ECM group order mystery - Page 3

Prime95 · 2022-09-24, 20:24

Finally, nailed the bugger.

Starting in 30.7, the curve starting point is incorrectly computed on every curve after the first curve. This is due to "leftover state" from stage 2.
The problem is lessened in 30.9 as often you'll see "switching back to FFTlen x" at the end of stage 2.

I do not know how this bug impacts the success rate for ECM finding a factor.

Prime95 · 2022-09-25, 02:09

Download 30.9 build 2 for a fix.

ATH · 2022-09-25, 11:19

Nice work!

Interesting that it could still find factors and even factors with group orders outside B1/B2.

I guess the different starting points made the curves correspond to different B1/B2 values than was intended?
This is just a guess. I do not really understand the math behind ECM in detail.

Denial140 · 2022-09-25, 14:20

Quote:

Originally Posted by ATH

I guess the different starting points made the curves correspond to different B1/B2 values than was intended?
This is just a guess. I do not really understand the math behind ECM in detail.

It sounds more likely to me that the different starting points correspond to different sigma values, which change the B1/B2 required to find the factor. For any starting point, you can run the curve to any B1/B2 you want in a similar fashion to P-1.

Prime95 · 2022-10-02, 19:24

Quote:

Originally Posted by Prime95

Download 30.9 build 2 for a fix.

Grrr, that fix was incomplete. Please download 30.9 build 3. Sorry.

bbb120 · 2022-10-08, 05:30

Quote:

Originally Posted by Prime95

Grrr, that fix was incomplete. Please download 30.9 build 3. Sorry.

why bug can also find a factor?
maybe you can find a new algorithm by studying this "bug"

charybdis · 2022-10-08, 12:01

Quote:

Originally Posted by bbb120

why bug can also find a factor?
maybe you can find a new algorithm by studying this "bug"

No, you can't.

ECM curves are determined by a random seed called the sigma value. The bug led to the incorrect sigma being displayed, so ECM was still being run, and was still just as likely to find a factor. Prime95 just wasn't running the exact curves it said it was running.

LaurV · 2022-10-10, 08:42

Quote:

Originally Posted by charybdis

No, you can't.

Shhhh! Let him try!
You never know...

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Group order not accepted for some GMP-ECM factors?	rwwh	FactorDB	2	2015-06-19 06:26
ECM curve group order	Brain	Miscellaneous Math	1	2010-12-08 01:00
What is this group?	wpolly	Math	1	2008-06-09 12:14
how to join a group	gian92	Software	0	2008-02-22 21:08
Lie group E8 mapped	ixfd64	Lounge	13	2007-03-23 15:06

2022-09-24, 20:24	#23
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 17743₈ Posts	Finally, nailed the bugger. Starting in 30.7, the curve starting point is incorrectly computed on every curve after the first curve. This is due to "leftover state" from stage 2. The problem is lessened in 30.9 as often you'll see "switching back to FFTlen x" at the end of stage 2. I do not know how this bug impacts the success rate for ECM finding a factor.

2022-09-25, 02:09	#24
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 3²×907 Posts	Download 30.9 build 2 for a fix.

2022-09-25, 11:19	#25
ATH Einyen Dec 2003 Denmark 19×181 Posts	Nice work! Interesting that it could still find factors and even factors with group orders outside B1/B2. I guess the different starting points made the curves correspond to different B1/B2 values than was intended? This is just a guess. I do not really understand the math behind ECM in detail.