gauss returns incorrect values for n > 1032 or so. The error increases gradually with n and is way off by n = 1080. This is not Float64 specific: I've confirmed that the same incorrect values are produced by gauss(Double64, n) with Double64 from DoubleFloats.jl. Not sure if it could be related to the accuracy issues for kronrod mentioned here: https://github.com/JuliaMath/QuadGK.jl/issues/93#issuecomment-1775755982
Here's an MWE using FastGaussQuadratures as reference:
Thanks for catching this; it looks like there was an overflow/underflow problem in my eigensolver for large n (as well as a bug where it was doing 1000x more iterations than needed). Should be fixed by #111.
gauss
returns incorrect values forn > 1032
or so. The error increases gradually withn
and is way off byn = 1080
. This is not Float64 specific: I've confirmed that the same incorrect values are produced bygauss(Double64, n)
withDouble64
from DoubleFloats.jl. Not sure if it could be related to the accuracy issues forkronrod
mentioned here: https://github.com/JuliaMath/QuadGK.jl/issues/93#issuecomment-1775755982Here's an MWE using FastGaussQuadratures as reference:
Scatter plot of
gauss
nodes against FastGaussQuadrature nodes:Using
quadgk
to confirm that the error is ingauss
, not in FastGaussQuadrature: