Problem with stiff ODE solved by Julia

There’s an ODE to be solved. It’s a chemical reaction system. There is 863 reactions and 230 species. The system is quite stiff and large. So I ues some algorithms advised by Solving Large Stiff Equations · DifferentialEquations.jl. I tried every algorithms, but the values never change at each step. In another word, the ODE solver cannot catch the small change.

So I wanna to know how I can sovle such ODE system. I just want to make the values change, so the time efficiency can be considered later.


This almost ceratinly means you coded your function incorrectly. if I had to guess, it would be that you used the in place form of f(du, u, p, t) but didn’t update du, but it’s hard to say for sure without seeing the code.

Here is the code. It seems du will update. Whatever algorithm I try, the value will not change.

function crnn(du,c,k,t)
    c = reshape(c,(230,1))
    k = reshape(k,(863,1))
    log_r = w_in * @.log(clamp(c,1e-35,1e15))
    log_r = log_r +  @.log(clamp(k,1e-35,1e15))
    r = @.exp(log_r);
    du = w_out' * r;
    return du;

using Symbolics
tspan = [0,1]
du0 = copy(conc[1,:])
jac_sparsity = Symbolics.jacobian_sparsity((du,u)-> crnn(du,u,k[1,:],0.0),du0,conc[1,:])
f = ODEFunction(crnn;jac_prototype=float.(jac_sparsity))
prob = ODEProblem(f, conc[1,:], tspan, k[1,:])
sol = solve(prob,KenCarp47(linsolve=KrylovJL_GMRES()))
for i in 1:230
    println(sol.u[2][i] - sol.u[1][i])
1 Like

Yeah. My guess was correct. You need du .= w_out' * r; Julia uses “pass by sharing” so when you say du = ... you aren’t changing du you’re just making a new du. Also the return du` isn’t doing anything here since you’re using the in-place formulation.


Could you please help me to solve my problem?

Please see the post: Forecasting time series data with neural ordinary differential equations in Julia

Could you please help?

It’s also unclear what is intended with w_out. Is it intentionally global?

Might help to do a MWE with a small, stiff, toy system. That can help separate syntax issues from the question of how to solve stiff systems.