I have a 1D experimental dataset and an ODE-based model. The model is able to describe experimental data reasonably well by manual fitting of the parameter:

The next step now is using LsqFit, at first trying to fit just one parameter:

Don’t use LsqFit.jl. We actively recommend against it because it’s generally not as stable as using Optimization.jl-based cost functions with a full optimizer.

Tomorrow is a project discussion with my boss, and it would be much better if we have this way of getting our material properties in addition/comparison to those used till now. Thus I may follow the advice and switch to another package later on, but today just trying to get results, quick and dirty.

I don’t need high precision, nor even precision estimation. Providing my own numeric differentiation for the Jacobian worked: Fit now converges to the same value for any reasonable initial parameters. Still would be nice to understand, why it didn’t work without it.