MATLAB Answers

Properties of adjacency matrix

17 views (last 30 days)
Marc Laub
Marc Laub on 17 Jul 2020
Edited: Bruno Luong on 28 Jul 2020
Hey
following thought about the adjacency matrix of a graph.
Is it possible to distinguish from the adjacency matrix of a graph if the whole system of points is interconnected, or if there are 2 or more subsystems, whcih have connections inside each subsystem but the subsystems are not connected to each other.
Which properties does the adjacency matrix need to represent a interconnected system, or where do i have to seed the "ones" in the matrox to ensure the whole system is interconnected?
I havent figured it out on my own my some example systems, but maybe someone has the answer on this already.
many thanks in advance

Answers (4)

Bruno Luong
Bruno Luong on 17 Jul 2020
Checkout conncomp
  3 Comments
Steven Lord
Steven Lord on 17 Jul 2020
What "special properties" did you need that prevented you from using graph or digraph? If you need the nodes and/or edges to have additional information associated with them, you can do that. See this documentation page for details.

Sign in to comment.


Christine Tobler
Christine Tobler on 27 Jul 2020
In terms of the adjacency matrix, a disconnected graph means that you can permute the rows and columns of this matrix in a way where the new matrix is block-diagonal with two or more blocks (the maximum number of diagonal blocks corresponds to the number of connected components).
If you want to compute this from scratch, you'll be better off using graph-style algorithms instead of matrix terminology, specifically a breadth-first search or depth-first search. These can be implemented in terms of the adjacency matrix, although it will be less efficient than the built-in used in the graph object. See https://en.wikipedia.org/wiki/Component_(graph_theory) which has some discussion of the algorithms involved.
  1 Comment
Bruno Luong
Bruno Luong on 27 Jul 2020
Here is a quick and dirty check of graph connexity using adjacent matrix. But I agree why not using stock function that is much better implemented (there is also some code on the FEX using graph algo, that once I checked out - can't remember the name - but I ended up using my own on implementation)
% A is the adjacent matrix, assumed to be symmetric (undirect graph)
n = size(A,1);
x = zeros(n,1);
x(1) = 1;
c = 1; % sum(x)
while true
x = A*x > 0 | x;
s = sum(x);
if s == c
break
end
c = s;
end
if c == n
fprintf('Single component\n');
else
fprintf('Multiple components\n');
end

Sign in to comment.


Bruno Luong
Bruno Luong on 27 Jul 2020
Edited: Bruno Luong on 27 Jul 2020
Here is a quick and dirty calculation of connexed component using adjacent matrix. It returns the same output as MATLAB conncomp function in BINS and BINSIZES.
% TMW example
s = [1 2 2 3 3 3 4 5 5 5 8 8 9];
t = [2 3 4 1 4 5 5 3 6 7 9 10 10];
G = digraph(s,t,[],20);
A = G.adjacency;
A = spones(A + A'); % no need to symmetrize for undirected graph
n = size(A,1);
bins = zeros(n,1);
k = 0;
while any(bins==0)
x = sparse(find(~bins,1), 1, true, n, 1, n);
y = x;
while true
y = A*y;
y = y > 0 & ~x;
if nnz(y) == 0
break
end
x = x | y;
end
k = k+1;
bins(x) = k;
end
% Formating output
nodes = (1:n)';
Comptable = table(nodes, bins);
CompSet = accumarray(bins, nodes, [], @(x) {sort(x)'});
binsizes = accumarray(bins, 1);
Display
plot(G,'Layout','layered')
k = length(CompSet);
fprintf('Graph has %d bins(s)\n', k);
Comptable
binsizes
for i=1:k
fprintf('Compnent %d: Nodes = %s\n', i, mat2str(CompSet{i}));
end

Bruno Luong
Bruno Luong on 27 Jul 2020
Edited: Bruno Luong on 27 Jul 2020
Another way - most direct perhaps link to matrix property - is using this property of Laplacian matrix:
"The number of connected components in the graph is the dimension of the nullspace of the Laplacian and the algebraic multiplicity of the 0 eigenvalue." from https://en.wikipedia.org/wiki/Laplacian_matrix
Not sure how is the numerical stability (probably not very reliable).
% TMW example
s = [1 2 2 3 3 3 4 5 5 5 8 8 9];
t = [2 3 4 1 4 5 5 3 6 7 9 10 10];
G = graph(s,t);
A = G.adjacency;
Use Laplacian
D = diag(sum(A)); % degree matrix
L = D - A; % laplacian matrix
[Q,R,P] = qr(L);
nc = full(sum(abs(diag(R)) < eps)) % number of components
  1 Comment
Bruno Luong
Bruno Luong on 28 Jul 2020
A basis of the null space of the Laplacian is actually the indicator (characteristic) functions of each connected component of G.

Sign in to comment.

Products


Release

R2019b

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!