Require P-1 and other factoring work to be done on reliable machines

[tha]

Originally Posted by Madpoo

That same user has checked in 2250 total "no factor by P-1" results...

Interesting, how many factors using P-1 did he turn in?

Also, could you PM me that list? I am considering double checking the P-1 work he did.

Given that 3 out of 35 results were from him I think we should consider rerunning your query using a larger dataset to see if we can smoke out more machines.

[Madpoo]

Originally Posted by tha

Interesting, how many factors using P-1 did he turn in?

Can't really say... these are coming from the old logs and unfortunately there's no way to tell whether it found the factor by TF, P-1, etc. You could make some educated guesses in some of them, I'm sure.

In total this user found 1375 factors.

Originally Posted by tha

Also, could you PM me that list? I am considering double checking the P-1 work he did.

Given that 3 out of 35 results were from him I think we should consider rerunning your query using a larger dataset to see if we can smoke out more machines.

I actually figured out I could output YAFU info to a single line of text with the -of option, so I did that and I'm getting something dodgy but working put together to figure out which of the P-1 runs he did where a factor was found later are actually P-1 "misses" or not. It's a little hodge podge of coding and very hacky, and I'd be embarrassed to show it to anyone, but it seems like it might work.

One oddity is that YAFU would sometimes show the factors out of order (usually the two highest anyway). I don't think that'll bother me too much but if it listed a really small factor last in the list of it's output it'd throw me off, because I'm definitely not doing any checking for stuff like that in my hurry to put this together.

[Madpoo]

Originally Posted by Madpoo

I actually figured out I could output YAFU info to a single line of text with the -of option, so I did that and I'm getting something dodgy but working put together to figure out which of the P-1 runs he did where a factor was found later are actually P-1 "misses" or not. It's a little hodge podge of coding and very hacky, and I'd be embarrassed to show it to anyone, but it seems like it might work.

I was closer to having that done than I thought...

Here's a list of exponents where I think "spicer" missed them? Someone else can make sure I'm right. :)
10409033
10853081
11144281
12111313
12560221
12782239
13016743
13055659
13539641

It has the 3 you previously noted and the extra one I spotted manually, plus 5 more. I found these purely by automation and I didn't double-check any myself, so I blame any false positives on the machine, of course. Definitely not my hodge podge of scripts.

[tha]

Originally Posted by Madpoo

It has the 3 you previously noted and the extra one I spotted manually, plus 5 more. I found these purely by automation and I didn't double-check any myself, ...

They all seem to be missed based on the public info in the Mersenne.org and mersenne.ca database.

I would like to take that full list and double check all the P-1 work he did.

[Madpoo]

Originally Posted by Madpoo

...It has the 3 you previously noted and the extra one I spotted manually, plus 5 more. I found these purely by automation and I didn't double-check any myself, so I blame any false positives on the machine, of course. Definitely not my hodge podge of scripts.

I started something running earlier to do YAFU to get all the breakdown of factor info on all of the "P-1 no factor, but a factor found later".

I then took those results and smushed it into a temp table and cross-checked with the actual P-1 info.

Unfortunately as I mentioned, YAFU was sometimes giving the list of factors in an order that wasn't smallest to largest. That breaks the process I was using which basically just snags the last 2.

Example, for exponent 1411427 with factor 1337361175439390575822903, YAFU gave me this output:
(((1337361175439390575822903-1)/2)/1411427)/3/1753/35597/14683/172357

My script picked up 14683 and 172357 but 35597 is the actual 2nd largest one, and that made a big difference. A couple of P-1 runs that I thought missed it were actually okay (B1 was only 15K in those cases).

So, can YAFU spit out factors sorted numerically? Otherwise I'll have to rework how I'm parsing the output... grab all the factors and sort them, blah blah. Ugh. I thought I had it figured out...

[axn]

Originally Posted by Madpoo

I started something running earlier to do YAFU to get all the breakdown of factor info on all of the "P-1 no factor, but a factor found later".

I then took those results and smushed it into a temp table and cross-checked with the actual P-1 info.

Unfortunately as I mentioned, YAFU was sometimes giving the list of factors in an order that wasn't smallest to largest. That breaks the process I was using which basically just snags the last 2.

Example, for exponent 1411427 with factor 1337361175439390575822903, YAFU gave me this output:
(((1337361175439390575822903-1)/2)/1411427)/3/1753/35597/14683/172357

My script picked up 14683 and 172357 but 35597 is the actual 2nd largest one, and that made a big difference. A couple of P-1 runs that I thought missed it were actually okay (B1 was only 15K in those cases).

So, can YAFU spit out factors sorted numerically? Otherwise I'll have to rework how I'm parsing the output... grab all the factors and sort them, blah blah. Ugh. I thought I had it figured out...

yafu trial((f-1)/2/p,B1)

Substitute f,p, and B1 with actual values. If it is tough to get B1 in there, use a large value like 10,000,000.

[R.D. Silverman]

Originally Posted by Madpoo

I started something running earlier to do YAFU to get all the breakdown of factor info on all of the "P-1 no factor, but a factor found later".

Why? What on Earth for?

A pointless exercize.

[tha]

Originally Posted by R.D. Silverman

Why? What on Earth for?

I do care about Mersenne primes. Recently, I even switched one machine to doing three first time LL tests. But I care about factors too. Even if a candidate exponent is proven composite, I like it if one or more factors are known. Just for the sake of it. And I dislike it if the database contains a wrong entry 'no factors below' or 'none within boundaries B1/B2' when that does not happen to be the case.

[R.D. Silverman]

Originally Posted by tha

I do care about Mersenne primes.

I hope you will be very happy with them.

But I care about factors too. Even if a candidate exponent is proven composite, I like it if one or more factors are known. Just for the sake of it.

The sake of what??? The way you describe it, it seems like an obsession.

Once a candidate has been proven composite, factors become irrelevant

[bloodIce]

Originally Posted by R.D. Silverman

Once a candidate has been proven composite, factors become irrelevant

What is considered proven composite by the golden standards?

[Madpoo]

Originally Posted by R.D. Silverman

Why? What on Earth for?

A pointless exercize.

Looking for anomalies. I guess it's not something everyone enjoys, but to each their own. But couldn't you just as easily have said the same thing about GIMPS as a whole?

Why? What on Earth for? ... A pointless exercize.

Clearly different people like doing different things. I like going through the data and looking for weird things, doing triple-checks, etc. I know it's like looking for a needle in a haystack when I think there's only a 0.0001% chance of a needle being there at all, but so what?

Meanwhile, I'm still waiting until I can clean up the results a bit, but I will say that I've identified a few users that seem to have pumped out a lot of P-1 misses.

So think of it this way: If I could find some user who did a lousy job at P-1 work, and those exponents are still awaiting double-checks, then re-running the P-1 *might* just save a double-check.