Pascal's OPN roadblock files

[Pascal Ochem]

Originally Posted by henryzz

As an update to this, I have run "pfn -COMP 90 -UT 3 -F" successfully.
I have been iterating runs of 110 that have returned 0 due to 8 then 3 then 1 then 2 composites. Hopefully, the next run will succeed. edit: no another 2 edit2: and yet another 2. Moving to larger runs as reading the factors is taking the majority of the time.
How much further I can push this will depend on the number of factorisations that are needed for larger bounds and whether they are snfs or gnfs candidates.

OK, so we really need to force 5^1. This can be done by replacing the -UT mechanism by a list of primes to branch on,
as in -SIZE mode. The list would just contain the prime 5. Then use " -U 5 1 " to branch only on the component 5^1.

The most recent version of the program is from 2014 and has no further optimizations, only additional modes
for the results in the other 2 papers with Michael (on the radical of an OPN and the bound Omega >= 2*omega+51).

And yes, to get beyond 10^1735, I just added 23 at the end of the list 127, 19,..,17.

[henryzz]

Originally Posted by Pascal Ochem

OK, so we really need to force 5^1. This can be done by replacing the -UT mechanism by a list of primes to branch on,
as in -SIZE mode. The list would just contain the prime 5. Then use " -U 5 1 " to branch only on the component 5^1.

The most recent version of the program is from 2014 and has no further optimizations, only additional modes
for the results in the other 2 papers with Michael (on the radical of an OPN and the bound Omega >= 2*omega+51).

And yes, to get beyond 10^1735, I just added 23 at the end of the list 127, 19,..,17.

That should be easy to switch to and should provide near-identical results. Currently I am working through around 100 composites from a run to 130. Probably won't do the NFS needed on some of them while the team factorisation is underway although I am currently running ecm on my GPU.

Are you able to share this version of the code? The -Y option that you mentioned in 2021 would be useful as it allows more exact branching than the version I have does. At least one of the roadblock circumventions you mentioned in your run to 2200 needed this.

I will add 23 onto the end of my future runs.

[Pascal Ochem]

Originally Posted by henryzz

That should be easy to switch to and should provide near-identical results.

No, the factor tree will be much smaller:
The factor 5^1 contributes to the abundancy at each step.
No branching on p^1 for p > 5.
No branching on 5^k for k > 1.

The latest version: https://www.lirmm.fr/~ochem/opn/p42n.cpp

[SuikaPredator]

More factors / remaining ~t35 composites.

Probably going to try c151~c160.

[mataje]

More factors with t40.

[henryzz]

Originally Posted by Pascal Ochem

No, the factor tree will be much smaller:
The factor 5^1 contributes to the abundancy at each step.
No branching on p^1 for p > 5.
No branching on 5^k for k > 1.

The latest version: https://www.lirmm.fr/~ochem/opn/p42n.cpp

I was being clueless. Of cause fixing 5^1 reduces the tree massively.

To get -U 5 1 working I had to add 5 to the list of forbidden factors for the COMPS switch. This is normally empty.
On the third attempt I had a successful run to 130(2 and then 1 composites from first 2 runs). I will push further. I expect there will be more composites that will limit pushing too much further without sustained effort.

[mataje]

More factors.

[VBCurtis]

Originally Posted by VBCurtis

The team-sieve has gathered over 100M relations. ETA is now 23 June, thanks to swellman pointing all his machines at this job the last few days.
Yield is down into the 3.9's, which shifts our end-Q to something in 165-170M range.

The half of the job running on NFS@home should finish around 17 June, so I expect to be able to try filtering & matrix-building before we reach 450M local relations.

Now at Q=127M, 283M relations found. ETA is now Thursday 22 June early-morning USA-pacific time.

I expect to be away from my computer 20-22 June for the world series of poker, so I'll run filtering on the 23rd to see where we are.

Yield on the nfs@home half of the job is about 8% lower than my test-sieve suggested- it looks like 450M relations rather than the ~480M I projected. Should we sieve those extra 30-40M relations locally, meaning run until the 24th, or should we add 10MQ to the sieve request at nfs@home?

[swellman]

Originally Posted by VBCurtis

Now at Q=127M, 283M relations found. ETA is now Thursday 22 June early-morning USA-pacific time.

I expect to be away from my computer 20-22 June for the world series of poker, so I'll run filtering on the 23rd to see where we are.

Yield on the nfs@home half of the job is about 8% lower than my test-sieve suggested- it looks like 450M relations rather than the ~480M I projected. Should we sieve those extra 30-40M relations locally, meaning run until the 24th, or should we add 10MQ to the sieve request at nfs@home?

I say we just bump up the sieving range on NFS@Home by 10MQ. My two cents.

[VBCurtis]

I second your motion. Since you're a gatekeeper on nfs@home, please add 10MQ to the sieve range.
That'll put us around 480-490M rels from nfs@home; I think we'll need 910-940M total, so the 450M we have planned locally should be plenty.

I'm pleased that the problems we had with database crashes on the 2330L job back in 2019 have been fixed; there have been no reboots of the cado-server package during this job.

I'm game to host future such hybrid CADO / nfs@home jobs in the future, particularly ones on the large end of 15e or small end of 16e.

[swellman]

Originally Posted by VBCurtis

I second your motion. Since you're a gatekeeper on nfs@home, please add 10MQ to the sieve range.
That'll put us around 480-490M rels from nfs@home; I think we'll need 910-940M total, so the 450M we have planned locally should be plenty.

I'm pleased that the problems we had with database crashes on the 2330L job back in 2019 have been fixed; there have been no reboots of the cado-server package during this job.

I'm game to host future such hybrid CADO / nfs@home jobs in the future, particularly ones on the large end of 15e or small end of 16e.

10MQ added to sieving range of the nfs@home job 131__897_5m1.