Parallel globally adaptive quadrature on the KSR-1 (Q1895918)

This paper develops and implements two new adaptive algorithms for one- dimensional quadrature problems. In the dynamic and asynchronous algorithm (DA) each pair of processors bisects the interval with the largest error estimate that is not currently being bisected by one of the other pairs of processors and then approximates the interval over the two halves of the interval concurrently. The dynamic and synchronous algorithm (DS) reduces the amount of list searching in DA and allows more than one processor to work on an interval by introducing synchronization points. Numerical comparisons with existing static implementations and parallel function evaluation implementations on a KSR1 show that DS gives good performance on all types of integrands tested, while DA gives some improvement over existing algorithms for some types of integrand behaviour.