Hans-Reinhard Mueller wrote:
> plot [300:800][0:2] tanh(x)
> There is a hole in the function between about 350 and 720. Similar for
> negative x, with added error that value goes to +1 for x < -720.
That behaviour is quite platform-dependent. It depends on low-level
implementation details of your computer's (or C compiler's) math
library. What you're seeing is the effect of gnuplot guessing that at
around 720 exp() (and thus, tanh()) will no longer yield well-behaved,
finite results, when in fact that already happens around 350.
To be able to fix this, we'ld need to know what version of gnuplot you
observed this on, on what platform.