-
Notifications
You must be signed in to change notification settings - Fork 3
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Associated 2-band OPs Golub Welsch? #29
Comments
There are "anti-Gaussian" quadrature methods which as far as I know interlace the points of Gaussian quadrature. Not sure if that is why you said "anti-Gauss–Lobatto" or if you just meant "the opposite of what Gauss–Lobatto does". |
I meant "opposite of what Gauss-Lobatto" (or Radau): instead of insisting on a specific point, I want to forbid it. Note its a bit delicate as for even points we get two points tending to (but not exactly) zero |
Is there a characterization of the non-negative weights that live in the kernel of the Hilbert transform? In the even case, does the pair of Gauss points breach |
I think the kernel is only
Not even close: julia> golubwelsch(Q[:,Base.OneTo(2)])[1]
2-element Vector{Float64}:
-0.3354101966249685
0.3354101966249685
julia> golubwelsch(Q[:,Base.OneTo(4)])[1]
4-element Vector{Float64}:
-0.8260064989841912
-0.10635236062957643
0.10635236062957698
0.8260064989841945
julia> golubwelsch(Q[:,Base.OneTo(6)])[1]
6-element Vector{Float64}:
-0.9146586174230956
-0.6894624123565193
-0.03501609669916683
0.0350160966991655
0.6894624123565204
0.9146586174230971 |
That's interesting: I can see why the odd symmetry forces a root at the non-negligible Dirac point mass, but had no idea that it would be "strong" enough to pull other points into zero-weight territory (in the even case). |
It sort of makes sense; since one typically expects finite section to converge |
Since two bands were moved out of the package I guess it's fine to close this for now. |
The two-band analogue of Chebyshev T is now implemented as
T = TwoBandJacobi(ρ, -1/2, -1/2, 1/2)
, orthogonal w.r.t.|x|/(sqrt(1-x^2)sqrt(x^2-ρ^2))
. It has the nice property that the Hilbert transform of its weight is 0, so the Hilbert transform is a simple map fromT
toQ = Associated(T)
, the associated orthogonal polynomials.The surprising thing is
Q
has an orthogonality measure with a delta function at 0, I believe its something likesqrt(1-x^2)sqrt(x^2-ρ^2)/|x| + ρ δ(x)
(where the first term is only valid forρ < |x| <1
). This has the annoying feature that Golub--Welsch gives extra points associated with theδ
:Anyone seen anything like this before and know if there's a modification to Gauss quadrature, i.e. an anti-Gauss–Lobatto that prevents certain points?
I just need this for the expansion in
Q
. I can work around it by using a Vandermonde matrix. But it's an interesting problem.@MikaelSlevinsky @TSGut @ajt60gaibb @marcusdavidwebb
The text was updated successfully, but these errors were encountered: