First time LL is no more

[firejuggler]

But... what about exponent that prp say they are probable prime? The communitty will probably snatch it fast but still... What will we do? a PRP-DC? or a LL (first time LL?) and a PRP-DC?

[kriesel]

Originally Posted by firejuggler

But... what about exponent that prp say they are probable prime? The communitty will probably snatch it fast but still... What will we do? a PRP-DC? or a LL (first time LL?) and a PRP-DC?

What I expect is this:
It takes years to occur. Someone's prime95 that's PrimeNet connected gets and completes the PRP test on the lucky exponent. That someone is considered the discoverer.
Someone gets the Cert assignment for it and completes it via prime95 and PrimeNet API.
The server spots the completed PRP confirmed and attempts to alert Woltman and a select few others of a probable new prime by email.
If that fails, Madpoo sees it during server maintenance and alerts George and Ernst.
A select few trusted participants run LL tests on the fastest most reliable hardware they can use, on Mlucas, mprime or prime95, CUDALucas, and on a Gpuowl V6.11-3xx version selected for speed for the relevant fft length and Jacobi check availability.
Assuming it's not a false alarm / bug, there's a preliminary announcement of a new find without specifics, followed by an official press release, timed for high coverage, soon enough to avoid leaks, and not on the first day of April.
And there's no EFF prize awarded, because M52 is not 100Mdigit or larger.
Then it takes several more years before M53 is apparently found, which has a shot at the 100Mdigit EFF prize.
https://www.mersenneforum.org/showpo...5&postcount=14

[ATH]

Originally Posted by firejuggler

But... what about exponent that prp say they are probable prime? The communitty will probably snatch it fast but still... What will we do? a PRP-DC? or a LL (first time LL?) and a PRP-DC?

These will only be the future Mersenne Primes, and we will run 2+ LL tests on different hardware and using different software Prime95/Mlucas/Gpuowl/CUDALucas like all the previous Mersenne Primes.

We will not find a PRP that is not prime.

[kriesel]

It looks like the server assignments change has been effective.
Just now I asked it for https://www.mersenne.org/assignments...chk=1&excert=1
and of the 1000 returned, sorting by type, just eleven are LL (1.1%).
These pending LL tasks assignments were issued not later than 2021-04-08.
(Assignments would not indicate any end-user manual changes from PRP to LL though. Hopefully those are very rare.)

Originally Posted by M344587487

This is a big deal, there should be cake.

Indeed!

Originally Posted by ATH

Lucas-Lehmer tests have served us well for over 24 years. They can rest now, but will be hopefully be called upon again soon for the honor of verifying the next Mersenne Prime.

2021 April - 1996 January = 25+. So yes, definitely over 24.

Originally Posted by ATH

we will run 2+ LL tests on different hardware and using different software Prime95/Mlucas/Gpuowl/CUDALucas like all the previous Mersenne Primes.

With paired runs or sharing of interim res64 values likely on larger exponents. The bigger the exponent, the more likely a lone LL test goes wrong before completion without such checks along the way. https://www.mersenneforum.org/showpo...6&postcount=14

[kriesel]

When a new Mersenne prime is found by GIMPS, by an LL test, the first confirmation check is usually George acquires the last interim save file from the lucky discoverer and reruns it from that point to completion, to reproduce the prime result, or not.

Let us suppose that someone running Gpuowl v6.11-x gets a lucky exponent assigned and finds M52.
And they were running PRP with proof generation, and the -cleanup option.

-cleanup           : delete save files at end of run

Or they found it with Gpuowl v7.x-y PRP with proof generation. And were not using the -noclean option.

-noclean           : do not delete data after the test is complete.

The last save file is gone. It cannot be provided to George for the usual confirmatory run of the last iterations.
All that remain are the result file and the gpuowl log file.
1) Since there's been a PRP proof file generated and uploaded, does it matter there are no save files on the discoverer's system?

Now suppose it was v7.1-x, and the proof file generated was bad.
2) How does that change things?

Now suppose the exponent is over 100Mdigit. It takes 2 weeks to run a primality test on a fast gpu, longer on a multicore Xeon. LL test reliability with Jacobi check is ~1-2% error for wavefront exponents, much higher (~10%; 20% without Jacobi) for 100Mdigit.
(Challenges of large exponents https://www.mersenneforum.org/showpo...6&postcount=14)
3) How will that affect how the prime discovery confirmation is performed?

4) What techniques may be used or added to make the next Mersenne prime discovery confirmation faster or more reliable?
(Interim res64 values as additional error check input https://www.mersenneforum.org/showpo...&postcount=20; what else?)

[slandrum]

Originally Posted by kriesel

It looks like the server assignments change has been effective

I've also noticed that the trailing edge DC assignments are filling in much faster than before, probably due to several factors, but a not insignificant one is that LL FTC requests are now returning LL DC assignments.

[kriesel]

Originally Posted by slandrum

I've also noticed that the trailing edge DC assignments are filling in much faster than before, probably due to several factors, but a not insignificant one is that LL FTC requests are now returning LL DC assignments.

That effect seems modest to me; ~15-25% to Mp#48*, declining below M56M.

[slandrum]

That's a little different than what I'm talking about. The number of DC assignments at the bottom that are fully assigned before you find an available DC assignment has been growing over the last 3 weeks. The number of exponents being assigned on average/day in cat0/cat1 has increased in the last 3 weeks.

It takes a little bit of time to see that in the cleared results, but we should start seeing that soon, the new rule only went into affect 3 weeks ago.

[P-A]

As LL will be finished at some point, would it be possible to have a countdown on the milestones page to count down to last LL test being finished?

[slandrum]

There are still people choosing to do FTC LL, overriding their assignments to do it.

[Uncwilly]

And there are people doing PRP-DC's, overriding their assignments to do it,