200
Comment:

← Revision 3 as of 20081114 13:42:11 ⇥
1072
converted to 1.6 markup

Deletions are marked like this.  Additions are marked like this. 
Line 3:  Line 3: 
Here is a useful proof of concept implementation:  * Here is a proof of concept implementation: 
Line 5:  Line 5: 
attachment:proofofconcept.sws attachment:proofofconcept.pdf  [[attachment:proofofconcept.sws]] [[attachment:proofofconcept.pdf]] 
Line 7:  Line 7: 
Notes: (1) I realized that reducing the entries in B modulo p^prec right before doing right_solve greatly speeds up the above proof of concept. (2) To make the above viable will require either modifying IML, modifying Linbox, or implementing Dixon padic lifting from scratch. I think implementing Dixon itself will be the best strategy, since we really need fine control over how it works, *and* we really want to use Pernet'ts FFLAS fast matrixmodp code, which IML doesn't use. * Here are slides from a talk by Michael Monagan (join work with Liang Chen) on the same problem, but the approach is different: [[attachment:Cyclotomic.pdf]] I think this approach is perhaps misguided (?), since it doesn't seem to mention LLL at all, and it's of course '''critical''' to understand how to apply LLL to this problem. There's no way around that. 
Implement padic solver with cyclotomic padic reconstruction algorithm
* Here is a proof of concept implementation:
Notes: (1) I realized that reducing the entries in B modulo p^prec right before doing right_solve greatly speeds up the above proof of concept. (2) To make the above viable will require either modifying IML, modifying Linbox, or implementing Dixon padic lifting from scratch. I think implementing Dixon itself will be the best strategy, since we really need fine control over how it works, *and* we really want to use Pernet'ts FFLAS fast matrixmodp code, which IML doesn't use.
* Here are slides from a talk by Michael Monagan (join work with Liang Chen) on the same problem, but the approach is different:
I think this approach is perhaps misguided (?), since it doesn't seem to mention LLL at all, and it's of course critical to understand how to apply LLL to this problem. There's no way around that.