Blended extended linear multistep methods for the accurate numerical integration of stiff initial value problems (Q1064748)

To construct codes which are efficient for low to moderate tolerances for stiff problems, the first author extended the standard backward differentiation formulas by using evaluations of the derivative \(y'=f(x,y)\) at two new step points. Subsequently, he constructed high order highly stable methods by extending second derivative methods in a similar way. In this paper, the authors show how to circumvent some major computational disadvantages of the extended second derivative methods. The authors adopt \textit{R. D. Skeel} and \textit{A. K. Kong}'s approach [ACM Trans. Math. Software 3, 326-345 (1977; Zbl 0399.65045)] to blend extended linear multistep methods with extended BDF formulas. The resulting method preserves the linear stability properties of the extended second derivative methods. It is shown how to implement these new formulas using only the Jacobian matrix rather than its square or the second derivative. A variable order/variable step package is described. Results from stiff DETEST are reported and discussed. This paper will be of interest to researchers involved in derivation formulas and construction of codes.