CRUS sieving questions

[storm5510]

rogue suggested I come here because I like to sieve. His srsieve2cl and the RTX 2080 I have do quite well. I have read most of the pages in this topic. There is a lot more to read in the primary topic.

[gd_barnes]

Originally Posted by storm5510

rogue suggested I come here because I like to sieve. His srsieve2cl and the RTX 2080 I have do quite well. I have read most of the pages in this topic. There is a lot more to read in the primary topic.

Great! Feel free to contribute however you want. Mark (Rogue) is the residential sieving expert on many different projects and writes/maintains many sieving programs. He will be able to answer whatever questions you have regarding sieving.

[storm5510]

Originally Posted by gd_barnes

Great! Feel free to contribute however you want. Mark (Rogue) is the residential sieving expert on many different projects and writes/maintains many sieving programs. He will be able to answer whatever questions you have regarding sieving.

Thank you for the reply! Is there a table for something similar where a person can pick what they wish to run?

[rogue]

Originally Posted by storm5510

Thank you for the reply! Is there a table for something similar where a person can pick what they wish to run?

Start here: http://www.noprimeleftbehind.net/cru...s-unproven.htm

Look for a conjecture that isn't reserved. Then go to
http://www.noprimeleftbehind.net/cru...e-reserves.htm
or
http://www.noprimeleftbehind.net/cru...e-reserves.htm

to see if there is a sieve file for the an unreserved range.

You will then need to look in this thread to see if anything is reserved for sieving. To save you time, look at conjectures with fewer than 1000 k remaining. I have or am sieving most of the ones with more than that number of k. Sieve from n to 2n. You will need to run a test with llr to determine target removal rate. I typically run a test for k*b^(1.7n)+/-1.

[storm5510]

Originally Posted by rogue

Start here: http://www.noprimeleftbehind.net/cru...s-unproven.htm

Look for a conjecture that isn't reserved. Then go to
http://www.noprimeleftbehind.net/cru...e-reserves.htm
.
You will then need to look in this thread to see if anything is reserved for sieving. To save you time, look at conjectures with fewer than 1000 k remaining. I have or am sieving most of the ones with more than that number of k. Sieve from n to 2n. You will need to run a test with llr to determine target removal rate. I typically run a test for k*b^(1.7n)+/-1.

A removal rate! This would depend on a "nash" value, I would think. LLR, I would much rather avoid. I have my own set of nash tables I created.

Below is a snip from the first link you provided. How do these column headers relate to k*2^n-1.

The second link appears to be finished sieves when any zip file on the right is selected.

[gd_barnes]

Originally Posted by storm5510

A removal rate! This would depend on a "nash" value, I would think. LLR, I would much rather avoid. I have my own set of nash tables I created.

Below is a snip from the first link you provided. How do these column headers relate to k*2^n-1.

The second link appears to be finished sieves when any zip file on the right is selected.

Base 2 is only a very small subset of what we do here. We test all bases <=1030 and both sides. Not just base 2. That is k*b^n-1 and k*b^n+1. Base 2 is mostly already sieved by either PrimeGrid or Yoyo. Yoyo does a lot of our sieving here on bases with < ~30 k's remaining.

Because of this, I'd suggest picking something with 30 to 100 k's remaining (that is not base 2) to get your feet wet. I'll recommend Riesel base 60 or R60 as we call it here. It has 33 k's remaining. See the k's remaining here: http://www.noprimeleftbehind.net/cru...60-reserve.htm

Riesel is the -1 side and Sierpinski is the +1 side.

So you would sieve 36*60^n-1, 1770*60^n-1, 4708*60^n-1, etc. with all of the 33 k-values shown on that page.

You can see on the page that R60 has already been tested to n=100K and it does not already have a sieve file. I would suggest sieving n=100K-250K or you could sieve n=100K-500K but the sieving job would be much longer.

To determine an optimum sieving rate, you'd need to be able to run a partial LLR test to estimate how long that test would take. Nash weight is meaningless when sieving multiple k's together. Software changes quite often these days. New and faster sieves are created, better versions of LLR are created, etc. Each time that happens, the optimum sieve depth can change. That's why the speed of an LLR test must be compared to the speed of your sieve to determine the optimum rate.

As per your final comment, if you see a sieve file on a base, it means it's already been sieved so of course you should choose something else.

[storm5510]

Originally Posted by gd_barnes

Base 2 is only a very small subset of what we do here. We test all bases <=1030 and both sides. Not just base 2. That is k*b^n-1 and k*b^n+1. Base 2 is mostly already sieved by either PrimeGrid or Yoyo. Yoyo does a lot of our sieving here on bases with < ~30 k's remaining.

Because of this, I'd suggest picking something with 30 to 100 k's remaining (that is not base 2) to get your feet wet. I'll recommend Riesel base 60 or R60 as we call it here. It has 33 k's remaining. See the k's remaining here: http://www.noprimeleftbehind.net/cru...60-reserve.htm

Riesel is the -1 side and Sierpinski is the +1 side.

So you would sieve 36*60^n-1, 1770*60^n-1, 4708*60^n-1, etc. with all of the 33 k-values shown on that page.

You can see on the page that R60 has already been tested to n=100K and it does not already have a sieve file. I would suggest sieving n=100K-250K or you could sieve n=100K-500K but the sieving job would be much longer.

To determine an optimum sieving rate, you'd need to be able to run a partial LLR test to estimate how long that test would take. Nash weight is meaningless when sieving multiple k's together. Software changes quite often these days. New and faster sieves are created, better versions of LLR are created, etc. Each time that happens, the optimum sieve depth can change. That's why the speed of an LLR test must be compared to the speed of your sieve to determine the optimum rate.

As per your final comment, if you see a sieve file on a base, it means it's already been sieved so of course you should choose something else.

How far do the n's need to go in the base 60 link above. What I see is 100K. I could sieve all of these fairly fast with a relatively large P. How would I reserve this group?

[gd_barnes]

Originally Posted by storm5510

How far do the n's need to go in the base 60 link above. What I see is 100K. I could sieve all of these fairly fast with a relatively large P. How would I reserve this group?

Sieving n=100K-250K is a good range.

Just checking: You will be sieving all of the k's at one time. Is that correct?

There are no formal reservations for sieving. Just state here in this thread that you will be sieving R60 for n=100K-250K.

[storm5510]

Originally Posted by gd_barnes

Sieving n=100K-250K is a good range.

Just checking: You will be sieving all of the k's at one time. Is that correct?

There are no formal reservations for sieving. Just state here in this thread that you will be sieving R60 for n=100K-250K.

Sieving R60 n=100K-250K. That was easy.

I started out using a P value is 100e9. I will probably increase this to get the final term count down more. Of course, raising this increases the time required. The k=36 sieve I ran last night took 40 minutes and ended with 6,091 terms. I am running it again with P at 250e9. I would rather they take longer. Not so much machine interaction. I will leave these in .abcd format. It is the default and far more compact than .npg. I won't post the sieves until all are complete.

Do you have a recommendation for a final P value?

[rogue]

Originally Posted by storm5510

Sieving R60 n=100K-250K. That was easy.

I started out using a P value is 100e9. I will probably increase this to get the final term count down more. Of course, raising this increases the time required. The k=36 sieve I ran last night took 40 minutes and ended with 6,091 terms. I am running it again with P at 250e9. I would rather they take longer. Not so much machine interaction. I will leave these in .abcd format. It is the default and far more compact than .npg. I won't post the sieves until all are complete.

Do you have a recommendation for a final P value?

I do not, but I would not be surprised if it were higher than 1e12. I use 1e12 for conjectures with thousands of k for n to 25000. With such a high n, I would expect 1e13 or higher.

[storm5510]

Originally Posted by rogue

I do not, but I would not be surprised if it were higher than 1e12. I use 1e12 for conjectures with thousands of k for n to 25000. With such a high n, I would expect 1e13 or higher.

I suspected as much. The series I am running jumps from 36 to 1770 and upward for the k's. There is no cover-all P for this. I think I will just set it at 1e12 for now and see how it goes.

Note: I think you meant 250,000 above.