Bug in LLR with a small exponent (< 34000)
 

I was doing some LLR runs on some smaller exponents recently, to see if I wanted to use my prime generator formula in the hunt for a top 5000 prime, and it has been failing on this one exponent. My errors occur on several different computers, so that pretty much eliminates hardware as the cause. My errors are:

[CODE]
Iter: 1/109814, ERROR: ROUND OFF (0.5) > 0.4
Continuing from last save file.
Unrecoverable error, Restarting with next larger FFT length...
Iter: 1/109814, ERROR: ROUND OFF (0.4999961853) > 0.4
Continuing from last save file.
Unrecoverable error, Restarting with next larger FFT length...
Iter: 1/109814, ERROR: ROUND OFF (0.4999818802) > 0.4
Continuing from last save file.
Unrecoverable error, Restarting with next larger FFT length...
Iter: 1/109814, ERROR: ROUND OFF (0.4999983311) > 0.4
Continuing from last save file.
Unrecoverable error, Restarting with next larger FFT length...
Iter: 1/109814, ERROR: ROUND OFF (0.4999961853) > 0.4
Continuing from last save file.
Unrecoverable error, Restarting with next larger FFT length...
Fatal error at setup : Number sent to gwsetup is too large for the FFTs to handle.
[/CODE]The exponent it was testing was only 33039.

Is there someway to contact the author?

Try posting in [URL="http://www.mersenneforum.org/showthread.php?t=15592"]this thread[/URL] over in the software forum. I don't know how responsive the author will be, but you're more likely to get a response there.

[QUOTE=Dubslow;276293]Try posting in [URL="http://www.mersenneforum.org/showthread.php?t=15592"]this thread[/URL] over in the software forum. I don't know how responsive the author will be, but you're more likely to get a response there.[/QUOTE]

Thanks, both George and Jean have emailed me about it. I am sending them more data.

Can you kindly post the exact number(s) so that we can try ourselves, using the current and older versions of LLR.

Thanks!

[QUOTE=Kosmaj;276899]Can you kindly post the exact number(s) so that we can try ourselves, using the current and older versions of LLR.

Thanks![/QUOTE]

Sure, paste this into a text file, save it, then run LLR on it. BTW, this is a base-10 project I was just messing around with. LLR will find all of the primes except for the last one (which is also prime). You will get roundoff errors for the last one.

[CODE]
24637167299:P:1:10:257
18446744073709551253 18
18446744073709551253 45
18446744073709551253 213
18446744073709551253 306
18446744073709551253 365
18446744073709551253 420
18446744073709551253 509
18446744073709551253 1021
18446744073709551253 1314
18446744073709551253 1459
18446744073709551253 2705
18446744073709551253 10203
18446744073709551253 10578
18446744073709551253 12941
18446744073709551253 16701
18446744073709551253 21073
18446744073709551253 30799
18446744073709551253 33039
[/CODE]

And, just wondering, is there a program where you can use a k > 2^64 for a Riesel Prime test? I can't seem to get any of my large k's to work with the ones I have.

I also can't seem to find a siever that works for k > 2^64 also.

[QUOTE=SaneMur;277205]And, just wondering, is there a program where you can use a k > 2^64 for a Riesel Prime test? I can't seem to get any of my large k's to work with the ones I have.

I also can't seem to find a siever that works for k > 2^64 also.[/QUOTE]

pfgw

Nothing that I know off of the top of my head.

[QUOTE=rogue;277213]pfgw[/QUOTE]

I am not so convinced pfgw can accept a k > 2^64. A simple test to demonstrate:

[CODE]9999999:M:1:2:258
18446744073709551617 2[/CODE]

...will generate the output:

1*2^2-1 is prime!

(It thinks 18446744073709551617 = 1 since 64 bits were overflowed.)

Calling
[code]
pfgw -q"18446744073709551617*2^2-1"
[/code]

will produce the correct result, the same as LLR V3.8.6 (RES64 1E1DF1A8370FB1D5).

There seems only a reading/converting-issue when calling "pfgw <filename>" with a newpgen format file.

Create a ABCD-format file like
[code]
ABC $a*2^$b-1
18446744073709551617 2
[/code]

and pfgw will do this one correct.


NewPGen can handle bigger k-values by using the Primorial Mode (k not prime):

So instead writing k as "2419275534239148308295" input the factored value "1043827764761*3*5*11*13*19*29*37*53".

See [url=http://www.mersenneforum.org/showthread.php?t=9755]here[/url] for some big values.

[QUOTE=kar_bon;277232]Calling
There seems only a reading/converting-issue when calling "pfgw <filename>" with a newpgen format file.
[/QUOTE]

Yes, but that is a BIG problem!

I was able to use NewpGen to create a sieved file using a factored form, thanks for the tip! Such a shame that I can't get any prime testing program to read it though. How are other people doing it?? I am perplexed!

I followed your link and would like to "join the hunt" for efficient k values for Riesel Prime generation. I am trying to get in touch with Robert, if you know how, pass the word.

That project is too cool to resist!

[QUOTE=kar_bon;277232]Calling
[code]
pfgw -q"18446744073709551617*2^2-1"
[/code]

will produce the correct result, the same as LLR V3.8.6 (RES64 1E1DF1A8370FB1D5).

There seems only a reading/converting-issue when calling "pfgw <filename>" with a newpgen format file.[/QUOTE]

Interesting. I'll look into it.