Accuracy of our work [k*2^n-1, k<300] - Page 4

Harvey563 · 2004-09-15, 16:14

I checked the reisel3.txt testfile with PRP3 up to n=~107012, & found the following false composites:

29302167*2^3646-1 is not prime. RES64: 8CF703F5D3C138F4. OLD64: A6E50BE17B43AAD9
28159983*2^3654-1 is not prime. RES64: C596551BEC2F8ED8. OLD64: 50C2FF53C48EAC86
15489885*2^8907-1 is not prime. RES64: 0AD1A34D20BF12F7. OLD64: 2074E9E7623D38E3
8324175*2^20024-1 is not prime. RES64: E5A012BF5900888D. OLD64: B0E0383E0B0199A6
9219177*2^20202-1 is not prime. RES64: BC80FAD46DA2F6A6. OLD64: 3582F07D48E8E3F0
12083025*2^24000-1 is not prime. RES64: 4935C1D06DD4A142. OLD64: DBA14571497DE3C4
10056945*2^24009-1 is not prime. RES64: 0F0C7257B224343D. OLD64: 2D255707166C9CB5
11521341*2^24010-1 is not prime. RES64: D5E6829B11201A1D. OLD64: 81B387D133604E54
11431839*2^24012-1 is not prime. RES64: FA4ACF6784EEA69B. OLD64: EEE06E368ECBF3CE
9921849*2^24016-1 is not prime. RES64: 1B102419C47CFA1D. OLD64: 51306C4D4D76EE54
9299241*2^24018-1 is not prime. RES64: F1A346AC2B55ECC9. OLD64: D4E9D4048201C658
10384821*2^24018-1 is not prime. RES64: 4CF30136596E7194. OLD64: E6D903A30C4B54B9
9956901*2^24022-1 is not prime. RES64: 7B999A3E2B4A0533. OLD64: 72CCCEBA81DE0F97
12932955*2^24022-1 is not prime. RES64: FFF504C8D078CB29. OLD64: FFDF0E5A716A6178
10761555*2^24023-1 is not prime. RES64: 0EEC3A40B91408AD. OLD64: 2CC4AEC22B3C1A05
11900691*2^24030-1 is not prime. RES64: 3DACBB48F027B37E. OLD64: B90631DAD0771A78
11664915*2^24040-1 is not prime. RES64: F9D8D495E348FD14. OLD64: ED8A7DC1A9DAF739
10075443*2^24043-1 is not prime. RES64: 5006E3C294EDD739. OLD64: F014AB47BEC985AA
10413675*2^24044-1 is not prime. RES64: DB0AC47F852C174E. OLD64: 91204D7E8F8445E7
11251275*2^24044-1 is not prime. RES64: A52B8FAB0FFA71D9. OLD64: EF82AF012FEF5589
11028303*2^24047-1 is not prime. RES64: CB91D616915A1690. OLD64: 62B58243B40E43AF
12408555*2^24047-1 is not prime. RES64: BD99065F75FCA7F6. OLD64: 38CB131E61F5F7E1
12463209*2^24047-1 is not prime. RES64: 7C8646A43CDBD48F. OLD64: 7592D3ECB6937DAC
12739695*2^24047-1 is not prime. RES64: 909AB7FE3F58E46E. OLD64: B1D027FABE0AAD48
12789399*2^24048-1 is not prime. RES64: B9CDD41F5E79A81B. OLD64: 2D697C5E1B6CF84E
12516447*2^24049-1 is not prime. RES64: E8CD4FB7072BE0A5. OLD64: BA67EF251583A1EE
8311875*2^35154-1 is not prime. RES64: D7A3466DE530BE24. OLD64: 86E9D349AF923A6B
8261077*2^38957-1 is not prime. RES64: F727A9BDF9B1DEE1. OLD64: E576FD39ED159CA1
10181715*2^39010-1 is not prime. RES64: 513D1C54DC36D4A2. OLD64: F3B754FE94A47DE4
12125391*2^39010-1 is not prime. RES64: C9D01CE995DDCB77. OLD64: 5D7056BCC1996263
8120565*2^45200-1 is not prime. RES64: 192E0C444765188B. OLD64: 4B8A24CCD62F49A0
8213205*2^45200-1 is not prime. RES64: BB8B9FB1AAEDE888. OLD64: 32A2DF1500C9B997
8311875*2^48997-1 is not prime. RES64: DB7D1B60A3B28640. OLD64: 92775221EB1792BD
6595767*2^50001-1 is not prime. RES64: 98F1CDB57D49A33A. OLD64: CAD5692077DCE9AC
7866207*2^50005-1 is not prime. RES64: 97746053C4E90D0E. OLD64: C65D20FB4EBB2729
8261079*2^57041-1 is not prime. RES64: 3820F722047B06E6. OLD64: A862E5660D7114B1
6456993*2^60060-1 is not prime. RES64: F561E4099F33AB76. OLD64: E025AC1CDD9B025F

These all appear to be cases with large k values, & no failures were found with n greater than 60060.

Harvey563

jocelynl · 2004-09-15, 18:13

Quote:

Originally Posted by Kosmaj

TTn
When I say "first version" I refer to the one of July 19, 2003.

That version is not on www.mersenne.org/gimps

Does anybody have that version ?
If so, please send it to George
or attach it on this forum.

Joss

Jean Penné · 2004-09-15, 20:27

George Woltman warned me Yesterday of the problems you encountered while using different versions of LLR. The most serious, indeed, is the LLRP4 problem.
Thanks to your "123missed.txt" file, I could make relevant tests, and the bugs
seem to be fixed (all the 123 numbers are now found prime by the corrected
program). Nevertheless, I wish to test this updated version on all the k*2^n-1
number from Chris Caldwell database before releasing it... So, I strongly suggest to all users to avoid using this program for the time beeing, the risk of missing prime numbers is too big. I am sorry for the drawback, but you can still use the previous versions (the very first IBDWTone of March 2004 excepted).

In fact, I found two distinct problems in LLRP4 :
The false negatives found with k values smaller than 2**22 are due to a pure LLR bug, and I corrected the LLR.c code accordingly.
The other false negatives found in the 123missed.txt file are the same as those causing problems to the PRP3 program, and I avoided this problem by forcing the "general mode" initialization for k values larger than 22 bits, which is only a provisory work around.
I hope this problem will not make too much double tests necessary...

Jean

Kosmaj · 2004-09-15, 22:08

Jean
Thank you for your help. I hope the problems will be resolved soon. If you need any help in testing your new version I can help and I'm sure Harvey and the others who have time and resources are willing to help too. I think it will be better if the tests can be done indepenedenlty by several testers.

BTW, can you tell us, are small k's, in particular k<100 affected by these bugs in LLR-P4?

As for other versions of LLR one false residue was found when testing 97*2^157221-1 using LLR of April 11 and LLR-P2. You can find more details here.

Finally, I have the first LLR of July 2003. I tried to attach it zipped but the file is too large. Will try to find another way.

Kosmaj · 2004-09-15, 22:30

I put two old versions of LLR on the 15k.org server.

LLR of July 2003, the first version of LLR, the most stable one, no missing primes found. Doesn't work for small n < 300. Used by prof Caldwell to verify k*2^n-1 primes reported to Top-5000.

LLR of April 11, 2004 The first stable IBDWT version. One false residue found when running on Pentium-4. No problems found on Athlon. Inefficient on both platforms when initial fft length guesses are wrong.

2004-09-16, 02:51

Kosmaj, it seems the error records are gone?
I am not going to post the problemed k again.
The error(missing primes) results were triggered by Thomas and I working together, while weighing small k in a primitive way. With the FIRST LLR using P4's, not AMD!
This was a particular problem with only certain k, I think.

2004-09-16, 03:36

I see, the first LLR, was released on Christmas eve. Created 12/20/02.
This is the version I speak of.
If anyone wants it, let me know where I can send it to.

And actually there was a different version of this version.
I was one of the first to start testing with LLR publicly, as George announced this on x-mas eve, and I started testing it an hour later. Luckily I saved all to flop. There were some beta testers before me that would know about the version I'm talking about. Steve Harvey is one, I believe.

Prime95 · 2004-09-16, 03:52

Quote:

Originally Posted by Harvey563

I've posted a prp3 test file of most of the proths on the top 5000 database, sorted from smallest to largest, to my website, http://www.geocities.com/harvey563/proth3.txt.

285441025 16895

does not belong in the list.

So far I have a corrected PRP3 that has worked on the reisel input file and 70% of the proth file.

2004-09-16, 07:59

Oh ok, since this is crucial to RMA.

I tested all of Riesel3.txt up to n = 10,000.
All found prime on a P4.
The most stable versions were of course 2002, and 2003.

Clue:
I could do anything to get LLRP41, to start the file.
But after starting up my version sucessfully, and then going back to LLR41, it started fine, and gave flawless positive results on all known problemed examples listed in this thread. I have noticed this before too.
This isn't all the programs fault, as far as I see.

Jean Penné · 2004-09-16, 14:00

While correcting the pure LLR bug in LLRP4, I saw that the program can detect the numbers that gave incorrect results with the present version, at init, that is to say before doing the test (no matter the number is prime or composite), so I will make an option in the ini file to allow users to test again only these numbers, and skip the test for the numbers which were already tested correctly (and which, I hope, are much more numerous). Unfortunatly, this feature can only work for k's up to 22 bits, so the big k tests must all be redone...

Also I have successfully tested the candidates in reisel3.txt up to the 3929th for now.

About the little k's (k<300) I can say that I tested successfully the version you are using on all the candidates in Wilfrid Keller "riesel" database.

Regards,
Jean

Prime95 · 2004-09-17, 19:20

The fixed prp3 can now be downloaded. There were two different bugs that affected k values around 7 to 9 digits long. I've made it about 90% of the way through the huge proth QA file.

You can get the versions from:

Windows: ftp://mersenne.org/gimps/prp3.zip
Linux: ftp://mersenne.org/gimps/prp3.tgz

The linux version is untested, I do not have Linux running on any P4s here.

LLRP4 was obviously impacted by this bug. I've sent the updated sources to Jean Penne.

2004-09-15, 16:14	#34
Harvey563 Apr 2004 3×61 Posts	*PRP3 errors on k2^n-1** I checked the reisel3.txt testfile with PRP3 up to n=~107012, & found the following false composites: 293021672^3646-1 is not prime. RES64: 8CF703F5D3C138F4. OLD64: A6E50BE17B43AAD9 281599832^3654-1 is not prime. RES64: C596551BEC2F8ED8. OLD64: 50C2FF53C48EAC86 154898852^8907-1 is not prime. RES64: 0AD1A34D20BF12F7. OLD64: 2074E9E7623D38E3 83241752^20024-1 is not prime. RES64: E5A012BF5900888D. OLD64: B0E0383E0B0199A6 92191772^20202-1 is not prime. RES64: BC80FAD46DA2F6A6. OLD64: 3582F07D48E8E3F0 120830252^24000-1 is not prime. RES64: 4935C1D06DD4A142. OLD64: DBA14571497DE3C4 100569452^24009-1 is not prime. RES64: 0F0C7257B224343D. OLD64: 2D255707166C9CB5 115213412^24010-1 is not prime. RES64: D5E6829B11201A1D. OLD64: 81B387D133604E54 114318392^24012-1 is not prime. RES64: FA4ACF6784EEA69B. OLD64: EEE06E368ECBF3CE 99218492^24016-1 is not prime. RES64: 1B102419C47CFA1D. OLD64: 51306C4D4D76EE54 92992412^24018-1 is not prime. RES64: F1A346AC2B55ECC9. OLD64: D4E9D4048201C658 103848212^24018-1 is not prime. RES64: 4CF30136596E7194. OLD64: E6D903A30C4B54B9 99569012^24022-1 is not prime. RES64: 7B999A3E2B4A0533. OLD64: 72CCCEBA81DE0F97 129329552^24022-1 is not prime. RES64: FFF504C8D078CB29. OLD64: FFDF0E5A716A6178 107615552^24023-1 is not prime. RES64: 0EEC3A40B91408AD. OLD64: 2CC4AEC22B3C1A05 119006912^24030-1 is not prime. RES64: 3DACBB48F027B37E. OLD64: B90631DAD0771A78 116649152^24040-1 is not prime. RES64: F9D8D495E348FD14. OLD64: ED8A7DC1A9DAF739 100754432^24043-1 is not prime. RES64: 5006E3C294EDD739. OLD64: F014AB47BEC985AA 104136752^24044-1 is not prime. RES64: DB0AC47F852C174E. OLD64: 91204D7E8F8445E7 112512752^24044-1 is not prime. RES64: A52B8FAB0FFA71D9. OLD64: EF82AF012FEF5589 110283032^24047-1 is not prime. RES64: CB91D616915A1690. OLD64: 62B58243B40E43AF 124085552^24047-1 is not prime. RES64: BD99065F75FCA7F6. OLD64: 38CB131E61F5F7E1 124632092^24047-1 is not prime. RES64: 7C8646A43CDBD48F. OLD64: 7592D3ECB6937DAC 127396952^24047-1 is not prime. RES64: 909AB7FE3F58E46E. OLD64: B1D027FABE0AAD48 127893992^24048-1 is not prime. RES64: B9CDD41F5E79A81B. OLD64: 2D697C5E1B6CF84E 125164472^24049-1 is not prime. RES64: E8CD4FB7072BE0A5. OLD64: BA67EF251583A1EE 83118752^35154-1 is not prime. RES64: D7A3466DE530BE24. OLD64: 86E9D349AF923A6B 82610772^38957-1 is not prime. RES64: F727A9BDF9B1DEE1. OLD64: E576FD39ED159CA1 101817152^39010-1 is not prime. RES64: 513D1C54DC36D4A2. OLD64: F3B754FE94A47DE4 121253912^39010-1 is not prime. RES64: C9D01CE995DDCB77. OLD64: 5D7056BCC1996263 81205652^45200-1 is not prime. RES64: 192E0C444765188B. OLD64: 4B8A24CCD62F49A0 82132052^45200-1 is not prime. RES64: BB8B9FB1AAEDE888. OLD64: 32A2DF1500C9B997 83118752^48997-1 is not prime. RES64: DB7D1B60A3B28640. OLD64: 92775221EB1792BD 65957672^50001-1 is not prime. RES64: 98F1CDB57D49A33A. OLD64: CAD5692077DCE9AC 78662072^50005-1 is not prime. RES64: 97746053C4E90D0E. OLD64: C65D20FB4EBB2729 82610792^57041-1 is not prime. RES64: 3820F722047B06E6. OLD64: A862E5660D7114B1 6456993*2^60060-1 is not prime. RES64: F561E4099F33AB76. OLD64: E025AC1CDD9B025F These all appear to be cases with large k values, & no failures were found with n greater than 60060. Harvey563

2004-09-15, 20:27	#36
Jean Penné May 2004 FRANCE 1001000100₂ Posts	LLRP4 problems George Woltman warned me Yesterday of the problems you encountered while using different versions of LLR. The most serious, indeed, is the LLRP4 problem. Thanks to your "123missed.txt" file, I could make relevant tests, and the bugs seem to be fixed (all the 123 numbers are now found prime by the corrected program). Nevertheless, I wish to test this updated version on all the k2^n-1 number from Chris Caldwell database before releasing it... So, I strongly suggest to all users to avoid using this program for the time beeing, the risk of missing prime numbers is too big. I am sorry for the drawback, but you can still use the previous versions (the very first IBDWTone of March 2004 excepted). In fact, I found two distinct problems in LLRP4 : The false negatives found with k values smaller than 2*22 are due to a pure LLR bug, and I corrected the LLR.c code accordingly. The other false negatives found in the 123missed.txt file are the same as those causing problems to the PRP3 program, and I avoided this problem by forcing the "general mode" initialization for k values larger than 22 bits, which is only a provisory work around. I hope this problem will not make too much double tests necessary... Jean

2004-09-15, 22:30	#38
Kosmaj Nov 2003 111000100110₂ Posts	Old Versions of LLR I put two old versions of LLR on the 15k.org server. LLR of July 2003, the first version of LLR, the most stable one, no missing primes found. Doesn't work for small n < 300. Used by prof Caldwell to verify k*2^n-1 primes reported to Top-5000. LLR of April 11, 2004 The first stable IBDWT version. One false residue found when running on Pentium-4. No problems found on Athlon. Inefficient on both platforms when initial fft length guesses are wrong.

2004-09-16, 02:51	#39
TTn 10100001010₂ Posts	missing? Kosmaj, it seems the error records are gone? I am not going to post the problemed k again. The error(missing primes) results were triggered by Thomas and I working together, while weighing small k in a primitive way. With the FIRST LLR using P4's, not AMD! This was a particular problem with only certain k, I think. Last fiddled with by TTn on 2004-09-16 at 02:51

2004-09-16, 03:36	#40
TTn 2×7²×97 Posts	ah I see, the first LLR, was released on Christmas eve. Created 12/20/02. This is the version I speak of. If anyone wants it, let me know where I can send it to. And actually there was a different version of this version. I was one of the first to start testing with LLR publicly, as George announced this on x-mas eve, and I started testing it an hour later. Luckily I saved all to flop. There were some beta testers before me that would know about the version I'm talking about. Steve Harvey is one, I believe. Last fiddled with by TTn on 2004-09-16 at 03:41 Reason: add

2004-09-15, 22:08	#37
Kosmaj Nov 2003 2×1,811 Posts	Jean Thank you for your help. I hope the problems will be resolved soon. If you need any help in testing your new version I can help and I'm sure Harvey and the others who have time and resources are willing to help too. I think it will be better if the tests can be done indepenedenlty by several testers. BTW, can you tell us, are small k's, in particular k<100 affected by these bugs in LLR-P4? As for other versions of LLR one false residue was found when testing 97*2^157221-1 using LLR of April 11 and LLR-P2. You can find more details here. Finally, I have the first LLR of July 2003. I tried to attach it zipped but the file is too large. Will try to find another way.

2004-09-16, 07:59	#42
TTn 3×19×151 Posts	tests valuable clue Oh ok, since this is crucial to RMA. I tested all of Riesel3.txt up to n = 10,000. All found prime on a P4. The most stable versions were of course 2002, and 2003. Clue: I could do anything to get LLRP41, to start the file. But after starting up my version sucessfully, and then going back to LLR41, it started fine, and gave flawless positive results on all known problemed examples listed in this thread. I have noticed this before too. This isn't all the programs fault, as far as I see.

2004-09-16, 14:00	#43
Jean Penné May 2004 FRANCE 2²×5×29 Posts	After the bad news, a better one for LLRP4 users. While correcting the pure LLR bug in LLRP4, I saw that the program can detect the numbers that gave incorrect results with the present version, at init, that is to say before doing the test (no matter the number is prime or composite), so I will make an option in the ini file to allow users to test again only these numbers, and skip the test for the numbers which were already tested correctly (and which, I hope, are much more numerous). Unfortunatly, this feature can only work for k's up to 22 bits, so the big k tests must all be redone... Also I have successfully tested the candidates in reisel3.txt up to the 3929th for now. About the little k's (k<300) I can say that I tested successfully the version you are using on all the candidates in Wilfrid Keller "riesel" database. Regards, Jean

2004-09-17, 19:20	#44
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 2·53·71 Posts	The fixed prp3 can now be downloaded. There were two different bugs that affected k values around 7 to 9 digits long. I've made it about 90% of the way through the huge proth QA file. You can get the versions from: Windows: ftp://mersenne.org/gimps/prp3.zip Linux: ftp://mersenne.org/gimps/prp3.tgz The linux version is untested, I do not have Linux running on any P4s here. LLRP4 was obviously impacted by this bug. I've sent the updated sources to Jean Penne.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Accuracy and Precision	davieddy	Math	0	2011-03-14 22:54
computer accuracy	lfm	Puzzles	34	2009-11-10 15:41
CPU Credit Accuracy	g0vegan	PrimeNet	1	2008-11-04 20:26
Verify Accuracy of Test	Numbers	PrimeNet	8	2005-07-31 08:16
Calculating sieving % accuracy	amcfarlane	Math	3	2005-01-02 19:34