Error using Rodas4()

[hurricane007]

Hello,

when I solve a PDE (basically dC/dt = -usgdC/dz + Dgd^2C/dz^2 + S) with alg_hints=[:stiff], I learnt the algorithm was Rodas4(), but I have never been able to solve the PDE with Rodas4() (also not possible with other algorithms such as KenCarp*()). anyone know the reason?

[hurricane007]

The only thing working is CVODE_BDF(), but this algorithm doesn't support "save_idxs" that I need for parameter estimation.

[anandijain]

can you post the reproducer so i can mess around with it?

[ChrisRackauckas]

That means your code is not ForwardDiff differentiable. You'll need to do Rodas4(autodiff=false) or KenCarp4(autodiff=false) to turn off automatic differentiation. See the FAQ for more details: https://diffeq.sciml.ai/stable/basics/faq/#Native-Julia-solvers-compatibility-with-autodifferentiation . Likely you're doing something like Array{Float64}(...) in your code, which of course will fail with this message.

[anandijain]

i suspect chris just enjoys the 1up...

[hurricane007]

can you post the reproducer so i can mess around with it?

It's a simulation of a fixed bed reactor, so the whole code to make it work is too long so I didn't post it. If you are interested I can post it. But Christ's method helped. Actually, I got this problem quite a lot of times. This time it seems to be due to:

for ii = 1:nGrid
    # usg = FA/CA [mo/m²⋅s]/[mol/m³] = Finert/(C_tot - C_CO2 - C_NO)
    usg[ii, ii] = Finert/(C_tot - u[ii, 1])
end

That means your code is not ForwardDiff differentiable. You'll need to do Rodas4(autodiff=false) or KenCarp4(autodiff=false) to turn off automatic differentiation. See the FAQ for more details: https://diffeq.sciml.ai/stable/basics/faq/#Native-Julia-solvers-compatibility-with-autodifferentiation . Likely you're doing something like Array{Float64}(...) in your code, which of course will fail with this message.

Thanks Chris, this solved the problem.