Hi all, I have always been frustrated by the behavior of UnivariateSolverUtils.bracket. The two problems I see are the following ones:
- the loop always uses increments of 1.0 to increase its search interval, - the loop always return the full current interval once it brackets a root. The fact 1.0 is hard-coded is hard-coded is a problem because many functions are limited to domains where 1.0 is far too large for searching. Rescaling functions just because the bracket method has an hard-coded constant is really a pain. The fact the loop always return the full interval is a waste of computing power. Suppose for example that some function has a root at x = 0.5 and that your initial guess is that the root is somewhere near x = 10. Then you will compute the pairs f(9), f(11), then f(8), f(12), then f(7), f(13) ... and finally f(0), f(20) and you will return the bracketing interval [0, 20], despite you already know the interval [0, 1] is also bracketing since you have already computed these values in the last two iterations. I have a fix for that and would like to commit it. It simply adds a new signature for "bracket" with a new parameter delta to replace the hard-coded 1.0 and some logic to select a better interval at the end. The existing signatures would simply delegate to it by calling it with delta = 1.0. However, implementing it implies the returned value of the interval is not the same as before (the interval will often be smaller). Do you think there would be any reason why this change would break user code? best regards, Luc --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org For additional commands, e-mail: dev-h...@commons.apache.org
