[Bug-gsl] bug in gsl_sf_synchrotron_1

[John Houck]

In gsl-1.7, the first derivative of gsl_sf_synchrotron_1(x)
seems to have a spike near

  x = 2.0*M_SQRT2 * GSL_SQRT_DBL_EPSILON
    = 4.21468e-08

Here are some computed values of the function and its first
differences in the neighborhood of this point:

    x                F(x)           delta-F(x)
4.212000000e-08  7.478926692e-03  -5.918277062e-07
4.213000000e-08  7.479518519e-03  -5.917340625e-07
4.214000000e-08  7.480110253e-03  -5.151888128e-07  <---
4.215000000e-08  7.480625442e-03  -5.915287484e-07
4.216000000e-08  7.481216971e-03  -5.914352158e-07

I encountered this problem in gsl-1.7, compiled with gcc-3.3.5
(Debian 1:3.3.5-13).  After installing gsl, I verified that all
the regression tests ran successfully.

-John