Two more pictures.
Here I compare the polynomial 32x32-interpolation with Sheldon's Kneser-sexp-implementation on the complex unit-circle of the height. The difference is not visible in the plot, it is about 1e-05 in numbers. (This difference is of the order of the error which the polynomial-method with 32x32 also produces, if tet(tet(x,0.5),0.5) - tet(x,1) is computed to check the consistency of two-times half-height-iteration vs one integer iteration: I get a difference of about 1e-5)
A plot of the differences (absolute difference between the evaluations of the two methods) shows two surprising *spikes*. Possibly there is a method- or an implementation-specific anomaly/bug?
(note: the h-parameter on the unit-circle 1..I..-1..-I..1 is replaced by the angle phi 0..2*Pi )
Gottfried
Here I compare the polynomial 32x32-interpolation with Sheldon's Kneser-sexp-implementation on the complex unit-circle of the height. The difference is not visible in the plot, it is about 1e-05 in numbers. (This difference is of the order of the error which the polynomial-method with 32x32 also produces, if tet(tet(x,0.5),0.5) - tet(x,1) is computed to check the consistency of two-times half-height-iteration vs one integer iteration: I get a difference of about 1e-5)
A plot of the differences (absolute difference between the evaluations of the two methods) shows two surprising *spikes*. Possibly there is a method- or an implementation-specific anomaly/bug?
(note: the h-parameter on the unit-circle 1..I..-1..-I..1 is replaced by the angle phi 0..2*Pi )
Gottfried
Gottfried Helms, Kassel