Changes between Version 1 and Version 2 of Ticket #20445, comment 17


Ignore:
Timestamp:
05/11/16 07:42:42 (6 years ago)
Author:
nthiery
Comment:

Legend:

Unmodified
Added
Removed
Modified
  • Ticket #20445, comment 17

    v1 v2  
    11Replying to [comment:15 tscrim]:
     2    CPU times: user 4min 23s, sys: 604 ms, total: 4min 23s
     3    Wall time: 4min 23s
     4
     5Great, I believe you can claim to be the first person ever who iterated through E8 in less than 5 minutes! (I was expecting that since 1. your computer was about twice as fast as mine last week, and 2. you took ~8 minutes to iter through E8 in chevie, and my code is about twice as fast as the chevie code.)
     6
    271. I would really like to see how to implement our own permutation group elements with only what we need. I would hope that to again result in some speedup. One question there: we have w(-\beta) = -w(\beta), so we would not need to record the complete permutation on all roots, but on the positive roots would be enough. That would speed several computations such as creating new elements and testing for equality), but it would have the drawback that we constantly need to work mod N (N=nr of positive roots), e.g., we would have such checks and mod's when multiplying two permutations.
    38
    4 This business sounds of the same nature as what we have for affine
    5 permutations (in window notation). Would there be a way to use the
    6 same implementation behind the scene?
    7 
    8 Cheers,
    9                                Nicolas
     92. We should get your parallelization to work with this so that people can then use many cores to actually do stuff with the elements in type E8.