mersenneforum.org Is it realistic to calculate and store the following info
 Register FAQ Search Today's Posts Mark Forums Read

 2020-09-17, 00:18 #1 Ensigm   Aug 2020 3·17 Posts Is it realistic to calculate and store the following info For every (composite) mersenne number, if it had a factor of x bits/digits, how likely would have the factor been already discovered. If we store by digits from 21 to 70 then there will be 50 floating point numbers for each exponent. Storing by bits for the same range will yield about 166. Either way it's not a huge demand for storage, so the only problem is to calculate it correctly. Last fiddled with by Ensigm on 2020-09-17 at 00:21
 2020-09-17, 02:08 #2 firejuggler     Apr 2010 Over the rainbow 2×1,217 Posts Yes it is possible but it is useless info. there is 3 method of factoring Mersenne. PM1, Trial factoring, and ECM. PM1 is probalistic. 53853801 has a factor. When I used it , the algo calculated that I had a 7.04% chance of finding a factor. It doesn't say anything about the size. TF has a 1/bit level chance of finding a factor. again unrelated to size. while ECM (eliptic curve method) work by size of factor. (called T level). If you complete T35, there is no chance of any factor below 35 digit on that particular M exist. Edit : Sorry for the multiple edit. I wanted to make myself clear. And sorry for anyone quoting me in between edit Last fiddled with by firejuggler on 2020-09-17 at 02:26
2020-09-17, 02:14   #3
VBCurtis

"Curtis"
Feb 2005
Riverside, CA

22×1,091 Posts

Quote:
 Originally Posted by firejuggler ]If you complete a T35 on a certain exponent and found no factor, then the chance of a sub 35 digit factor is all but nill.
This isn't quite true- when a T35 is finished, there is a 1/e chance of missing a 35 digit factor (if one exists).

It's more accurate to say that after a T35 is done, chances of a 30 or smaller digit factor is nil (1/e^-6 or worse). A 33 digit factor has about a 10% chance (1/e^2, roughly) to be missed by a T35.

2020-09-17, 07:38   #4
Ensigm

Aug 2020

3×17 Posts

Quote:
 Originally Posted by firejuggler while ECM (eliptic curve method) work by size of factor. (called T level). If you complete T35, there is no chance of any factor below 35 digit on that particular M exist.
I thought T35 only exclude a large majority of factors below 35 digits so that it makes more sense to search for factors with 35-40 digits? Someone knows more about ECM can correct me.

Quote:
 Originally Posted by firejuggler TF has a 1/bit level chance of finding a factor. again unrelated to size.
If a bit level has been TF'ed, then if a factor is in that level, the probability of discovering that factor would be 100%.

2020-09-17, 16:01   #5
VBCurtis

"Curtis"
Feb 2005
Riverside, CA

22×1,091 Posts

Quote:
 Originally Posted by Ensigm If a bit level has been TF'ed, then if a factor is in that level, the probability of discovering that factor would be 100%.
This is correct- he meant the odds of finding a factor have nothing to do with the size of the input number. That's why the data you want to collect would not be useful- none of the 3 methods he mentioned have odds changed by the size of the input.

2020-09-17, 19:54   #6
Ensigm

Aug 2020

638 Posts

Quote:
 Originally Posted by VBCurtis This is correct- he meant the odds of finding a factor have nothing to do with the size of the input number. That's why the data you want to collect would not be useful- none of the 3 methods he mentioned have odds changed by the size of the input.

I never mentioned the size of the input number. What I was talking about can be paraphrased as, in your words, "the chance of missing an x-digit factor (if one exists)". For each exponent, this will produce a line chart with bit or digit as the x-axis and probability as the Y axis.

This can be useful when people manually do curves with bounds that are not the typical values of T25, T30, etc. Under current rules I assume they are rounded down to the closest typical value? (Such as if I do 100 curves with B1=1e5, B2=1e7, they will only be counted as 100 T25 curves.)

2020-09-17, 20:05   #7
xilman
Bamboozled!

"πΊππ·π·π­"
May 2003
Down not across

277E16 Posts

Quote:
 Originally Posted by Ensigm I never mentioned the size of the input number. What I was talking about can be paraphrased as, in your words, "the chance of missing an x-digit factor (if one exists)". For each exponent, this will produce a line chart with bit or digit as the x-axis and probability as the Y axis. This can be useful when people manually do curves with bounds that are not the typical values of T25, T30, etc. Under current rules I assume they are rounded down to the closest typical value? (Such as if I do 100 curves with B1=1e5, B2=1e7, they will only be counted as 100 T25 curves.)
Good question.

What I do with the ECMNET server running at 83.217.167.17:8194 is first to multiply the number of curves by the B1 reported by a client and use that as a measure of work done. Each candidate in the server has its own record of the amount of work done by that measure. That total is used to set an appropriate B1 for a client when the candidate is allocated to it.

 2020-09-17, 21:09 #8 VBCurtis     "Curtis" Feb 2005 Riverside, CA 110C16 Posts mersennne.org does something similar- I run ECM with B1=80k, B2=100*B1, and every 5 curves I run earns credit of 8 standard B1=50k curves. It's not a perfect conversion, but it works well enough for B1 near the expected T-level. Where it breaks down is for B1 choices far smaller than useful- e.g. M1277; 1000 curves at B1 =1e6 are useless there, but get credited at just over 1 curve @850M. We hope people don't do that...

 Similar Threads Thread Thread Starter Forum Replies Last Post Bulldozer Software 7 2019-08-30 14:37 ewmayer Lounge 2 2008-01-15 16:18 HiddenWarrior Data 18 2005-10-11 03:53 Fusion_power Puzzles 7 2003-08-31 01:37 Paulie Software 4 2003-07-16 14:19

All times are UTC. The time now is 11:40.

Tue Oct 20 11:40:02 UTC 2020 up 40 days, 8:51, 0 users, load averages: 1.87, 2.17, 2.21