Question on unweighted Gauss-Hermite

[karenml]

This might be obvious, but why do you do
_w .*= sqrt(π)/sum(exp.(-_x.^2).*_w)

(in line 54) to the unweighted Gauss-Hermite quadrature weights?

Thanks!

[dlfivefifty]

exp.(-_x.^2).*_w gives the Gauss quadrature weights so sum(exp.(-_x.^2).*_w) will give what they think is the integral of 1 * exp(-x^2). We know this should be sqrt(π) so this line fixes any issue with getting the scaling wrong.

Note this was present in @ajt60gaibb original implementation:

bfd8eac#diff-28a0f2d6660b3d3cf920bd063823ccfcd0d64bf99303eb221485646ac5fe881aR27

So maybe he knows why this fix is needed?