11 Mar
2014
11 Mar
'14
9:50 a.m.
On 10.03.2014 22:18, Rhys Ulerich wrote:
I think they could also be generalized to generic SDIRK implementation, parametrized only by its coefficients.
One challenging thing would be getting the caching correct to permit reuse of operator factorizations across substeps. If SDIRK drives the API design, a BDF-ready API should fall out but not necessarily vice versa.
We already have a generic implementation of arbitrary RK schemes, based on template meta programming. Maybe similar techniques from this implemetation can be apply for SDIRK. Sounds really interesting to me.