And onwards...

bayanne · 2004-01-17, 19:59

Hey, this is all good fun, but whatabout moving the bit depth upwards. Do we need to analyse out own results, so that we can then say take it up to 2^52. What's going to be the procedure for taking a range for the future at these higher bit depths.....

Xyzzy · 2004-01-17, 21:43

We can either use one of the standalone factoring programs, or wait for George to remove the limitations in Prime95...

We do need to figure out how to reverse engineer the .cmp file format so we can make new worktodo.ini files...

I have all the raw data files and I will put them online in an hour or so... Maybe someone out there can help with figuring out what has been done and what we need to do...

I'll post here once the files are online...

Xyzzy · 2004-01-17, 23:37

Raw files:

http://www.mersenneforum.org/lmh/data/

ET_ · 2004-01-18, 19:50

I worked on the range 195M_200M and the following is the initial results.txt file:

Code:

M195000023 has a factor: 23790002807
M195000079 no factor from 2^0 to 2^50, WX1: B830ACC7
M195000107 no factor from 2^0 to 2^50, WX1: B823ACC9
M195000109 no factor from 2^0 to 2^50, WX1: B825ACCB
M195000119 no factor from 2^0 to 2^50, WX1: B82FACC8
M195000131 has a factor: 14040009433
M195000161 no factor from 2^0 to 2^50, WX1: B830ACCB
M195000193 no factor from 2^0 to 2^50, WX1: B828ACD2
M195000241 no factor from 2^0 to 2^50, WX1: B82FACCE
M195000259 no factor from 2^0 to 2^50, WX1: B841ACD3
M195000269 no factor from 2^0 to 2^50, WX1: B822ACD1
M195000271 no factor from 2^0 to 2^50, WX1: B824ACD3
M195000307 no factor from 2^0 to 2^50, WX1: B81FACD0
M195000317 has a factor: 147810240287
M195000359 no factor from 2^0 to 2^50, WX1: B82BACD0
M195000373 no factor from 2^0 to 2^50, WX1: B839ACD2
M195000383 no factor from 2^0 to 2^50, WX1: B843ACCF
M195000391 has a factor: 87750175951
M195000419 no factor from 2^0 to 2^50, WX1: B83EACCC
M195000433 has a factor: 18720041569
M195000493 has a factor: 1170002959
M195000511 no factor from 2^0 to 2^50, WX1: B820ACCE

Then I decided to rework the range using my own factoring program, and in the same range obtained the following:

Code:

M195000023 has a factor: 23790002807
M195000079 no factor from 2^1 to 2^50.
M195000107 no factor from 2^1 to 2^50.
M195000109 has a factor: 182394911954023
M195000119 no factor from 2^1 to 2^50.
M195000131 has a factor: 14040009433
M195000161 no factor from 2^1 to 2^50.
M195000193 no factor from 2^1 to 2^50.
M195000241 no factor from 2^1 to 2^50.
M195000259 has a factor: 2839593771559
M195000259 has a factor: 39497692460969
M195000269 has a factor: 6578139074447
M195000271 has a factor: 129175979521241
M195000307 no factor from 2^1 to 2^50.
M195000317 has a factor: 147810240287
M195000359 no factor from 2^1 to 2^50.
M195000373 no factor from 2^1 to 2^50.
M195000383 has a factor: 113005061953097
M195000391 has a factor: 87750175951
M195000419 has a factor: 87324697635743
M195000433 has a factor: 18720041569
M195000493 has a factor: 1170002959
M195000511 no factor from 2^1 to 2^50.

All the factors found are certified primes.

As you can see, Prime95 missed 7 factors among the first 22 exponents tested.

My humble question is: should we go on trial-factoring Mersenne numbers with this version of prime95, missing some exponents and forcing in future some LL-test and refactoring that could be avoided, or wait until a new version of (say) Factoring95 is released?

It's fun to factor numbers, but maybe we're giving Will Eddington an incomplete batch of data, and this worries me.

Luigi

Xyzzy · 2004-01-18, 23:09

Every factor we find just reduces the number left to be checked...

Even if we are missing factors, we've eliminated so many that someone could run the same work behind us in a fraction of the time...

Once we have the data files set up properly, creating a new worktodo.ini will be trivial...

Xyzzy · 2004-01-19, 00:25

Okay, the factor data file is done... All the factors were verified... We have found 3,663,418 factors so far... You can download the factors file here:

http://www.mersenneforum.org/lmh/

hbock · 2004-01-19, 00:29

I've just repeated this on my system (Athlon 1.2 , 21.3.1 and 21.4.1 tested):

Code:

M195000023 has a factor: 23790002807
M195000079 no factor from 2^0 to 2^50, WX1: B830ACC7
M195000107 has a factor: 1380243907364191
M195000109 has a factor: 182394911954023
M195000119 no factor from 2^0 to 2^50, WX1: B82FACC8
M195000131 has a factor: 14040009433
M195000161 no factor from 2^0 to 2^50, WX1: B830ACCB
M195000193 no factor from 2^0 to 2^50, WX1: B828ACD2
M195000241 no factor from 2^0 to 2^50, WX1: B82FACCE
M195000259 has a factor: 2839593771559
M195000259 has a factor: 39497692460969
M195000269 has a factor: 6578139074447
M195000271 has a factor: 129175979521241
M195000307 no factor from 2^0 to 2^50, WX1: B81FACD0
M195000317 has a factor: 147810240287
M195000359 no factor from 2^0 to 2^50, WX1: B82BACD0
M195000373 no factor from 2^0 to 2^50, WX1: B839ACD2
M195000383 has a factor: 113005061953097
M195000391 has a factor: 87750175951
M195000419 has a factor: 87324697635743
M195000433 has a factor: 18720041569
M195000493 has a factor: 1170002959
M195000511 no factor from 2^0 to 2^50, WX1: B820ACCE

So prime95 found all factors plus a 51 bit factor (M195000107 has a factor: 1380243907364191).
We know that not all factors will be found but at least the smallest ones should. Splitting the work in two or more runs helps to find more factors and maybe all if only one bit depth is checked (b1=b2). Unfortunately this takes much longer time.

ColdFury · 2004-01-19, 02:41

It should be trivial to write a stand-alone factoring program using GMP. It would be no-where as fast as Prime95, but the factoring is already pretty fast at such high exponents.

nfortino · 2004-01-19, 03:19

Quote:

Originally posted by Xyzzy
We do need to figure out how to reverse engineer the .cmp file format so we can make new worktodo.ini files...

I have been working on reverse engineering the .cmp format, and I was successfully able to create a nofactor.cpm for the normal GIMPS data, and decompress it using decomp, getting back the same file. I went to do the same with this new data and found that 100923007 is in neither factor nor nofactor, nor is it in the raw data file (and yes, it is prime). Unfortunately, my method requires that every prime above the start point be in nofactor, or factor. Up to 100923007, my program worked perfectly.

cheesehead · 2004-01-19, 21:01

Quote:

Originally posted by ET_
My humble question is: should we go on trial-factoring Mersenne numbers with this version of prime95, missing some exponents and forcing in future some LL-test and refactoring that could be avoided, or wait until a new version of (say) Factoring95 is released?

It's fun to factor numbers, but maybe we're giving Will Eddington an incomplete batch of data, and this worries me.

As long as we cannot certify that no possible factor remains below the specified test limit bit level, we should not report unsuccessful TF tests to Will.

We should report only the exponents for which (verified) factors are found.

And as long as we can't certify that the reported factors include the smallest factor for their exponents, we should tag each report to Will with a disclaimer that the smallest factors may not have been discovered (as has been the case with Prime95 for a while).

Xyzzy · 2004-01-23, 03:58

http://www.mersenneforum.org/lmh/

factor.zip has been updated...

4,412,421 factors...

2004-01-17, 19:59	#1
bayanne "Tony Gott" Aug 2002 Yell, Shetland, UK 163₁₆ Posts	And onwards... Hey, this is all good fun, but whatabout moving the bit depth upwards. Do we need to analyse out own results, so that we can then say take it up to 2^52. What's going to be the procedure for taking a range for the future at these higher bit depths.....

2004-01-17, 21:43	#2
Xyzzy Aug 2002 5×1,697 Posts	We can either use one of the standalone factoring programs, or wait for George to remove the limitations in Prime95... We do need to figure out how to reverse engineer the .cmp file format so we can make new worktodo.ini files... I have all the raw data files and I will put them online in an hour or so... Maybe someone out there can help with figuring out what has been done and what we need to do... I'll post here once the files are online...

2004-01-17, 23:37	#3
Xyzzy Aug 2002 5·1,697 Posts	Raw files: http://www.mersenneforum.org/lmh/data/

2004-01-18, 23:09	#5
Xyzzy Aug 2002 5×1,697 Posts	Every factor we find just reduces the number left to be checked... Even if we are missing factors, we've eliminated so many that someone could run the same work behind us in a fraction of the time... Once we have the data files set up properly, creating a new worktodo.ini will be trivial...

2004-01-19, 00:25	#6
Xyzzy Aug 2002 10000100100101₂ Posts	Okay, the factor data file is done... All the factors were verified... We have found 3,663,418 factors so far... You can download the factors file here: http://www.mersenneforum.org/lmh/

2004-01-19, 02:41	#8
ColdFury Aug 2002 2⁶×5 Posts	It should be trivial to write a stand-alone factoring program using GMP. It would be no-where as fast as Prime95, but the factoring is already pretty fast at such high exponents.

2004-01-23, 03:58	#11
Xyzzy Aug 2002 5×1,697 Posts	http://www.mersenneforum.org/lmh/ factor.zip has been updated... 4,412,421 factors...