Missed factors

TheMawn · 2014-01-09, 00:08

I found this page while browsing around mersenne.ca.

Is there any way to find out who TF'ed these exponents (or see if any of them belong to me). Also, is a factor missed in TF guaranteed to show up in P-1?

ewmayer · 2014-01-09, 03:20

Quote:

Originally Posted by TheMawn

is a factor missed in TF guaranteed to show up in P-1?

No.

Brian-E · 2014-01-09, 09:43

Note that most if not all of these cases are likely to have been caused by a bug which apparently existed in early versions of the trial factoring software causing factors to be missed. Almost all the cases listed on that page do indeed seem to be related to TF work which would have been carried out years ago.

EDIT: Now I'm doubting what I just wrote. Some further browsing of earlier discussion about it seems to suggest that the known bug only meant that sometimes a found factor would not be the lowest factor, but it would not be the case that this would cause further unnecessary factoring work (or LL tests!) to be carried out on the Mersenne number. Can anyone confirm that or provide more information about the known historical bug in TF?

axn · 2014-01-09, 11:43

Quote:

Originally Posted by Brian-E

Note that most if not all of these cases are likely to have been caused by a bug which apparently existed in early versions of the trial factoring software causing factors to be missed.

That is not what is happening. These are P-1 misses, not TF misses. Here, there has been a P-1 with sufficient bounds which should've found the factor, but didn't. Later, a TF did find the factor (mostly, the TF was done by GPU).

TheMawn · 2014-01-09, 16:04

Ah. The wording "P-1 factors that should have been found" confused me. I understood "Factors found by P-1 that should have already been found" which is not the case. Thanks.

Brian-E · 2014-01-09, 21:33

Quote:

Originally Posted by TheMawn

Ah. The wording "P-1 factors that should have been found" confused me. I understood "Factors found by P-1 that should have already been found" which is not the case. Thanks.

Yes, I suffered from the same misunderstanding.

Since, like you, I'm intrigued by the question of what might be going on, does anyone know of any possible reason why the P-1 factoring could ("legitimately" or otherwise) have gone wrong in this way?

ewmayer · 2014-01-09, 22:23

Could be any number of things - a low-frequency bug in the powering algo or the gcd, a restart-from-interrupt bug, data corruption, etc. It is important to discover if the frequency is similar to that of LL test failures, if the bad results correlate with specific software releases, etc. In other words, George needs to comment.

Brian-E · 2014-01-10, 10:23

Thanks. I can see that with software of this complexity, running on numerous unknown machines under the control of unknown humans, such a statistical analysis may be the only feasible way of discovering what is going on.

Whether it is happening frequently enough to make such an investigation worth anyone's trouble is another question.

2014-01-09, 00:08	#1
TheMawn May 2013 East. Always East. 11·157 Posts	Missed factors I found this page while browsing around mersenne.ca. Is there any way to find out who TF'ed these exponents (or see if any of them belong to me). Also, is a factor missed in TF guaranteed to show up in P-1?

2014-01-09, 09:43	#3
Brian-E "Brian" Jul 2007 The Netherlands 2·11·149 Posts	Note that most if not all of these cases are likely to have been caused by a bug which apparently existed in early versions of the trial factoring software causing factors to be missed. Almost all the cases listed on that page do indeed seem to be related to TF work which would have been carried out years ago. EDIT: Now I'm doubting what I just wrote. Some further browsing of earlier discussion about it seems to suggest that the known bug only meant that sometimes a found factor would not be the lowest factor, but it would not be the case that this would cause further unnecessary factoring work (or LL tests!) to be carried out on the Mersenne number. Can anyone confirm that or provide more information about the known historical bug in TF? Last fiddled with by Brian-E on 2014-01-09 at 09:53

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
missed factor?	tha	Data	79	2021-11-19 15:55
Factor missed by TF	bcp19	PrimeNet	15	2015-08-10 11:57
More missed factors	lycorn	Data	76	2015-04-23 06:07
P-1 Missed factor	tha	Data	7	2014-04-30 20:54
Missed small factors	dswanson	Data	63	2004-11-24 04:30

2014-01-09, 16:04	#5
TheMawn May 2013 East. Always East. 11×157 Posts	Ah. The wording "P-1 factors that should have been found" confused me. I understood "Factors found by P-1 that should have already been found" which is not the case. Thanks.

2014-01-09, 22:23	#7
ewmayer ∂²ω=0 Sep 2002 República de California 26754₈ Posts	Could be any number of things - a low-frequency bug in the powering algo or the gcd, a restart-from-interrupt bug, data corruption, etc. It is important to discover if the frequency is similar to that of LL test failures, if the bad results correlate with specific software releases, etc. In other words, George needs to comment.

2014-01-10, 10:23	#8
Brian-E "Brian" Jul 2007 The Netherlands 2×11×149 Posts	Thanks. I can see that with software of this complexity, running on numerous unknown machines under the control of unknown humans, such a statistical analysis may be the only feasible way of discovering what is going on. Whether it is happening frequently enough to make such an investigation worth anyone's trouble is another question.