Re: [Bug-gsl] bug report regarding gsl_diff_*

[Brian Gough]

Zegenhagen Stefan writes:
 > I think I've found a bug that makes the numerical differentiation
 > routines quite unusable in some cases.
 > The solution could be to either a stepsize relative to x in the
 > calculation of the higher order derivative(something like 
 > 
 >      double h = x * GSL_SQRT_DBL_EPSILON;
 > 
 > in the function header). This solution should avoid the restricted
 > precision problems.

Thanks for the bug report.

I agree with you, the fixed size algorithm used there is flawed.
Ideally it should be adaptive.  I've added this information to
BUG#13 in the BUGS file.

-- 
Brian Gough