I have implemented Bigfloat arithmetic for the Exponential Integrals and some
further test for special values including tests for Bigfloatzero.
Furthermore I have extended the test file. The numerical results for Complex
Bigfloats are verified for all Exponential Integrals within an accuracy of 62 to
64 digits.
Remarks:
1.
The Bigfloat routine does not add some extra digits to fpprec. So we lost 1 to 2
digits of precision. This can be further improved.
2.
There is no optimization for speed. It was my aim to get an algorithm which will
work and calculate correct results.
3.
The accuracy of the numerical tests of the test file are optimized for my system
(GCL 2.6.8). On other systems a lot of the numerical tests could fail or even
produce more accurate results.
4.
The examples for testing the symmetry give with one exception exact results (0.0
or 0.0b0). Because we have numerical routines that seems to be by accident.
Perhaps other compiler can not verify this results.
I have attached the code and a test file to the feature request on
Sourceforge.net.
Dieter Kaiser