Graph Theory: Cutsets, Bridges & Robustness

The Basics of Social Network Analysis

Part 1: Edge and Node Connectivity (Cutsets & Bridges)

Edge Cutsets & Connectivity

  • We can disconnect a connected graph by removing a subset of its edges.
  • Edge Cutset:
    • Any set of edges whose removal disconnects the graph into two or more components.
  • Minimum Edge Cutset:
    • An edge cutset containing the minimum possible number of edges needed to disconnect the graph.
  • Edge-Connectivity (\(\lambda(G)\)):
    • The number of edges in a minimum edge cutset.
  • Bridge:
    • A single edge whose removal disconnects the graph (a minimum edge cutset of size 1, so \(\lambda(G) = 1\)).

Edge Cutsets: A Visual Example

  • In our 5-node graph on the right:
    • Edges A-C and B-C are highlighted in red (Before).
  • Removing both edges:
    • Disconnects \(\{C, D, E\}\) from \(\{A, B\}\) (After).
    • This is an edge cutset of size 2.

Bridges: A Cutset of Size 1

  • Bridge:
    • A single edge whose removal disconnects the graph (so \(\lambda(G) = 1\)).
  • In our graph on the right, the edge C-D is highlighted in red.
    • Removing edge C-D completely disconnects \(\{A, B, C\}\) from \(\{D, E\}\).
    • Bridges represent the most critical, yet vulnerable transmission links in any network.

Special Properties of Bridges

  • In a connected graph, a bridge (such as the red C-D edge on the right) possesses two interesting mathematical properties (Buckley and Harary 1990):
  • 1. The No-Cycle Rule:
    • A bridge never lies on any cycle of the graph.
    • If it did, removing it would still leave the rest of the cycle intact, failing to disconnect the graph.
  • 2. The Intermediation Bottleneck:
    • The bridge is an edge on every path connecting at least one pair of nodes (e.g., any node in \(\{A, B, C\}\) to any node in \(\{D, E\}\)).
    • Every packet of information flowing between these clusters must pass through this single bottleneck.

Local Bridges

  • In a classic paper on “The Strength of Weak Ties” (1973), Mark Granovetter developed the concept of a local bridge:
    • An edge that, if removed, does not disconnect the entire graph, but significantly increases the shortest path length between its end nodes.
  • Sociological Power:
    • Local bridges act as crucial “shortcuts” that facilitate efficient, cross-group communication.
    • They are the primary channels through which novel information diffuses between separate social clusters.

Local Bridge: A Visual Example

  • Let’s look at the edge C-D highlighted in red:
    • It connects the left cluster \(\{A, B, C\}\) to the right cluster \(\{D, E, F\}\).
  • Is it a bridge? No. If removed, the graph remains connected via the bottom path (\(B-G-H-I-E\)).
  • Is it a local bridge? Yes!
    • Observed distance: \(d(C, D) = 1\).
    • If removed, the distance increases to: \[d(C, D) = 6 \quad (C-B-G-H-I-E-D)\]

Node Cutsets & Connectivity

  • We can also disconnect a graph by removing nodes (vertices).
  • Removing a node also removes all edges incident to it—nodes “take their links with them”!
  • Node Cutset (Vertex Cutset):
    • Any set of nodes whose removal disconnects the remaining graph into two or more components.
  • Minimum Node Cutset:
    • A node cutset containing the minimum possible number of nodes needed to disconnect the graph.
  • Node-Connectivity (\(\kappa(G)\)):
    • The number of nodes in a minimum node cutset.
  • Articulation Node (Cutpoint):
    • A single node whose removal disconnects the graph (so \(\kappa(G) = 1\)).

Articulation Nodes: A Visual Example

  • In our house graph on the right, node E is highlighted in red.
  • Removing node E:
    • Removes all of its incident edges: \(B-E, E-C, E-D\).
    • Completely disconnects node \(B\) from the rest of the network.
    • This is a node cutset of size 1 (Node \(E\) is an articulation node).

Special Properties of Articulation Nodes

  • An articulation node (or cutpoint, such as E on the right) has unique properties mirroring those of bridges:
  • 1. Node Equivalent of a Bridge:
    • A single actor upon whom the entire connectivity of the network depends (\(\kappa(G) = 1\)).
  • 2. The Path Intermediator:
    • It stands as an inner node on every single path connecting some node \(v\) to the rest of the network (e.g., node \(E\) is on every path connecting \(B\) to \(\{A, C, D\}\)).
  • 3. The End-Node Rule:
    • A node of degree \(1\) (an end-node, like \(B\)) can never be an articulation node; removing it always leaves the remaining graph connected.

Node Cutset: An Edge Cutset Comparison

  • Let’s look at node \(A\) in our house graph on the right.
  • Separating \(A\) with an edge cutset requires removing 2 edges: \(C-A\) and \(D-A\) (red).
  • Removing nodes can be more economical: deleting the single node \(E\) destroys three edges at once (\(B-E\), \(E-C\), \(E-D\)) and disconnects node \(B\) from the rest of the network.
  • Key insight: Node removals are doubly destructive — a node “takes its links with it,” so one node removal can do the work of several edge removals.

k-Connected Graphs & Robustness

  • \(k\)-Connected Graph:
    • A graph is \(k\)-connected (or \(k\)-node connected) if the minimum number of nodes required to disconnect it is \(k\).
    • 1-Connected: Removing 1 node disconnects the graph (contains an articulation point).
    • 2-Connected: Requires removing at least 2 nodes to disconnect it.
    • 3-Connected: Requires removing at least 3 nodes to disconnect it.
  • Menger’s Theorem (1927):
    • If a graph is \(k\)-connected, there are at least \(k\) node-independent paths (paths that do not share any common inner nodes) connecting each non-adjacent pair of vertices.
  • Social Cohesion: Higher \(k\)-connectivity indicates a more robust, cohesive, and redundant communication structure.

1-Connectivity: A Visual Example

  • 1-Connected Graph:
    • A graph where removing exactly 1 node can disconnect the network.
  • Before (Left):
    • A 7-node symmetric bow-tie graph, with the central bridge node (node 7) highlighted in red.
  • After (Right):
    • The central node 7 is removed, leaving two completely separate connected components (triangles).

2-Connectivity: A Visual Example

  • 2-Connected Graph:
    • A graph where it requires removing at least 2 nodes to disconnect it.
  • Before (Left):
    • A 6-node cycle with a diagonal chord (edge 2-5), with nodes 2 and 5 highlighted in red.
  • After (Right):
    • Removing nodes 2 and 5 leaves the graph split into two disconnected components: {1, 6} and {3, 4}.

3-Connectivity: A Visual Example

  • 3-Connected Graph:
    • A graph where it requires removing at least 3 nodes to disconnect it.
  • Before (Left):
    • A 3-regular 6-node prism graph (two concentric triangles), with nodes 1, 2, and 6 highlighted in red.
  • After (Right):
    • Removing nodes 1, 2, and 6 leaves the single edge 4-5 and completely isolates node 3 (disconnected).

Graph Cutsets: Quantitative Cohesion

  • Cohesion can be directly translated to graph-theoretic cutsets:
    • Cohesive Groups: Have large edge-cutsets and node-cutsets. Many connections or individuals must be removed to disrupt or fragment the group.
    • Fragmented Groups: Have small cutsets (such as bridges or single cutpoints). They are vulnerable to rapid disruption if key facilitators are removed.
  • Any network can be decomposed into hierarchical subgraphs with specific levels of \(k\)-connectivity.

The Connectivity Inequality

  • For any undirected graph \(G\), node-connectivity (\(\kappa\)), edge-connectivity (\(\lambda\)), and minimum degree (\(\delta\)) obey a fundamental structural hierarchy: \[\kappa(G) \le \lambda(G) \le \delta(G)\]
  • A Strict Inequality Example (\(1 < 2 < 3\)):
    • In our 9-node “double-clique” graph on the right:
      • Minimum Degree (\(\delta = 3\)): The least connected nodes (\(3, 4, 7, 8\)) have exactly 3 edges.
      • Edge-Connectivity (\(\lambda = 2\)): Removing \(\{1-9, 2-9\}\) separates the left clique from the waist. No single bridge exists.
      • Node-Connectivity (\(\kappa = 1\)): Removing the central cutpoint (node 9) completely disconnects the graph.

Wrap-Up: Cutsets & Robustness

  • Cutsets & Robustness: Bridges, local bridges, articulation nodes, \(k\)-connectivity, and the hierarchy \(\kappa \le \lambda \le \delta\).
  • Note: Directed connectivity — the weak, unilateral, strong, and recursive flavors and weakly vs. strongly connected components — is now covered in the Centrality in Networks deck, right before directed closeness.
  • Next up: Subgroups, cliques, and cohesive subsets — the densely connected groups within the network.