Poly select and planning for 2,2210M

[VBCurtis]

Originally Posted by fivemack

I think that's just not true; the yields you get are definitely lower for large skew, but it will run.

(sieving the demonstrated polynomial from 400M to 400M+100k with 16e and 33-bit large primes gives a yield of 40 relations)

My understanding is that GGNFS internally adjusts skew down to Q; I'll see if I can find the source of my [mis]understanding.

If you set skew to 410M, do you get the same 40 relations? A "no" would conclusively reject my claim.

[Max0526]

Originally Posted by swellman

# norm 8.388914e-21 alpha -9.216995 e 7.639e-16 rroots 5
skew: 20979898852.98
c0: 27649131585183391812916066541481716806536590015577581424
c1: 27330385748632515681889738973504668871302682226
c2: -2155465644405759903230110401342981943
c3: -213599886258400208384412088
c4: 4879065781035618
c5: 121716
Y0: -96818882190388663272841084874018768984485
Y1: 74413548803525738089

I can't spin it up, sorry. I can drop skew down to 10 mln but then the E score also drops:

R0: -96818882190389519247725813097586565418501
R1: 148827097607051476178
A0: 6833616701827777569778742549146341079870193298849419692
A1: 13689944665495682939372373982326074456184996569
A2: -2148090690147041132745084413722189879
A3: -427648439376739194153191792
A4: 19488261277504392
A5: 973728
skew: 10892841023.33
# size 6.258e-21, alpha -9.564, combined = 7.121e-16 rroots = 5

[Max0526]

Originally Posted by Max0526

I can drop skew down to 10 mln but then the E score also drops:

I meant 10 bln of course.

[Max0526]

Originally Posted by VBCurtis

I finally finished my deg 6 CADO run:

skew: 402972.111
c0: 2118696065289163977962187507742615436240832
c1: -45195693963724945835426077433010009156
c2: -298923418394166123716233260770564
c3: 328843002406215196716601281
c4: 2641231812968768558129
c5: -1065578287083828
c6: -276312960
Y0: -8247746467927728998793444647326297
Y1: 69974668288047521820527
# MurphyE (Bf=1.718e+10,Bg=1.718e+10,area=1.288e+16) = 1.17e-08

Cownoise says skew 586407.73298 score 1.21942919e-15

Two close polys:

Y0: -8247744810367786591523747762682721
Y1: 139949336576095043641054
c0: -121697515694159644891681490270913604269342
c1: 3234923352759311823303330074168072691
c2: 133402768957865817059990225314786
c3: -573052357776445442366809569
c4: -5025398088308725392418
c5: 4419400381850832
c6: 2210503680
skew: 255103.69305
# lognorm 58.78, E 49.61, alpha -9.17 (proj -3.12), 6 real roots
# MurphyE = 1.17147226e-15

and

Y0: -16495489966970231872306633493333038
Y1: 69974668288047521820527
c0: -58164215341232454269144736611727941171251680
c1: 358962284111646560756801536406561670778
c2: 1083864026708650513009029013236827
c3: -1096238294845069101359840050
c4: -2526340671503694979769
c5: 550374252942234
c6: 69078240
skew: 1115632.55307
lognorm 59.65, E 49.78, alpha -9.88 (proj -2.66), 4 real roots
MurphyE = 1.15731893e-15

[fivemack]

Originally Posted by VBCurtis

My understanding is that GGNFS internally adjusts skew down to Q; I'll see if I can find the source of my [mis]understanding.

If you set skew to 410M, do you get the same 40 relations? A "no" would conclusively reject my claim.

No, I get thirteen relations.

I think you are getting confused with the way that GGNFS 15e internally adjusts alim/rlim down to Q0 (IE if you do -a -f 20000000, it will sieve with alim=20000000 if the value in the poly file is larger than that)

[VBCurtis]

Originally Posted by fivemack

No, I get thirteen relations.

I think you are getting confused with the way that GGNFS 15e internally adjusts alim/rlim down to Q0 (IE if you do -a -f 20000000, it will sieve with alim=20000000 if the value in the poly file is larger than that)

Thank you for doing the footwork to confirm my misunderstanding. Your explanation is the best I have, until I get around to perusing some of the places where esoteric GGNFS discussions have taken place 'round here.

[VBCurtis]

My second deg-6 run has finished:

n: 1035494967981608996094968430385789545942273148368197179174157979458916926996249361835599034841457568721713517$
skew: 1483907.861
c0: -758486316282603838580261525697958736410558065
c1: 89244789800220344402617658427797448531
c2: 3372372111141991619249066738092247
c3: -1439537140116844521216509025
c4: -1528375860162091192646
c5: 125567209134150
c6: -25137000
Y0: -12078430430581541489961342383064362
Y1: 12185711440192546653435119
# MurphyE (Bf=1.718e+10,Bg=1.718e+10,area=1.288e+16) = 1.11e-08

This score is lower than the first one, so I didn't bother with cownoise.
Test-sieving on 2330M indicates that deg 5 is far superior to deg 6 on the polys we found for that project (~1.18 deg 5 vs ~1.38 deg 6 is about 20% faster for deg 5). So, I'm moving my CADO search to deg 5 next time I have some free cycles; I expect a 9e-16 deg 5 to be faster than either of my deg 6 polys here.

[swellman]

Originally Posted by VBCurtis

My second deg-6 run has finished:

n: 1035494967981608996094968430385789545942273148368197179174157979458916926996249361835599034841457568721713517$
skew: 1483907.861
c0: -758486316282603838580261525697958736410558065
c1: 89244789800220344402617658427797448531
c2: 3372372111141991619249066738092247
c3: -1439537140116844521216509025
c4: -1528375860162091192646
c5: 125567209134150
c6: -25137000
Y0: -12078430430581541489961342383064362
Y1: 12185711440192546653435119
# MurphyE (Bf=1.718e+10,Bg=1.718e+10,area=1.288e+16) = 1.11e-08

This score is lower than the first one, so I didn't bother with cownoise.
Test-sieving on 2330M indicates that deg 5 is far superior to deg 6 on the polys we found for that project (~1.18 deg 5 vs ~1.38 deg 6 is about 20% faster for deg 5). So, I'm moving my CADO search to deg 5 next time I have some free cycles; I expect a 9e-16 deg 5 to be faster than either of my deg 6 polys here.

I’m torn - I want to help with the poly search but being restricted to msieve-GPU gives little hope of ever producing anything useful for a job this big. I’ve been traveling but when I get home later today I will terminate poly searching unless I find that a fantastic result fell out.

[VBCurtis]

Gimarel found quite a few good polys for 2330M using msieve (admittedly, with his secret-sauce using multiples of 120120 or 210210- I forget which- as leading coefficients).

I'd start somewhere like 50 million if I were using msieve, and if I had a functioning CUDA install I'd be using msieve in addition to CADO.

[Gimarel]

My best polys so far. I hope to get above 1e-15.

# norm 1.191764e-20 alpha -8.739251 e 9.542823e-16 rroots 5
skew: 290147042.90
c0: -135776935562388704990549019644794533664814428164000
c1: 5937379427987227585128437919182971104650500
c2: 5804073041546668110449407917351361
c3: -333095482498862424229058609
c4: -135588904276392300
c5: 683242560
Y0: -19784296655912504351830419322570633016087
Y1: 144157419788673482041

# norm 1.142823e-20 alpha -9.668513 e 9.431845e-16 rroots 3
skew: 58035263.11
c0: -4986872452312652038145241813323771956740728170410
c1: -246617563086234831662512103167398893438877
c2: 26937900914384572962697443420308252
c3: -65711817531926823292081111
c4: -1328932861673785270
c5: 37383746400
Y0: -18093085778901249382350215393204523934312
Y1: 161147338183822935653

# norm 1.152128e-20 alpha -8.465413 e 9.360483e-16 rroots 3
skew: 262473092.38
c0: -242951724554279944917921601254250434171076371823856
c1: 208050745037498983680860166923829227243064
c2: 14449538634104684658097780043593177
c3: -75840018042149775269085489
c4: 169111997159595514
c5: 687566880
Y0: -19759347904801229668001852015718067377447
Y1: 217477571882419666331

[swellman]

Well sure enough, when I returned home from my travels msieve-GPU had not found anything useful. I’m pulling the plug on my search temporarily, but I will likely later take up the suggestion by VBCurtis to search a higher range, e.g. c5>50M. And I must play with CADO poly search - I don’t have much experience with CADO but it is very powerful.