Does NewPGen have a bug?

[chunsonnn]

I was sieving k=173 from n=1M to n=2M, and I found this message by NewPGen:

p=2215115304221 divides n=1798220.

However 2,215,115,304,221 is composite (that number is 8627 * 256765423), so that message should not have popped up. Why didn't NewPGen remove n=1798220 when it was at p=8627 or at p=256765423?

In case anyone's wondering, I was using NewPGen version 2.82 on a Pentium 4 computer. The problem also appears on another Pentium 4 machine, so I'm pretty sure it's not a hardware problem.

Thank
Cristiano

[axn]

Originally Posted by chunsonnn

I was sieving k=173 from n=1M to n=2M, and I found this message by NewPGen:

p=2215115304221 divides n=1798220.

However 2,215,115,304,221 is composite (that number is 8627 * 256765423), so that message should not have popped up. Why didn't NewPGen remove n=1798220 when it was at p=8627 or at p=256765423?

In case anyone's wondering, I was using NewPGen version 2.82 on a Pentium 4 computer. The problem also appears on another Pentium 4 machine, so I'm pretty sure it's not a hardware problem.

Thank
Cristiano

p=2215115304211 divides 173*2^1798220-1

[paulunderwood]

Isn't k=173 already sieved very deeply by PrimeGrid?

[ATH]

Neither 8627 nor 256765423 is a factor of 173*2^1798220-1, so it must have been p=2215115304211 as axn pointed out.

[pinhodecarlos]

Originally Posted by paulunderwood

Isn't k=173 already sieved very deeply by PrimeGrid?

Yes and tested up to n=2.04M by RPS.

[rogue]

Why are you using newpgen for this form? Use srsieve/sr1sieve. They are much, much faster than newpgen for this form.

[storm5510]

Does NewPGen have a bug? Possibly. I noticed for any given range of n's, NewPGen will remove more n's, and at lower values of p, than the srXsieve family.

Either NewPGen is removing n's it should not or srXsieve family is not removing n's it should. A combination of both is also possible. I have no way to determine which case it may be.

[VBCurtis]

Originally Posted by storm5510

Does NewPGen have a bug? Possibly. I noticed for any given range of n's, NewPGen will remove more n's, and at lower values of p, than the srXsieve family.

Either NewPGen is removing n's it should not or srXsieve family is not removing n's it should. A combination of both is also possible. I have no way to determine which case it may be.

Please elaborate, with example files. The OP's post is surely a typo, while yours is a serious allegation.

[storm5510]

Originally Posted by VBCurtis

Please elaborate, with example files. The OP's post is surely a typo, while yours is a serious allegation.

"Serious allegation." I do not consider it as being such. I am not condemning either. Just a simple comparison.

I can provide example files. It may take a couple of days. I am running a wide-range sieve on my HP to 1.5-trillion. I can replicate this with NewPGen on my laptop. My i7 is running a double-instance of LLR so I cannot use it.


Edit: I will allow the srXsieve on the HP to finish, and then run the NewPGen test after it. sr1sieve estimates completion on March 17 at around 19:00 UTC.

[pepi37]

Originally Posted by storm5510

"Serious allegation." I do not consider it as being such. I am not condemning either. Just a simple comparison.

I can provide example files. It may take a couple of days. I am running a wide-range sieve on my HP to 1.5-trillion. I can replicate this with NewPGen on my laptop. My i7 is running a double-instance of LLR so I cannot use it.


Edit: I will allow the srXsieve on the HP to finish, and then run the NewPGen test after it. sr1sieve estimates completion on March 17 at around 19:00 UTC.

Why is needed to wait so many days: take sequence you know produce "problems" runn it on NewPgen few minutes, and run it under srxsieve few minutes ( to reach same depth)
So all work can be done in 10 minutes.
Or tell us what sequence have problem and I would like to comparison for you,
For that you can only need seconds to write reply on my post

[storm5510]

I scaled this test down dramatically. The parameters are:

Series: 7*2^n-1
n: 350,000
N: 650,000
p: (default)
P: 50e9

It took srsieve/sr1sieve about 30 minutes to run this. NewPGen, over four hours. NewPGen does not start storing factors until it p passes 2^32. So, I set the ceiling for srsieve to this value. The staring p for sr1sieve was the same.

During the sieve NewPGen pulled out 5,603 factors, sr1sieve pulled out 840.
Yet, the compete files for LLR were exactly the same. 7,092 elements each. Some may be familiar with fc, short for file compare. I compared both LLR files. The only difference was in the header line at the very top.

All the files are stored in the attached zip file. There is also an image attachment with NewPGen set into a PowerShell window where I ran srsieve and sr1sieve.

Q: How is it possible for NewPGen to flag so many more factors than sr1sieve and both results files be virtually identical?