Double-checking attempts and successes

[Madpoo]

Originally Posted by Viliam Furik

But the bad residues are not matching. In the above post, I was describing the probability of two bad residues randomly matching.

There are some truly bizarre example cases out there... exponents that had a LOT of bad runs. Some like this, I'm guessing the user was testing the same exponent on multiple machines or something? Many of which were bad. Or it was one of those odd cases where the machine wasn't clearing the worktodo so it just kept doing the same exponent repeatedly. Either way... chatmate sure kept at it:
M2397103

Then this one... EDNY just couldn't get a matching one no matter how hard they tried:
M16073131

Things like this too... and one of the reasons I like to do independent verification of results. Who's to say the user kept trying to get a match to their first one and finally gave up and forged a "match"? (they didn't, but still...)
M23057101

Now come the odd ones... multiple bad results that have matching residues. Yes, it's extremely improbable, and how would you catch it if it happened? Well, that's why we also require different shift counts, because sometimes users submit the same result multiple times (and from different accounts). It's accidental almost all the time... user found an old results file and wasn't sure it was submitted, so they manually resubmit but didn't logon to their account, or whatever else.

M17650979

Now for the really weird ones:
M4589707
M39988591
M42525269 (well, this user had a known issue on these last 3)
M43714607
M45111893

Those first 2 are definitely a surprise. In the case of M4589707 there was already one different res64 from a different user, but then those 2 "matching" from Klaus with different shift counts? Pretty sketchy.

The 2nd one, M39988591, justified my whole thing about making sure all verifications are done by independent people... the same person shouldn't be allowed to turn in their own self check. Why? Well, it's why we mask the last hexits of the residue, to prevent cheating. Someone would have to turn in as much as 256 guesses before getting a match (well, on average more like 128) and we'd spot that super easy. Even one wrong "guess" where only the last couple hexits of a residue changed would be a dead giveaway. But the user who did the first one would know it and can fake a result with a different shift count, and there you go.

Now... I'm not saying Klaus or Kali cheated, but you tell me... how else would that happen? Same partial residue, same user, supposedly different shift counts so it wasn't a case of submitting the same result twice... yeah, no other way I can think of besides forging it.

What's to stop someone from creating multiple accounts and cheating that way, by submitting them from different accounts so they're harder to spot? Yeah... it's a problem. As computers get faster though, we'll spot check previous results and eventually we may catch others.

As it is, I've personally done my own test on all exponents below 4M or 5M or whatever, just because it was trivial at the time. Imagine that someday maybe it'll be trivial to do the same up to 30M, 40M, or more. Eventually get through all of the LL tests, leaving us with the (harder to forge?) PRP + cert results.

[ixfd64]

Off-topic, but glad to see you here again, Aaron. I was wondering where you went.

[Uncwilly]

I have seen him around behind the scenes picking up some TC and QC exponents. No vacation in Club Fed.

[tuckerkao]

Originally Posted by Madpoo

M42525269 (well, this user had a known issue on these last 3)
M43714607
M45111893

You cannot have a good C-LL with the residue of all 0's. Best to run the PRP tests on those exponents. A true P-LL requires the P-PRP result and the successful certification.

[slandrum]

Originally Posted by tuckerkao

You cannot have a good C-LL with the residue of all 0's. Best to run the PRP tests on those exponents. A true P-LL requires the P-PRP results.

How many Mersenne primes have you found with LL?

[tuckerkao]

Originally Posted by slandrum

How many Mersenne primes have you found with LL?

I ran both the PRP and LL for M1257787 once to make sure my machine performed correctly, but I haven't tried on other exponents that are larger than 1M.

I ran 1 LL test on M103374163 a long time ago, but no one has tried to double check it anyway.

[kriesel]

Originally Posted by slandrum

How many Mersenne primes have you found with LL?

No mersenne prime has yet been found by PRP first then LL confirmation.
Since PRP began around 77.5M, only one known might have been found by PRP, but was found by LL first.
https://www.mersenne.org/report_expo...exp_hi=&full=1
Hundreds of false positive 0 res64 have been reported for LL, mostly by CUDALucas <v2.06.
https://www.mersenneforum.org/showpo...40&postcount=4

[tuckerkao]

ATH is working on the Double Checks of LL tests with no first time LL results - M106314457 and M106316611.

[Zhangrc]

Originally Posted by tuckerkao

ATH is working on the Double Checks of LL tests with no first time LL results - M106314457 and M106316611.

More likely they are PRP proofs. The first time LLs seem to take months to finish.

[tuckerkao]

Originally Posted by Zhangrc

More likely they are PRP proofs. The first time LLs seem to take months to finish.

There are no finished PRP tests for those 2 exponents either.

[ATH]

Originally Posted by tuckerkao

ATH is working on the Double Checks of LL tests with no first time LL results - M106314457 and M106316611.

I grabbed them as they expired and they were already registered as LL not PRP, so I had to use LL to get them, but I will do a PRP proof on them instead.