20130907, 19:40  #12  
Banned
"Luigi"
Aug 2002
Team Italia
29×167 Posts 
Quote:
I am "stuck on sieving" because my old PARI script was run up to 1G (50" for a full test on K), then I passed the survivors on Ernst's program up to 20G, and each candidate required less than 15 minutes to reach 20G. A test on MM34 with pfgw requires about one hour, so I decided to presieve candidates up to 800G (less than one hour) to eliminate some more Ks. As the MM exponent grows, the time required by pfgw grows more, so the bigger the exponent, the higher the sieve level. Finally, I only sieve when I have spare time (that's why it seems I'm "stuck"). Let me try your script, and see if the method can be reworked. Just a last question: did you test if LLR is faster than pfgw on these numbers? Luigi 

20130907, 20:16  #13 
"Serge"
Mar 2008
Phi(4,2^7658614+1)/2
2×3×1,657 Posts 
Well, I meant "stuck" figuratively. And surely not just you but the whole team,  the plural "you". I'd expect that more people ventured into actually, testing, without a need to nudge.
Re: your question about PFGW vs LLR: I didn't check; in LLR a typical test for MM35 is ~40 minutes. We can benchmark on this toy plain looking example: Code:
ABC $a*$b^$c$d 1433553 2 1398270 2867105 
20130907, 20:55  #14  
Banned
"Luigi"
Aug 2002
Team Italia
29×167 Posts 
Quote:
I am not fluent with PARI, so I may have lost something in translation... You assign M=2000000, start the forprime cycle with p=5 to create the primes array, then reassign M to p (this is the value I change in your script after each break). What am I losing? Luigi Last fiddled with by ET_ on 20130907 at 21:00 

20130907, 21:00  #15 
"Serge"
Mar 2008
Phi(4,2^7658614+1)/2
2×3×1,657 Posts 
After the break, you use
p=K[1]; This value is also saved in the sieve file (in the 2nd line). And yes, you have to rename the file (or change the script slightly). It is easier for you to rename it externally, right after it is done writing. 
20130907, 21:08  #16  
Banned
"Luigi"
Aug 2002
Team Italia
29×167 Posts 
Quote:
In my opinion, the fle should shrink after each break, as the sieve proceeds. It didn't. Maybe I let the whole thing run for too little time? Luigi 

20130907, 21:39  #17 
"Serge"
Mar 2008
Phi(4,2^7658614+1)/2
2×3×1,657 Posts 
It is an interactive script  you don't call it. You throw in batches of lines into a working pari shell.
You can, of course, modify it to be parametrized and sieve to a certain limit: replace "while(p=nextprime(p+1)," with "while((p=nextprime(p+1))<some_limit," and run it through. It was not my intention to make it a finished script. 
20130909, 01:19  #18 
"Serge"
Mar 2008
Phi(4,2^7658614+1)/2
2·3·1,657 Posts 
Oh, when it rains, it pours.
This one is luckier still than the previous one. But there we go: 374568*(2^30213771)+1 is prime. MM3021377 division test is in progress... ...it doesn't divide MM3021377 (no surprise here) Last fiddled with by Batalov on 20130909 at 05:31 
20130909, 08:20  #19 
Banned
"Luigi"
Aug 2002
Team Italia
29×167 Posts 
Nice one! You are rocking on MM34, MM35, MM36 and MM37!
May I ask you to communicate the range of exponents you test from time to time, so that I can keep the ranges updated? I've seen that you used OpenPFGW for this one, using Mp as helper for N1. Would you mind sharing to the forum the command line you used? Meanwhile, I am using your script to extend the sieving limits to MM38, MM47 and MM48 for the first 2,000,000 candidates. I plan to update the table for all double Mersennes > MM33, so that we can concentrate on searching candidates instead of "sieving". Luigi 
20130909, 19:33  #20 
"Serge"
Mar 2008
Phi(4,2^7658614+1)/2
23326_{8} Posts 
News from subproject Deep Sieving
Luigi,
I will wrap up all results and email them to you. (Too big for attachments.) Re: the proof. You prepare a separate file, let's call it "Help139", where you put a single line Code:
2^13982691 Code:
pfgw f1 l t hHelp139 q"507568*(2^13982691)+1" I've proven many nearrepunits for M.Kamada's site and had used very large helper files from time to time. In the helper file, every line should be a prime (and formulae are accepted, too). If you run a tc proof, you put factors for both N1 and N+1 side in the same file. The routine will simply use try them after short trial factoring (that's why you don't have to put the 507568 factors, 2 * 2 * 2 * 2 * 31723: a) they are small, b) the N1 factorization is already way over 33%.) Here's another example: http://primes.utm.edu/primes/page.php?id=115087 (note that all factors were supplied only once) 
20130910, 09:57  #21  
Banned
"Luigi"
Aug 2002
Team Italia
4843_{10} Posts 
Quote:
Luigi 

20130910, 20:02  #22  
"Åke Tilander"
Apr 2011
Sandviken, Sweden
2·283 Posts 
Quote:


Thread Tools  
Similar Threads  
Thread  Thread Starter  Forum  Replies  Last Post 
Deep Sieving MM49 in parallel  ET_  Operazione Doppi Mersennes  22  20160728 11:23 
Deep Hash  diep  Math  5  20121005 17:44 
The news giveth, the news taketh away...  NBtarheel_33  Hardware  17  20090504 15:52 
Question on going deep and using cores  MercPrime  Software  22  20090113 20:10 
Deep Sieving 10m Digit Candidates  lavalamp  Open Projects  53  20081201 03:59 