LowRankFactorizedNEP creates a dense matrix

[jarlebring]

One of the constructors of LowRankFactorizedNEP creates a dense matrix.

NonlinearEigenproblems.jl/src/low_rank_nep.jl

Lines 38 to 42 in 0368485

	function LowRankFactorizedNEP(L::AbstractVector{S}, U::AbstractVector{S},
	f) where {S<:AbstractMatrix}
	A = L .* adjoint.(U);
	return LowRankFactorizedNEP(L,U,A,f)
	end

This will be slow. Better to use e.g. LowRankApprox.LowRankMatrix

[jarlebring]

Workaround

using LowRankApprox
import NonlinearEigenproblems.LowRankFactorizedNEP
function NonlinearEigenproblems.LowRankFactorizedNEP(L::AbstractVector{S}, U::AbstractVector{S},
                              f) where {S<:AbstractMatrix}
    A=map(i-> LowRankMatrix(L[i],U[i]),1:size(L,1))
    rnk = mapreduce(u -> size(u, 2), +, U)
    spmf=SPMF_NEP(A, f, align_sparsity_patterns=false, check_consistency=false);
    return LowRankFactorizedNEP(spmf,rnk,L,U)
end