20211112, 08:30  #1 
Feb 2020
Germany
41_{8} Posts 
Triple prime search
Hi,
currently I am interested in triple primes. This is mainly because the record primes listed in Chris Caldwell's database are still quite small and therefore quick to check. While choosing a 'n' to sieve, I asked myself whether there is a list or some coordinated prime search project for triple primes where one could check if the 'n' was already searched or up to which upper limit the exponent was searched. I know that there are several similar projects / private range reservations etc. for other prime numbers. If there is no such project, would it be of interest to create one? regards hunson 
20211112, 10:07  #2 
"Vincent"
Apr 2010
Over the rainbow
2^{2}·11·61 Posts 
Look at http://www.pzktupel.de/ktuplets.htm
4111286921397 * 2^66420 + d, d = 1, 1, 5 seeem to be the largest one know Also look at this thread https://www.mersenneforum.org/showth...ighlight=tuple 
20211112, 10:50  #3 
"Matthew Anderson"
Dec 2010
Oregon, USA
3C5_{16} Posts 

20211112, 16:04  #4 
Feb 2020
Germany
3×11 Posts 
Thanks for the answers.
@firejuggler: The page you mentioned is about that what I thought about. I image a project that summarizes the ranges that where tested with the count of primes, twins, triples, sieve depth etc. so that anyone who wants to search for triple primes can easily see whether the range that he choose was already worked on. Similar to the CarolKyneaProject (https://www.mersenneforum.org/showthread.php?t=21216). I have a chemistry background and me and some friends back in uni imagined how useful a journal would be that does not show the latest and greatest in organic chemistry, but the things not to try :) The title would be something like "Don't waste your time trying this ...". So in that fashion one could image a summery of the ranges searched for triple primes. This would be especially interesting for "nice" exponents like k*2^55555, k*2^54321 or palindromic exponents etc. I am certainly not the only one that finds primes with these desirable. But since there is no public documentation about the sieving / checking efforts one could easily waste time on these if they were already checked up to a certain bound. hunson 
20211114, 07:47  #5 
Jun 2009
2^{2}·173 Posts 
Hi hunson, welcome!
I do hold the current triple record. It's true, these numbers aren't really that big compared to twins or single primes records. It's not my goal to discourage you from trying. These records are meant to be broken, after all. Still you need to be aware that finding a constellation of e.g. three primes is much harder than finding a single prime. But what really kept me from going higher is the fact that in a triplet there will always be one prime that can't be verified using LLR but will require a general prime proving software such as PRIMO. At 20.000+ digits you are looking at several months of computing on a powerful machine just to get that proof without knowing for sure it will be possible to prove it. There are limitations to the software that might make it impossible to prove a number prime even when it's smaller than PRIMOs current limit. Never had that happen to me though. You could also try to find numbers of a special form so you can prove them thanks to knowing enough "helper primes" but that will slow down the sieving process. I'll be happy to point you in the right direction if you want to go this way. There's information and software available in this forum. I'll have to look for the right threads if you like. TL;DR: I have considerable computing power and after doing trial runs & calculations I decided to not spend years trying to go higher. But that's just me. If you want to try, more power to you. Just be aware of the amount of work you're taking on. If you find enough people to help, it might be a fun project to run. 
20211114, 12:27  #6  
Feb 2020
Germany
100001_{2} Posts 
Thanks for your input PuzzlePeter.
I do not shoot for a new record prime, I do not have the computing power for this. I am very pleased with a top 10, like with my twinprime which was on the 9th place at the time. This is all just a fun project for me. I am aware that I have to check the +5 form with PRIMO, that's mainly the reason why a ~12700 digit number is my goal. It should be still manageable to check this within a month or so on reasonable new hardware I guess. The idea for a project page with an overview of the checked ranges/exponents might be still useful for someone who wants to find a prime in the top 15 like me!? Quote:
Last fiddled with by hunson on 20211114 at 12:28 

20211115, 08:46  #7  
Just call me Henry
"David"
Sep 2007
Cambridge (GMT/BST)
2×2,969 Posts 
Quote:


20211115, 12:29  #8  
Jun 2009
2^{2}·173 Posts 
Quote:
Yes, I just needed some time to hunt down the thread. It starts around here: https://mersenneforum.org/showthread.php?t=16705&page=8 In short: The idea was to start with a pair of precomputed twins. I'll call them s and s+2. These will be of the form k*b^n+d. I then wanted to use the following formula for the candidates: N=(s+2)*(a*s^23) + {1, +1, +5} with "a" being the running variable. N=a*s^33*s+2*a*s^26 Sorting into terms with and without "a": N=(63*s)+a*(2*s^2+s^3) + {1, +1, +5} N=(s+2)*(a*s^23)1 can be tested using N+1 test and known prime factor s+2 N=(s+2)*(a*s^23)+1 can be tested using N1 test and known prime factor s+2 N=(s+2)*(a*s^23)+5 =a*s^33*s+2*a*s^26 +5 = a*s^33*s+2*a*s^21 can be tested using N+1 test and known prime factor s N=(s+2)*(a*s^23)+7 =a*s^33*s+2*a*s^26+7 = a*s^33*s+2*a*s^2+1 can be tested using N1 test and known prime factor s which would give a quadruplet So for triplets and quadruplets every number can be proven using one of the twin primes as a helper and the software PFGW. But how to do the sieving? I am a really bad programmer, but Robert Gerbicz was so nice and wrote his software "Polysieve" Polysieve uses the form sorted into terms with and without "a" and calls them P(s) and a*Q(s) respectively. So for the example above we get P(s)=62*s and Q(s)=3*s^2+s^3 Polysieve will ask for the degrees of P and Q, the coefficients and the "c" values (1, +1, +5 in our example for a triplet, additionally +7 for quadruplet) and after sieving to your chosen limit will write a candidate file in ABC format that can be tested using PFGW. There will probably be some questions left. Just ask, here or in PM. In case the "Germany" under your name means that German is your native language  so is mine. That might make the PMs a lot easier Last fiddled with by PuzzlePeter on 20211115 at 12:34 

20211115, 17:50  #9  
"Robert Gerbicz"
Oct 2005
Hungary
7^{2}×31 Posts 
Quote:
You would need T(N)=O(log(N)^5/loglog(N)^2) bit operations (including sieve) to find a prp triplet near N=2^n, but for large N the cost of the "slow" primality test for the ugly k*2^n+5 number using elliptic curves (say with Primo) is still faster than T(N). Last fiddled with by R. Gerbicz on 20211115 at 17:55 Reason: renamed variables 

20211115, 19:55  #10  
Jun 2009
2^{2}·173 Posts 
Quote:


20211125, 04:20  #11 
"Matthew Anderson"
Dec 2010
Oregon, USA
5·193 Posts 
Hi again all,
Here are two links that seem to be germane to this discussion. Regarding 3 primes with the pattern (1, 1, 5). https://primes.utm.edu/glossary/xpage/PrimeTriple.html http://www.pie.de/ktuplets.htm#largest3 Also, the first few small ones are in OEIS.org at A007529 Hope that helps. Regards, Matt Last fiddled with by MattcAnderson on 20211125 at 04:26 Reason: add third link 
Thread Tools  
Similar Threads  
Thread  Thread Starter  Forum  Replies  Last Post 
Twin prime search?  MooooMoo  Twin Prime Search  115  20100829 17:38 
Exponential prime search  Citrix  Prime Sierpinski Project  37  20090817 06:32 
Benson's prime search  Kosmaj  Riesel Prime Search  67  20090118 21:59 
Prime Search on PS3?  Kosmaj  Riesel Prime Search  6  20061121 15:19 
Prime Search forum  geoff  Forum Feedback  3  20060726 03:17 