Reserved for MF - Sequence 276

[kar_bon]

The intention of creating my pages at rieselprime.de 10 years ago was to collect data spreaded over the whole forum and display a condensed version of those information and to easily find all related stuff.

So for the old 15k-search project those days to say a special k-val was searched up to a given n-val, without looking dozens of post or thread in here.

As there're many projects working on such problems it's no managable anymore like I did a long time.

I saw the possibility of a Wiki software to give users the chance to update those page by themselves.
The problem is you have to know how to edit and that's in my opinion to many users reason enough not doing so.

The page at MersenneWiki is an example how this could be done more userfriendly without knowing much about the Wiki interna.

It would also help to collect and summarize information about users doing their own searches but haven't their own page to dislpay them.

For ALQs Zimmerman, Clavier and seqs reserved by others than yoyo could put their data in there, too, so things like 276 won't be missed then.

Whishful thinking I know, but the internet was started because someone had an idea to make work easier.

[LaurV]

As we used to say, every kick in the butt is a step ahead...

This is good opportunity for Ryan to step in...

[yoyo]

For the sequence 276, does it make sense to run ecm to 65 digits (B1=850M)?

[VBCurtis]

For a C201, I would run about 25% of a T65 at 850M, and then a couple thousand curves at 3e9 (say, 0.2T65 worth). I have no justification- that's merely what I would choose to acheive roughly half a T65.

I'd like to see what the Bayesian tool has to say!

[swellman]

Originally Posted by Dubslow

I agree there's a lot of fragmentation, but unfortunately there's no real good solution.

Incidentally, do we the forum want to take on 276 (perhaps at the expense of 4788)? It should be mostly doable except for linear algebra. That would require significant resources that maybe not even Tom can bring to bear (even if he were willing).

Ryan took back 4788 in late March, though progress has slowed since.

I'm willing to help with AS276, though yoyo would be the logical player for ECM. Can this effort fit into the 15e siever? It's a daunting task!

[wombatman]

Originally Posted by VBCurtis

For a C201, I would run about 25% of a T65 at 850M, and then a couple thousand curves at 3e9 (say, 0.2T65 worth). I have no justification- that's merely what I would choose to acheive roughly half a T65.

I'd like to see what the Bayesian tool has to say!

Bayesian gives the following, assuming 48000 curves at B1=260e6 (a t60):

 7900 curves with B1 =   2.9e9

Note that this assumes only 1 thread of ECM running, so it's going to underestimate somewhat, but that gives you some idea of what's optimal.

[yoyo]

I will run 20000 curves with B1=850M on it.
Don't know if this is sufficient?

I can run also much more, even on higher B1.

[swellman]

As a suggestion, how about a total of 40000 curves @B1=85e7 and then 8000 curves @B1=29e8? It's probably a bit of overkill but should certainly provide proper due diligence in the preprocessing stage.

Thoughts?

[VBCurtis]

In the spirit of the Bayesian tool, I like doing effort similar to what you suggest, but with more at 29e8 and less at 85e7. How about 20k at 85e7 for now, then 10k (or 15k) at 29e8?

Some forumites can do some 29e8 curves while yoyo tackles the 85e7 set; or can his network do the big-bound curves too?

The CADO root-opt tools seem to take about a digit of difficulty off, compared to 2014-15 msieve. Scores have been 10-15% better than most msieve best-polys for each digit length above 160; if that holds at 200, that "free" digit should really help the matrix step.

[yoyo]

My Minion network can run also 29e8 on this C201.

[swellman]

Originally Posted by yoyo

My Minion network can run also 29e8 on this C201.

So 20000 curves @B1=85e7, followed by 10000 curves @B1=29e8. I like it.

Anyone else?