I had my own copy of t1200 with some factors divided out and wanted to divide out from that all the factors from the new checkfacts.txt. In the end I found it easier to start from t1200 again, using the n,x in phi_n(x) as keys for sorting. Having a straightforward implementation of Bernstein's algorithm would occasionally be nice to have when you want to get rid of a long list of know factors without thinking too hard about how to do it efficiently.

[QUOTE=Pascal Ochem;250919]The only factors I have are 5, 11, and 5582516651485570586701.
This number is in the tree of 97, so it is not needed for me: the factor 11 is forbidden at this point of the proof.
Thus, the composite cofactor might have received only few ECM. I started ECM on it at 3e6.[/QUOTE]

I also have no other factors. [strike]However, this number has been in the More Wanted ECM server for some time, and is at about 1.5t45. Any new ECM should continue with B1=43M.[/strike]

See update below

[QUOTE=akruppa;250942]I had my own copy of t1200 with some factors divided out and wanted to divide out from that all the factors from the new checkfacts.txt. In the end I found it easier to start from t1200 again, using the n,x in phi_n(x) as keys for sorting. Having a straightforward implementation of Bernstein's algorithm would occasionally be nice to have when you want to get rid of a long list of know factors without thinking too hard about how to do it efficiently.[/QUOTE]As you're doing the checkfacts.txt I didn't bother downloading that file.

I did, however, run all the explicitly given factors >= P15 in all my tables against t1200.txt --- about 100K primes in total --- without finding anything new. Unfortunately, I don't record the explicit values of the largest prime factor of each composite in the tables so it's possible that there may still be some in there. To be honest, I didn't expect to find anything but it seemed worth spending a few minutes to be sure.

Paul

[QUOTE=akruppa;250929]Does anyone happen to have an implementation of Bernstein's factoring with product trees lying around somewhere? It doesn't have to be exceptionally fast, just work. I want to divide known factors out of t1200.txt.[/QUOTE]
You can use the code in msieve which computes a batch gcd using a remainder tree; it's in common/batch_factor.c

Edit: 32-bit factors only, and by default it computes prime factors on the fly

As a matter of interest, how much ECM work has been done on the t1200 file? Clearly it's not very much if there are still C101 entries present.

Paul

I've done about two dozen curves at B1=1M so far. The factors I find are mostly in the 30-35 digit range so B1=1M should be reasonable.

[QUOTE=akruppa;250978]I've done about two dozen curves at B1=1M so far. The factors I find are mostly in the 30-35 digit range so B1=1M should be reasonable.[/QUOTE]Ok, thanks. I may give it a try. No promises.


Paul

[QUOTE=xilman;250967]As a matter of interest, how much ECM work has been done on the t1200 file? Clearly it's not very much if there are still C101 entries present.

Paul[/QUOTE]

Why? Are you use only ECM to factor c101 numbers?

[QUOTE=wblipp;250948]I also have no other factors. However, this number has been in the More Wanted ECM server for some time, and is at about 1.5t45. Any new ECM should continue with B1=43M.[/QUOTE]

CORRECTION. I just realized I've seen this number before. This number was of special interest to you already a year ago and has had 9000 curves by yoyo@home at both 43e6 and 11e7. Additional work should be from 11e7.

SNFS would be reasonable if there were a good polynomial. The obvious quartic polynomial is pretty small though. Does anyone see a better polynomial? This number is

Phi_5(Phi_5(3737657091169))

3737657091169 = Phi_3(3348577) / 3

3348577 = Phi_3(3169) / 3

3169 = Phi_3(97) / 3

[QUOTE=R. Gerbicz;250984]Why? Are you use only ECM to factor c101 numbers?[/QUOTE]If C101 are present, then one should be able to assume that ECM has done at most to the point where it is equally costly to factor them with ECM and an alternative such as QS and NFS. This would suggest to me that ECM up to about t30 has been applied. Alex has recently confirmed my supposition.

Perhaps my standards are different, but a t30 test is "not very much". In the interests of openness, my pet project of generalized Cullen & Woodall numbers has not had very much ECM work done either. Some have been completed to t40 but a good number are only half way through the t30 level. Another three months on my current resources should fix them.


Paul

[QUOTE=xilman;250992]If C101 are present, then one should be able to assume that ECM has done at most to the point where it is equally costly to factor them with ECM and an alternative such as QS and NFS. This would suggest to me that ECM up to about t30 has been applied. Alex has recently confirmed my supposition.[/QUOTE]

OK, I understand you.