Sun Jan 23 17:12:00 EST 2011

State space DSL: next

It seems the state space form is quite natural, so let's make it the
central abstraction.

  - The architecture is taylored according to the requirement for the
    output to be code.  The state space model fits well here.

  - All behaviour is purely 
functional: state feedback is handled in
    the top level combinator.

  - We want composition of state space models.  The reason is that
    different forms of models require different forms of analysis, so
    it makes sense to partition the solution into building blocks that
    can be individually tested and combine them into a single

  - Apart from simple composition, we probably also need:
          - domain nesting, i.e. matrices and vectors
          - nested iteration, i.e. function evaluation & optimizatiobn