one of the most powerful methods is an implicit Runge-Kutta method. In such a method one has to solve at every time step a system of nonlinear equations of the form \[ R(Y_n)=0, \qquad R(Y_n):= Y_n-(e\otimes I)y_{n-1}- h_n(A\otimes I)F(Y_n)\tag{2} \] where \(A\) denotes the \((s,s)\)-matrix containing the parameters of the \(s\)-stage Runge-Kutta method, \(y_{n-1}\) denotes the approximation to \(y(t_{n-1})\), \(e\) is the \(s\)-dimensional vector with unit entries, \(I\) is the \((d,d)\)-identity matrix, \(h_n\) is the stepsize \(t_n-t_{n-1}\) and \(\otimes\) denotes the Kronecker product. The \(s\) components \(Y_{n,i}\) of the \(sd\)-dimensional solution vector \(Y_n\) represent numerical approximations to the \(s\) exact solution vectors \(y(t_{n-1}-c_ih_n)\), where \(c\) denotes the abscissa vector with components distinct and positive, \(i\) ranges from 1 to \(s\). Solving the equations (2) by a modified Newton process requires a lot of computational effort for high-dimensional problems. The authors work with a Crout decomposition \(A=TU\), where \(U\) is unit upper triangular. It is shown that with this matrix \(T\) stiff error components in the numerical solution are strongly damped. The authors show that for a large class of Runge-Kutta methods this procedure can be carried out and that the diagonal entries of the matrix \(T\) are positive. This means that the linear systems that are to be solved have a non-singular matrix.