Quote:
Originally Posted by EdH
If you ran two searches across the same region with vastly different P, would they find the same polynomials, with the larger P finding fewer of them? Or, would the found polynomials be different?

From params.c90:
choose lc(g) with two prime factors in [P,2P]
So, if you double P (call it Q=2P), it's highly unlikely a polynomial will have this property in both [P,Q] and [Q,2Q]. I've changed P by 25% and had the same best polynomial pop out of the two runs; I've changed P by 2x and never (in just a handful of trials) had the same poly come out of both runs.
Each adrange is producing hundreds of candidate polys; over the course of a poly select run you're finding thousands and keeping the best 60 or 90 or whatever. That doesn't mean you should inflate P so high that you only find a few hundred polys; things don't work that way.
I've been selecting P values such that the lastplace Exp value (the one published during sizeopt) is a minimum. I figure that's the P that finds the most "good" poly candidates, while trying to optimize for 5th place or 1st place is more a matter of luck.