http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51133
--- Comment #4 from Jason Dick <dickphd at gmail dot com> 2011-11-15 05:59:27 UTC --- I have attached a tiny program which I used to test these polynomials. I believe the test is entirely robust provided the first two Hermite polynomials are correct. On my machine, this test fails at n=85 because the normalization factor inside the square root becomes too large. Computing the normalization factor in logarithmic space instead allows us to push up to n=87, but then GSL can no longer integrate the function (likely to to a limit of floating point accuracy).