Bases 33-100 reservations/statuses/primes

[gd_barnes]

Originally Posted by Mathew Steine

I have finished the test
primes
7698*42^26751-1
12039*42^27032-1
7614*42^27539-1
7614*42^29019-1 (double)

k's to remove: 3
extra primes: 1
other: when sieving sr2sieve stated that 2304 is algebraic not really sure what this means.

Attached are the results. Hopefully I did not miss anything.

Everything looks good. Thanks for posting it all.

One hint for future reference that you may have already inferred from our PM exchange: For the initial testing, you don't need to use the -t or -tp switch. Then when you have all of the PRP's, you can run a separate test using -t for Sierp or -tp for Riesel to prove just those (3 or 4 in this case). That should save you some (or a lot) of CPU time and reduce the size of your results file substantially.

Get with me again the next time you start on one. There's a command that you can put at the beginning of a sieve file to tell PFGW to stop searching a k-value once a prime is found for it...additional CPU time savings.

As for k=2304 having algebraic factors as shown by sr2sieve, don't worry about that. It has to do with a k-value that is a perfect square containing algebraic factors for all even n-values and some of those n-values are still remaining in the sieve file. That is: 2304*42^(2q)-1 = (48*42^q-1)*(48*42^q+1). It means that you could manually remove all even n-values in your sieve file for k=2304 (to save even more CPU time). In other words for k=2304, a prime must have an odd n-value.


Gary

[rogue]

Originally Posted by gd_barnes

I too have never experenced PFGW saying a true prime was composite. Perhaps the new GWNUM libraries did something. How were you able to isolate it to that? If so, I wonder if a similar bug exists in LLR or Prime95?

The problem seems to be in gwnum. I am working with George to isolate the issue. PFGW 1.x (with gwnum 23.8) works fine without -a1. I can't speak for LLR or Prime95. Although they cannot do primality tests for this base, it is possible that this bug might manifest itself in those programs.

[rogue]

Originally Posted by rogue

The problem seems to be in gwnum. I am working with George to isolate the issue. PFGW 1.x (with gwnum 23.8) works fine without -a1. I can't speak for LLR or Prime95. Although they cannot do primality tests for this base, it is possible that this bug might manifest itself in those programs.

I have verified that the problem was introduced in gwnum 25.12. I can change PFGW to address the problem, but instead of getting a new release I will wait until George has more information. I do not know if he will fix this in gwnum 25.x as he is working on gwnum 26.x.

[Mathew]

I forgot to mention that I would like to continue with Riesel base 42. I would like to reserve all k's until n=50k.

[gd_barnes]

Mark,

After a double check primes run to n=2.5K combined with all primes for n=2.5K-40K, Riesel base 58 checks out at 330 k's remaining at n=40K. Good work.


Gary

[rogue]

Originally Posted by rogue

I have verified that the problem was introduced in gwnum 25.12. I can change PFGW to address the problem, but instead of getting a new release I will wait until George has more information. I do not know if he will fix this in gwnum 25.x as he is working on gwnum 26.x.

Although the issue is in gwnum 25.x, George is too busy with higher priority items. I have discovered that there is a switch I can use to tell gwnum to use an irrational FFT instead of a rational FFT. The irrational FFT is safer, but probably slightly slower. I haven't verified that. I will be releasing 3.2.4 today or tomorrow. Note that the number in question is the only one out of tens of thousands, if not hundreds of thousands or millions of numbers that has a problem.

Gary, eventually I will modify PFGW to use George's 26.x gwnum library. At that time, it would be prudent to take all of the small primes (n < 10000) for a number of bases and re-test them both for PRP and primality as a means of testing gwnum 26.x. I suspect that small FFTs pose more problems than large ones.

[MyDogBuster]

Riesel Base 39

Tested from n=10K-15K. 818 primes found (see attached file)

4514k's remaining + the 1 I am not testing = 4515 left

Releasing base

[gd_barnes]

Originally Posted by rogue

Gary, eventually I will modify PFGW to use George's 26.x gwnum library. At that time, it would be prudent to take all of the small primes (n < 10000) for a number of bases and re-test them both for PRP and primality as a means of testing gwnum 26.x. I suspect that small FFTs pose more problems than large ones.

That sounds reasonable. It won't take very long for most of the bases and we could just choose a few with medium to high conjectures and for smaller and higher bases to get a good cross section of them.

[MyDogBuster]

Riesel Base 72 (last k) tested to n=200K.

Continuing

[MyDogBuster]

Reserving Sierp Base 97 as a new base to n=10K

[gd_barnes]

Guys, I appreciate your enthusiasm greatly. That said, in the thread where I created the new script, I was kindly asking that people only do one new base per person per day and to reserve them ahead of time. Can we please stick with that? I know it's fun to do new bases but we have an awful lot of work available on existing bases and it does take quite a bit of my time to update the pages for so many new ones.


Thank you,
Gary