Double checking

[Kosmaj]

Thomas and I filled gaps of all Ks from the 4th Drive appearing on his list, except those that you did.

Keller doesn't accept proxy reports (I tell you, you tell him) but I'll report all of them soon.

[gd_barnes]

2 of the 7 100k ranges are complete for 300<k<=1001 for 50K<n<=100K. The 300's and 400's are done. I found the following missing primes:

349*2^56031-1
373*2^82231-1
497*2^66912-1

There were no incorrect primes.

I double-checked my double-check and confirmed that they were indeed prime.

The error rate continues its steady climb. There was a 1.96% error rate for 25K<n<=50K. For 50K<n<=100K for the 300's and 400's, when adding 3 errors that were previously found and now corrected on our site, it's a total of 6 errors in 208 primes = 2.88%. For all of the k's that I've searched so far for 50K<n<=100K including errors found previously, it's 10 errors in 432 primes = 2.31%.

The 500's and 900's are nearing completion and will be done by the weekend. There are some missing primes in the 900's and I'll post them when the ranges are done.

When this is all done, we'll have all k<=1001 double-checked up to n=100K!

Edit: Thanks for info. Kosmaj!


Gary

[Kosmaj]

Next you can double check some Ks from the 5th Drive. With exception of k=57 which has no gaps, the status of other four k<=135 is not clear because Sax hasn't reported his completed ranges, but you can work on k>200 with exception of k=285 which I mostly did and will finish it in a few days. Here is the list of remaining eight k>=213 (some have multiple small gaps):

 k      nmin    nmax
213	100	250
237	230	260
249	150	200
249	230	260
261	135	150
261	155	260
267	155	160
267	180	205
267	260	300
273	100	120
273	130	150
273	155	170
273	175	180
279	100	130
279	135	145
279	155	165
279	230	260
291	130	165

[gd_barnes]

Originally Posted by Kosmaj

Next you can double check some Ks from the 5th Drive. With exception of k=57 which has no gaps, the status of other four k<=135 is not clear because Sax hasn't reported his completed ranges, but you can work on k>200 with exception of k=285 which I mostly did and will finish it in a few days. Here is the list of remaining eight k>=213 (some have multiple small gaps):

 k      nmin    nmax
213    100    250
237    230    260
249    150    200
249    230    260
261    135    150
261    155    260
267    155    160
267    180    205
267    260    300
273    100    120
273    130    150
273    155    170
273    175    180
279    100    130
279    135    145
279    155    165
279    230    260
291    130    165

Cool! I'll start on these in 1-2 weeks. Since I'm well past halfway on my current double-check effort, I want to finish most of it up first. Most of the ranges here are relatively small except for k=213 and 261 so it shouldn't be too bad of an effort.


Thanks,
Gary

[gd_barnes]

2 more of the 7 100k ranges are complete for 300<k<=1001 for 50K<n<=100K for a total of 4 of the 7 100k groupings complete. The 500's and 900's are done. I found the following missing primes:

937*2^51701-1
951*2^62093-1

There were no incorrect primes.

The range of 500<k<600 had no errors. Oddly this range of k has had NO errors for any k for any n all the way up to n=100K! At least one small range of k has been accurate all the way through.

Once again, I double-checked my double-check and confirmed that they were prime.

The total # of errors and primes for the 300s, 400s, 500s, & 900s for n=50K-100K including errors found before is: 8 errors in 430 primes = 1.86%. The errorless 500s helped drop it back under 2%.

Kosmaj, this leaves 600<k<900 to go for this effort. They are collectively about 65-70% done so it should be ~5 days before completion. At that time, I'll have 2 cores to start sieving on the k's and ranges that you suggested.


Gary

[gd_barnes]

Kosmaj,

All double-checking for 300<k<=1001 for n=50K-100K will be done in 2-3 days. I have taken one core off of it and am using the other core to now sieve the k's<300 for the ranges that you suggested. I don't have an ETA on completion of it yet but a wild guess would be 4-5 weeks on one 2.66Ghz P4 core.


Gary

[gd_barnes]

I have completed all k-ranges for the double-check for 300<k<=1001 for n=50K-100K. The ranges left to report were for k's in the 600s, 700s, and 800s. This was the most problematic area. I have found the following additional 5 missing primes:

619*2^62209-1
675*2^67456-1
749*2^60308-1
781*2^63537-1
883*2^72151-1

There were no incorrect primes and I double-checked my finds to confirm primality.

This makes a total of 10 errors found with this particular effort. That in addition to 7 errors found previously for n=50K-100K makes a total of 17 errors for this n-range. With the above bad k-ranges, the error-rate continues its steady climb:

n=0-10K; 1 error out of 7065 primes; 0.01%
n=10K-25K; 9 out of 940; 0.96%
n=25K-50K; 14 out of 716; 1.96%
n=50K-100K; 17 out of 732; 2.32%

We can now officially say that all k<=1001 are double-checked up to n=100K!!

Anon, if you happen to read this, sometime in the spring of 2008, I might hit you up with helping me coordinate a double-check for n=100K-260K for 300<k<=1001 using Prime Grid/BOINC/LLRnet in some fashion. Do you think that would be doable? With the above statistics, we definitely need to get all Prime Search ranges double-checked. n=100K-260K for 351 k's will be a HUGE effort that certainly needs some DC.


Gary

[em99010pepe]

Gary,

Please count with my help....

Carlos

[gd_barnes]

Originally Posted by em99010pepe

Gary,

Please count with my help....

Carlos

Thanks, Carlos. That would be great. I'll let you know. I'm shooting to start the effort in March-April 2008 after my 12 heavy-weight k's are done LLRing to n=400K. (Very rought estimate.)


Gary

[mdettweiler]

Originally Posted by gd_barnes

I have completed all k-ranges for the double-check for 300<k<=1001 for n=50K-100K. The ranges left to report were for k's in the 600s, 700s, and 800s. This was the most problematic area. I have found the following additional 5 missing primes:

619*2^62209-1
675*2^67456-1
749*2^60308-1
781*2^63537-1
883*2^72151-1

There were no incorrect primes and I double-checked my finds to confirm primality.

This makes a total of 10 errors found with this particular effort. That in addition to 7 errors found previously for n=50K-100K makes a total of 17 errors for this n-range. With the above bad k-ranges, the error-rate continues its steady climb:

n=0-10K; 1 error out of 7065 primes; 0.01%
n=10K-25K; 9 out of 940; 0.96%
n=25K-50K; 14 out of 716; 1.96%
n=50K-100K; 17 out of 732; 2.32%

We can now officially say that all k<=1001 are double-checked up to n=100K!!

Anon, if you happen to read this, sometime in the spring of 2008, I might hit you up with helping me coordinate a double-check for n=100K-260K for 300<k<=1001 using Prime Grid/BOINC/LLRnet in some fashion. Do you think that would be doable? With the above statistics, we definitely need to get all Prime Search ranges double-checked. n=100K-260K for 351 k's will be a HUGE effort that certainly needs some DC.


Gary

Congratulations on getting it finished! That was a lot of heavy lifting. But the error rate is quite appalling.

As for using PrimeGrid/BOINC/LLRNet to help do the 300<k<1001 doublecheck sometime later: Yes, it might be doable--would this be considered part of RPS, or would it be considered an outside effort? If it would be part of RPS, then we obviously already have Kosmaj's ruling that RPS should stay independent (i.e. no PrimeGrid)--but if it's separate from RPS (since those are Prime Search ranges, thus not technically RPS), then we might be able to go that route.

[gd_barnes]

Originally Posted by Anonymous

Congratulations on getting it finished! That was a lot of heavy lifting. But the error rate is quite appalling.

As for using PrimeGrid/BOINC/LLRNet to help do the 300<k<1001 doublecheck sometime later: Yes, it might be doable--would this be considered part of RPS, or would it be considered an outside effort? If it would be part of RPS, then we obviously already have Kosmaj's ruling that RPS should stay independent (i.e. no PrimeGrid)--but if it's separate from RPS (since those are Prime Search ranges, thus not technically RPS), then we might be able to go that route.

Good questions. I can't speak for Kosmaj but I don't think he'd have a problem. I think his objection would come in more if it were k < 300 and/or if we ended up getting reportable primes out of it. Kosmaj, feel free to comment on this if you want.

I would consider this to be project-independent. It's neither RPS nor Prime Search nor any other project. I'm just choosing to coordinate it here because I know people here and 15k.org now lists ALL known Riesel k's. For that matter, I could have coordinated it through one of the other Prime search projects here at mersenneforum.org.


Gary