Discussion related to cado-nfs

[Alexander Kruppa <alexander.kruppa@inria.fr>]

----- Original Message -----
> From: "Paul Leyland" <paul.leyland@gmail.com>
> To: cado-nfs-discuss@lists.gforge.inria.fr
> Sent: Thursday, 24 October, 2013 5:42:24 PM
> Subject: Re: [Cado-nfs-discuss] Documentation and/or parameter suggestions  
>   for SNFS
> 
> (TL;DR --- having troubles setting up a SNFS factorization with
> cado-nfs 2.0 beta)

Yes, I agree that, if SNFS factorizations are a supported feature (which I 
guess they are with the Python script), they need to be documented properly.


> On Thu, 2013-10-24 at 10:06 +0100, Paul Leyland wrote:
> Further to this, I'm trying to set up to factor 235*8^235+1 which has
> just started under "Msieve v. 1.52 (SVN 923M)".  The msieve
> polynomial file, working very nicely, is
> 
>         n:
>         
> 9344879323330582132476647523076036766251058644308032256575482675551196064229458278655388065516061224981191913520205352585413584634611067426349293720079701
>         Y0: -166153499473114484112975882535043072
>         Y1: 1
>         c0: 1
>         c1: 0
>         c2: 0
>         c3: 0
>         c4: 0
>         c5: 0
>         c6: 1880
>         skew: 0.284647271438178
>         
> and I've saved that as msieve.poly on the system where cado-nfs is
> installed

This polynomial is actually already in CADO form; msieve, CADO, and afaik 
ggnfs all use the same format. It's merely missing the common root m:

m: 166153499473114484112975882535043072

When I add this line, cadofactor.py accepts the polynomial and starts sieving.
We don't try to recover the common root from any linear polynomial or via 
polynomial gcd, so the common root m must be specified explicitly.


>         scripts/cadofactor/parameters.snfs for that.
> 
> but that file isn't in the beta tarball ====> documentation bug at
> least.

Yes, it is. I'll try to make a few example parameter files for degree 4, 5, 
6, and a few SNFS sizes each.
Sieving parameter choice for SNFS can vary wildly, depending on the degree 
and coefficient size on the algebraic polynomial you get, but there should be 
a handful of working examples provided as a starting point.



> The doc in scripts/cadofactor/README tells me to
[...]
> which I did.  After unsuccessful attempts to convert to CADO-NFS
> format
> by hand I discovered /installed/bin/misc/convert_poly which is not
> obviously documented.  After experimentation I used
> 
>         convert_poly -if msieve < msieve.poly > c8_235.poly
> 
> with the result
>         n: 0
>         c0: 0
>         Y0: 0
>         m: 0
> 
> which is not very helpful  8-(

This is probably a bug, I'll have to check.


> Tried again with the msieve c8_235.fb file as input to convert_poly.
> The contents this time are
> 
> [...]
> with result
> 
>         n:
>         
> 9344879323330582132476647523076036766251058644308032256575482675551196064229458278655388065516061224981191913520205352585413584634611067426349293720079701
>         c6: 1880
>         c5: 0
>         c4: 0
>         c3: 0
>         c2: 0
>         c1: 0
>         c0: 1
>         Y1: 1
>         Y0: -166153499473114484112975882535043072
>         m: 166153499473114484112975882535043072

This is the ggnfs/msieve/CADO format again, which should work, but missing 
skewness, as you point out.

         
> Adding a line
> 
>         skew: 0.284647271438178
> 
> to c8_235.poly and restarting cadofactor.py produced
> 
>         ...
>         Info:Client Launcher: Running clients: localhost4 (Host
>         localhost, PID 4307), localhost3 (Host localhost, PID 4304),
>         localhost2 (Host localhost, PID 4301), localhost (Host
>         localhost, PID 4298)
>         Info:Polynomial Selection: Starting
>         Info:Polynomial Selection: No polynomial was previously found
>         Warning:Polynomial Selection: Polynomial in
>         file /home/pcl/cado_fact/c8_235.poly has no Murphy E value
>         Info:Polynomial Selection: New best polynomial from
>         file /home/pcl/cado_fact/c8_235.poly: Murphy E = 0
>         Info:Polynomial Selection: Adding workunit
>         c8_235_polyselect_0-100000 to database
>         Info:Polynomial Selection: Adding workunit
>         c8_235_polyselect_100000-200000 to database
>         ...
> 
> and now I'm stuck.  What next?
> 
> Note that I blew away the entire $(workdir) and any orphan
> polyselect2l
> process between each successive run of cadofactor.py
> 
> Thanks for any assistance which may be forthcoming.
> 
> Paul (this one)

The fact that it starts running polyselect jobs is a hint that admin < admax. 
Can you check again that both are set to 0 to prevent it from starting a 
polynomial search?

Alex