I know that to generate a block-diagonal matrix in Matlab the command blkdiag
generates such a matrix:
There is a submission on the File Exchange that can do this: (Block) tri-diagonal matrices.
You provide the function with three 3D-arrays, each layer of the 3D array represents a block of the main, sub- or superdiagonal. (Which means that the blocks will have to be of the same size.) The result will be a sparse matrix, so it should be rather efficient in terms of memory.
An example usage would be:
As = bsxfun(@times,ones(3),permute(1:3,[3,1,2]));
Bs = bsxfun(@times,ones(3),permute(10:11,[3,1,2]));
M = blktridiag(As, zeros(size(Bs)), Bs);
where full(M)
gives you:
1 1 1 10 10 10 0 0 0
1 1 1 10 10 10 0 0 0
1 1 1 10 10 10 0 0 0
0 0 0 2 2 2 11 11 11
0 0 0 2 2 2 11 11 11
0 0 0 2 2 2 11 11 11
0 0 0 0 0 0 3 3 3
0 0 0 0 0 0 3 3 3
0 0 0 0 0 0 3 3 3
This could be one approach based on kron, tril & triu -
%// Take all A1, A2, A3, etc in a cell array for easy access and same for B
A = {A1,A2,A3,A4}
B = {B1,B2,B3}
%// Setup output array with the A blocks at main diagonal
out = blkdiag(A{:})
%// logical array with 1s at places where kth diagonal elements are to be put
idx = kron(triu(true(numel(A)),k) & tril(true(numel(A)),k),ones(size(A{1})))>0
%// Put kth diagonal blocks using the logical mask
out(idx) = [B{1:numel(A)-k}]
Sample run with k = 1
for 2 x 2
sizes matrices -
>> A{:}
ans =
0.3467 0.7966
0.6228 0.7459
ans =
0.1255 0.0252
0.8224 0.4144
ans =
0.7314 0.3673
0.7814 0.7449
ans =
0.8923 0.1296
0.2426 0.2251
>> B{:}
ans =
0.3500 0.9275
0.2871 0.0513
ans =
0.5927 0.8384
0.1629 0.1676
ans =
0.5022 0.3554
0.9993 0.0471
>> out
out =
0.3467 0.7966 0.3500 0.9275 0 0 0 0
0.6228 0.7459 0.2871 0.0513 0 0 0 0
0 0 0.1255 0.0252 0.5927 0.8384 0 0
0 0 0.8224 0.4144 0.1629 0.1676 0 0
0 0 0 0 0.7314 0.3673 0.5022 0.3554
0 0 0 0 0.7814 0.7449 0.9993 0.0471
0 0 0 0 0 0 0.8923 0.1296
0 0 0 0 0 0 0.2426 0.2251