Design and implementation of DIRK integrators for stiff systems (Q1399153)

The stiff integrators presented in this paper, unlike typical diagonally implicit Runge-Kutta (DIRK) methods, have a zero in the \((1,1)\) position of the coefficient matrix. Furthermore, the last row of this matrix is identical with the vector of output weights so that the method possesses the `first same as last' property. This makes it possible to benefit from a stage order two structure. Based on these and a number of other design considerations, two new DIRK methods are derived with orders 3 and 4, respectively. In each case, embedded methods of lower order are provided for stepsize control purposes. The new methods are implemented and compared with existing DIRK methods, using standard test problems.