How to update individual sparse matrices mapped to sparse diagonal-matrix efficiently?

General Usage
1

I have a simple MWE below that includes a diagonal sparse matrix A and its sub-diagonal matrices A_1,A_2,A_3. Assuming that A is going to be updated in values only (pattern is fixed), what is the efficient way to update A_1,A_2,A_3 correspondingly cause the below code for updating them is not efficient by using nzval field of the sparse matrices?

#### Define section
A = sparse([100.0 200 0 0 0 0 0; 16 17 0 0 0 0 0; 0 0 300 4 5 0 0; 0 0 6 7 8 0 0; 0 0 13 14 15 0 0; 0 0 0 0 0 9 10; 0 0 0 0 0 11 12]);
n0=1;n1=3;n2=6;n3=7;
A_1 = A[n0:n1-1,n0:n1-1];
A_2 = A[n1:n2-1,n1:n2-1];
A_3 = A[n2:n3,n2:n3];

#### Update section
# First, update A values
# Second, update the individual matrices
A_1 .= A[n0:n1-1,n0:n1-1];
A_2 .= A[n1:n2-1,n1:n2-1];
A_3 .= A[n2:n3,n2:n3];
2

Maybe just use @views:

@views begin
    A_1 = A[n0:n1-1,n0:n1-1];
    A_2 = A[n1:n2-1,n1:n2-1];
    A_3 = A[n2:n3,n2:n3];
end

Then these will be views, not copies, and when you update A the changes will be immediately reflected in A_1 etcetera.

1 Like
3

Thanks for your reply.

Yes, but actually, I need to use these matrices with KLU which makes conflict with that as below. So, I am trying to seek another way manually.

A = sprand(10,10, 0.5)
julia> klu(@view(A[1:5,1:5]))
ERROR: MethodError: no method matching klu(::SubArray{Float64, 2, SparseMatrixCSC{Float64, Int64}, Tuple{UnitRange{Int64}, UnitRange{Int64}}, false})