Pascal's OPN roadblock files

[wombatman]

Running 297903607^23-1

[wombatman]

Running 8970971^29-1

Having some trouble with the SNFS poly. FactorDB gives:

n: 4288882318725178503864985939002570343870783101076222294692132636882637343988982346014447015553752612504342250262840696959794627098440959386503033787331436025332337203640782929696973892811770840659915530
m: 58102827030430867738060703578743851
deg: 5
skew: 0
type: snfs
c5: 6476760099930193480511831281
c0: -1
rlim: 18610400
alim: 18610400
lpbr: 29
lpba: 29
mfbr: 58
mfba: 58
rlambda: 2.7
alambda: 2.7

but GGNFS doesn't like the skew of 0.

Phi gives:

n: 2580482901793422593005519906534751048235270635043096719781698476519957422481996978800942266020020022345900390998498545851996437063629641295043489044129593444543441973751813443723
# 8970971^29-1^29, difficulty: 208.59, skewness: 24.58, alpha: 0.00
# cost: 2.95737e+017, est. time: 140.83 GHz days (not accurate yet!)
skew: 24.579
c5: 1
c0: -8970971
Y1: -1
Y0: 521238776308011431982978168044507303749321
m: 521238776308011431982978168044507303749321
type: snfs

This one works but seems to give low relations. Any thoughts?

[axn]

For the first one, the correct skew would be 0.00000274.

For the second one, the larger rational side coefficient implies that you should use larger rational side parameters.

[wombatman]

Much obliged for the response.

On the 1st one, with the skew set appropriately, GGNFS still gives:

gnfs-lasieve4I14e (with asm64): L1_BITS=15, SVN $Revision$
Please set a skewness

For the second, factsmieve sets the parameters as follows:

rlim: 21300000
alim: 21300000
lpbr: 29
lpba: 29

I'm still slowly learning both the theory and the practical applications here, so I'm not too good at determining whether these are set properly.

[chris2be8]

Don't bother with the poly provided by factordb. c5: 6476760099930193480511831281 is ridiculous for snfs. In general the smaller the coefficients are the better it will sieve.

phi generated a reasonable poly. How may relations per special Q does it give? A rule of thumb (originally from Fivemack) is that if you are getting less that 2 relations per Q you should go to a larger siever or raise LPB[AR] and/or MFB[AR].

There was a "ggnfs pearls of wisdom" thread to collect such advice. It's worth reading.

Chris

[wombatman]

I'd have to check to get an exact number, but it was something like 1.5 relations/Q or so, which seemed really low to me. I'll try and track that wisdom thread--I can definitely use any of that I can find!

[henryzz]

The other problem with factordb polys is that it gives you the whole number to factor even if it has very small factors. msieve will find those factors and complain at you.

[chris2be8]

Originally Posted by henryzz

The other problem with factordb polys is that it gives you the whole number to factor even if it has very small factors. msieve will find those factors and complain at you.

I think using msieve compiled without ECM will stop it finding small factors when you don't want it to. But it's better to remove the small factors first. It might be useful for SNFS around 85 digits.

Chris

[chris2be8]

Originally Posted by wombatman

I'd have to check to get an exact number, but it was something like 1.5 relations/Q or so, which seemed really low to me. I'll try and track that wisdom thread--I can definitely use any of that I can find!

In that case you would be better off raising LPBR and LPBA to 30 (and MFB[AR] to 60). That should double the yield, but it will nearly double the number of relations you need to collect. Raising just LPBA and LPBR would raise yield and relations needed a bit less.

In practice the job would still work with a yield around 1.5 per Q. It would take a little longer than with better parameters though (I've run a few like that by mistake).

Chris

[henryzz]

Originally Posted by chris2be8

I think using msieve compiled without ECM will stop it finding small factors when you don't want it to. But it's better to remove the small factors first. It might be useful for SNFS around 85 digits.

Chris

It will still do trial division I think.

[wombatman]

I upped LPBR to 30 (totally arbitrary choice), and the yield went from ~1.5 to ~2 relations/Q, so that did indeed help. It still hasn't finished gathering relations yet, but we'll see how it does with the matrix building.

Originally Posted by chris2be8

In that case you would be better off raising LPBR and LPBA to 30 (and MFB[AR] to 60). That should double the yield, but it will nearly double the number of relations you need to collect. Raising just LPBA and LPBR would raise yield and relations needed a bit less.

In practice the job would still work with a yield around 1.5 per Q. It would take a little longer than with better parameters though (I've run a few like that by mistake).

Chris