mersenne.ca - Page 92

James Heinrich · 2023-01-27, 02:13

Quote:

Originally Posted by LordJulius

when the 'known factors' or 'either' box checked, the results TF column will show TFs below 2^67.

Thanks, fixed.

harlee · 2023-01-28, 16:51

I was looking at the default results returned by Factoring Beyond First Factor and Exponents that were poorly P-1 factored and I have a couple of questons. For the same exponent the two pages returns different Pminus1 bounds:

Factoring Beyond First Factor (Optional Fixed B1 field is blank)

Pminus1=1,2,4180471,-1,100000,100000,67,"44685391937063190031,1112388632373104350711"

Exponents that were poorly P-1 factored

Pminus1=1,2,4180471,-1,220000,6600000,67,"1112388632373104350711,44685391937063190031"

Previously tested with B1=45000 B2=630000

Why is the default B1 and B2 bounds different? As of the B2 bound, I know that for v30.8b? if B2 is either equal to 0 or B2 is equal to B1 then Prime95 will calculate the B2 bounds. Just did a quick test on the latter Pminus1 line (B1=220000, B2=6600000) and v30.8b15 ignored the B2 bound and calculated its own B2 bound. Anyway, is B2=6600000 originally set for v30.7 and earlier Prime95 software?

James Heinrich · 2023-01-28, 19:42

Quote:

Originally Posted by harlee

Maybe add an option to return B2 bounds for ether v30.7 and earlier versions or v30.8 and later versions.

At best I have a poor-man's version of the bounds calculations from v30.7, I do not have any means to calculate v30.8+ bounds.

morefactors: uses "generous" bounds: B1 is calculated using the standard bounds calculation, using the "max" variant of B1, then doubled.

pm1_worst: uses "massive" bounds: Pfactor "tests saved" is set as high as 10 (<5M) down to no less than 2 (>40M). Pminus1 bounds again use the B1_max value, but multiplied by the number of tests saved. If existing B1 of existing P-1 test large, it will ensure that the new B1 is at least 20% bigger than previous B1.

Clearly these bounds calculations are not suitable for first-time wavefront P-1 work, they're designed as a starting point for doing 2nd/3rd/Nth P-1 test on previously-tested exponents (which sometimes have already had 10 or more P-1 tests of varying quality).

harlee · 2023-01-28, 21:19

Thank you for the explanation about the differences.

Andrew Usher · 2023-01-29, 02:38

Of course, it is just as useful for 30.8 if you allow the program to select the optimal B2 - you'll get better P-1 in about the same time. But, as James said, those are just starting points and you're almost expected to change them; he can't predict the exact effort you'd like.

lisanderke · 2023-01-29, 12:17

On the Factoring Beyond First Factor page it would be useful to be able to sort between values of B1 previously done, for example: show only exponents with latest P-1 run of B1 = 1 000 000 or only show exponents with B1 between 100 000 and 1 000 000 etc.

When trying to sort for exponents that have only had stage 1 done (B1=B2), there's no way to distinguish between exponents with B1 done to 100 000 or 1 000 000 000 000. This makes for some manual work to take out exponents from the generated worktodo entries that would have very long runtimes compared to others when doing, for example, P-1.

harlee · 2023-01-29, 13:28

Quote:

Originally Posted by Andrew Usher

Of course, it is just as useful for 30.8 if you allow the program to select the optimal B2 - you'll get better P-1 in about the same time. But, as James said, those are just starting points and you're almost expected to change them; he can't predict the exact effort you'd like.

I'm picking my own B1 bounds and allowing v30.8b15 to select the optimal B2 based on one worker, one cpu, 4096 MB to P-1 (two laptops - one 9 years old and the other is 7 years old). I was just bringing up differences that I noticed on the two pages - same exponent different bounds.

Andrew Usher · 2023-01-29, 13:32

Quote:

Originally Posted by lisanderke

On the Factoring Beyond First Factor page it would be useful to be able to sort between values of B1 previously done, for example: show only exponents with latest P-1 run of B1 = 1 000 000 or only show exponents with B1 between 100 000 and 1 000 000 etc.

That is what the 'Poorly P-1 Factored' page does, now that it supports factored exponents.

lisanderke · 2023-01-29, 17:38

@Andrew
The probability calculation used on the poorly P-1 page is not based on actual work done. Especially low exponents that I'm interested in will give a skewed probability since they often have high T-level (lots of ECM done) far above what primenet would have TF'ed them to. Also, when sorting by 'stage 1 only' that page seems to list exponents with B1=B2=0, which would mean no P-1 done. Using the poorly P-1'ed page instead of the factoring beyond first factor also loses me some other features from the latter.
PS: There are some other obvious reasons why I won't use that page that I forgot to list here, such as I can't sort by actual bounds, the probabilities given there are not useful to me since they assume no P-1 done on the exponent...

Question for James: The probability calculation on the poorly P-1 page assumes Primenet TF-level, but when generating worktodo in the form of pminus1= it seems to take actual TF done? For consistency, wouldn't it be better to calculate the probability based on actual TF done?

Different observation and another request for James:
Looking at the 'factor probability' as listed on a 'P-1 results' table for an exponent, I wonder if it would be possible to calculate both probability taking into account TF done and the probability given the T-level (if any ECM was done on said exponent) and then showing whichever one gives the lowest probability. Currently the probability of having found a factor with whatever bounds ran seems to only be based on TF done, which for small exponents is not (at all) accurate.

James Heinrich · 2023-01-29, 17:52

Poor P-1 page always calculates probability based on Primenet-default TF level, regardless of what TF has actually been done, or ECM, or previous P-1 effort. This is to give a consistent evaluation of how "good" the existing P-1 bounds are relative to the exponent, without being thrown off by somebody TF'ing the exponent 10 bit levels higher or whatever the case may be. For cases where extraordinary TF/P-1/ECM effort has been applied to an exponent, notably for very-low exponents, this may deviate significantly from the actual chance of finding a factor.

I have no idea how to account for ECM into P-1 probability.

lisanderke · 2023-01-29, 17:55

Quote:

Originally Posted by James Heinrich

I have no idea how to account for ECM into P-1 probability.

I was referring to this

2023-01-28, 16:51	#1003
harlee Sep 2006 Odenton, MD, USA 208₁₀ Posts	I was looking at the default results returned by Factoring Beyond First Factor and Exponents that were poorly P-1 factored and I have a couple of questons. For the same exponent the two pages returns different Pminus1 bounds: Factoring Beyond First Factor (Optional Fixed B1 field is blank) Pminus1=1,2,4180471,-1,100000,100000,67,"44685391937063190031,1112388632373104350711" Exponents that were poorly P-1 factored Pminus1=1,2,4180471,-1,220000,6600000,67,"1112388632373104350711,44685391937063190031" Previously tested with B1=45000 B2=630000 Why is the default B1 and B2 bounds different? As of the B2 bound, I know that for v30.8b? if B2 is either equal to 0 or B2 is equal to B1 then Prime95 will calculate the B2 bounds. Just did a quick test on the latter Pminus1 line (B1=220000, B2=6600000) and v30.8b15 ignored the B2 bound and calculated its own B2 bound. Anyway, is B2=6600000 originally set for v30.7 and earlier Prime95 software? Last fiddled with by harlee on 2023-01-28 at 17:04 Reason: Delete on line, added exponenet link and previous bounds

2023-01-29, 17:38	#1010
lisanderke "Lisander Viaene" Oct 2020 Belgium 2⁴×7 Posts	@Andrew The probability calculation used on the poorly P-1 page is not based on actual work done. Especially low exponents that I'm interested in will give a skewed probability since they often have high T-level (lots of ECM done) far above what primenet would have TF'ed them to. Also, when sorting by 'stage 1 only' that page seems to list exponents with B1=B2=0, which would mean no P-1 done. Using the poorly P-1'ed page instead of the factoring beyond first factor also loses me some other features from the latter. PS: There are some other obvious reasons why I won't use that page that I forgot to list here, such as I can't sort by actual bounds, the probabilities given there are not useful to me since they assume no P-1 done on the exponent... Question for James: The probability calculation on the poorly P-1 page assumes Primenet TF-level, but when generating worktodo in the form of pminus1= it seems to take actual TF done? For consistency, wouldn't it be better to calculate the probability based on actual TF done? Different observation and another request for James: Looking at the 'factor probability' as listed on a 'P-1 results' table for an exponent, I wonder if it would be possible to calculate both probability taking into account TF done and the probability given the T-level (if any ECM was done on said exponent) and then showing whichever one gives the lowest probability. Currently the probability of having found a factor with whatever bounds ran seems to only be based on TF done, which for small exponents is not (at all) accurate. Last fiddled with by lisanderke on 2023-01-29 at 17:43

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Gaussian-Mersenne & Eisenstein-Mersenne primes	siegert81	Math	15	2023-06-30 18:04
Small inconsistencies between mersenne.org and mersenne.ca factor databases	GP2	mersenne.ca	44	2016-06-19 19:29
mersenne.ca (ex mersenne-aries.sili.net)	LaurV	mersenne.ca	8	2013-11-25 21:01
Mersenne Wiki: Improving the mersenne primes web site by FOSS methods	optim	PrimeNet	13	2004-07-09 13:51

2023-01-28, 21:19	#1005
harlee Sep 2006 Odenton, MD, USA 11010000₂ Posts	Thank you for the explanation about the differences.

2023-01-29, 02:38	#1006
Andrew Usher Dec 2022 773₈ Posts	Of course, it is just as useful for 30.8 if you allow the program to select the optimal B2 - you'll get better P-1 in about the same time. But, as James said, those are just starting points and you're almost expected to change them; he can't predict the exact effort you'd like.

2023-01-29, 12:17	#1007
lisanderke "Lisander Viaene" Oct 2020 Belgium 1110000₂ Posts	On the Factoring Beyond First Factor page it would be useful to be able to sort between values of B1 previously done, for example: show only exponents with latest P-1 run of B1 = 1 000 000 or only show exponents with B1 between 100 000 and 1 000 000 etc. When trying to sort for exponents that have only had stage 1 done (B1=B2), there's no way to distinguish between exponents with B1 done to 100 000 or 1 000 000 000 000. This makes for some manual work to take out exponents from the generated worktodo entries that would have very long runtimes compared to others when doing, for example, P-1.

2023-01-29, 17:52	#1011
James Heinrich "James Heinrich" May 2004 ex-Northern Ontario 10B5₁₆ Posts	Poor P-1 page always calculates probability based on Primenet-default TF level, regardless of what TF has actually been done, or ECM, or previous P-1 effort. This is to give a consistent evaluation of how "good" the existing P-1 bounds are relative to the exponent, without being thrown off by somebody TF'ing the exponent 10 bit levels higher or whatever the case may be. For cases where extraordinary TF/P-1/ECM effort has been applied to an exponent, notably for very-low exponents, this may deviate significantly from the actual chance of finding a factor. I have no idea how to account for ECM into P-1 probability.