Least squares solvers - Optimization Toolbox

Hello,

I am working on an optimization of a discretized differential function that depends on several parameters.

I am using the lsqcurvefit solver and the trust-region-reflective algorithm to optimize one parameter. The code runs but I still get this warning, does someone have an idea how to get over it? I tried to increase the difference (ub, lb), but the code stutters.

"Warning: Derivative finite-differencing step was artificially reduced to be within bound constraints. This may adversely affect convergence. Increasing distance between bound constraints, in dimension 1, to be at least 1.5799e-09 may improve results".

Thanks in advance for your help.

[Torsten]

https://de.mathworks.com/matlabcentral/answers/382157-warning-derivative-finite-differencing-was-artificially-reduced-to-be-within-bound-constraints

Answer 1

[Vaibhav]

Hi Sana,

I understand that this error could surface when the tolerances set are not realistic considering the scale of data. Additionally, if the imposed limits on certain parameters make it practically impossible to find a solution, this error could occur.

One possible solution can be adjusting the bounds and tolerances to be more realistic and considering centring and scaling for the fit.

You can refer to the below MATLAB answer for more information:

https://www.mathworks.com/matlabcentral/answers/382157-warning-derivative-finite-differencing-was-artificially-reduced-to-be-within-bound-constraints

Hope this helps!