yafu bugs - Page 8

jrk · 2012-09-16, 18:43

Quote:

Originally Posted by Dubslow

Code:

lanczos error: only trivial dependencies found

This is the problem.

Try running -nc again with only 15M relations.

Dubslow · 2012-09-16, 19:10

Quote:

Originally Posted by jrk

This is the problem.

Try running -nc again with only 15M relations.

Oh, I didn't even see that.

How would too many relations cause only trivial dependencies? Or is it that the BL algorithm fails somehow?

I'll truncate the rel file and try again.

Dubslow · 2012-09-16, 20:22

Quote:

Originally Posted by Dubslow

Oh, I didn't even see that.

How would too many relations cause only trivial dependencies? Or is it that the BL algorithm fails somehow?

I'll truncate the rel file and try again.

Thanks again, that did the trick, although the matrix took quite a bit longer to solve. (50 minutes; though it was actually smaller, it was also about +20 weight/col more dense.)

jasonp · 2012-09-17, 01:42

Nobody is sure, but even with other packages an NFS matrix will become impossible to solve with enough oversieving.

Dubslow · 2012-09-17, 01:47

Quote:

Originally Posted by jasonp

Nobody is sure, but even with other packages an NFS matrix will become impossible to solve with enough oversieving.

Talk about counter-intuitive.

What about with Gaussian Elimination? Theoretically, as long as there are more columns than rows...

LaurV · 2012-09-17, 02:34

x+y=7
2x+2y=14
3x+3y=21
etc.
Solve?

Dubslow · 2012-09-17, 02:46

Quote:

Originally Posted by LaurV

x+y=7
2x+2y=14
3x+3y=21
etc.
Solve?

There's an infinite number of solutions, though I'm not sure what this has to do with anything, because were not solving for the null space of ((1,1),(2,2),(3,3)) by finding (x,y), and this certainly has more rows than columns...

LaurV · 2012-09-17, 03:37

Grrr, scrap that! I tried to play smart a$$ before the morning coffee...
Right now I understood what you wanted to say...

Kosmaj · 2012-09-17, 05:25

Has anybody tried to factor 1000+ digit numbers with Yafu? (Of course, sieving methods cannot be used, but Yafu is still better than plain ecm to automate the process, and keep track of found factors.)

The other day I started with 3 numbers with about 1600 digits, everything was working great, and I reached t30 without any problems.

Then I switched to numbers with about 1000-1020 digits, a few of them were all right, then problems began. On one number Yafu crashed with an "unrecognized character on input line" kind of error, on the next one with "too many allocs or frees" error. I'm not sure but it seems it happened when Yafu was about to start ecm.
The above was with Yafu-1.32.1 on linux (no external ecm), with the same version on Windows-7 Yafu crashed on the very first of these numbers (which Yafu on linux was able to process).

This is all happening in the batch mode, with "pretest -plan custom" and "-pretest_ratio" 0.02 or 0.03.
BTW, is there a way to tell Yafu to stop ecm on a particular fixed level, say t30, independent of the size of the input number? "-preset_ratio" works but it accepts only 2 fractional digits, if I try 0.025 it is converted to 0.03.
Even with preset_ratio accepting more fractional digits, an option like "preset_level t30" will be better.

Thanks!

jasonp · 2012-09-17, 10:53

Quote:

Originally Posted by Dubslow

Talk about counter-intuitive.

What about with Gaussian Elimination? Theoretically, as long as there are more columns than rows...

It isn't that simple; if there are more columns than rows but some of those columns are linear combinations of each other, then you may not have a nontrivial nullspace. My theory is that this is what happens when you have too much oversieving, matrix columns start to look more and more alike because the relations that contribute to them get mixed together too much.

bsquared · 2012-09-17, 13:42

Quote:

Originally Posted by Kosmaj

Has anybody tried to factor 1000+ digit numbers with Yafu? (Of course, sieving methods cannot be used, but Yafu is still better than plain ecm to automate the process, and keep track of found factors.)

The other day I started with 3 numbers with about 1600 digits, everything was working great, and I reached t30 without any problems.

Then I switched to numbers with about 1000-1020 digits, a few of them were all right, then problems began. On one number Yafu crashed with an "unrecognized character on input line" kind of error, on the next one with "too many allocs or frees" error. I'm not sure but it seems it happened when Yafu was about to start ecm.
The above was with Yafu-1.32.1 on linux (no external ecm), with the same version on Windows-7 Yafu crashed on the very first of these numbers (which Yafu on linux was able to process).

This is all happening in the batch mode, with "pretest -plan custom" and "-pretest_ratio" 0.02 or 0.03.
BTW, is there a way to tell Yafu to stop ecm on a particular fixed level, say t30, independent of the size of the input number? "-preset_ratio" works but it accepts only 2 fractional digits, if I try 0.025 it is converted to 0.03.
Even with preset_ratio accepting more fractional digits, an option like "preset_level t30" will be better.

Thanks!

Other people have reported problems when the input size is right around 1024 characters. Coincidentally, most of the character buffers in yafu are initialized to 1024 characters. They are supposed to grow as necessary, and given that 1600 digit inputs work they seem to be doing that, but it looks like there are issues when the inputs are close to the overflow point. I'll look into it.

There is no way that I know of to have yafu stop at a certain t-level, but I don't think it would be hard to add.

Thanks for the feedback.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Running YAFU via Aliqueit doesn't find yafu.ini	EdH	YAFU	8	2018-03-14 17:22
Where to report bugs	Matt	Software	1	2007-02-20 19:13
Possible Prime95 bugs	JuanTutors	Software	9	2006-09-24 21:22
RMA 1.7 beta bugs	TTn	15k Search	2	2004-11-24 22:11
RMA 1.6 fixes LLR bugs!	TTn	15k Search	16	2004-06-16 01:22

2012-09-17, 01:42	#81
jasonp Tribal Bullet Oct 2004 DCE₁₆ Posts	Nobody is sure, but even with other packages an NFS matrix will become impossible to solve with enough oversieving.

2012-09-17, 02:34	#83
LaurV Romulan Interpreter Jun 2011 Thailand 22073₈ Posts	x+y=7 2x+2y=14 3x+3y=21 etc. Solve?

2012-09-17, 03:37	#85
LaurV Romulan Interpreter Jun 2011 Thailand 22073₈ Posts	Grrr, scrap that! I tried to play smart a$$ before the morning coffee... Right now I understood what you wanted to say...

2012-09-17, 05:25	#86
Kosmaj Nov 2003 2·1,811 Posts	Has anybody tried to factor 1000+ digit numbers with Yafu? (Of course, sieving methods cannot be used, but Yafu is still better than plain ecm to automate the process, and keep track of found factors.) The other day I started with 3 numbers with about 1600 digits, everything was working great, and I reached t30 without any problems. Then I switched to numbers with about 1000-1020 digits, a few of them were all right, then problems began. On one number Yafu crashed with an "unrecognized character on input line" kind of error, on the next one with "too many allocs or frees" error. I'm not sure but it seems it happened when Yafu was about to start ecm. The above was with Yafu-1.32.1 on linux (no external ecm), with the same version on Windows-7 Yafu crashed on the very first of these numbers (which Yafu on linux was able to process). This is all happening in the batch mode, with "pretest -plan custom" and "-pretest_ratio" 0.02 or 0.03. BTW, is there a way to tell Yafu to stop ecm on a particular fixed level, say t30, independent of the size of the input number? "-preset_ratio" works but it accepts only 2 fractional digits, if I try 0.025 it is converted to 0.03. Even with preset_ratio accepting more fractional digits, an option like "preset_level t30" will be better. Thanks!