Does the NetworkX function return a value between zero and one?

No. It returns the raw sum of endpoint-degree products.

Can graphs with identical degrees have different s values?

Yes. The worked example changes which degrees are connected while preserving every node's degree.

Does a high s value prove a graph is scale-free?

No. Such an interpretation requires a stated theoretical definition and comparison context; the raw number alone is insufficient.

AI Glossary / Graphs and symbolic AI / Intermediate

S metric

Definition

The s metric is a graph statistic obtained by summing the product of the endpoint degrees over all edges. For a simple undirected graph, s(G) = sum over edges {u,v} of degree(u)×degree(v), counting each edge once. It describes one aspect of how a graph's connectivity is arranged, particularly how strongly high-degree endpoints are linked together.

What it means

Two networks can contain the same number of nodes and edges, and even the same degree sequence, while connecting those degrees differently. The s metric can help distinguish those arrangements.

An edge between nodes of degrees ten and eight contributes 80, whereas an edge between degrees ten and one contributes 10. Thus the statistic weights connections involving high-degree endpoints more heavily. Its raw value is not a probability, a distance between graphs, or a score with a universal maximum of one. Interpretation requires context about graph size, degree sequence, and any normalization used.

How it works

Calculate every node's degree in the chosen graph. For each edge, multiply the two endpoint degrees and add the products. In an undirected graph, avoid summing both u - v and v - u as separate edges unless an alternative convention explicitly requires it.

The supplied NetworkX function returns this unnormalized sum using graph degrees. Its implementation does not use an edge-weight attribute in the calculation. This entry focuses on simple undirected graphs so that degree and edge counting are unambiguous.

For meaningful structural comparison, often hold the degree sequence fixed or state a suitable reference construction. The original research discusses normalizations relative to a maximum over a specified graph family; that is a separate calculation from the raw NetworkX result.

Three practical examples

Network comparison: distinguish alternative wiring arrangements with the same degree sequence when examining the placement of highly connected nodes.
Graph-generation studies: track how degree-preserving rewiring changes a structural statistic without changing the number of connections at each node.
Graph-learning diagnostics: describe one connectivity feature of benchmark graphs alongside other measures rather than relying only on node and edge counts.

Think of it this way

Think of scoring every road connection by multiplying the number of roads at its two junctions, then adding those scores. Roads connecting two busy junctions contribute more. The analogy does not measure traffic volume, travel time, or reliability: those quantities are absent unless separately represented and analyzed.

Worked example

Construct a fictional graph with six nodes and edges A - B, A - C, A - D, B - E, and C - F. Node A has degree 3; B and C have degree 2; D, E, and F have degree 1.

The edge contributions are 3×2 = 6 for A - B, 3×2 = 6 for A - C, 3×1 = 3 for A - D, 2×1 = 2 for B - E, and 2×1 = 2 for C - F. Adding them gives s = 19.

Now construct another graph with edges A - B, B - C, A - D, A - E, and C - F. The degree of every labeled node is unchanged: A still has 3, B and C still have 2, and the other nodes still have 1. Its contributions are 6, 4, 3, 3, and 2, giving s = 18.

The first arrangement scores higher because it connects the degree-three node directly to both degree-two nodes. The second places the degree-two nodes in a chain. Both graphs are connected trees with six nodes and five edges, so neither node count, edge count, nor the degree sequence alone explains the difference.

This one-point difference does not establish that the first network is better. If the application concerns latency, physical construction cost, or resilience to particular failures, additional information and analysis are needed. The statistic answers a structural question about degree products, not every operational question about a network.

Illustrative case study

Fictional case: Cedar Graph Classroom compares two proposed dependency diagrams. A student initially labels the larger s value as “more reliable.” Another student notes that the calculation contains no failure probabilities, capacities, or recovery policies.

The group revises its report to state the exact edge lists, degree sequence, and raw s values. It then treats reliability as a separate question and defines what failures and outcomes would be examined. This prevents a descriptive statistic from being turned into an unsupported business conclusion.

The case is a teaching scenario, not evidence that increasing or decreasing s improves a real system.

Common misconceptions and mistakes

Raw s values should not be compared casually across graphs of different sizes or degree distributions. A larger graph or higher degrees can increase the sum even without the structural interpretation intended by the comparison.

Do not equate the s metric directly with the degree assortativity coefficient. They are related in certain fixed-degree settings, but the correlation coefficient includes a different normalization and has different interpretation and edge cases.

The word metric also does not make s a mathematical distance between two graphs. It assigns one value to one graph, and different graphs can share that value.

Business use

The s metric can add a compact descriptor to a broader network analysis. Its transparency is useful: reviewers can trace the total to individual edges and understand which connections contribute most.

Record whether the graph is directed, weighted, or allows parallel edges, and make any conversion explicit before using the simple undirected interpretation. If reporting a normalized variant, define the denominator and permitted graph family. Avoid presenting a raw library return value as though it were already scaled for comparison across unrelated networks.

Check your understanding

An undirected edge joins nodes of degrees four and three. How much does that edge contribute to the stated s metric?

Answer: It contributes 12, counted once. The degrees are computed in the whole graph being analyzed, not only in an isolated drawing of that edge.

Related terms

Assortativity
Rich Club
Normalization
Degree sequence
graph statistic
degree-preserving rewiring
graph topology
scale-free network

Learn next

Verify both edge lists and degree products in the worked example. Then perform another degree-preserving rewiring and recalculate s. Study assortativity and the original paper's normalized structural measures next, keeping their definitions separate from the unnormalized sum.

Assortativity

Sources

NetworkX, s_metric documentation and implementation: https://networkx.org/documentation/stable/reference/algorithms/generated/networkx.algorithms.smetric.s_metric.html Li and colleagues, Towards a Theory of Scale-Free Graphs: https://arxiv.org/html/cond-mat/0501169v2 Official formula and source code were read; original research context was consulted. Graphs, arithmetic, and fictional case are original.

Previous: S chart Next: S-set