Patterns?

[ceselb]

Reading some old threads, I see that some people think there's patterns in the factors found with sieving.

I've made an excel spreadsheet with 1G < p < 80G, 3M < n < 20M for 4847.

If anyone wan't to try something with it, get it here. (http://www.visit.se/~ceselb/4847_3-20M_1G-80T.zip) (zipped, , 569kb)

I could do more if there's any demand, just ask.

[ceselb]

Not very pretty, but here it goes.

http://www.visit.se/~ceselb/4847_gaps.png

[ceselb]

http://www.visit.se/~ceselb/4847_pn.png

[MikeH]

I think I understand the effect we are seeing with these plots.

Looking at the upper plot, there are a number of horizontal bands close to the y-axis. These would seem to suggest a clustering of factors. However, what I think we are seeing here are the effects of duplicate and excluded factors.

For best illustration take the two most prominent bands around 67T and 78T. If you take a look at my results_duplicates_excluded_marked.txt file, there are huge quantities of excluded factors in these ranges.

In the higher ranges I believe the huge numbers of excluded factors have come from sievers using proth sieve submitting factors in the factrange.txt and factexcl.txt files. In some of the lower ranges, the anomalies will have come from the version of SoBSieve that didn't have the sob.dat bitmap enabled, and was producing far too many factors.

Ceselb, try regenerating the plots with my results_no_duplicates.txt, I suspect all the anomalies will disappear.

...and for anyone that hasn't already tried submitting the factrange.txt and factexcl.txt files, please don't submit factexcl.txt - nothing in it will ever score since they have already been removed from the sob.dat file, and it just takes up space on the server. As for factrange.txt, a few of the small factors might score, but really aren't that useful. If we are all going to make a habit of submitting small factors, then really we should change the sob.dat file to cover n right down to 1000!

[ceselb]

I'm fairly sure I did use no_duplicates.
I also removed any factors in that file below 300000.

I'll take another look in a few days.

[Keroberts1]

perhaps some analysis on a number of factors per say 40 bias maybe there would be some grouping found in this or perhaps a smaller (maybe bigger) scale