P-1 factoring anyone?

[petrw1]

Originally Posted by markr

There are only eight others in 58M with poor P-1, so here they are:
.
.
.
58020301,68,500,50000
58020709,68,100,10000
58021409,68,100,10000
.
.

Not sure how I missed these 3???

[James Heinrich]

Originally Posted by petrw1

Not sure how I missed these 3???

You didn't do it the easy way

[garo]

Thanks for the responses guys and thanks Mr. P-1 for a through post. I've thought about creating a mathematical model which assumes a certain probability that an LL machine will do a proper P-1 or a stage-1 only or nothing and then assuming that everything will have a proper P-1 by the time doublechecks roll around. And then seeing whether it made sense to do P-1 before the LL wave gets there. I didn't have much time and the analysis is non-trivial especially because the probability of finding a factor through P-1 is variable. I did a "half-P-1" on three exponents in the 52M range but am not pursuing this any further and going back to 53M.

[petrw1]

Originally Posted by James Heinrich

You didn't do it the easy way

[petrw1]

Originally Posted by James Heinrich

You didn't do it the easy way

It appears either your "easy way" is also missing some ... or I am not understanding. I am assuming exponents in a small range with the same TF and B1/B2 will have about the same chance of finding a factor.

I looked at range 7,000,000.
Your report found NONE with B1=B2=30,000 below 7,001,650

Yet
http://www.mersenne.org/report_facto...=1&B1=Get+Data
found all of these with P-1 of B1=B2=30,000:
7001069,63,30000,30000
7001083,63,30000,30000
7001233,63,30000,30000
7001341,63,30000,30000
7001359,63,30000,30000
7001377,63,30000,30000
7001437,63,30000,30000
7001537,63,30000,30000
7001639,63,30000,30000

[James Heinrich]

Originally Posted by petrw1

It appears either your "easy way" is also missing some

My page (currently) show only unassigned exponents, stuff that you can immediately grab and throw into your worktodo. Those "missing" ones are already assigned, e.g. M7001069

I can modify my page to also show currently-assigned exponents, if you think that would be useful?

[petrw1]

Originally Posted by James Heinrich

My page (currently) show only unassigned exponents, stuff that you can immediately grab and throw into your worktodo. Those "missing" ones are already assigned, e.g. M7001069

I can modify my page to also show currently-assigned exponents, if you think that would be useful?

Ah!!!! Ok.

No I think it is better the way you have it now

[KingKurly]

Originally Posted by James Heinrich

My page (currently) show only unassigned exponents, stuff that you can immediately grab and throw into your worktodo. Those "missing" ones are already assigned, e.g. M7001069

I can modify my page to also show currently-assigned exponents, if you think that would be useful?

Perhaps as a checkbox that defaults off? (Or on, depending how you word it: to be clear, the default should be to exclude currently-assigned exponents, I think we "all" agree on that.) Sometimes I look at the page and for some reason I find myself surprised when my exponents are missing... well that's because they're assigned to me, so of course they're missing! Ha!

Another suggestion, perhaps offer different sorting options? Right now it seems to sort based on percentages (which is a good default), which confused me because I was trying to find a particular range of exponents that I thought should be at the bottom (based on exponent) but were near the top (based on percentage). Or perhaps this would confuse the look/feel of the page too much... meh. Decisions, decisions...

Out of curiosity, what percentage did my work in the M198xxxx (1.98M) range wind up at? There seems to be a cap at 9.99%, and it would seem that my B1=1M, B2=30M exceeds that cap. I'm sure it's massive overkill, but it sure does find a bunch of factors! But perhaps I can scale it back a bit, speed up the process, and still keep my high rate of effectiveness.

[James Heinrich]

Originally Posted by KingKurly

Out of curiosity, what percentage did my work in the M198xxxx (1.98M) range wind up at?

You can calculate probabilities on the probability page. For those bounds, you should get around 14.3%.

Also, I have updated "the easy way" to allow up to 19.999% in the search, and also a default-off option for displaying assigned+unassigned exponents (unfortunately I can't easily flag which ones are assigned without making two queries to the server, but I do put a warning on the page).

[Mr. P-1]

Originally Posted by garo

Thanks for the responses guys and thanks Mr. P-1 for a through post. I've thought about creating a mathematical model which assumes a certain probability that an LL machine will do a proper P-1 or a stage-1 only or nothing and then assuming that everything will have a proper P-1 by the time doublechecks roll around. And then seeing whether it made sense to do P-1 before the LL wave gets there. I didn't have much time and the analysis is non-trivial especially because the probability of finding a factor through P-1 is variable. I did a "half-P-1" on three exponents in the 52M range but am not pursuing this any further and going back to 53M.

I've also thought about creating such a mathematical model, and also concluded that the analysis is non-trivial. Nevertheless it think it is obviously true that it is better for machines with at least a reasonable amount of memory to do P-1 in advance of LL machines, simply from the observation that only about half of all exponents behind the LL wave which have been P-1ed got a stage 2.

[James Heinrich]

Is there an easy way to find where the LL and DC wavefronts are?