20200615, 13:33  #1 
Mar 2006
Germany
17·167 Posts 
Msieve Statistical Data for factored numbers
Seeing projects like (Homogeneous) Cunningham, Brent or x^y+y^x for factoring large numbers using NFS@Home or only private crunching using Msieve, there're always log files with data almost always using pastebin to show.
What about a data collecting effort and therefore comparing such data in a tabular manner? I've created a sample page in the Wiki for the Homogeneous Cunningham number 11^281  5^281 with some data only shown as text so far. There's also a file of the msievelog uploaded as ASCII text. Links to the forum post and to the FactorDB data are also available. My question: To compare the data perhaps from the poly file, what data should be included in a template, which can hold the data and can be shown in a table with sorting/comparing? Some years ago I've create a (now outdated) page for HCN: Pages with such numbers using templates can be used to create reservations, wanted factorizations and overview tables to be created easily by filling such template. And all data are available in one place. Suggestions? Last fiddled with by kar_bon on 20200615 at 13:34 
20200615, 15:01  #2  
"Dylan"
Mar 2017
2×3×83 Posts 
Quote:
For a template, I'm thinking we should have similar data as shown on Makoto Kamada's website for near repunit number factorization:


20200615, 21:40  #3 
Mar 2006
Germany
2839_{10} Posts 
Are SNFS/GNFS diff. calculable? Otherwise only a parameter to fill in is possible here.
ECM depth could be collected like a list: #curves at any B1. Reservations, dates should be standard as in other templates. Long numbers can be written like in FactorDb: <first 10 digits>...<last 10 digits> or something if the digitlength is greater than 20 to keep a table view short. The whole number is available via FactorDb link or in the Wiki as separate page/text. What about the best poly or the poly taken to factor the number? E score the only important value? 
20200616, 04:45  #4  
Nov 2016
100001101001_{2} Posts 
Quote:
I think you should list the primes of these forms first, then the list of the factorizations of the numbers of these forms, e.g. there are currently no primes pages "repunit b" (primes of the forms (b^p1)/(b1) with prime p) and "homogeneous Cunningham M a b" (primes of the form (a^pb^p)/(ab) with prime p) and "homogeneous Cunningham P a b" (primes of the form (a^p+b^p)/(a+b) with odd prime p) in the wiki. 

20200616, 04:47  #5  
Nov 2016
2,153 Posts 
Quote:


20200616, 04:53  #6  
Nov 2016
4151_{8} Posts 
Quote:


20200616, 07:05  #7 
Mar 2006
Germany
17·167 Posts 
sweety439:
 learn how to quote a post or even don't: 3 posts with same quoting in 8 minutes is redundant, you can edit your posts in a 5 minute period when send  stop talking about 'we/we should', as almost always in your posts and stop asking people do something  I will not factorize any number in the Wiki, I try to collect and show data for factoring efforts  I can't implement any type of number you given, this is not doable  Why should I create pages for the form (a^p+b^p)/(a+b), if the general form a^p+b^p is the same with small factors included? Specialising is more work to do.  This thread is a first sight for collecting suggestions and nothing yet done in this direction in the Wiki. Perhaps only links to the FactoDB are better than doing the same as there like listing all factors. Keep in mind: someone has to keep those pages current. 
20200616, 14:15  #8  
"Dylan"
Mar 2017
2·3·83 Posts 
I'll try to answer each of kar_bon's questions in turn.
Quote:
For SNFS the difficulty is based on the original number (so for the number I just mentioned it's ~189.72). Of course, whether SNFS or GNFS is quicker depends on a number of factors (factors found previously, degree of the SNFS polynomial, etc.). Showing all curves makes sense  although perhaps in a table it's a bit too much to show  maybe just show the highest B1 curves, and then on the page show all curves known. For showing the number  I thought some more about it, and since we can find the number elsewhere, it would seem redundant to show the full form in a table. Besides, if the number has a numerical form, then one could calculate it. But what about numbers that don't have that, or depend on the previous number's factors (home primes and aliquot sequences)? Quote:
There are also anorm and rnorm scores which can be seen by using the poly generator on cownoise. But this appears only for Cunningham numbers, Brent numbers and OPN's  I am not sure how this would be calculated for other, more general numbers. At least one can test both sides (algebraic and rational) to see which one sieves better. I'm sure someone with more experience (VBCurtis, EdH, RDS) would have more information on the e score. 

20200616, 14:45  #9  
Nov 2016
100001101001_{2} Posts 
Quote:


20200616, 15:00  #10 
"Dylan"
Mar 2017
2·3·83 Posts 
The reason why it’s base 10 is because all pages give the difficulty in terms of digits, which is base 10. Same with the rule of thumb with GNFS (increasing the difficulty by 56 digits doubles the factoring effort). One can simply convert over to base 2 difficulty simply by multiplying the base 10 value by log(10)/log(2). Although I don’t see why you want to do that.

20200616, 15:56  #11  
"Curtis"
Feb 2005
Riverside, CA
19·223 Posts 
Quote:
I am very interested in collecting data about anorm and rnorm values the literature and the smart people around here indicate that lim's and LP values should be chosen based somewhat on those values, but I have never done so I just guess and iterate over what finds speed. If that data is gathered for SNFS jobs, I feel like I'll personally gain insight into possible parameter choices that may find speed (e.g. when to try a larger lim or more unbalanced lim choice than standard practice). 

Thread Tools  
Similar Threads  
Thread  Thread Starter  Forum  Replies  Last Post 
Condition on composite numbers easily factored (Challenger)  baih  Factoring  1  20191010 11:14 
Condition on composite numbers easily factored  baih  Factoring  16  20190929 15:48 
Runs of factored numbers  henryzz  Factoring  8  20170309 19:24 
Use Msieve NFS for small numbers?  skan  Msieve  8  20130226 20:35 
Same team factored RSA challenge numbers in recent time  koders333  Factoring  12  20060331 16:06 