mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   Wagstaff PRP Search (https://www.mersenneforum.org/forumdisplay.php?f=102)
-   -   There are no new Wagstaff primes with exponent below 10 million (https://www.mersenneforum.org/showthread.php?t=24185)

GP2 2019-03-19 16:06

There are no new Wagstaff primes with exponent below 10 million
 
In an effort that ran until 2013, a group including Tony Reix, Vincent Diepeveen, Paul Underwood and Jeff Gilchrist tested all Wagstaff numbers below 10 million. In 2010 they [URL="https://mersenneforum.org/showthread.php?t=13108"]found one Wagstaff prime, with exponent 4031399[/URL].

This used the Vrba-Reix algorithm as implemented in the LLR software package by Jean Penné. This is a heuristic method that is not mathematically proven.

Starting last August, I began double-checking this range and today the testing reached the 10 million milestone. As might have been expected, no new primes have been found.

Note this is not double-checking in the strict sense of verifying matching residues, since it uses the standard 3-PRP test and mprime software, updating to new versions and new builds as they become available. For exponents up to about 4.6 million the testing used a slightly-modified 29.4 version that printed out 2048-bit residues instead of 64 bits. Internally, the program has always used full-sized residues, so the algorithm was unchanged, it was merely a matter of printing out the extra bits. After this, the testing used mprime versions 29.5 and 29.6, which print 2048-bit residues as a standard feature.

These are "type 5" residues, which means the value calculated is [c]3^(2^p) mod (2^p + 1)[/c] and the least-significant 2048 bits are recorded. Since mprime uses Gerbicz error checking, there is reason to believe the results are reliable. As discussed [URL="https://mersenneforum.org/showthread.php?t=24094"]in another thread[/URL], prior to version 29.6 build 2 there was a flaw in mprime's implementation of Gerbicz error checking that could allow an undetected error to be introduced in the last few iterations on machines with unreliable hardware. However I am using c5 servers on AWS, which have ECC memory and a very strong track record of error-free results.

I am evaluating whether to continue double-checking beyond 10 million, since the computations are obviously getting more expensive. Currently axn is testing the 10.0 to 10.1 million range.

The next two Wagstaff primes with exponents 13347311 and 13372531 [URL="https://mersenneforum.org/showthread.php?t=18569"]were found by Ryan Propper in 2013[/URL]. He searched at least parts of the 10M, 11M, 12M and 13M ranges using 3-PRP testing and PFGW software, but it is not known whether the coverage was exhaustive. Unfortunately Ryan does not recall exactly which exponents he tested. Nonetheless my hunch at this point is that there are no new Wagstaff primes below 14 million.

[URL="http://mprime.s3-website.us-west-1.amazonaws.com/wagstaff/"]A mini-website[/URL] contains links to flat files with known factors and 2048-bit residues.

diep 2019-03-19 16:19

Many thanks for the report. I did TF above 10M - what nowadays with GPU's is way faster and deeper to achieve. Yet we didn't test a single exponent above 10mbit.

Of course you won't find anything new until those 2 PRP's that suddenly were posted. out of the blue.

Bobby Jacobs 2019-07-26 18:21

Have you checked up to 13347311 yet? I would like to know if 13347311, 13372531 are really the next Wagstaff prime exponents.

Batalov 2019-07-27 00:14

[QUOTE=GP2;511141]Nonetheless my hunch at this point is that there are no new Wagstaff primes below 14 million.[/QUOTE]
I would not bet on it.

Sometimes there are gaps and holes in large-size computations, and when someone gets around to fill up the gaps - a new prime is found. This happened in one of the PrimeGrid projects before, iirc. This also happened in Gaussian-Mersenne search while closing a gap that was left undone.

GP2 2019-07-27 08:52

[QUOTE=Bobby Jacobs;522352]Have you checked up to 13347311 yet? I would like to know if 13347311, 13372531 are really the next Wagstaff prime exponents.[/QUOTE]

It's effectively paused for the time being. Some very small-scale testing is still being done in the 10.1 million range, but at a much slower rate.

One issue is that GPUs like the recent Radeon VII have proven to be startlingly more cost-effective for Mersenne testing than CPU-based testing, perhaps by up to an order of magnitude. If the gpuOwL program is eventually adapted to do Wagstaff testing in addition to Mersenne testing, then that would be the way to go, rather than using mprime. Sadly, the Radeon VII has already been discontinued, but there will surely be other suitable GPUs. So I'd rather wait and get more bang for the buck.


All times are UTC. The time now is 07:44.

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2022, Jelsoft Enterprises Ltd.