Sloan Ordering Gives Incorrect Results

[Allen Barnett <allen@…>]

I was trying to use Sloan ordering to reduce the bandwidth on a couple of my finite element meshes. It did not give very good results; the bandwidth after reordering was exactly equal to number of vertexes! So, I tried the example program here: http://www.boost.org/doc/libs/1_53_0/libs/graph/example/sloan_ordering.cpp I expected the results given in the comments in the example file:

/*
  Sample Output
  #####################################
  ### First light of sloan-ordering ###
  #####################################

  original bandwidth: 8
  original profile: 42
  original max_wavefront: 7
  original aver_wavefront: 4.2
  original rms_wavefront: 4.58258

  Starting vertex: 0
  Pseudoperipheral vertex: 9
  Pseudoperipheral radius: 4

  Sloan ordering starting at: 0
    0 8 3 7 5 2 4 6 1 9
    bandwidth: 4
    profile: 28
    max_wavefront: 4
    aver_wavefront: 2.8
    rms_wavefront: 2.93258

  Sloan ordering without a start-vertex:
    8 0 3 7 5 2 4 6 1 9
    bandwidth: 4
    profile: 27
    max_wavefront: 4
    aver_wavefront: 2.7
    rms_wavefront: 2.84605

  ###############################
  ### sloan-ordering finished ###
  ###############################
*/

Instead I got:

#####################################
### First light of sloan-ordering ###
#####################################

original bandwidth: 8
original profile: 58
original max_wavefront: 7
original aver_wavefront: 4.2
original rms_wavefront: 4.58258

Starting vertex: 0
Pseudoperipheral vertex: 9
Pseudoperipheral radius: 4

Sloan ordering starting at: 0
  0 7 4 9 1 6 2 5 3 8 
  bandwidth: 8
  profile: 55
  max_wavefront: 6
  aver_wavefront: 3.6
  rms_wavefront: 3.94968

Sloan ordering without a start-vertex:
  8 5 4 9 1 7 6 2 0 3 
  bandwidth: 9
  profile: 66
  max_wavefront: 7
  aver_wavefront: 4.2
  rms_wavefront: 4.66905

###############################
### sloan-ordering finished ###
###############################

You can see that the current results are in line with what I was experiencing with my own meshes, namely, no reduction in bandwidth or profile.

Sloan ordering seems to have first appeared in boost 1.30.0. I tried that version too, and it gives essentially the same results as I see with 1.53. So, perhaps it's the comments in the example program that need to be updated and Sloan ordering is not appropriate for my meshes.

[Jeremiah Willcock]

(In [83877]) Fixed 32/64-bit size issues; fixes #8434