MultiFactorial Prime Search
 

The new Multifactorial search page can be found [URL="http://mfprimes.cba.pl"]here[/URL]. This is a site run by Marian Otremba. Please work with him if you want to participate in the search.

My [URL="http://www.mersenneforum.org/rogue/multifactorial.html"]page[/URL] will still be available for a while but I won't be making updates. Once I complete the search on 7! to 600,000 I will likely take it down.

I have verified that primes are missing from Ken's site for various types. Sigh! I have only tested for b < 8000 so far. Hopefully nothing larger is missing.

Are there any ranges you'd like help on?

There seem to be a gap with n=16. If you want to sieve and test from 100001!16 thru 210000!16 (or further) that would be great.

No problem. I'll get start with n=16 from b=100001 to 210000.

Edit: Sorry, dumb question time. How do I start a new search? Do I need to set up a PFGW ABC file first?

[QUOTE=wombatman;442065]No problem. I'll get start with n=16 from b=100001 to 210000.

Edit: Sorry, dumb question time. How do I start a new search? Do I need to set up a PFGW ABC file first?[/QUOTE]

You could use the -h option of mfsieve to tell you. :smile:

Use -M16 -n10e4 -N21e4 -P1e14 -m1e14 -omf16.out.

You will want to vary -t (# of threads) and -C (# of chunks) to find the optimal performance.

Before you start run a PRP test for a 180000!16+1 (or thereabouts). You will want to sieve until the removal rate is slower than the time for that PRP test. Note that if you have multiple threads, then you will have to adjust the removal rate accordingly.

I'm ashamed to admit that I did use -h and misunderstood what it was telling me... :help:

Thanks for the instructions. :smile:

I have more bad news. For the two types I ran up to 100,000 I found missing primes for one of those types for n > 50,000. That combined with what I already knew was missing means that about 25 percent of the ranges listed on his site are missing primes and that is only on the small range of numbers I have retested. It appears that it will take me a number of months to re-run everything. If anyone else wants to throw their hat into the ring, I would appreciate it. I will try to get my new webpage up later today.

Note that I'm not blaming Ken in any way. His webpage is based upon what others have reported to him. There is no way for him to know that results submitted to him are wrong.

I have some good news. Now that OpenCL is working again fn my iMac, I can get back to an OpenCL version of the sieve. So far the testing is going smoothly, but I have further to go.

I have more bad news. I have tested to over 40,000 on each of the types. Of the 24 I'm testing, 17 are missing primes with a total of 21 of 48 with missing primes when including both + and -.

Right now I'm focused on type 2 up to 100,000. Although there is nothing missing from that type, the problems I've seen with the other types force me to retest it. Hopefully that will complete by the end of October, but it is difficult to gauge how long it will take because of how quickly the numbers get larger.

It sounds like it may be more sensible just to start over.