Fast Breeding (guru management)

[jyb]

Originally Posted by jyb

SNFS-182.5 (octic) C174 HCN (8+3,750L), ECM to t55. For 14e.

Interestingly 14e actually came out a little bit better than 15e for this one. I also tried using 3LP on the algebraic side. Yield was substantially higher, but it was also commensurately slower, so it ended up being a little bit worse in total (projected) time, to say nothing of the extra relations which would probably be needed.

Mike, I believe the octic you tried last fall had a difficulty of 175. I'll be interested to see whether this one ends up sieving better than yours. Certainly the test sieving I did suggests it should be okay, but of course your test sieving ended up misleading you, so perhaps the same thing will happen here.

[unconnected]

QUEUED AS C171_11040_10138

C171 from 11040:i10138 for 14e queue

n: 354225149254554986860192908811463962691956702506502254924592920677644828794402181249756192592010065434808034711619046330370429918761354901147288863312191867942395462161293
skew: 109771994.05
c0: 130122329926881277862630188680456328241448800
c1: 6229157937504323888359085530752201336
c2: 23910461580493994724481567820
c3: -847619197460505156903
c4: -3594682303798
c5: 34920
Y0: -1589428643886288767634525387300483
Y1: 711661560001010207
rlim: 134000000
alim: 134000000
lss: 0
lpbr: 31
lpba: 31
mfbr: 62
mfba: 62
rlambda: 2.6
alambda: 2.6

Suggesting sieving range 20M-180M. I'll take the LA.

[fivemack]

Originally Posted by fivemack;524129
The best polynomial I got after an overnight search was

[code

n: 4490929289993305463941782391402569555718836701920884946505771634999063887745952445887475578875492666806455138495828705759696808152009844859518266437837184140423450006769373462680041385370778722521
# norm 2.782481e-19 alpha -8.139046 e 6.806e-15 rroots 3
skew: 1546232048.23
c0: -840275467664477384115030092092308912794938768818000
c1: -401623096793192586248782919178513646926916
c2: 1364798544769124532296446769656552
c3: 181953540629697374159247
c4: -493865460396230
c5: 170352
Y0: -121394160494208958424033844696497192557
Y1: 28307512268007061
lpbr: 32
lpba: 32
mfbr: 64
mfba: 96
alambda: 3.4
rlambda: 2.6
alim: 400000000
rlim: 400000000
[/code]

with a yield of 0.54 for 15e 400M..400M+1k ... doing trial sieving with a longer range and with 16e now.

15e, 400M..400M+10k : 6791 rels, 1.77723s/r
16e, 400M..400M+10k : 15752 rels, 1.44237s/r

[Max0526]

Originally Posted by fivemack

The best polynomial I got after an overnight search was

n: 4490929289993305463941782391402569555718836701920884946505771634999063887745952445887475578875492666806455138495828705759696808152009844859518266437837184140423450006769373462680041385370778722521
# norm 2.782481e-19 alpha -8.139046 e 6.806e-15 rroots 3
skew: 1546232048.23
c0: -840275467664477384115030092092308912794938768818000
c1: -401623096793192586248782919178513646926916
c2: 1364798544769124532296446769656552
c3: 181953540629697374159247
c4: -493865460396230
c5: 170352
Y0: -121394160494208958424033844696497192557
Y1: 28307512268007061
lpbr: 32
lpba: 32
mfbr: 64
mfba: 96
alambda: 3.4
rlambda: 2.6
alim: 400000000
rlim: 400000000

This poly spins up a bit:

Y0: -121394160494208635119633489923116438285
Y1: 56615024536014122
c0: -105585521921181701103640345310081612864493498154458
c1: -92594919639107685756663746064550136764561
c2: 685324456456926747152371015765412
c3: 159613702437217943737487
c4: -968274759937420
c5: 681408
skew: 797775329.89
# size 2.306e-19, alpha -8.139, combined = 6.860e-15 rroots = 3

[jyb]

QUEUED AS 7p4_350

SNFS-253.53 C170 HCN (7+4,350), ECM to t55. For 14e.

n: 79914170303630821636026797890048444556975345685648263846659949185388632991848813783896287921441624696035480039352728725218949312274080628143077882775012019402751447382601
# 7^350+4^350, difficulty: 253.53, skewness: 1.00, alpha: 2.24
skew: 1.000
c6: 1
c5: -1
c4: 1
c3: -1
c2: 1
c1: -1
c0: 1
Y1: -1267650600228229401496703205376
Y0: 1798465042647412146620280340569649349251249
rlim: 134000000
alim: 134000000
lpbr: 31
lpba: 31
mfbr: 93
mfba: 62
rlambda: 3.6
alambda: 2.6

Trial sieving 2K blocks.

 Q		Yield
---		-----
20M		4196
60M		2741
100M		2337
140M		1907
180M		1666
220M		1779
260M		1640
300M		1540

Recommend sieving special Q on rational side, 20M - 260M.

[RichD]

QUEUED AS 6315307_37m1

C240 from the OPN t600 file.

n: 468606385875692014929972652289661433000588213753651450054257348630859358470526020304983526315899188651321848158981995731337834216429975710602749558580783965017051317749288679772059003373383296582945027203369312190011522773516180264957644281
# 6315307^37-1, difficulty: 251.61, skewness: 0.07, alpha: 0.00
# cost: 9.3915e+18, est. time: 4472.14 GHz days (not accurate yet!)
skew: 0.074
c6: 6315307
c0: -1
Y1: -1
Y0: 63440530225929597678780901886081018950249
type: snfs
rlim: 134000000
alim: 268000000
lpbr: 32
lpba: 32
mfbr: 64
mfba: 64
rlambda: 2.7
alambda: 2.7

Trial sieving 5K blocks.

  Q  Yield
 20M 10279
 60M  8659
100M  8164
200M  6506
300M  5526
350M  5442

[jyb]

QUEUED AS 8p3_750M

Originally Posted by jyb

Interestingly 14e actually came out a little bit better than 15e for this one. I also tried using 3LP on the algebraic side. Yield was substantially higher, but it was also commensurately slower, so it ended up being a little bit worse in total (projected) time, to say nothing of the extra relations which would probably be needed.

Mike, I believe the octic you tried last fall had a difficulty of 175. I'll be interested to see whether this one ends up sieving better than yours. Certainly the test sieving I did suggests it should be okay, but of course your test sieving ended up misleading you, so perhaps the same thing will happen here.

Sieving for 8+3,750L went pretty much exactly as the test sieving suggested, so that's good. And the resulting matrix was pleasingly small. I wonder what went wrong with yours last fall.

Anyway, 8+3,750M should have the same characteristics, and test sieving seems to bear that out:

SNFS-182.5 (octic) C174 HCN (8+3,750M), ECM to t55. For 14e.

n: 184297901378704001051400947472809906021306982539424534628765632569392246268068872237931555882481992079429062159719172385089952471555323821903942659019581500261176936799545001
skew: 1.63299316185545
lss: 0
c8: 81
c7: -324
c6: -1080
c5: 5184
c4: 2304
c3: -20736
c2: 0
c1: 24576
c0: 4096
Y1: 109561042308169728
Y0: -37778931863804450319011
rlim: 134000000
alim: 134000000
lpbr: 31
lpba: 31
mfbr: 62
mfba: 62
rlambda: 2.6
alambda: 2.6

Trial sieving 10K blocks.

 Q		Yield
---		-----
20M		24338
35M		22771
50M		20578
65M		20002
80M		18499
95M		20576
110M		18300
125M		18062
140M		14654
155M		14955
170M		15713
185M		14566
200M		15109
215M		13158
230M		13317
245M		12245

Recommend sieving special Q on algebraic side, 20M - 160M.

[jyb]

QUEUED AS 3m2_583

GNFS-169 HCN (3-2,583), ECM to t55. For 14e.

n: 1366463024891822563809372215137224364930400443353450102734654456891580326641979241300486675985611995537181353277021413975685136338885053006379712923373209136737763977077
# norm 2.099723e-16 alpha -7.979449 e 3.950e-13 rroots 3
skew: 6254633.06
lss: 0
c0: 50754827591310587647932677033512537585800
c1: 9365380533380879500250541765296876
c2: -4219874468410266636786625036
c3: -1148096476954940267646
c4: -95565614298981
c5: 16041168
Y0: -153487276973815962661538994747767
Y1: 545946643683344837
rlim: 134000000
alim: 134000000
lpbr: 31
lpba: 31
mfbr: 62
mfba: 62
rlambda: 2.6
alambda: 2.6

Trial sieving 2K blocks.

 Q		Yield
---		-----
20M		3547
50M		3486
80M		2425
110M		4146
140M		2958
170M		2566
200M		2495
230M		2632

Recommend sieving special Q on algebraic side, 20M - 180M.

[fivemack]

[QUOTE=jyb;524680]

n: 184297901378704001051400947472809906021306982539424534628765632569392246268068872237931555882481992079429062159719172385089952471555323821903942659019581500261176936799545001
skew: 1.63299316185545
lss: 0
c8: 81
c7: -324
c6: -1080
c5: 5184
c4: 2304
c3: -20736
c2: 0
c1: 24576
c0: 4096

I can't help noticing that c_{8-n} is divisible by 2^n, which sounds as if you could pull a factor two out to the rational side ... did you try that and find it sieved worse?

[VBCurtis]

Originally Posted by jyb

Sieving for 8+3,750L went pretty much exactly as the test sieving suggested, so that's good. And the resulting matrix was pleasingly small. I wonder what went wrong with yours last fall.

Well, my test-sieving indicated the f siever would be 50% faster than e, so I used f to run the factorization (I think, it has been a while). But f test-sieves weirdly, including sec/rel that depends on the size of the Q-range tested; I kept the production ranges close in size to the test ranges (I think 10k vs 5k), but my testing still wasn't accurate.

Next time I run an octic, I'll stick to e, and based on your results I expect normal & predictable outcomes. I think my lesson learned is about f siever, rather than octics.

[jyb]

[QUOTE=fivemack;524800]

Originally Posted by jyb

n: 184297901378704001051400947472809906021306982539424534628765632569392246268068872237931555882481992079429062159719172385089952471555323821903942659019581500261176936799545001
skew: 1.63299316185545
lss: 0
c8: 81
c7: -324
c6: -1080
c5: 5184
c4: 2304
c3: -20736
c2: 0
c1: 24576
c0: 4096

I can't help noticing that c_{8-n} is divisible by 2^n, which sounds as if you could pull a factor two out to the rational side ... did you try that and find it sieved worse?

Hmm, no I didn't try that and I obviously should have. I did just now, and it sieves a little bit better with your suggested change (like maybe about 7-8% better). This number has been queued, but it hasn't yet had any work units handed out. It would be possible to remove it and submit the change, if we can get Greg to cancel the work units before anything is handed out. It's not clear to me whether that would be considered worthwhile, given the modest improvement, though other things being equal I would of course prefer to see resources used as efficiently as is practical. I guess I'll have to leave that decision to those who are managing the queue.