ggnfs sqrt problem

[hallstei]

Hi,

I am trying to factor a c129 from the xyyxf project.

(The number is 689248390756594092329903555441060933187672009587156714157714459911067848381094129751231275207217559323589590068177837430265561551)

However, when I run the sqrt program, I either get the trivial factorization 1*N, or the error message posted beneath.

Any idea on how I can fix this?

ggnfs/Version reads:

0.77.1-20060513
Sun May 14 09:18:57 UTC 2006

The system is a Pentium 4 running Ubuntu 6.06.

Cheers,

Hallstein

--------------------------------

...
Reading relations from rels.bin.23...
The final square should be: 130680677895632184865139317000365589321504868538119170411575793060617878224349779001039329223849671202761057909376525885051074735
We used 2277256 (a,b) pairs.
Error: Odd exponent found: AFB[11] has exponent 1!

[R.D. Silverman]

Originally Posted by hallstei

Hi,

I am trying to factor a c129 from the xyyxf project.

(The number is 689248390756594092329903555441060933187672009587156714157714459911067848381094129751231275207217559323589590068177837430265561551)

However, when I run the sqrt program, I either get the trivial factorization 1*N, or the error message posted beneath.

Any idea on how I can fix this?

ggnfs/Version reads:

0.77.1-20060513
Sun May 14 09:18:57 UTC 2006

The system is a Pentium 4 running Ubuntu 6.06.

Cheers,

Hallstein

--------------------------------

...
Reading relations from rels.bin.23...
The final square should be: 130680677895632184865139317000365589321504868538119170411575793060617878224349779001039329223849671202761057909376525885051074735
We used 2277256 (a,b) pairs.
Error: Odd exponent found: AFB[11] has exponent 1!

I don't use GGNFS, but I strongly suspect that this is a linear algebra
error and not a sqrt error. Is there a way to verify the LA solution
sets?

[kubus]

Originally Posted by hallstei

However, when I run the sqrt program, I either get the trivial factorization 1*N, or the error message posted beneath.

Any idea on how I can fix this?

There is a bug in sqrt. Check this: http://tech.groups.yahoo.com/group/ggnfs/message/2005. If this is your problem then a solution is to get the latest cvs verision of ggnfs.

[frmky]

It can also be completed using the latest version of msieve. See
this post for details: http://www.mersenneforum.org/showpos...&postcount=403

Greg

[hallstei]

Originally Posted by frmky

It can also be completed using the latest version of msieve. See
this post for details: http://www.mersenneforum.org/showpos...&postcount=403

Greg

Thanks for the tip!

I tried this just now. But when I ran msieve, I got an enormous amount of errors messages:

90% of the error messages were on the form "error -10 reading relation X"

The remaining were "error -5..."

Then, in the linear algebra stage, it segfaulted. The last message to be written to the log was

"using 32 quadratic characters above 134217362"

[frmky]

This is a large GNFS factorization, isn't it? Version 1.19 doesn't support 64-bit a values, which you have in abundance. The next version of msieve, hopefully due any day now, will support 64-bit a values and therefore should work.

Greg

[fivemack]

Originally Posted by frmky

This is a large GNFS factorization, isn't it? Version 1.19 doesn't support 64-bit a values, which you have in abundance. The next version of msieve, hopefully due any day now, will support 64-bit a values and therefore should work.
Greg

I have managed to complete a couple of (slightly smaller) GNFS factorizations with 64-bit a values using 1.19; in my case I lost about 30% of the relations to the 'error -10' and 'error -5' messages, but the code proceeded to completion anyway. msieve seems quite good at discarding wrong relations and proceeding, so I don't think that's the issue here.

I haven't had a segfault in linalg from msieve, though that may be something to look forward to when my C200 SNFS finally accumulates enough relations to build a matrix. ggnfs is failing on that run: procrels dies during the make-more-files step roughly at the point where the total size of the rels.bin.* files exceeds 2^32 bytes, in such a way as to destroy the rels.bin.*. Thankfully, I save unprocessed relations.

[jasonp]

Originally Posted by frmky

Version 1.19 doesn't support 64-bit a values, which you have in abundance. The next version of msieve, hopefully due any day now, will support 64-bit a values and therefore should work.

It's coming, it's coming :) I'll try to get a release out tonight.