Differentialequations.jl v6.9.0: Multi-GPU Implicit Ode Solves, SciPy/R Bindings

Chris's ODE solver suite is a huge improvement over decades old Fortran libraries. It is also really pushing the boundary on what is possible to do with GPUs beyond the standard neural network workloads.

This release provides capabilities that allow for significant improvements in the capabilities of simulations in pharma, engineering and physics-informed machine learning.

I'm guessing you still have to describe the differential equation structure in Julia ? Can it handle black box functions that provide the state and derivatives (like an FMU in model exchange) ?

Yes, there really isn't a good way to do a black box binary. We will be setting things up with eFMU though which has the equations, and that we can generate GPU code for through ModelingToolkit quite easily. We will need to get that setup as a standard feature when eFMU is standardized

Thanks. I'll check out what eFMU is.

It's the planned extensions to the FMU: https://www.researchgate.net/publication/336775272_Standardi...

I found this package to be highly useful. Great to see that it is progressing like this. Still waiting for fractional derivatives.