Alternatives if Algebra Solver fails to return root?

Modeling
1

Hello,

The implementation of the algebra_solver fails to return root values for a function I am investigating. I initially augment the algebra_solver with a bisection method, i.e. bisection method pre-solves the function, then uses the output as the initial value for the algebra solver… This seems to work better, but it is not a be-all-end-all, i.e. I am still getting wrong roots even if the bisection method hones into the approximate roots.

The function I am solving works well with bracketing methods-- I have tested other root solving routines on MATLAB such as regula falsi, secant and newton-raphson, but gradient based methods fail for some combination of the target function parameters-- I understand that the algebra_solver uses first-order derivatives as well.

What else should I consider for this problem?

2

You should probably figure out why something like Newton-Raphson doesn’t work, instead of looking for more methods. Maybe it isn’t differentiable or has vanishing gradient.

1 Like
3

Good point, I’ll see what I can do, thank you.

4

Sorry to hear you are having problems. It would be easier for us to help you if you provided the actual system you are trying to solve. One thing I have experience with is that algebra_solver can fail when the target is not defined over full \mathrm{R}^n (e.g. when there are some bounds on the allowed parameter values). In such cases it helps to reparametrize so that the variables you solve for are not bounded.

2 Likes