mersenne.ca

[tuckerkao]

Originally Posted by Viliam Furik

Now they do. But at the time of writing my post, there was only the bigger one. Sadly, I didn't make a screenshot.

It could be that the F-PM1 of M27288977 found the larger composite factor first, then divided them into 2 smaller prime factors later.

[Luminescence]

Originally Posted by tuckerkao

It could be that the F-PM1 of M27288977 found the larger composite factor first, then divided them into 2 smaller prime factors later.

Yes, gpuowl found the large composite factor. I use Mihais script to upload the results and when it uploaded that factor it took unusually long. When I looked at my results page it showed up twice and on the exponent page it was split into two factors.

[petrw1]

Originally Posted by James Heinrich

Note that those are mersenne.org URLs and this is a mersenne.ca thread.
I have no control over the generation of those reports.

Oops! my bad.

[Viliam Furik]

Originally Posted by tuckerkao

It could be that the F-PM1 of M27288977 found the larger composite factor first, then divided them into 2 smaller prime factors later.

I don't see how it could be like this. The one factor which was displayed was the bigger prime factor of the newly found composite one.

[James Heinrich]

Originally Posted by Luminescence

Yes, gpuowl found the large composite factor. I use Mihais script to upload the results and when it uploaded that factor it took unusually long. When I looked at my results page it showed up twice and on the exponent page it was split into two factors.

This is normal. When a composite factor is found, it is split into prime factors (which could take several seconds to process) and each prime factor is inserted into the database. An unfortunate side-effect is that on primenet when the two prime factors are recorded, they're entered in the history log with the reported result line, which shows the composite factor rather than the prime factor being recorded. It's far from ideal, but it's the result of new things being hacked onto an old system that originally didn't consider that composite factors might be reported.

[SethTro]

Originally Posted by James Heinrich

They are now. There's a new row near the bottom of the page for "Exponent Limits" which is a plaintext report, one exponent per line, which includes these columns (all data tab-separated):

exponent	factored	tfnf	b1	b2	pm1_prob

"factored" will be 1 if there are known factor(s) (but not how many, unfortunately) and 0 if none known
"tfnf" is how far this exponent has been fully TF'd (fully complete bitlevels, not stopping after finding a factor)
"b1","b2" are the best P-1 bounds (by probability) run on this exponent. Will be blank if no P-1.
"pm1_prob" is % probability of factor for best P-1 run, if applicable. Note that this probability is calculated based on Prime95-default TF levels for the exponent, not the actual TF level the exponent has had run, to provide an even basis of comparison (higher TF means lower P-1 probability).

Hope that's useful.

This is very useful. Thanks James for taking the time to help me.

I double checked ~100 exponents and found a small set of discrepancies for small results.

https://www.mersenne.ca/exponent/97301
These all have a strictly large P-1 done (B1=100M, B2=2000M).

97301   0       67      16777216        1677721600      48.937   
97327   0       67      16777216        1677721600      48.938
97381   0       67      16777216        1677721600      48.939

[James Heinrich]

Originally Posted by SethTro

found a small set of discrepancies for small results

It seems that the P-1 probability algorithm breaks down somewhat when presented with extreme cases of exponent size, bounds, TF. I have fixed the 3 specific cases you listed, and if any other cases exist, they should be self-fixed when an exponent page is loaded but the page will need to be reloaded to see the changed data, and a message is displayed to that effect (see attached).

Thanks for catching this.

edit: I batch ran through 167 exponent in the 90-100k range that had B1=2^24 so those ones should already be fixed.

[techn1ciaN]

I was recently looking at M340705633 (from the strategic DC thread) on mersenne.ca (page), which says that the exponent's TF release levels are 2⁷⁷ for PrimeNet and 2⁸¹ for GPU72. I checked the mersenne.org math page, though, and that seems to indicate that the PrimeNet TF release level is actually 2⁷⁸ (and I understand that GPU72's release level is always ((PrimeNet release level) + 4), so that should then be 2⁸²). There's not a fluke error for this particular exponent because I checked out several others in the 340 M range and the same problem was present.

Were PrimeNet's release levels recently tweaked and mersenne.ca hasn't been updated to reflect this? If no, what else could the problem be here?

[SethTro]

Another feature request to our intrepid leader, James, may he program forever.

is there any chance you can add a column on the status page for number of untested exponents (e.g. no LLR / PRP tests complete)

https://www.mersenne.ca/status/tf/0/0/2/0

[James Heinrich]

Originally Posted by techn1ciaN

I checked the mersenne.org math page, though, and that seems to indicate that the PrimeNet TF release level is actually 2⁷⁸
Were PrimeNet's release levels recently tweaked and mersenne.ca hasn't been updated to reflect this? If no, what else could the problem be here?

I don't know when PrimeNet's TF thresholds were updated (probably browsing through source code commonc.h for recent versions of Prime95 would reveal the change version/date). The threshold numbers I had were from 2011.

For reference, this is what's defined in commonc.h (lines 85-100):

#define FAC82	1071000000L
#define FAC81	842000000L
#define FAC80	662000000L
#define FAC79	516800000L
#define FAC78	408400000L
#define FAC77	322100000L
#define FAC76	253500000L
#define FAC75	199500000L
#define FAC74	153400000L
#define FAC73	120000000L
#define FAC72	96830000L
#define FAC71	77910000L
#define FAC70	60940000L
#define FAC69	48800000L
#define FAC68	38300000L
#define FAC67	29690000L

I have updated mersenne.ca to use the numbers from Prime95 v30.7b9 source, which doesn't quite match up with the Math page, but I've sent an email to George asking for clarification as to which numbers are more accurate (I would guess the source code and the database table has simply been overlooked). Perhaps don't take the numbers on either the Math page or mersenne.ca as authoritative until George has confirmed what the numbers should be (I'll post back here).

[James Heinrich]

Originally Posted by SethTro

Another feature request to our intrepid leader, James, may he program forever.
is there any chance you can add a column on the status page for number of untested exponents (e.g. no LLR / PRP tests complete)
https://www.mersenne.ca/status/tf/0/0/2/0

My data on number PRP/LL tests (especially as it relates to whether an exponent is considered cleared or not) is incomplete at best.
I could potentially make use of the data here which is available in 1M granularity which means it would be meaningless on my charts on the two highest zoom levels.