6

I am using SimulatedAnnealing method for a simulation based optimization problem and was wondering if someone can enlighten me on the following features:

  1. Is there some guidelines on optimal size of "SearchPoints" based on the size of the solution space knowing that the objective function is non-linear?

  2. It is intuitive that the number of iterations will depend on the objective function, but I have noticed that the number of iterations seem to be a multiple of number of variables. is this true? Sample code is provided in Simulated Annealing Convergence

  3. I think it is difficult to prove global optimization results, but was wondering if any one has ideas on alternative ways to do that.

Community
  • 1
brama
  • 898
  • 5
  • 18
  • 1
    Does anyone have a comment or a response? – brama Feb 25 '14 at 22:47
  • 5
    I'm voting to close this question as off-topic because it is a question about the simulated annealing method for global optimization, rather than a question about Mathematica's implementation. – dr.blochwave Sep 25 '15 at 18:53
  • 3
    @blochwave Not sure. I could benefit from an answer to points one and two – Dr. belisarius Sep 25 '15 at 20:13
  • 1
    @belisarius, true but I still think that (1) is a SA question rather than the specific MMA implementation. – dr.blochwave Sep 25 '15 at 20:54
  • The docs only has this to say "The default number of starting points, given by the option "SearchPoints", is min(2d, 50), where d is the number of variables." - Reference. That even answers (2) to a point. – dr.blochwave Sep 25 '15 at 20:55
  • @blochwave: 1) min(2d,50) is a default number for SearchPoints, not the optimal number I was looking for. 2) The document defines SearchPoints as number of initial points and has nothing to do with the number of iterations it will take to converge. So it does not answer (2) in any way. 3) Agree that the question is more to do with SA than MMA, but what is the benefit of putting a question on hold, when any responses to that question can benefit a larger audience. Thank You. – brama Sep 30 '15 at 14:06
  • @brama I think you should try scicomp.stackexchange.com in the first instance - see for example this question on convergence and then come back if you still have questions about the particular details of the MMA implementation. – dr.blochwave Sep 30 '15 at 14:52
  • it's december 2016... MMA documentation does not provide any further explanation on Simulated Annealing Parameters. – Fabio Dec 12 '16 at 06:45

0 Answers0