Single-iteration Sobolev descent for linear initial value problems (Q350814)

This paper deals with the numerical solution of initial value problems (IVPs) for linear scalar differential equations where the unknown function \( y = y(x)\), \( x \in I=[0,1]\) satisfies the differential equation \( D y \equiv y^{(n)}+ p_{n-1} y^{(n-1)}+ \ldots + p_0 y = f \) in \(I\) where \( p_j \) are continuous in \(I\) together with the initial conditions \( y^{(j)}(0) = y_j\), \( j=0, \ldots , n-1\). The proposed approach introduces a suitable Hilbert space \(X\) and a functional \(K\) in this space so that the unique solution of the IVP minimizes this functional. Then, it is proved that a steepest descent based on the gradient induced by the inner product of \(X\) converges in one iteration to the unique solution. A simple example with a uniform discretization of a second-order test problem is presented by including a MATLAB code and the results of some numerical experiments. The authors note that the proposed approach can be extended to other differential problems showing the application to a particular second-order boundary value problem.