All the .fld files in this directory are representations of seismic
P-wave velocity in the lower mantle of the earth.  The data is 
derived from a spherical harmonic expansion and so can be converted
to several different kinds of AVS data structure.  The value is defined
only in the region of the lower mantle -- between two spherical
shells at 3441km and 5700km from the centre.    The range of the
data is between -21 and 21.

lm1-uniform.fld		Sampled on a 51x51x51 grid.  Points outside
			the lower mantle are given a value of -800.

lm1-scatter.n.fld	Sampled on a grid in spherical polar co-ordinates.  
			The cartesian co-ordinates of the grid points are 
			added to complete a 1D "scatter" AVS field. The data 
			has been downsized from the 61x31x10 grid by n.

lm1-irreg.fld		Sampled on a 61x31x10 grid in spherical polar 
			co-ordinates.  The cartesian co-ordinates of the grid
			points are added to complete a 3D irregular AVS field.

There are several illustrative networks to demonstrate the merits and
demerits of each type of field.

1) Isosurface -- iso.net.  Can be used with uniform or irregular data.
   The disadvantage of the uniform sampling is that an isosurface is
   always drawn between the data and the non-data.

2) Arbitrary slicer -- still one of the best ways to look at fine
   detail in 3D data.  Compare uniform and irregular datasets.

3) 3D voxel rendering -- tracer.net and shade.net.  In general volume
   rendering only works on uniform data.  The colormap "hole.cmap" makes
   the bulk of the data around 0 fairly transparent so the structure can 
   be seen.

4) Example of the scatter_to_ucd module -- scatter.net This is one way
   to use scatter fields.  Note that the time varies as a high power of
   the data size -- try it with the most downsized field first!

5) UCD. ucd-*.net  The irregular field is converted to a UCD structure by
   the "field_to_ucd" module.  ucd-rslice and ucd-crop give excellent
   3D renditions showing the 3D structure as well as the data.  

6) ucd-iso.net.  Compare with iso.net.

7) UCD Voxel rendering -- ucd-tracer.net.  This is the only way to do voxel
   rendering of irregular 3D data in AVS.  Keep the downsize low though!

Some vague conclusions.  

a) The bounded nature of the data makes the uniform field
representation nearly useless.  The only technique that works is 3D
voxel rendering.  The trick is to assign an out-of-band value to the
underined region and assign that range a zero opacity in thr
colourmap.

b) Scatter fields are not much use as there is no explicit information
on the spatial locality of the points. That is you can't find the
neighbouring points easily. Scatter_to_ucd adds this information, but
at a heavy and unbounded cost in CPU time.

c) Provide the locality information by defining a field of true
curvilinear co-ordinates, with a mapping from computational to
physical space.  You (nearly) lose the ability to do voxel rendering
but gain much more.  You get a clear view of the domain of your data
using volume bounds -- or even more clearly by converting it to UCD.