Escaping local minima costs something.
At high temperature, uphill moves can survive. As the temperature falls, the algorithm becomes increasingly conservative. The randomness is not noise added after the fact; it is part of the search strategy.
Optimization · CMNA Laboratory
Sometimes the algorithm should make a bad move on purpose.
Define a two-dimensional objective in R, then watch simulated annealing accept, reject, wander, cool, and remember the best point it has seen.
Bring your own landscape
Let R wander before it settles.
Write f(x) and x0 in R, choose temperature, cooling rate, and seed, then play the accepted and rejected moves.
Runtime
Preparing R…RDefine
f and two-dimensional x0| n | temperature | proposal f | accepted? | best f |
|---|
Book§6.3.3 · Simulated Annealing · p. 187Book details
TeachingInstructor notes and chapter contextOpen teaching material
Numerical ZooWorked problems