Clarkson's Algorithm: Higher Dimensional Experiments


Here we report a computational comparison of the LP algorithm and LP+Clarkson algorithm for higher dimensions (27 and 35). The instances we use below for the comparison are produced by Mathieu Dutour Sikirić who used redundancy removal for the study of Voronoi polytopes and C-type domain theory.

All input data files can be found in CtypeData . See Instructions for doing the experiments by yourself.

Remark that with floating-point arithmetic, the computation may not produce the correct result (numerical failure), and furthermore, the numerical irregularities may yield segmentation faults.

d n Float Exact (Rational)
LP algorithm LP+Clarkson Speedup LP algorithm LP+Clarkson Speedup
CTYPE 7
ctype7_0 (1%ess) 27 2898 52s 0.66s 79 212s 12s 18
ctype7_1 (1%ess) 27 2898 120s 0.68s 176 168s 9s 19
ctype7_2 (1%ess) 27 2898 106s 0.8s 133 211s 12s 18
ctype7_3 (1%ess) 27 2898 126s 0.69s 183 174s 9s 19
ctype7_4 (1%ess) 27 2898 144s 0.66s 218 194s 11s 18
ctype7_5 (1%ess) 27 2898 (numerical failure) 0.7s * 294s 18s 16
ctype7_6 (1%ess) 27 2898 117s 0.6s 195 393s 9s 44
CTYPE 8
ctype8_0 (0.4%ess) 35 9075 (seg. fault) (seg. fault) * 103841s 172 s 604

back


Written by Komei Fukuda (ETH Zurich, Switzerland) return
Last updated: January 26, 2021