[Proj] Problems with Pittman geodesic??

[Hamish]

Gerald wrote:
> While checking out the accuracy of Vincenty vs. Pittman
> I was fairly consistently getting agreement to micron or
> better.
> 
> BUT not always!!!
> 
> I would appreciate anyone who has a Pittman geodesic routine
> to test the following inverse (WGS84 ellipsoid):
> 
> point 1 at 0 lat, 0 lon and point 2 at 45N 90E.
> 
> With my version of Pittman I get a distance of 9993541.5348708
> AND I get the same distance while changing the longitude of the
> second point from about 89.6 to a hair over 90.
> 
> At 0,0/45,89.5:
> Pittman: 9970963.0100082
> Vincenty:9970963.01000801
> 
> At 0,0/45,89.6
> Pittman: 9993541.5348708
> Vincenty:9978847.65947167
> 
> Pittman remained constant in this longitude interval
> 
> At 0,0/45,90
> Pittman:  9993541.5348708
> Vincenty:10010386.3610382
> 
> Between 90.0000001 and 90.000001 Pittman finally came back
> into near agreement with Vincenty.
> 
> Note: I did nothing to the Pittman FORTRAN loaned from Mugnier
> other than compile it with gfortran and link to a C driver.


I would guess that some internal variable is defined as a REAL (float)
instead of a DOUBLE PRECISION (double), and so the effect you are seeing
is quantization due to variable precision.


Hamish