mersenneforum.org Reserved for MF - Sequence 4788
 Register FAQ Search Today's Posts Mark Forums Read

 2021-08-28, 02:02 #3081 swellman     Jun 2012 62478 Posts Completed the 40-50M range. Best cownoise score was 3.0276e-16. Fairly blah. Searching 50-60M next.
 2021-08-29, 05:42 #3082 bur     Aug 2020 79*6581e-4;3*2539e-3 41810 Posts Completed 6e6-7e6, best poly was 3.066e-16 cownoise. Seeing swellman's result, I'll stop the poly search now since it seems the 3.8e-16 will be very hard to beat. I did 1e6-2e6 and 5e6-7e6. Btw, did you finish the range you did twice with different P? Last fiddled with by bur on 2021-08-29 at 05:43
2021-08-29, 10:52   #3083
swellman

Jun 2012

323910 Posts

Quote:
 Originally Posted by bur Btw, did you finish the range you did twice with different P?

40-41M w/P=10M, nq=6^6 gave a best poly of 2.709e-16 per cownoise.

40-41M w/P=20M, nq=6^6 gave a best poly of 2.578e-16.

40-41M w/P=10M, nq=6^7 gave a best poly of 2.898e-16 and the 10th best poly scored 2.586e-16.

(All used incr of 4620 and adrange 4x that.)

So nq=6^7 helped in this case, producing the highest scoring poly. But it took a week rather than a day and the results were still rather unimpressive.

My current search of 50-60M (P of 10M, nq is 6^6 and incr is 4620) should finish up on Wednesday. I’m prepared to search up to 300M but that will take a long time.

2021-08-29, 11:51   #3084
charybdis

Apr 2020

22·33·5 Posts

Quote:
 Originally Posted by bur Seeing swellman's result, I'll stop the poly search now since it seems the 3.8e-16 will be very hard to beat. I did 1e6-2e6 and 5e6-7e6.
Sean was using incr=4620 so his 10M range was more comparable to one of your 1M ranges.

And about the "hard to beat" bit...

Code:
n: 5272066026958413205513021090082556639441277154855572268239336980532402881465013381219738819137405617219594641652778228990107677072837959240400905630530715435664638237254654768053674178165309345879869729405448588458952991
skew: 331125.033
c0: 3394949349018158588040756570853974067801857
c1: -6653335906218714378144808303035495969
c2: 74040391174021644872973194328541
c3: -276448343494407068460138157
c4: -4606872840152057004918
c5: 1393805639345406
c6: 816114600
Y0: -226657691871392565816046108637660098
Y1: 69216809380453058295685069
Cownoise says skew 534276.01166 score 4.17772314e-16! This one actually does look like a possible outlier; let's see if Gimarel can make it even better.

Second place from 25M-40M was
Code:
n: 5272066026958413205513021090082556639441277154855572268239336980532402881465013381219738819137405617219594641652778228990107677072837959240400905630530715435664638237254654768053674178165309345879869729405448588458952991
skew: 259828.731
c0: 291564310755372745873385870920629996500880
c1: -11770010484070210819325995157141005616
c2: 804097425342022282773053964128739
c3: -1607028399777058166539822136
c4: -15556228087219154844031
c5: -1494532571792100
c6: 584942400
Y0: -237694813432371337441531043422949258
Y1: 47124166781711191105282781
which scores 3.78642741e-16, so with some spin it might also beat the previous 3.85 - but that's purely academic now.

2021-08-29, 12:56   #3085
Gimarel

Apr 2010

11·17 Posts

Quote:
 Originally Posted by charybdis Sean was using incr=4620 so his 10M range was more comparable to one of your 1M ranges. And about the "hard to beat" bit...
I couldn't improve the previous two. I'll try these now.

2021-08-29, 16:33   #3086
Gimarel

Apr 2010

11×17 Posts

Quote:
 Originally Posted by charybdis Cownoise says skew 534276.01166 score 4.17772314e-16! This one actually does look like a possible outlier; let's see if Gimarel can make it even better.
Sorry, I didn't find anything better.

2021-08-29, 22:22   #3087
swellman

Jun 2012

41·79 Posts

Quote:
 Originally Posted by charybdis Sean was using incr=4620 so his 10M range was more comparable to one of your 1M ranges. And about the "hard to beat" bit... Code: n: 5272066026958413205513021090082556639441277154855572268239336980532402881465013381219738819137405617219594641652778228990107677072837959240400905630530715435664638237254654768053674178165309345879869729405448588458952991 skew: 331125.033 c0: 3394949349018158588040756570853974067801857 c1: -6653335906218714378144808303035495969 c2: 74040391174021644872973194328541 c3: -276448343494407068460138157 c4: -4606872840152057004918 c5: 1393805639345406 c6: 816114600 Y0: -226657691871392565816046108637660098 Y1: 69216809380453058295685069 Cownoise says skew 534276.01166 score 4.17772314e-16! This one actually does look like a possible outlier; let's see if Gimarel can make it even better.
Nice one! Too bad it couldn’t be increased but that’s a breathtaking score.

For reasons of learning and stubborn pride, I’m going to finish my search range in a vain effort to improve on it.

2021-08-29, 23:18   #3088
charybdis

Apr 2020

22·33·5 Posts

Quote:
 Originally Posted by Gimarel Sorry, I didn't find anything better.
No worries, thanks for checking!

 2021-08-30, 18:09 #3089 bur     Aug 2020 79*6581e-4;3*2539e-3 2·11·19 Posts Oh, that's impressive... :D It did beat the RSA-220 poly by 7.5%. It was nice learning a lot about polynomial search. Now let Ryan Propper work his server farm magic.
 2021-08-31, 03:33 #3090 VBCurtis     "Curtis" Feb 2005 Riverside, CA 117118 Posts Not just yet- we have to pick sieve parameters first! I used lpb0=34 and lpb1=35 for the GNFS-207 team sieve two years ago, and both LP bounds were higher than they needed to be. I believe those are big enough for this job. The default C220 params file uses 500M/800M for lim's. Are those big enough? mfb0 should be test-sieved at 66-67-68, and mfb1 should be test-sieved at 99-100-101-102. I used 66 and 99 for the C207, but this job is over four times tougher so likely 67 and 100-101 will be best. One of the CADO publications observed that a very large ncurves on the 2LP side helped, say 50 or 75 (again, test-sieve?) ncurves1 of 20ish is likely fine (default is 21). We'd like to determine yield for a couple of test Q values, say at 100M intervals; once we estimate yield across the sieve region we can choose starting Q such that ending Q is expected to be 8 * starting Q.
 2021-08-31, 14:32 #3091 charybdis     Apr 2020 22·33·5 Posts I'll do some test-sieving. Unsurprisingly, the 4.18e-16 poly is clearly better than the 3.85e-16 poly; I won't bother testing any of the others.

 Similar Threads Thread Thread Starter Forum Replies Last Post RichD Aliquot Sequences 529 2021-11-05 17:27 kar_bon Aliquot Sequences 136 2021-10-21 16:17 RichD Aliquot Sequences 476 2021-10-04 20:47 prism019 GPU to 72 6 2020-09-21 22:11 petrw1 Lone Mersenne Hunters 82 2010-01-11 01:57

All times are UTC. The time now is 14:40.

Sat Nov 27 14:40:55 UTC 2021 up 127 days, 9:09, 0 users, load averages: 0.80, 1.01, 1.01