I am using the eigs() function (from the Arpack package) to find the eigenvalues of a sparse matrix (eigen() doesn't work for spare matrices). Apparently, eigs() fails to find all the eigenvalues, even in a very simple case:
using Arpack
M = spdiagm(0 => [1,2,3])
eigs(M, nev = 3)
The output of the last line is a vector contacting only 2 eigenvalues, "2" and "3" ("1" is missing). What am I missing? Is there a different function for computing the eigenvalues of a sparse matrix (the actual sparse matrix is much large than the above M).
It actually shows at warning:
julia> eigs(Matrix(M), nev = 3);
┌ Warning: Adjusting nev from 3 to 2
└ @ Arpack c:\JuliaPkg\Julia1.5.0\packages\Arpack\o35I5\src\Arpack.jl:82
Looking at the source code this can return a maximum of LinearAlgebra.checksquare(M) - 1
values.
What you could try to do is to use a BandedMatrix
instead which is also sparse:
julia> m=BandedMatrix(0=>1:3)
3×3 BandedMatrix{Int64,Array{Int64,2},Base.OneTo{Int64}}:
1 ⋅ ⋅
⋅ 2 ⋅
⋅ ⋅ 3
julia> eigen(m)
Eigen{Float64,Float64,Array{Float64,2},Array{Float64,1}}
values:
3-element Array{Float64,1}:
1.0
2.0
3.0
vectors:
3×3 Array{Float64,2}:
1.0 0.0 0.0
0.0 1.0 0.0
0.0 0.0 1.0
eigs
uses an iterative method that is only practical or encouraged for large sparse matrices and when you only want a small number of eigenvalues (small compared to the size of the matrix).
The error message you got in eigen
is a little unfortunate. Preferably, it would have also mentioned that if your array is small, you should just collect
it into a dense matrix, eg
julia> using SparseArrays, LinearAlgebra
julia> M = spdiagm(0 => [1,2,3])
3×3 SparseMatrixCSC{Int64,Int64} with 3 stored entries:
[1, 1] = 1
[2, 2] = 2
[3, 3] = 3
julia> eigen(collect(M))
Eigen{Float64,Float64,Array{Float64,2},Array{Float64,1}}
values:
3-element Array{Float64,1}:
1.0
2.0
3.0
vectors:
3×3 Array{Float64,2}:
1.0 0.0 0.0
0.0 1.0 0.0
0.0 0.0 1.0
If your matrix is too big to fit into memory and you must use a sparse matrix, then I suspect you don't need all the eigenvalues.
The technical post webpages of this site follow the CC BY-SA 4.0 protocol. If you need to reprint, please indicate the site URL or the original address.Any question please contact:yoyou2525@163.com.