Traditional step size controllers make the tacit assumption that the cost of a time step is independent of the step size. This is reasonable for explicit and implicit integrators that use direct solvers. In the context of exponential integrators, however, an iterative approach, such as Krylov methods or polynomial interpolation, to compute the action of the required matrix functions is usually employed. In this case, the assumption of constant cost is not valid. This is, in particular, a problem for higher-order exponential integrators, which are able to take relatively large time steps based on accuracy considerations. In this paper, we consider an adaptive step size controller for exponential Rosenbrock methods that determines the step size based on the premise of minimizing computational cost. The largest allowed step size, given by accuracy considerations, merely acts as a constraint. We test this approach on a range of nonlinear partial differential equations. Our results show significant improvements (up to a factor of 4 reduction in the computational cost) over the traditional step size controller for a wide range of tolerances.
翻译:传统的职级大小控制器暗中假定一个时间步骤的费用与职级大小无关。 这对使用直接溶剂的直线和隐含混集体来说是合理的。 但是,在指数集集体的情况下,通常使用一种迭代法,如Krylov 方法或多式内插法来计算所需的矩阵函数的动作。 在这种情况下,假设不变成本是无效的。这对更高级级指数集体来说尤其是一个问题,它们能够根据准确性因素采取相对较大的时间步骤。 在本文中,我们考虑对指数化罗森布洛克法采用一个适应性的职级大小控制器,根据尽可能降低计算成本的前提来决定职级大小。根据精确性考虑而允许的最大职级大小只是作为一种制约作用。我们用一系列非线性部分差异方程式来测试这一方法。我们的结果显示,对于一系列广泛的宽度,对传统的职级控制器来说,比传统的职级大小控制器有了显著的改进(降幅为4倍)。