🚀 go-pugleaf

To summerize some earlier posts, Russell Blyth reported an error in
graphing the function  x^(1/log x) over the range [0,10] : for one point
in this range f(x) is 3E-15, rather than the correct value of 10.  Fred
Worth was unable to reproduce this TI-82 error on his T1-85; but the error
did appear at another point in the range.  Elwood Gruschow found four such
errors in the range [1,185] - apparently a serious contender for the title.
He also noted the importance of the range, since only a finite number of
function values are calculated.

Specifically, 127 equally spaced values of x are calculated by

		x = xMin + N(xMax - xMin)/126  ;   N = 0,..., 126  .

Fred Worth's bad point corresponds to N = 3; while Elwood Gruschow's
correspond to N = 33, 42, 60, 65 .

This still leaves unanswered the question I raised in my earlier post.

To calculate  x^(1/log x) the TI-85 in fact calculates exp(ln x/log x) .
(See the file UNIVPOWR.TXT in the archive) . But the graph of this
function holds no surprises.  f(x) = 10  for all calculated values of x
in both ranges.   What is happening ???