Changes between Initial Version and Version 1 of Ticket #21579, comment 16


Ignore:
Timestamp:
09/30/16 12:15:02 (3 years ago)
Author:
cpernet
Comment:

Legend:

Unmodified
Added
Removed
Modified
  • Ticket #21579, comment 16

    initial v1  
    11Nice catch. This seems like we hit the unfortunate case of the randomized algorithm. Still I don't understand why:
    22* only the constant coefficient is erroneous. Normally (i.e. with random matrices), the bitsize of the coefficients grows a bit with the degree, before decreasing, hence, there should be other coefficients for which the theoretical bound was not met when the early termination was decided. Maybe this is due to the special shape of your instance. Could you check whether the bitsize of the coeffs of degree>0 is lower than that of the determinant ?
    3 * why is the erroneous factor (-5) so small.
     3* the erroneous factor (-5) is so small.