(42S) HP 42S/ DM42: Nelder Mead Optimization

[rawi]

Hi,

concerning the termination criterion: I do not think that it is faster to use the differences in the X and Y than in the f(X,Y). Near the optimum most derivable functions are rather flat with a derivate well below 1, so that the differences in the function are smaller than the differences in the variables. So it should be slower. But I thought if you set a threshold you want to set the exactness of your input factors.

Concerning the domain of FU you are right. If the square route would be negative the procedure would stop. I put not enough care on the function which was just for demonstration purposes.

Concerning not derivable functions: These are a problem to all optimization functions. It can be that the simplex does not get along the valley and the size of the simplex is shrinked to virtually zero. But in this case both termination criteria would stop the procedure. A possible way to handle such cases is to start the procedure again at the solution reached so far.

Best

Raimund