integrator failed

[MatWur-S530113]

Hello,

today I try to factor a number using nfs (for the first time) in msieve. I choosed a number (C116), put it in the worktodo.ini and started 'msieve -np'. After a few minutes of calculating I got an error as a screen-output:

integrator failed 1.073703e-013 3.829712e-018
error: size score computation failed

Then msieve continued as if nothing has happened. No hint about this error in the log-file.
Have I to pay attention about this error?

And if the polynomial selection has finished, what I have to do to sieve this number with 2 cores of my computer? Atm I'm thinking of this procedure:
1) copy worktodo.ini, msieve.dat.p and msieve.fb to another directory with a second instance of msieve.
2) start worker1 with 'msieve -ns 1,10000' and worker 2 with 'msieve -ns 10001,20000'
3) merge 2 *.dat files and process 'msieve -nc' for this dat-file
4) if sieving still needed I restart with 2) and new lines '20001,30000' and '30001,40000'. This I will do until I get a result

Any objections about this plan, maybe something is completely wrong.
In the meantime I got the described error again (with other numbers after 'integrator failed'). hm ?

best regards,

Matthias

[jasonp]

The next version will greatly reduce (hopefully eliminate) the integrator problems that you see. The errors are not fatal, just annoying. As for the sieving, using msieve will be 5x slower than using GGNFS (for a C116 you'd be saving close to a week of runtime).

[MatWur-S530113]

ok, in the meantime I finished the poly selection with msieve and got a polynomial. As you suggested, I will ignore the 'integrator' errors.
Of course it sounds very interesting to use GGNFS, but I have no plan how to use it. I downloaded the win-32bit Core2 version Jeff Gilchrist is offering at his page (thanks for it btw, I can not compile such binaries for myself. I only have a Pascal-compiler...). Can I use now the poly I got by msieve or should I run a new poly-selection with the tools of GGNFS? Hm, this looks more complicate than using msieve. Is there something like a 'step by step introduction' how to use GGNFS to factor a number?

best regards,

Matthias

[jasonp]

The msieve polynomial is usable within GGNFS (you'll have to create a text file with the polynomial in it, that GGNFS can read). See this thread for the central clearinghouse of people comparing notes on running GGNFS. Yes, it's a lot more complex than just starting msieve, but there is no substitute for the tools that GGNFS contains.

[Jeff Gilchrist]

Originally Posted by MatWur-S530113

Of course it sounds very interesting to use GGNFS, but I have no plan how to use it. I downloaded the win-32bit Core2 version Jeff Gilchrist is offering at his page (thanks for it btw, I can not compile such binaries for myself. I only have a Pascal-compiler...). Can I use now the poly I got by msieve or should I run a new poly-selection with the tools of GGNFS? Hm, this looks more complicate than using msieve. Is there something like a 'step by step introduction' how to use GGNFS to factor a number?

Now that I recently went through the process to figure out how to do it myself, I should write up a little guide in an easy to find location for people. But basically if you are doing this on Windows the easiest way is to install cygwin (http://www.cygwin.com/) so you get some UNIX tools and perl, then you can use the GGNFS binaries I compiled which includes a perl script called factMsieve.pl. You edit that file at the top to tell the script where you find your GGNFS binaries, you create a file.poly file which is slightly different format from what msieve generates, and you run "factMsieve.pl file.poly" and it handles the rest for you, using the faster lattice siever in GGNFS and then switches back to msieve to do the post-processing.

As an example, when msieve finishes generating a polynomial, it creates an msieve.fb file which looks something like this:

N 542199118360122645149018222712631403787793963895701957093138322222784637474878020968003860466806398192243491945916177216280574963
SKEW 168279.97
R0 -6331679219837515495939907
R1  156433515642929
A0 -71923817076071821269252524150400
A1  1552973346849705835047567360
A2  1842020125749195022561
A3 -78887013322320122
A4  348969912
A5  53280

You need to copy that file to your GGNFS directory and rename it to file.poly (anything you want with .poly at the end). Then edit the file to look like this to be used by GGNFS:

n: 542199118360122645149018222712631403787793963895701957093138322222784637474878020968003860466806398192243491945916177216280574963
Y0: -6331679219837515495939907
Y1:  156433515642929
c0: -71923817076071821269252524150400
c1:  1552973346849705835047567360
c2:  1842020125749195022561
c3: -78887013322320122
c4:  348969912
c5:  53280
skew: 168279.97
type: gnfs

So you are adding ":" and changing "R" to "Y" and "A" to "c" and adding "type: gnfs". Then assuming you have factMsieve.pl configured properly you just run inside a cygwin console:

./factMsieve.pl file.poly

[MatWur-S530113]

@ jason: thank you for you explanations, I already startet to read the thread. I hope to get some answers there.

@ Jeff: of course thank you, too. I very appreciate these first steps. I will give cygwin a try and use the same directory-structure you have postet in the other thread. But it sounds complicate, I will need some time to figure out how it works. I'm sure I will come back with some (stupid) questions .

best regards,

Matthias