- golden rectangle search
- fastest search algorithm to find the minimum of unimodal function
- assume the function is: y = f(x)
- tau = (1. + sqrt(5.))/2 ; golden rectangle constant of 1.61803
- lower bound = x1
- higher bound = x4
- minimum lies between x1 and x4
- one interation
- set x3 = (x4 - x1) * tau
- set x2 = x1 + (x4 - x3)
- evaluate y1 = f(x1), y2 = f(x2), y3 = f(x3), y4 = f(x4)
- eliminate x1 if
- y3 gt y2
- eliminate x4 if
- y2 gt y3
- each iteration reduces the interval to 61.8 % of its original interval
- reduction factor after n intervals
- 1 = 61.8 %
- 5 = 9.0 %
- 10 = .81 %
- 15 = .073 %
- 20 = .006 %
- 25 = .00059 %
- 30 = .000053 %
- 15 intervals is OK for most solutions
- 10 intervals is sufficient for rough solutions
- 20 intervals is plausible for precise solutions
- number of function calls = (number of intervals + 3)
- nested golden rectangle search
- nesting of the search is needed to solve multiple things at once
- for example: solve for
- weight between 450 pounds and 600 pounds
- mast side slop between 5 degrees and 20 degrees
- lap time for the optimum VMG courses (upwind and downwind)
- boat angle between 10 degrees and 80 degrees upwind
- boat angle between 100 degrees and 170 degrees downwind
- boat speed between (.5 * TrueWindSpeed) and (5 * TrueWindSpeed)
- total function calls = (10 + 3){weight} * (10 * 3){slop} * 2{Upwind-Downwind} * (12 + 3){VMG} * (15 *3){SPEED}
- total function calls = 76,050 for above solution
design . analysis . optimization . tuning . race strategy . rules . ice . clubs . venues
Thursday, August 14, 2014
Golden rectangle search
solve for one variable