good news from Payam Samidoost

[jjjjL]

I just got back into town last night and recieved some good news in the form of an email from Payam Samidoost. Anyone familar with the main Sierpinski problem homepage (http://www.prothsearch.net/sierp.html) will recognize him as the current searcher of k=4847.

Well, in short, he has offered his k value to our project and I am glad to have it. Hopefully he will also join us with his awesome computing power that was able to get him to over n=2million! :eek:

We will be continuing the search on k=4847 along with k=33661 once k=27653 is up to 3 million. Hopefully other search participants will follow Payam's example and allow us to search their k values so that the lower bounds can remain more uniform and primes can be found faster.

Thanks again to Payam Samidoost. This is more positive news for the project. :)

-Louie

[FoBoT]

good deal! :)
i won't pretend to understand much of what you said, i am but a simple moderate stats whore :o

[pointwood]

I have no idea either, but it sound good to me :thumbs: :D

[smh]

Well, i do understand what you mean :)


Originally posted by jjjjL
Well, in short, he has offered his k value to our project and I am glad to have it. Hopefully he will also join us with his awesome computing power that was able to get him to over n=2million! :eek:

A big part of the range was done by Joseph Arthur (who now is working on another number). He seems to have a lot of resources


Originally posted by jjjjL
We will be continuing the search on k=4847 along with k=33661 once k=27653 is up to 3 million. Hopefully other search participants will follow Payam's example and allow us to search their k values so that the lower bounds can remain more uniform and primes can be found faster.

How will you do this? First 33661 upto the level of 4847 and then together with lowest N first?

K=4847 has still one of the highest N's tested. Most K's are not even tested upto 1M, so does it make sence to bring 4847 to higher levels ASAP (a reason could be that it's the lowest of the 17 remaining numbers)

I strongly believe that one more number can be eliminated with an N below 1 million, hope it's done by this project. It could give us another boost. (what will happen with the name of this project? :confused: )

[CrazySaint]

We'll just call it Sixteen Or Bust ;)

[MAD-ness]

It will just be a reminder of how much we have gotten done, the name itself reminds us of where we started. ;)

[nuutti]

I think that we should redo every number from n=1.
(Double checking + we really can't) trust anyone’s results.
+ It does not take very long (like less than week) with our
current resources.

Yours,

Nuutti

[Mystwalker]

If we can't trust other people's results, how should anyone trust ours? Apart from found primes, it would be wasted computing power.
Plus, checking those n's once again will indeed take weeks, as it's the n that makes the number big, not the k. I don't know how the algorithm works, but I guess it would take long to get to n=2M.

Once we can be sure the algorithms is implemented correctly, I think we can trust those results, too...

[smh]

Originally posted by nuutti
I think that we should redo every number from n=1.
(Double checking + we really can't) trust anyone’s results.
+ It does not take very long (like less than week) with our
current resources.


There is something to say for it. A fast P4 can do the first 100 or 200K alone. Even above 200K the numbers still take less then an hour to test (IIRC), and there aren't many numbers to check since the weight of most numbers is fairly low. (ie. less candidates survive sieving)

But even if we redo part of a range we still have to compare residues. And The part which we do first time has to be double checked too.

IIRC the error rate on GIMPS is about 1,8%. Current numbers under test with SoB are much lower then the numbers GIMPS is testing so the error rate will probably be a bit lower, but still.

There could also be a bug in the code, so to be absolutely sure everything should be double checked, like with GIMPS. Only they have 1 range, the Sierpinski problem has currentely 17 ranges.

[nuutti]

>If we can't trust other people's results, how should anyone
>trust ours?

In some cases we does not even have logs from previous runs.
May be someone has made typo when he has reported
that he has cheked up to 654,000 when he meaned 645,000.
Then there is one very small prime betweed these two numbers
and we search 10 years that number before we will found next prime.
Not good idea. Worst case is too bad.
In some cases we have logs and when we redo when can
compare residues, We have to doublecheck every number anyway. (When test shows that number is not prime we have residue and that residue is same every time when we test that numbers, some someone after us can check numbers again ja compare our residues to his residues.)


This project is not matter of weeks. This can last a decade.

Yours,

Nuutti

[Firebirth]

Since this is relatively small numbers, the sequences have probably been tested a couple of times before, by people before, eager like us, to solve the conjecture. Therefore, where it really begins to kick ass (and we have a chance of working onsomething new), is (my wild guess) when we reach the point where a single number takes days!

- Further if we don't believe the people haven't lied about their numbers - there is really some CPU power out there, and I think we should pay them some respect.

[smh]

- Further if we don't believe the people haven't lied about their numbers - there is really some CPU power out there, and I think we should pay them some respect.

I think we can trust the people who tested the numbers. I think they all found some big primes (in the top 5000 list) and some of them are also working on the Riesel project (same as Sierpinski but for numbers of the form K*2^N-1) and eliminated some candidates there (this project is even harder since there are more numbers left over below the lowest known Riesel number)

The problem is more if we can trust their computers. Some are over clocked pretty far, some keep on crashing in while running a test, others run buggy programs etc,etc...

Because of the Double check for Mersenne numbers we know that (IIRC) 1,8% of the tests is wrong

[nuutti]

>I think we can trust the people who tested the numbers. I think
>they all found some big primes (in the top 5000 list) and some
>of them are also working on the Riesel project (same as
>Sierpinski but for numbers of the form K*2^N-1) and eliminated
>some candidates there (this project is even harder since there
>are more numbers left over below the lowest known Riesel
>number)


Yes, Most of them are well known and respected prime serachers. But some of them have just visited on the www.prothsearch.net
and participated to project and submitted some results and after that quitted + you don't need to submit any logs. You just tell that
you have searched up to some number. (I know this because I have participated to Riesel search and even eliminated one number : 443857*2^369457-1 ).
I have been coordinating one project and sometimes people mail a log to me and say. Here is checked range. When I have checked the log I have noticed that he has checked it only partially. When I ask explanation they can answer something like this : "I used many computer and when I splitted the task to those computers I made an error".

So peoples make errors all the time and that is normal.
The Project shoud take that on account.

Yours,
Nuutti Kuosa