Riesel base 2 discussion - Page 6

gd_barnes · 2009-04-22, 06:45

OK, very good. That would be a good excercise. To be more specific:

I would suggest trying it at P=1P intervals for a range of P=10G. That is:

P=3000T-3000.010T
P=4000T-4000.010T
P=5000T-5000.010T
(etc.)

Do that and when you find a range that you start finding factors, then go to halfway between the two ranges and try again. Keep doing that until you narrow it down to the actual sieve depth. Example:

Let's assume that you find factors in P=5000T-5000.0010T, then try 4500T-4500.010T. If you find factors there, split the difference with the 4000T-4000.010T range and try 4250T-4250.010T, etc.

Do the above and once you've determined which P=1P range there are missing factors, it will only take you log(1P)/log(2) = ~50 attempts to get to the exact sieve depth. But we don't really need to do ~50 attempts, we could narrow it down to the closest P=1M range in ~30 attempts or the closest P=1G range in ~20 attempts. Sieving an extra P=1M or even 1G for this would not take much time so that should be sufficient.

The key here is that a big enough range is tested so that it has a chance to find factors. Look at the expected # of factors that sr2sieve gives for that. If it is 10 or higher, then you should be good.

I'll be curious what you come up with.

Gary

KEP · 2009-04-22, 06:52

Thanks for your advice, I'll follow it and get back to you as soon as I've something to report to you

Kenneth

Dougal · 2009-08-25, 21:58

any news on weather this is going again?id love to do some testing on this,id be willing to fire 4-6 cores at it for about 6 months then maybe cut back to 2-4 cores for the fore-seeable future.

gd_barnes · 2009-08-27, 08:31

Quote:

Originally Posted by Dougal

any news on weather this is going again?id love to do some testing on this,id be willing to fire 4-6 cores at it for about 6 months then maybe cut back to 2-4 cores for the fore-seeable future.

As far as we are concerned, Riesel Sieve is dead. I removed all links in the CRUS web pages to Riesel Sieve many months ago because it has no info. Any references refer to the "former Riesel Sieve project", which is what it now is. See your PM for more info.

Within 1-2 months, I am hoping to have all NPLB and CRUS servers moved over to my machines. At that time, I/we may consider setting up a server for processing part of Riesel Base 2. If we do that, a PRPnet server would likely be best because the inherent problems that PRPnet has are related to too many k/n pairs being returned at once. This would not be a problem for tests at n>3M.

Keep in mind that tests at n>3M take ~2-4 hours each. Testing at that level is definitely not child's play! :-) The chance of a prime with only 4-6 cores running for 6 months is extremely remote.

Gary

ValerieVonck · 2010-03-05, 10:21

Anyone still interested to reboot this project?

Dougal · 2010-03-05, 11:57

im working on this at the moment,il have to send gary the results some of these days.i think it should be restarted,but first you need someone willing to run it,which would require alot of work.

ValerieVonck · 2010-03-05, 12:19

IMHO, we should set several goals before rebooting this project.
E.g.:

1) I have still a dat file, so tests can be made from it
2) We should then determine a cut off point from where to restart the llr testing e.g. 2M for all k's because no one has access to the residues to verify them
3) Should this a k by k seach, test a k until we find a prime, rinse & repeat?
4) Or llr tests for all k's simultanesoly (?)?
5) Hosting of the llr files: @ mersseneforum.org?
6) Should this be co-ordinated in a separated thread?
7) Who will step up for the coordination? I am willing to help to coordinate this
8) Short Term: can be probably done @ mersenneforum, self llr crunching
8b) Medium Term: proper website & llrnet implemenation
8c) Long Term: Boinc project?
9) Should we do a handover to Primegrid?
=> They have the resources (people and hardware)
10) Should we sieve? IMHO, no, maybe if we reach 10M
11) Should we do P-1? IMHO, maybe, but the p-1 effort I coordinated @ Rieselsieve yielded a very low number of factors, ok a factor found is an LLR-test spared, maybe other B1-B2 bounds?

These are my $0.02 ...

So the gauntlet is down again

Anyone up for a challenge ?

(Excuse when I offended someone)
Regards
C

kar_bon · 2010-03-05, 12:51

right, first 'we' should determine the source to start with (pairs, results, factors).

a new thread should also be right to give that big project an own start-point.
and as we can see, for many other projects there'll be questions/suggestions only for that project, so collect them all under one thread!

BOINC or LLR(net)? some of those here certainly know what i think and LLR is a good place to start! more details later!

rogue · 2010-03-05, 13:42

If this were a small effort, then I think that hosting somewhere else would be fine, but because it has such high visibility, I suggest that you turn it over to PrimeGrid. The BOINC client is being used by SoB.

KEP · 2010-03-05, 13:53

Please remember everyone, that the testing depth of 2.947 M, includes a complete doublecheck up to that depth. So a good starting point would be n=2.947M...

I personally and several other has asked numerous of times to get PG to do an effort in trying to tackle this conjecture, since the testings has grown to a huge testing time and also since they have the needed resources to come somewhere... however the conclusion everytime from Rytis is that "they want have anything to do with us either", so unless we can find a way to convince Rytis and the others in the PG crew that Rieselsieve is in fact dead and not just under reconstruction, then PG should be excluded as a feasible alternative...

Regarding sieving, it makes no sence, since sieving appeared to be ended around 5.4P (5400 T).

Now a question for you Dougal, how much work have you done since I send you the untested k/n pairs for Riesel base 2 conjecture?

And of course, I wouldn't mind a drive, where one can pick up a few ranges here and there and do whatever one can support this almost megatesting project with. But maybe we can save some time, if someone can get in contact with one or two of the then responsible people at rieselsieve, and have them send us the exact testing depth aswell the exact doublecheck testing depth.

Regards

KEP

Dougal · 2010-03-05, 22:08

KEP,i split the file into 5 smaller files,but i only have 4 cores on it,so 1 file is completely untested.i only have about 2000-3000 tests done.

2009-04-22, 06:45	#56
gd_barnes May 2007 Kansas; USA 101×103 Posts	OK, very good. That would be a good excercise. To be more specific: I would suggest trying it at P=1P intervals for a range of P=10G. That is: P=3000T-3000.010T P=4000T-4000.010T P=5000T-5000.010T (etc.) Do that and when you find a range that you start finding factors, then go to halfway between the two ranges and try again. Keep doing that until you narrow it down to the actual sieve depth. Example: Let's assume that you find factors in P=5000T-5000.0010T, then try 4500T-4500.010T. If you find factors there, split the difference with the 4000T-4000.010T range and try 4250T-4250.010T, etc. Do the above and once you've determined which P=1P range there are missing factors, it will only take you log(1P)/log(2) = ~50 attempts to get to the exact sieve depth. But we don't really need to do ~50 attempts, we could narrow it down to the closest P=1M range in ~30 attempts or the closest P=1G range in ~20 attempts. Sieving an extra P=1M or even 1G for this would not take much time so that should be sufficient. The key here is that a big enough range is tested so that it has a chance to find factors. Look at the expected # of factors that sr2sieve gives for that. If it is 10 or higher, then you should be good. I'll be curious what you come up with. Gary Last fiddled with by gd_barnes on 2009-04-22 at 06:47

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Riesel Base 5 LLR	em99010pepe	Sierpinski/Riesel Base 5	8	2010-06-08 21:21
Discussion about new Riesel base 3 subforum	KEP	Conjectures 'R Us	20	2008-11-15 10:15
Sierpinski/Riesel Base 10	rogue	Conjectures 'R Us	11	2007-12-17 05:08
Sierpinski / Riesel - Base 23	michaf	Conjectures 'R Us	2	2007-12-17 05:04
Riesel Base 5 discussion	rogue	Sierpinski/Riesel Base 5	2	2005-12-21 08:00

2009-04-22, 06:52	#57
KEP Quasi Admin Thing May 2005 1111000110₂ Posts	Thanks for your advice, I'll follow it and get back to you as soon as I've something to report to you Kenneth

2009-08-25, 21:58	#58
Dougal Jan 2009 Ireland 2×3×31 Posts	any news on weather this is going again?id love to do some testing on this,id be willing to fire 4-6 cores at it for about 6 months then maybe cut back to 2-4 cores for the fore-seeable future.

2010-03-05, 10:21	#60
ValerieVonck Mar 2004 Belgium 841₁₀ Posts	Anyone still interested to reboot this project?

2010-03-05, 11:57	#61
Dougal Jan 2009 Ireland 2×3×31 Posts	im working on this at the moment,il have to send gary the results some of these days.i think it should be restarted,but first you need someone willing to run it,which would require alot of work.

2010-03-05, 12:19	#62
ValerieVonck Mar 2004 Belgium 349₁₆ Posts	IMHO, we should set several goals before rebooting this project. E.g.: 1) I have still a dat file, so tests can be made from it 2) We should then determine a cut off point from where to restart the llr testing e.g. 2M for all k's because no one has access to the residues to verify them 3) Should this a k by k seach, test a k until we find a prime, rinse & repeat? 4) Or llr tests for all k's simultanesoly (?)? 5) Hosting of the llr files: @ mersseneforum.org? 6) Should this be co-ordinated in a separated thread? 7) Who will step up for the coordination? I am willing to help to coordinate this 8) Short Term: can be probably done @ mersenneforum, self llr crunching 8b) Medium Term: proper website & llrnet implemenation 8c) Long Term: Boinc project? 9) Should we do a handover to Primegrid? => They have the resources (people and hardware) 10) Should we sieve? IMHO, no, maybe if we reach 10M 11) Should we do P-1? IMHO, maybe, but the p-1 effort I coordinated @ Rieselsieve yielded a very low number of factors, ok a factor found is an LLR-test spared, maybe other B1-B2 bounds? These are my $0.02 ... So the gauntlet is down again Anyone up for a challenge ? (Excuse when I offended someone) Regards C

2010-03-05, 12:51	#63
kar_bon Mar 2006 Germany 3²×17×19 Posts	right, first 'we' should determine the source to start with (pairs, results, factors). a new thread should also be right to give that big project an own start-point. and as we can see, for many other projects there'll be questions/suggestions only for that project, so collect them all under one thread! BOINC or LLR(net)? some of those here certainly know what i think and LLR is a good place to start! more details later!

2010-03-05, 13:42	#64
rogue "Mark" Apr 2003 Between here and the 2⁴×397 Posts	If this were a small effort, then I think that hosting somewhere else would be fine, but because it has such high visibility, I suggest that you turn it over to PrimeGrid. The BOINC client is being used by SoB.

2010-03-05, 13:53	#65
KEP Quasi Admin Thing May 2005 2×3×7×23 Posts	Please remember everyone, that the testing depth of 2.947 M, includes a complete doublecheck up to that depth. So a good starting point would be n=2.947M... I personally and several other has asked numerous of times to get PG to do an effort in trying to tackle this conjecture, since the testings has grown to a huge testing time and also since they have the needed resources to come somewhere... however the conclusion everytime from Rytis is that "they want have anything to do with us either", so unless we can find a way to convince Rytis and the others in the PG crew that Rieselsieve is in fact dead and not just under reconstruction, then PG should be excluded as a feasible alternative... Regarding sieving, it makes no sence, since sieving appeared to be ended around 5.4P (5400 T). Now a question for you Dougal, how much work have you done since I send you the untested k/n pairs for Riesel base 2 conjecture? And of course, I wouldn't mind a drive, where one can pick up a few ranges here and there and do whatever one can support this almost megatesting project with. But maybe we can save some time, if someone can get in contact with one or two of the then responsible people at rieselsieve, and have them send us the exact testing depth aswell the exact doublecheck testing depth. Regards KEP

2010-03-05, 22:08	#66
Dougal Jan 2009 Ireland 2×3×31 Posts	KEP,i split the file into 5 smaller files,but i only have 4 cores on it,so 1 file is completely untested.i only have about 2000-3000 tests done.