dnorm: in the 3D version we normalize
based on the overlapping window rather than segment
(essentially based on npmin rather than segmax)

Reason: the overlapping window can be large and then the
normalization based on the segment size was not sufficient.
But that means that the segments are smaller than dnorm.

#######################################################
HISTORY:

21/feb/09 (Antonio Galea)
  - static initialization of spline coefficients
  - avoid calling exp() and erf() multiple times for
    spline and derivatives
  - do not compute first and second derivatives unless
    required

02/03/03 (jh) - added deviation site file to the output

03/11/00 (jh) - re-written I/O operations, several bug fixes
including 3D topo parameters, and 2nd RST derivatives

12/07/00 (MN) - added field selection parameter for sites lists
                (allows selecting column in site_lists)

copied from Mark 15.may.1996,
just a beta version, not fully tested