A matricial exponentially fitted scheme for the numerical solution of stiff initial-value problems (Q1309755)

For the ordinary differential equation initial value problem, \(y' = f(t,y)\) given \(y(0)= y_ 0\), the scheme \(y_{n+1} = y_ n + (e^{hJ_ n} - I)J^{-1}_ nf_ n\) is considered. To enable the implementation of this method, an algorithm is given for computing the matrix exponential and a local error estimate is derived, based on computing two half steps for each full step. Various cost/accuracy statistics are collected on a number of difficult test problems using a variable step version of the scheme and some comparisons are made with other methods.