20050309, 19:43  #1 
P90 years forever!
Aug 2002
Yeehaw, FL
16353_{8} Posts 
Curve Counts
According to http://www.loria.fr/~zimmerma/records/ecm/params.html the curve counts listed for the 2 and 2+ pages is too high. The values I use came from ECMNet's pages several years ago.
I suspect I should convert my pages to use these new values. Opinions anyone? 
20050309, 19:56  #2 
"Nancy"
Aug 2002
Alexandria
2,467 Posts 
I wanted to post something about curve counts for a while but I'm still running tests on how often gmpecm finds factors vs. computed values (looking very good so far) and also haven't had the time to write something up yet. I'm pretty sure that the expected curve numbers computed by GMPECM 6 are pretty close to the truth, and that the old figures were too high. I also think the effect of curves with smaller bounds on larger factor sizes should be considered and the bound level should be increased when the probability of missing a factor of the target size with *all* curves done so far is <exp(1). This will further reduce the number of curves to do at each level.
I'll try to write something more coherent when the latest batch of test curves finish and I have a moment to put together the results. Alex 
20050309, 21:36  #3  
P90 years forever!
Aug 2002
Yeehaw, FL
11×673 Posts 
Quote:


20050309, 22:15  #4 
"Nancy"
Aug 2002
Alexandria
2,467 Posts 
Yes.
The expected number of curves to find a p30 with B1=50k, B2=100*B1 is 5268. Doing 262 curves does 262/5268 = .0497 of the required work, so only 616 additional curves at B1=250k should be needed to get the probability of a miss <exp(1). The effect from curves with smaller bounds, say B1=11k, is very small and can be ignored. Similarly, if someone does curves at a high bound before the expected number of curves at lower levels are done, the effect of the "large" curves on smaller factor sizes should be considered. This will also avoid people waste cpu time on small bounds to "fill in gaps" when a lot of curves at high bounds have been done already. There are no actual gaps, they are merely an artifact of the bookkeeping method. I think it might be better to not count curves but a fraction or percent, where "p45 nn% done" means that the probability of missing a p45 with the curves that have been done so far is exp(nn/100). If people do curves with varying B2 parameters, for example by finetuning gmpecm, the "standard curve" with B2=100*B1 become a rather arbitrary unit that everyone need to convert to. Normalizing the unit to "100% = all finished" is probably easier in the long run. Comments cordially invited! Alex 
20050310, 02:14  #5  
P90 years forever!
Aug 2002
Yeehaw, FL
1110011101011_{2} Posts 
Quote:
BTW, I have to plead guilty to some sloppy bookkeeping. In my tables, GMPECM curve counts emailed to me are simply doubled no matter what B2 is chosen. I do appreciate the forum threads here which aggregate GMPECM curves and compute the multiplier properly. 

20050310, 10:20  #6 
Aug 2002
Termonfeckin, IE
3^{2}·307 Posts 
Alex,
I think I'll take your suggestion for updating the Cunningham tables in the forum here. I'm also leaning towards removing the 2 and 2+ tables altogether (with the exception of 2LM) since having these tables on both George's pages and the forum is just confusing and adds work and the probability of an error. 
20050310, 12:06  #7  
Jun 2003
The Texas Hill Country
2101_{8} Posts 
Quote:


20050310, 13:23  #8  
"Mark"
Apr 2003
Between here and the
6,277 Posts 
Quote:
Since George's pages only have 2+/2, maybe he could redirect his links to the thread in this forum. The format of the tables in this forum, plus all of the other information available, such as which are reserved, make this forum a better resource. Also, maybe Paul Zimmerman can point his Cunningham project to that thread in this forum. Again, the format is far easier to understand than those on his website. I like onestop shopping and to have only 1 place to coordinate all Cunningham work is a good thing. 

20050310, 13:43  #9 
P90 years forever!
Aug 2002
Yeehaw, FL
11×673 Posts 
I have no problem with letting someone else maintain the 2 and 2+ tables in the forums. I'll do whatever everyone thinks is best.

20050310, 13:49  #10 
Jul 2004
Potsdam, Germany
3×277 Posts 
My short answer would be:
A lot of people can maintain those tables, but only few (or maybe only one?) can improve prime95/mprime. 
20050310, 14:13  #11 
Aug 2002
Termonfeckin, IE
3^{2}·307 Posts 
For historical reasons, I'm inclined to let George handle the 2+ and 2 tables. Plus, believe me, updating all the tables is a pain and very timeconsuming and if I have two less tables to handle I'll be a lot happier :)
Regardless, it is always possible to give other people  including George  mod privileges on the forum so if we do decide to host everything on the forum, we should be able to divide the work. I did an update on the the tables last week but I'm still waiting for an email response or two before I finish the job and upload the tables that are now residing on my HD. BTW, I wanted to ask Xyzzy if it is possible to enable HTML on the forum. Maybe I have asked before and he has said no already but I don't remember. 
Thread Tools  
Similar Threads  
Thread  Thread Starter  Forum  Replies  Last Post 
post counts  Raman  Forum Feedback  67  20160107 18:35 
Why different shift counts are important  Madpoo  Data  6  20151011 03:55 
I don't understand these relation counts  fivemack  Msieve  4  20121029 11:26 
Dedidicated rig; What counts in Prime Crunching?  kaffikanne  Hardware  10  20091212 23:46 
Case of the disappearing post counts...  PrimeCruncher  Lounge  7  20041031 22:07 