mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > PrimeNet

Reply
 
Thread Tools
Old 2021-08-17, 05:02   #1
Zhangrc
 
"University student"
May 2021
Beijing, China

127 Posts
Default Suggestion for combining PRP-CF with P-1

Yesterday I did a PRP-CF assignment

PRP=1,2,11517659,-1,99,0,"276423817,2211390529"

I thought that if a prior P-1 test was done, it would save the PRP test, so I changed it to

PRP=1,2,11517659,-1,99,2,3,1,"276423817,2211390529"

Still no P-1 was done. I supposed that it was related to "99" in the assignment, because the Prime95 client thought that there were no more factors below 2^99, so no P-1 was needed. I changed it to

PRP=1,2,11517659,-1,67,2,3,1,"276423817,2211390529"

This time, P-1 was done, but the PRP test had no proof.

Later I found that

PRP=1,2,11517659,-1,67,2,"276423817,2211390529" would do both P-1 and PRP with proof. (deleting "3,1,")

Could the server release the assignments like this, say "67,2" instead of "99,0"?

(Also I've spotted a bug. If I get the assignment from https://www.mersenne.org/workload/, it shows

PRP=1,2,11517659,-1,99,2,3,1,"276423817,2211390529"

which could result in neither P-1 nor proof.)

Last fiddled with by Zhangrc on 2021-08-17 at 05:03
Zhangrc is offline   Reply With Quote
Old 2021-08-17, 06:23   #2
Uncwilly
6809 > 6502
 
Uncwilly's Avatar
 
"""""""""""""""""""
Aug 2003
101×103 Posts

29·349 Posts
Default

I do PRP-CF work, but I am not looking for more factors. If you want to find more factors, feel free to do more P-1. If the server started to ask me to do P-1 on this machine (as part of the PRP-CF), I would stop doing PRP-CF on it. I would take this CPU over to doing TF.

If you like doing both all the time, you can write a script that will update your assignments.
Uncwilly is offline   Reply With Quote
Old 2021-08-17, 06:42   #3
Zhangrc
 
"University student"
May 2021
Beijing, China

7F16 Posts
Default

Quote:
Originally Posted by Uncwilly View Post
I do PRP-CF work, but I am not looking for more factors.
These cofactors have no prior P-1 done. P-1 would save more time by finding factors, thus eliminating the need for PRP. And it only take a few minutes (maybe less than a minute, if you have a computer better than Delo's) to do a P-1.

By doing P-1, you test more cofactors (with an increased chance of being PRP) using less time, and you have factors as by-products.

Last fiddled with by Zhangrc on 2021-08-17 at 06:47
Zhangrc is offline   Reply With Quote
Old 2021-08-17, 07:21   #4
axn
 
axn's Avatar
 
Jun 2003

2·23·113 Posts
Default

I am not sure what you mean by "P-1 would save more time by finding factors, thus eliminating the need for PRP". The PRP has to be always run to see if the cofactor is (probable) prime or not. And then every time someone finds a new factor, you have to run another PRP test. Maybe you're saying that "PRP, P-1 factor, Another PRP" can become "P-1 factor, PRP"?

Actually, there is a test by which you can do a single PRP and generate a special residue, and if any future factor is found, the server can quickly validate if the new cofactor could be potentially prime using a cheap calculation using the special residue. But I dont think GIMPS has implemented it.
axn is online now   Reply With Quote
Old 2021-08-17, 08:11   #5
LaurV
Romulan Interpreter
 
LaurV's Avatar
 
"name field"
Jun 2011
Thailand

100110010100002 Posts
Default

Short (offtopic) note: You are wasting your time (and mine! ) by doing P-1 to those limits in 11M range. See the "thinking out loud" thread for details. Of course, anybody is free to do whatever he wants, but if you do it, do it properly.
LaurV is offline   Reply With Quote
Old 2021-08-17, 15:41   #6
Zhangrc
 
"University student"
May 2021
Beijing, China

1778 Posts
Default

Quote:
Originally Posted by LaurV View Post
See the "thinking out loud" thread for details
I've seen it, which said that we should do P-1 to adequate bounds, say B1=5M, B2=150M.

I do not consider PRP-CF as my main work type. I do it in order to fit the "Computer must have returned at least x results in the last 120 days" requirement in order to get smaller exponents.

If possible, I suggest that we should only assign PRP-CF assignments when adequate amount of TF, P-1 or ECM have been done.
Zhangrc is offline   Reply With Quote
Old 2021-08-17, 17:34   #7
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

3·11·233 Posts
Default

Quote:
Originally Posted by Zhangrc View Post
we should only assign PRP-CF assignments when adequate amount of TF, P-1 or ECM have been done.
The definition of "adequate" here is very hard to pin down.
Prime95 is offline   Reply With Quote
Old 2021-08-21, 03:55   #8
Happy5214
 
Happy5214's Avatar
 
"Alexander"
Nov 2008
The Alamo City

23·97 Posts
Default

Quote:
Originally Posted by Zhangrc View Post
Yesterday I did a PRP-CF assignment

PRP=1,2,11517659,-1,99,0,"276423817,2211390529"

I thought that if a prior P-1 test was done, it would save the PRP test, so I changed it to

PRP=1,2,11517659,-1,99,2,3,1,"276423817,2211390529" (My emphasis added)

Still no P-1 was done. I supposed that it was related to "99" in the assignment, because the Prime95 client thought that there were no more factors below 2^99, so no P-1 was needed. I changed it to

PRP=1,2,11517659,-1,67,2,3,1,"276423817,2211390529"

This time, P-1 was done, but the PRP test had no proof.

Later I found that

PRP=1,2,11517659,-1,67,2,"276423817,2211390529" would do both P-1 and PRP with proof. (deleting "3,1,")

[...]
Where I added the emphasis is why it didn't generate a proof. It was doing a type-1 PRP test, basically meaning it was doing a Fermat PRP test on the actual cofactor. Therefore, there was no proof and no Gerbicz error checking. We generally recommend type-5 PRP tests for PRP-CF exponents, which basically run a normal Fermat PRP test on Mp, but test for congruence with 3^d instead of 1. With that, you get the proof, GEC, and the same residue when you find a new factor (assuming you don't hit a PRP on the cofactor).

Quote:
Originally Posted by Zhangrc View Post
I do not consider PRP-CF as my main work type. I do it in order to fit the "Computer must have returned at least x results in the last 120 days" requirement in order to get smaller exponents.

If possible, I suggest that we should only assign PRP-CF assignments when adequate amount of TF, P-1 or ECM have been done.
First, I did not know PRP-CF results counted toward the assignment limits. Second, exponents in this range with known factors are being plowed through along a wavefront, much like the DC and FTC wavefronts up ahead, regardless of other work done. This is way ahead of the ECM range.

Last fiddled with by Happy5214 on 2021-08-21 at 03:56 Reason: Clipping
Happy5214 is offline   Reply With Quote
Old 2021-08-21, 04:54   #9
Zhangrc
 
"University student"
May 2021
Beijing, China

127 Posts
Default

Quote:
Originally Posted by Happy5214 View Post
We generally recommend type-5 PRP tests for PRP-CF exponents, which basically run a normal Fermat PRP test on Mp, but test for congruence with 3^d instead of 1. With that, you get the proof, GEC, and the same residue when you find a new factor (assuming you don't hit a PRP on the cofactor).
Did you mean this?
Attached Thumbnails
Click image for larger version

Name:	3.JPG
Views:	56
Size:	145.6 KB
ID:	25502  

Last fiddled with by Zhangrc on 2021-08-21 at 04:54
Zhangrc is offline   Reply With Quote
Old 2021-08-21, 09:00   #10
Happy5214
 
Happy5214's Avatar
 
"Alexander"
Nov 2008
The Alamo City

23·97 Posts
Default

Quote:
Originally Posted by Zhangrc View Post
Did you mean this?

(Thumbnail attached.)
Pretty much, though pardon me for leaving off moduli in my original answer.
Happy5214 is offline   Reply With Quote
Old 2021-08-21, 11:29   #11
Zhangrc
 
"University student"
May 2021
Beijing, China

127 Posts
Default

Quote:
(Also I've spotted a bug. If I get the assignment from https://www.mersenne.org/workload/, it shows

PRP=1,2,11517659,-1,99,2,3,1,"276423817,2211390529"

which could result in no proof.)
So it should be type 5 instead of 1, and that's a server bug. Maybe I should move it to "server problems" thread.

Last fiddled with by Zhangrc on 2021-08-21 at 11:30
Zhangrc is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Algorithm for combining Carmichael numbers devarajkandadai Number Theory Discussion Group 1 2019-09-21 05:40
combining different PCs to search same exponent ssybesma Information & Answers 13 2019-05-07 20:32
Combining CPUs Edmond Lounge 11 2017-07-03 16:31
Combining Msieve with CADO NFS mfeltz Msieve 10 2016-03-16 21:12
Combining NewPGen files? roger Riesel Prime Search 4 2008-01-15 00:01

All times are UTC. The time now is 08:54.


Mon Dec 6 08:54:11 UTC 2021 up 136 days, 3:23, 0 users, load averages: 1.44, 1.52, 1.52

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.