I have an optimisation problem for which the objective function is not differentiable. I am looking for a global optimum using the Simulated Annealing technique in MATLAB. I am facing the problem of a very large search domain. I want to narrow down the search domain before applying Simulated Annealing. Is there a method to reduce the search domain before I apply Simulated Annealing?
Asked
Active
Viewed 34 times
2
-
Not differentiable does not say much. $|x|$ is not differentiable but is trivial to minimize. – Johan Löfberg Sep 15 '17 at 19:39
-
The objective function is a CFD code – user14352 Sep 16 '17 at 01:08
-
and the constraints, which this question really is about? – Johan Löfberg Sep 16 '17 at 06:24
-
The variables are range bound. No linear or non-linear constraints. – user14352 Sep 16 '17 at 15:32
-
OK. Range bounds does not give you much to do. The general theory you are looking for is bound or constraint propagation in global optimization. – Johan Löfberg Sep 16 '17 at 17:07
-
All right. Is there any text, article, or document I can refer to for further understanding? – user14352 Sep 17 '17 at 02:05
-
Just a random article with the correct key-words https://arxiv.org/pdf/1706.08601.pdf – Johan Löfberg Sep 17 '17 at 06:45