Polynomial Request Thread

[Max0526]

Originally Posted by RichD

One more to try. This may not "spin-up" to an acceptable value.
I should be done with my range by the end of the week..

R0: -85201023096095502462195246075989797733
R1: 15072793197528296573
A0: 164515308847335041245918218232138873461226320000
A1: 24893089190908560312936283088389362735880
A2: 106473286264769714352581049587106
A3: -10046975453646539492044043
A4: -1220109891798340
A5: 4655304
skew 234764745.72, size 2.183e-19, alpha -8.347, combined = 6.408e-15 rroots = 5

A bit better after scaling, but not a 7 yet:

Y0: -85201023096095558110947731350460745249
Y1: 30145586395056593146
c0: 20552925626733838358539059941131231603103136080
c1: 6223075645329039701640044567706058833330
c2: 53292283232552524317165929776647
c3: -10046957434429099257132363
c4: -2440391657420360
c5: 18621216
skew: 117721989.97
# size 2.315e-019, alpha -8.520, combined = 6.708e-015 rroots = 5

[VBCurtis]

I'll finish my GPU-msieve run to 1M 'cause I said I would, but I think CADO is just better at poly select.
I ran 500k-510k on CADO overnight, which took 40ish thread-hours on CPU.

n: 20901242090929750331460152543809850619619290452256403363055163961224444523016833873892080737966534318102450035876136293050152061659131510274015648390725105024055454805289274799895497557473573901631
skew: 81678244.604
c0: 17454608775923076542170191197367958335585302280
c1: -3597613049655018147902433963741039992857
c2: 62474104924329530466242328603952
c3: 1962455287259675504373633
c4: -5663904865157888
c5: -4070880
Y0: -132652270861043793523404821016733722071
Y1: 73944454545426597057419
skew 81678244.60, size 2.374e-19, alpha -7.783, combined = 6.917e-15 rroots = 5

The last line is the result of invoking msieve -nc1 to see what score it gave.
I'll continue both CADO and msieve, for now.

[swellman]

Originally Posted by VBCurtis

I'll finish my GPU-msieve run to 1M 'cause I said I would, but I think CADO is just better at poly select.

From what I’ve seen posted here, it’s hard to disagree. My msieve-GPU results on this C197 have contributed nothing to the effort, including last night’s results. But hope springs eternal.

I will finish c5 through 500k because it’s important to finish a job, but my expectations are low. I’m playing with parameters again tonight in attempt to get better production if not higher e-scores.

Which composite is on deck, assuming we are almost done with the C197?

[Max0526]

Originally Posted by VBCurtis

I'll finish my GPU-msieve run to 1M 'cause I said I would, but I think CADO is just better at poly select.
I ran 500k-510k on CADO overnight, which took 40ish thread-hours on CPU.

n: 20901242090929750331460152543809850619619290452256403363055163961224444523016833873892080737966534318102450035876136293050152061659131510274015648390725105024055454805289274799895497557473573901631
skew: 81678244.604
c0: 17454608775923076542170191197367958335585302280
c1: -3597613049655018147902433963741039992857
c2: 62474104924329530466242328603952
c3: 1962455287259675504373633
c4: -5663904865157888
c5: -4070880
Y0: -132652270861043793523404821016733722071
Y1: 73944454545426597057419
skew 81678244.60, size 2.374e-19, alpha -7.783, combined = 6.917e-15 rroots = 5

The last line is the result of invoking msieve -nc1 to see what score it gave.
I'll continue both CADO and msieve, for now.

Skew value can be better:

Y0: -132652270184646064681478657332028539527
Y1: 73944454545426597057419
c0: 8764470660078061180004984919116618804464294760
c1: 1979525637499648007811807357392774873881
c2: -113453349487284995307981304430848
c3: -1751809529556075066257281
c4: 5850094215212288
c5: 4070880
skew: 135252441.05
# size 2.374e-19, alpha -7.783, combined = 6.979e-15 rroots = 5

[Max0526]

Originally Posted by swellman

From what I’ve seen posted here, it’s hard to disagree. My msieve-GPU results on this C197 have contributed nothing to the effort, including last night’s results. But hope springs eternal.

I will finish c5 through 500k because it’s important to finish a job, but my expectations are low. I’m playing with parameters again tonight in attempt to get better production if not higher e-scores.

Which composite is on deck, assuming we are almost done with the C197?

Well, hold on a second. Some info on the last three polyselect projects.
1) The initial poly with E=1.438e-13 for C177_134_73 was found by my CADO and was optimized by msieve, CADO, and scaling before E became 1.527e-013 -- a new C177 record. There was an increase in E of 6.2% during spinning the poly. That would hardly be possible without using msieve.
2) The initial poly with E=1.102e-13 for C179 (cofactor of 13*2^914-1) was found by VBCurtis's CADO and was optimized by msieve, CADO, and scaling before E became 1.226e-013 -- a new C179 record that is also better than the current C178 record. There was an increase in E of 11.3% during spinning the poly. Again, that would hardly be possible without using msieve.
3) The initial poly for the current C197

N: 20901242090929750331460152543809850619619290452256403363055163961224444523016833873892080737966534318102450035876136293050152061659131510274015648390725105024055454805289274799895497557473573901631
R0: -86825671372509677292865083630970354043
R1: 2045148860419396619
A0: 2312331198521591434462181124108904018182857200248
A1: 77239043600173382348632075745581721624878
A2: -678505319363690260419212557463911
A3: -3627182455388757596945318
A4: 5686908312307908
A5: 4235760
skew 349240680.19, size 2.226e-19, alpha -8.300, combined = 6.704e-15 rroots = 5

with E=6.704e-15 was found by RichD's msieve and was optimized by msieve, CADO, and scaling before E became 7.867e-015 -- a new C197 record. There was an increase in E of 17.3% during spinning the poly. Now, that would definitely be impossible without using msieve.
So please don't give up on msieve, it actively participates in generating polys with record high E scores.
And I guess the next record poly will be found by swellman's msieve, to keep everything even. :-)

[VBCurtis]

Max-
We're referring to GPU-msieve for the first phases vs CADO. Whatever you do for root-opt and polish is not what we mean.

My most recent point about CADO being flat better is that a 2-process overnight run produced a 6.91, while the best msieve produced was 6.70. You're right that Rich's polished better, but msieve in aggregate among Rich, Sean and I has been run for ~15 GPU days. If I give CADO's size-opt 60 thread-days I'm pretty sure it'll turn up a 7.x or five, and it's fairly likely that polishing will improve on the current best.

If Rich's raw poly proves the best, then you're quite right that we shouldn't give up on msieve-GPU just yet!

[swellman]

Point taken Max. I meant finding the best “first hit” seems to favor CADO, especially on these bigger numbers. Maybe it’s just my perception. Certainly your alchemy bounces one method off the other, so both tools are needed.

Anyway I’m still punching through with msieve.

[RichD]

It is almost like watching a ping-pong match. Msieve serves one up across the net. It is returned by CADO with a slight improvement. Wait, did Msieve send one back?

Now it is CADO's turn to serve one up. Msieve puts a slight spin on it. Did it hit the edge of the table? :-)

[Max0526]

I think it would be beneficial (for my scaling afterwards, anyways) to have msieve's c5 coefficient to be a multiple of 60 (same as in CADO) instead of the default multiple of 12.
Is it possible to modify?

[chris2be8]

You could write a script to run:
msieve -np1 60,60 ...
msieve -np1 120,120 ...
msieve -np1 180,180 ...

over whatever range you wanted.

Chris

[RichD]

I am finished with the 4-5M range. The last push didn't even get into a 6-handle.