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
How does eigs work? #645
Comments
Because you want the ∞-dimensional eigenfunctions, not the spurious ones coming from truncation. Checking the norm is of the tail is a convenient proxy for convergence. |
Ok, if the operator |
The true eigenfunctions have vectors with an ∞ number of entries, so they are technically different than the finite dimension eigenvectors. But these vectors decay exponentially fast so effectively they are finite-dimensional. A simple example are eigenfuctions of u'' with Dirichlet conditions using Chebyshev series. These are of course just sine functions, whose Chebyshev coefficients are known explicitly in terms of Bessel functions. They are not polynomial however and so you technically need an infinite number of coefficients. |
Thanks! If you are interested I can turn your explainations into a docstring for |
Can you point me to some literature that explains why we expect exponential decay of coefficients? |
Sure! Though long term this should be replaced by an infinite-dimensional implementation of shifted simultaneous iteration, which is why not much effort has gone into this yet.
This is a fairly standard result that follows since the eigenvectors are in all Sobolev spaces. Roughly, if we have Ax = λx and |
I am trying to understand how
eigs
works. I understand that it builds a size(n,n)
approximation of the operator and computes eigenvalues of it. What I don't understand is the idea behindpruneeigs
. Why do we need to throw away some eigenvectors? And why is it a good criterion to keep only those vectors whose last four components have a small norm? Or do I misunderstand something?ApproxFun.jl/src/Extras/eigs.jl
Line 53 in 5d84172
The text was updated successfully, but these errors were encountered: