This is the result of some tinkering around at one time when investigating
yet another density estimation idea.
I found it useful to use normal mixtures as examples, including the
Marron-Wand densities.

			Martin Mächler, ETH Zurich, Switzerland
			March 1997,  June 2002
---------

o  The Marron-Wand densities are not regular data sets, e.g., they can't be
   used without the normix package. ==> hence in ./R/ and not ./data/

o  bug in qnorMix() --- the interval for uniroot is sometimes too small
   --> tests/ex.R  -- fixed fo 1.0-6

   HOWEVER: improve  qnorMix() for long input vectors:
     - sort(p)
     - compute  qnormMix( range(p) ) = qnorMix( p[c(1, n)] )
       use splinefun() in between as good starting values for uniroot()!