Hi,
I am going to build an open source physics modelling and simulation framework named liblamp (Lamp = Lamp abstract managed physics) for the .NET architecture. Project web site: http://liblamp.sourceforge.net
Of course physics simulation will call for numerical methods (e.g. solving of ODEs). The framework will contain a numerics part where interfaces (or contracts as you can say) are defined that all simulation modules can rely on. Hovever, implementation of the numerical part will be dedicated to exchangable modules which can be own implementations or wrappers around existing implementations such as Math.Net.
For the future a .NET language called Phy# based on the Lamp framework features (for example unit systems) is planned. Here we might rely on Math.Net symbolics to create the best language for implementing physical models and simulators ever ;-)
Of course, i will contribute to Math.Net if I see any numarical feature i need for liblamp unimplemented.