0
$\begingroup$

The neighbourhood of node $v$ in graph $G$ is the subgraph of $G$ induced by all vertices adjacent to $v$.

enter image description here enter image description here enter image description here

The number of edges between neighbours divided by the number of pairs of neighbours is the so-called clustering coefficient $c$. The first node has $c = 6 / \binom{6}{2} = 0.4$, the second $c = 2 / \binom{5}{2} = 0.2$, the third $c = 0$

You now may add nodes at larger distances from node $v$ to its neighbourhood when they lie on minimal cycles going through $v$:

enter image description hereenter image description here

My first question is: What's the official name of such "closed neighbourhoods"? For tilings the closed neighbourhood of a vertex is essentially its vertex configuration, but what about the general case? Would one call the closed neighbourhood "vertex configuration", too?

In any case, the concept of clustering coefficients now can be generalized. Let me first note that the name "clustering coefficient" is not the best that could have been chosen. It is true that triangles are complete graphs $K_3$ and thus are clusters, but they are also cycles, i.e. cycle graphs $C_3$, and so "cycle coefficient" would have been an equivalent name – or to be specific: "3-cycle coefficient".

The concept of clustering coefficients – considered as cycle coefficients – can be generalized quite naturally: The $k$-cycle coefficient $c_k$ of a node $v$ is the number of induced minimal $k$-cycles $C_k$ it lies on, divided by the number of possible such $k$-cycles. The overall cycle coefficient $\mathbf{c}$ can then be defined as the sum $\mathbf{c} = \sum_{k=3}^n c_k$ and cannot exceed 1 which is achieved exactly for the nodes of the complete graph $K_n$, for which $c_k = \delta_{k3}$. For the cycle graph $C_n$ the cycle coefficient of all nodes is $\mathbf{c} = 1 / n!$. For trees we have of course $\mathbf{c} = 0$.

Reference request: Under which name (and where) are such "cycle coefficients" investigated in the literature?

$\endgroup$
7
  • 1
    $\begingroup$ I think one obstacle here is the definition of a `minimal cycle'. It's obvious how to choose these in sufficiently nice embeddings of a sufficiently connected graph into a surface (as you have done with the tiling examples), but for more general graphs we need a different notion of what cycles to count. $\endgroup$ Commented Sep 21, 2020 at 13:48
  • $\begingroup$ @BrandonduPreez: What would be wrong with this definition: A cycle is minimal iff the subgraph induced by it is a cycle, too? In other words: A cycle is minimal iff it is isomorphic to the subgraph induced by it. $\endgroup$ Commented Sep 21, 2020 at 15:47
  • 1
    $\begingroup$ @BrandonduPreez: Since you seem to be interested in distances in graphs you might want to have a look at this question The distance distribution of graphs. And maybe also at this interactive visualization tool, especially for distances in graphs: SRGX $\endgroup$ Commented Sep 21, 2020 at 15:56
  • $\begingroup$ That definition is fine, but it is worth noting that even for 2-connected graphs in the plane, the cycles you end up with are not the same as those considered by the vertex configuration (you can have chords that lie `outside' the faces incident with a vertex). $\endgroup$ Commented Sep 21, 2020 at 15:57
  • $\begingroup$ Thanks for the recommend, I'll give it a look. $\endgroup$ Commented Sep 21, 2020 at 15:57

0

You must log in to answer this question.

Browse other questions tagged .