20150525, 10:14  #1 
May 2015
111_{2} Posts 
ECM odds
I apologize for the remedial newbie question.
For M<1e6 and the GIMPS choices for B1/B2/number of curves tested, I would like to estimate the odds that the last curve of one B1/B2 range finds a factor, versus the first curve of the next range. Can someone provide a reference/equation to help? 
20150525, 15:48  #2 
"William"
May 2003
New Haven
3×787 Posts 
The question is more complicated that it sounds. If a factor of a particular size exists, the probability of finding that factor is the same for every curve of the set. But that includes refinding factors already found by earlier curves in the set. And you seldom know that a factor of a particular size exists, so working through a set curves without finding any such factor makes it more likely that a factor of that size does not exist. These are not particularly complicated probability ideas, but there are several things to keep track of. We could help you work through them if you want.
But the answer you probably want can be figured out in a simpler manner. Recommended change points happen because the probability of finding a factor per unit of computing power has just become higher for the bigger curve. So right at the change over the probabilities are the same. (In fact, they are nearly the same for a broad range around the change point, but that's not necessary for this argument). So if the new curve takes "k" times more computing power, it is "k" times more likely to find a factor. If it were not so, the recommended change point would be early or later  at the point where it IS true. So if you only want the relative odds  that's how to do it. If you want to know the actual values of these probabilities, we need to dig deeper into the issues of the first paragraph. 
20150525, 17:12  #3  
Nov 2003
2^{2}·5·373 Posts 
Quote:
The OP's question may be answered by reading my joint paper with Sam Wagstaff Jr.: A Practical Analysis of ECM, Math. Comp. 

20150525, 20:44  #4  
May 2015
7 Posts 
Quote:
It would take some time to recreate your Bayesian analysis for these specific bounds, but it seems clear that the first curve in the next range will have a better chance of finding a factor well before you finish the previous range, even if you account for the longer run time. It would be interesting to come up with an optimal scheme to fill out the ranges nonsequentially. 

20150525, 22:04  #5  
"William"
May 2003
New Haven
939_{16} Posts 
Quote:
I also laid the groundwork for why he would need your paper. I suspected he would not want that level of detail, but it appears I was wrong about that. I'm a little to slow to actually recommend the paper because of the error in the rho function at the beginning. It's a great paper for the sufficiently savvy to get the concept, and you have assured me that the results used the correct rho function. But the error can be a stumbling block for anyone hoping to get numeric results from the paper. 

Thread Tools  
Similar Threads  
Thread  Thread Starter  Forum  Replies  Last Post 
What are the odds?  petrw1  PrimeNet  0  20161006 22:40 
Odds  Fred  Software  4  20160308 03:05 
Seems to defy the odds....  petrw1  Factoring  6  20130319 00:21 
odds in genetics.  science_man_88  Science & Technology  10  20101109 22:01 
ARE THE ODDS CORRECT..Please help  lpmurray  Lounge  4  20050209 10:38 