Remark on "Algorithm 395: Student's t-Distribution [S14]"

@article{Hill1981RemarkO,
  title={Remark on "Algorithm 395: Student's t-Distribution [S14]"},
  author={Geoffrey W. Hill},
  journal={ACM Trans. Math. Softw.},
  year={1981},
  volume={7},
  pages={247-249}
}
For small y ( < c m a x say) the precision lost in evaluating ln(1 + y) corresponds to a relative error about e l y , where e denotes the relative magnitude of processor roundoff. The alternative summation of the logarithmic series until the R th term is negligible, (yR/(R + 1) < e), accumulates roundoff error resulting in an average relative error of about eJ-R. The maximum of these relative errors is minimized, as in Algorithm 465, by choosing c m a x ffi R-~/2, where R is determined for a p… CONTINUE READING