TS?

[mauro3]

I might do some work on ODE/DAE time steppers coming spring or summer. I was wondering whether it is planned to wrap TS and, if so, any idea when? (BTW, thanks for all the good work so far!)

[JaredCrean2]

There are a few things in the queue ahead of TS (some Matrix and KSP items), but getting the basic TS functionality working shouldn't be too difficult. Are there any features you are particularly interesting in?

[mauro3]

Two things:

I'm implementing/porting some of the PETSc-ODE solvers to Julia (SciML/ODE.jl#72) and it would be good to check them.

I'd be generally interested in making some/all of the PETSc-ODE solvers available in Julia, similarly to the Sundials.jl package.

Thanks for the quick reply!

[JaredCrean2]

Sounds reasonable. I will update this issue once work on TS gets started.

[JaredCrean2]

Everything needed to do explicit time marching methods for ODEs is in place. Implicit should be ready in a week or two. See src/ts.jl.

[mauro3]

Cool, thanks!

[JaredCrean2]

Sorry this took so long, but implicit works now. Take a look and let me know if anything else is needed.

[mauro3]

Thanks so much!

My GSoC student @obiajulu may try to wrap the PETSc ODE-solvers such that they can be compared to others using https://github.com/mauro3/IVPTestSuite.jl . We'll have a look at the tests in https://github.com/JuliaParallel/PETSc.jl/blob/master/test/ts.jl and see whether we can make sense of them. Otherwise we might request some more examples ;-)

[JaredCrean2]

Cool. IVPTestSuite looks interesting. It will be nice to have a good comparison of the available methods and packages.

[ChrisRackauckas]

So TS is now available? What does the interface look like? I'd like to wrap it into DifferentialEquations.jl to test with everything else.

[JaredCrean2]

That's right. The interface is all in ts.jl. You have to cast your equation in a particular form described in the TS chapter of the Petsc manual, but the main idea is you have to supply Petsc with functions to evaluate the right hand side and left hand side functions, and possibly their jacobians (depending on what method you use). This is done with the functions set_rhs_function, set_rhs_jac, set_lhs_function, set_lhs_jac. You will need to create a TS object, set the time-stepping parameters (using set_times), and call solve! to execute the time-stepping. The tests contain examples of solving the 1D heat equation using forward and backward Euler here. You can use the Petsc options database to set the type of problem (linear of nonlinear) and method (forward Euler, backward Euler, etc), or you can pass them into the TS constructor.