Achieving tolerance proportionality in software for stiff initial-value problems (Q1262094)

The paper presents a basis for a uniform interpretation of the user's accuracy requirement in codes for solving stiff initial value problems. It is shown that it is possible to design the local error control strategy in a code in such a way as to achieve the tolerance proportionality. The strategy is based on a control of the local error per unit step. Next, the strategy is discussed for two stiff solvers: LSODE (Hindmarsh, 1980) and SECDER (Addison, 1979). The numerical results show that the strategy leads to a significant improvement in tolerance proportionality at litte extra cost.