I clicked on the link and it goes to this post in the forum,
Oh, I saw the other post but missed the (click to expand). So according to the code, it is just (y_comp - y_exact) / eps(). That’s one of the things I tried, I still get worse errors.
The other library is Arb; it’s a multiprecision library, and from the code it seems to be set to a precision of 500. I’ve used it, I will take those values as ‘exact’. I was also comparing against an ‘eact’ value, so that should not be the problem.
so far the Fortran part is just a port, i.e., no attempts at changing the accuracy
It’s a different compiler, but I assume it should give pretty much the same accuracy as the Julia version.
… you need to run with
-ffree-line-length-nonewith gfortran
My bad. Did not read the README.