GGNFS questions and Answers

[ValerieVonck]

Originally Posted by akruppa

For some reason sqrt found only trivial factorisations. This might be a bug, perhaps block-Lanczos produces only trivial dependencies due to duplicate rows or something like this. Maybe Sam (trilliwig) can tell more.

Alex

Ok I will PM // Mail him ....

[trilliwig]

Originally Posted by CedricVonck

I know but what are the odds that of the factors is 1 !!!??
What does it mean? Can some run a doublecheck on this number?

1 is a factor of all positive integers. :) But yes, it's not a prime factor. You should expect a sqrt run to turn up the trivial factorization 1*N about half the time. Still, with 32 dependencies, the probability that they all turn up the trivial factorization should be close to 0, unless something makes the behavior non-random....

It's hard to tell what went wrong without seeing the sqrt output logs. But the value of n and the poly coefficients seem ok. Sqrt didn't crash since the ggnfs.log you posted contains r1 and r2 outputs for each sqrt run. And so, based on past experience with polynomials with really small coefficients, the most likely error is that the iterative portion of sqrt didn't work correctly, and left some CRT residues not equal to 1. And I don't know enough about the sqrt process to fix this.

Anyway, you seem to be using an old version of factLat.pl that computes the SNFS difficulty incorrectly if Y1 is not 1. So I would propose using a more up-to-date version of factLat.pl. It should then default to saner parameters that should run in under a day, and it'll get you another 32 bites at the apple.

You can get the latest snapshotted script from http://sourceforge.net/project/showf...roup_id=140917 and get the latest source .bz2. Decompress it using 'tar jxpvf <filename>' and get factLat.pl out of the ggnfs/tests directory.

--
Sam

[ValerieVonck]

Sam,

Thx for the reply but I have 2 problems:

1) I don't have the courage to do the factorization again (too loud fan & the pc is my bedroom).
2) I don't have linux. I am running windows

So, if anyone can mail the script (PM me for my mail adress), I like to crunch small (s)(g)nfs numbers.

Cheers,
C.

[smh]

Originally Posted by CedricVonck

Sam,

Thx for the reply but I have 2 problems:

1) I don't have the courage to do the factorization again (too loud fan & the pc is my bedroom).
2) I don't have linux. I am running windows

So, if anyone can mail the script (PM me for my mail adress), I like to crunch small (s)(g)nfs numbers.

Cheers,
C.

You could try to use procrels -prune 0.01 and run the matrix again.

Attached is a fairly new version of factlat.pl. Don't know if it's the latest and greatest but should be newer than the one you are using.

[ValerieVonck]

Ok .. I will make a backup first

[ValerieVonck]

Did run procrels wth prune 0.01 (1%) => the matrix is about the same weight (34M).... I am giving up this number. I will try other smaller numbers..

Again tx to all

[smh]

Originally Posted by CedricVonck

Did run procrels wth prune 0.01 (1%) => the matrix is about the same weight (34M)....

Of course, you only threw away 1 percent of the relations.

Originally Posted by CedricVonck

I am giving up this number. I will try other smaller numbers..

Give it another try with the new script. It will choose a much smaller factorbase.A SNFS difficulty of ~142 is considered easy IMHO

-> ___________________________________________________________
-> |        This is the factLat.pl script for GGNFS.          |
-> | This program is copyright 2004, Chris Monico, and subject|
-> | to the terms of the GNU General Public License version 2.|
-> |__________________________________________________________|
-> This is client 1 of 1
-> Working with NAME=cv...
-> SNFS_DIFFICULTY is about 142.2400426620279512910725265478295261169.
-> Selected default factorization parameters for 153 digit level.
-> Selected lattice siever: ../../bin/gnfs-lasieve4I13e.exe
-> Creating param file to detect parameter changes...
=> "../../bin/makefb.exe" -rl 2400000 -al 2400000 -lpbr 27 -lpba 27 -3p -of cv.fb -if cv.poly
Making factor base...Generating AFB with norms upto 2400000...
Making algebraic factor base.
Checking p=13...(total=0)
Elapsed time: 13 seconds.
Factor base sucessfully created.
-> minimum number of FF's: 394910
-> Q0=1200000, QSTEP=100000.
-> makeJobFile(): q0=1200000, q1=1300000.
-> makeJobFile(): Adjusted to q0=1200000, q1=1300000.
-> Lattice sieving algebraic q-values from q=1200000 to 1300000.
=> "../../bin/gnfs-lasieve4I13e.exe" -k -o spairs.out -v -n0 -a cv.job

I'm wondering why the script uses parameters which are specified for a difficulty of 153 though.

[ValerieVonck]

I know but for the moment I have already reserved an other smaller (ggnfs) number. The big problem is that I can't leave my PC on 24/7 and I cannot move it downstairs => the living room is becoming quite cramped with stuff from my other hobby (vinyl records )

I will see that I continue this number or I probably on a good def-par // def-nm-file ... (although I got them from source forge).

CU

[smh]

There's no need to run your pc 24/7.

If you're running windows with activeperl, the job file isn't properly updated after a ctrl-c. Just update the .job file manually with the last special-q and rename spairs.out to spairs.add. You do need a windows version of cat.exe to let the script concatenate the two files, otherwise you'll loose all relations since the last procrels run (and you don't need to update the .job file because these special-q can be resieved.)

[ValerieVonck]

Finally, after approx. 2 days of work with a GNFS run on a c115 number the ggnfs suite produced the 2 cofactors

[ValerieVonck]

Can the following number be re-written as a SNFS job?

(71·10^160-17)/9
= 89 · 3491 · 78791 · 163337 · 494927 · 20992251737· 25597988189016413 ·

[74183751132125462572090856533831080834569538057811826337065936337992030749034836183025904719268995337840525177697<113>] (compsite cofactor)

Can anyone please give a good example of a job file?
And if possible a good explanation?
Thank you.