All things doublecheck!!

[axn]

Originally Posted by michaf

I suspect that all known residues will be in by now.

Quite possible. But another week won't cause much harm one way or the other. Just wanted to give everyone a chance to get their work recorded as "first".

Originally Posted by michaf

I suggest we indeed open the pairs without residue for first time testing then (if only it looks weird that we haven't tested some low pairs...)

Agreed. We'll let the Sierpinki's continue to 20k. Probably we can do the same for Riesels also. However I would still like to wait another week for everyone to have one last chance to submit their hidden residues After that, we can go full steam ahead ( behind?)

Originally Posted by michaf

As far as the double check goes, I think we need to get it up to a point where we can say that we have at least confidence that no little primes are left in there (let's say that is all below 20k)

Masser has already done this for Sierpinskis, and I have done this for Riesels. So we can safely say that there are no "small" primes left behind. So the main focus is to get all results recorded systematically.

Originally Posted by michaf

And besides, I don't think running the checks upto 20k will take too much of a time; already on 6k now

It is fun to watch a single day's testing to mess up all the graphs, isn't it?

[michaf]

I mostly agree to your post two above; ( you type too quick....)

I think we DO have a good go at the prp-error rate, be it human or computational... waaay too many primes were missed.

But then, get your DC efforts in LLR-Net, so it shows what has been done

[axn]

Originally Posted by ltd

Error rates are not a problem with these short running tests. Even at PSP where n is>2880000 there is no need for DC.

It is more about tests never done due to errors.

I did not know that there has been an effort by masser to check the low sierpinski numbers eles i would not have released them.

I promise next time i will ask before doing something stupid.

Lars

Meh! I wanted to do that myself. As michaf noted, the "min n untested" column has been bugging me to no end. In fact I would suggest that you go ahead and do the same for Riesels also. After one week, we can release 20k < n < 100k for all k's (both S & R).

You're right about the need for dc - which is why I am hesitant to start a new queue just for dc in the LLRNet server. But it would be nice to have residues for all first time tests.

[michaf]

Originally Posted by axn1

It is fun to watch a single day's testing to mess up all the graphs, isn't it?

Yeah!

pity it will be off the charts again in 30 days...

[axn]

Originally Posted by michaf

But then, get your DC efforts in LLR-Net, so it shows what has been done

If you look at the "DC min untested" column, you'll see that the dc effort is indeed being recorded there. Once I have sweeped the Riesels to 55k, I'll get the residues updated in there. However, these are currently not considered as first time tests -- hence the "min n untested" column will stay where they are.

I know, I know, it is all backwards But at least, I hope you understand what is going on, right?


PS:- Yeah, sometimes I do type too fast

[ltd]

Originally Posted by axn1

Meh! I wanted to do that myself. As michaf noted, the "min n untested" column has been bugging me to no end. In fact I would suggest that you go ahead and do the same for Riesels also. After one week, we can release 20k < n < 100k for all k's (both S & R).

You're right about the need for dc - which is why I am hesitant to start a new queue just for dc in the LLRNet server. But it would be nice to have residues for all first time tests.

I will release the low n for riesel now.

Lars

[michaf]

Originally Posted by axn1

You're right about the need for dc - which is why I am hesitant to start a new queue just for dc in the LLRNet server. But it would be nice to have residues for all first time tests.

What we can do too, once we have all the first time residues, is to set up an occasional double-check into llr-net, every 100 numbers there is 1 dc or summat like that. At least we then do have the opportunity to check the error-rate, and it will not get a lot of load on us either (just get geoff to make his sieve run 1% faster again, and we will shave the time off that way )

[michaf]

Can LLR-net be set up like this:

for each residue above 100k, release one from the lower part?

that way we still move up, and we get all the residues eventually as a bonus.

[masser]

Hi all,

I'm out of town until Tuesday, so I won't be able to send the residues until then. (Expect them Tuesday evening, axn1) All of the Sierpinski's with n < 37 k have been doublechecked by me. In another 2 weeks I'll complete the double check to n = 65 k.

I have no problem with any of the double check candidates being placed in the LLRNET queue. If it turns out that we're triple-checking some candidates, what harm can be done? The tests are reasonably short on faster machines and it's always possible that I made some "human error" somewhere and we missed a prime again. (unlikely, but possible)

At any rate, the Sierpinski double check candidates with n > 65k should certainly be entered into the queue sometime in the near future.

masser

[konrad127123]

I won't be home until the 15th and so can't send in my residues untill then.

[ltd]

So far we have done 43295 double checks and all result were identical.

Is it possible that we split up the double check things into another thread?

Lars