mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Data > Marin's Mersenne-aries

Reply
 
Thread Tools
Old 2021-09-23, 14:31   #56
masser
 
masser's Avatar
 
Jul 2003
Behind BB

23·13·17 Posts
Default

Quote:
Originally Posted by Uncwilly View Post

Or other things can happen....
masser is offline   Reply With Quote
Old 2021-09-23, 14:42   #57
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

17·349 Posts
Default

Prediction is hard, especially about the future.
kriesel is offline   Reply With Quote
Old 2021-09-23, 22:37   #58
greenskull
Xebeche
 
greenskull's Avatar
 
Apr 2019
🌺🙏🌺

6668 Posts
Default

Quote:
Originally Posted by kriesel View Post
which states in part:
...
later in date predicted.
I will read it when I take a vacation and have more spare time :)

PS The forecast error is best taken into account depending on the distance from which the forecast was made.
Since making a error for a couple of weeks from a distance of 1 year and from a distance of 1 month are completely different things.

Last fiddled with by greenskull on 2021-09-23 at 22:46
greenskull is offline   Reply With Quote
Old 2021-09-24, 00:58   #59
tuckerkao
 
"Tucker Kao"
Jan 2020
Head Base M168202123

3×191 Posts
Default

Quote:
Originally Posted by greenskull View Post
My prediction is November 22, 2021
95%, ±82 days
50%, ±25 days

Let's see :)
There are only 17 outstanding exponents left until M#48 becomes official, so the prediction has been really accurate.

Wouldn't it be easier if other users just run the PRP tests for these exponents, less likely to encounter the Gerbicz errors than the Jacobi errors.

Last fiddled with by tuckerkao on 2021-09-24 at 01:06
tuckerkao is online now   Reply With Quote
Old 2021-09-24, 01:23   #60
Uncwilly
6809 > 6502
 
Uncwilly's Avatar
 
"""""""""""""""""""
Aug 2003
101×103 Posts

13×19×41 Posts
Default

Quote:
Originally Posted by tuckerkao View Post
Wouldn't it be easier if other users just run the PRP tests for these exponents, less likely to encounter the Gerbicz errors than the Jacobi errors.
Uncwilly is offline   Reply With Quote
Old 2021-09-24, 02:03   #61
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

17·349 Posts
Default

required reading
kriesel is offline   Reply With Quote
Old 2021-09-24, 02:09   #62
tuckerkao
 
"Tucker Kao"
Jan 2020
Head Base M168202123

3·191 Posts
Default

Quote:
Originally Posted by kriesel View Post
Encounter Jacobi errors in LL are much more likely than Gerbicz errors in PRP. Also the PRP certifications of those exponents won't take longer than 1 hour for most users.

Last fiddled with by tuckerkao on 2021-09-24 at 02:25
tuckerkao is online now   Reply With Quote
Old 2021-09-24, 02:35   #63
LaurV
Romulan Interpreter
 
LaurV's Avatar
 
"name field"
Jun 2011
Thailand

22·11·223 Posts
Default

Methink encountering errors has nothing to do with what math you use to spot the error and to recover from that error. Errors pop up if you have shitty hardware, or try to push it too hard.
LaurV is offline   Reply With Quote
Old 2021-09-24, 12:25   #64
Happy5214
 
Happy5214's Avatar
 
"Alexander"
Nov 2008
The Alamo City

23×97 Posts
Default

Quote:
Originally Posted by tuckerkao View Post
Encounter Jacobi errors in LL are much more likely than Gerbicz errors in PRP. Also the PRP certifications of those exponents won't take longer than 1 hour for most users.
I would have guessed the opposite. On bad hardware, Gerbicz should catch more errors than Jacobi (key word is catch; the actual number of bad iterations should be roughly equal either way), but also have the ability to correct the bad iterations more efficiently (wrt granularity/frequency of checks and the probability of an error being caught), thus producing more reliable results.
Happy5214 is offline   Reply With Quote
Old 2021-09-24, 14:09   #65
firejuggler
 
firejuggler's Avatar
 
"Vincent"
Apr 2010
Over the rainbow

268510 Posts
Default

Lets hope it id before the new year, and i'll be happy.
firejuggler is online now   Reply With Quote
Old 2021-09-24, 14:29   #66
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

17·349 Posts
Default

Quote:
Originally Posted by Happy5214 View Post
Gerbicz should catch more errors than Jacobi (key word is catch; the actual number of bad iterations should be roughly equal either way), but also have the ability to correct the bad iterations more efficiently (wrt granularity/frequency of checks and the probability of an error being caught), thus producing more reliable results.
Correct. Jacobi symbol check is done infrequently, only ~twice a day because it is computationally costly. Its detection rate is only ~50% of errors. If done after every LL iteration it would cost more than it might save, and still only detect ~75% of errors. For Mersennes it yields +1 or -1 IIRC; there are only 3 possible values in general, 1, -1, 0, so an error going undetected is quite probable. This post describing the Jacobi symbol check led to some quick implementation, and consideration of what other checks might be available.
Robert Gerbicz' post describing the check for PRP followed days later. Both Jacobi and GEC are typically applied by the software authors to use around 0.2% of computing time by default, with some user control for more or less cost & frequency.

Gerbicz' check is highly effective, since it can return one of a great many values, and only one will match.
Empirically, PRP/GEC has shown ~24ppm error in completed tests, ~800 times lower than LL. And some of that error count was from outside / after the GEC check, handling the final residue produced. Prime95 got hardened against that after some such errors were identified. See https://www.mersenneforum.org/showpo...40&postcount=4
These number-theory based checks are in addition to and independent of other inexpensive measures, such as round-off magnitude checking, or sum of inputs vs. sum of outputs.
kriesel is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Let's finish primality verification through Mp#49*, M(74 207 281) firejuggler Marin's Mersenne-aries 61 2021-12-01 23:45
Mission Creep davieddy PrimeNet 14 2011-12-10 20:55
Doing more about LL test verification jasong Software 13 2010-03-10 18:53
Mission Accomplished garo Soap Box 13 2009-01-22 20:10
First mission GP2 Completed Missions 2 2003-09-28 23:16

All times are UTC. The time now is 13:11.


Wed Dec 8 13:11:17 UTC 2021 up 138 days, 7:40, 1 user, load averages: 1.52, 1.70, 1.68

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.