Real Instituto y Observatorio de la Armada en San Fernando.
Using DE200/LE200 JPL data base we compress astronomical ephemerides finding the ``best approximation'' in the Chebyshev sense. It is done by solving a linear programming problem. When reproducing the data base, the maxima and minima values of the error are equal in absolute value over the considered interval.
However, this polynomial approximation does not guarantee the desired error in interpolated points of the data base. Then a direct evaluation of the polynomial could not be reliable.
Uniform approximation for continuous functions does ensure the desired error over the entire interval. By previously interpolating the data base we get a closer approximation to a continuous function. The smoothness of the resulting discrete function error corresponding to the new polynomial approximation in our opinion ensures the desired error over the entire interval.