Accuracy of TRBDF2 vs matlab's ode23tb?

Specific Domains Modelling & Simulations
10

This seems to be plenty fast and working just fine as long as you don’t expect the du to be exactly 0, but rather a number <abstol. The Rosenbrock methods get your du correct to floating point precision, but TRBDF2 only computes the du to within numerical tolerance.

I’m also not seeing the timing differences between TRBDF2 and Rodas5P if you don’t set dtmax (which is an option that should basically never be used).

julia> @btime sol = solve(prob, Rodas5P(); reltol =1e-6, abstol =1e-6);
  432.221 μs (674 allocations: 66.91 KiB)

julia> @btime sol = solve(prob, TRBDF2(); reltol =1e-6, abstol =1e-6);
  381.361 μs (580 allocations: 53.39 KiB)