Different solutions of fmincon when changing x0

Question

Daniela Würmseer 2022-3-31

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1685029-different-solutions-of-fmincon-when-changing-x0

编辑： Matt J 2022-4-26

采纳的回答： Torsten

在 MATLAB Online 中打开

I wanted to ask if it is normal that fmincon can have different solutions with different x0?

I implement an Algoritjm by myself and use fmincon as follows:

options = optimoptions('fmincon','Display','iter','Algorithm','sqp');
x = fmincon(fun,x0,A,b,Aeq,beq,lb,ub,nonlcon,options)

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Torsten 2022-3-31

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1685029-different-solutions-of-fmincon-when-changing-x0#answer_931234

编辑：Torsten 2022-3-31

I wanted to ask if it is normal that fmincon can have different solutions with different x0?

There can be several reasons for fmincon to converge to different "solutions" dependent on the starting point. But - sure - the optimization algorithm implemented in "fmincon" tries to find local extrema. If a function has more than one or is ill-behaved, it depends on the starting point to which of the local extrema the algorithm converges.

2 个评论
显示无隐藏无

Shaily_T 2022-4-26

Could you please elaborate more on an ill-behaved function? Is it inherent to the function or can be dependent on our codes as well? I mean for example something wrong in our code.

And is there a way to determine which of the probabilities you mentioned is the case in our function? I mean being ill-behaved or having more than one extrema.

Thank you in advance!

Matt J 2022-4-26

编辑：Matt J 2022-4-26

在 MATLAB Online 中打开

You can look at the function values at the two solutions to see if they are nearly the same. If so, then plainly you have multiple minima. Better still, given the two solutions x1 and x2,you could plot the 1D function

g(t)=fun(t*x1+(1-t)*x2) % t from -1 to 2

If the graph is very flat between t=0 and t=1 or is very bumpy, it is a likely sign that your graph has multiple local/global minima.

请先登录，再进行评论。