BVProblem with constraints #3323
Conversation
|
I think this is mostly ready for review @ChrisRackauckas . Some of the problems in the tests don't seem fully well-posed, so I still need to work on that a bit, but the code itself should be correct. |
| `constraints` are used to specify boundary conditions to the ODESystem in the | ||
| form of equations. These values should specify values that state variables should | ||
| take at specific points, as in `x(0.5) ~ 1`). More general constraints that | ||
| should hold over the entire solution, such as `x(t)^2 + y(t)^2`, should be | ||
| specified as one of the equations used to build the `ODESystem`. Below is an example. |
There was a problem hiding this comment.
It's odd that this is here. Problem constructors are only supposed to take values: constraints are structural. They should be added to the system I would think? @baggepinnen thoughts?
There was a problem hiding this comment.
Can the 0.5 in x(0.5) ~ 1 be a parameter? I can imagine this being something one would like to change without re-simplifying the system.
There was a problem hiding this comment.
Would it make sense to have a BVPSystem that contains a ConstraintSystem like the OptimizationSystem?
There was a problem hiding this comment.
Sounds good, I will work on these conversions.
| # Cartesian pendulum from the docs. | ||
| # DAE IVP solved using BoundaryValueDiffEq solvers. | ||
| # let | ||
| # @parameters g |
There was a problem hiding this comment.
What's up with these ones?
There was a problem hiding this comment.
The interface for specifying boundary conditions is different for the BVDAE solvers at the moment (you specify the time points as a separate argument), was gonna wait until they get updated to be the same.
Followup to #3251. Performance is something like 30x worse than generating the BVProblem using manually-defined f, bc, need to investigate why. Will uncomment the BVDAE solver tests when they get updated to use the same boundary condition format as the others.