line_graph#
- line_graph(G, create_using=None)[source]#
Returns the line graph of the graph or digraph
G
.The line graph of a graph
G
has a node for each edge inG
and an edge joining those nodes if the two edges inG
share a common node. For directed graphs, nodes are adjacent exactly when the edges they represent form a directed path of length two.The nodes of the line graph are 2-tuples of nodes in the original graph (or 3-tuples for multigraphs, with the key of the edge as the third element).
For information about self-loops and more discussion, see the Notes section below.
- Parameters:
- Ggraph
A NetworkX Graph, DiGraph, MultiGraph, or MultiDigraph.
- create_usingNetworkX graph constructor, optional (default=nx.Graph)
Graph type to create. If graph instance, then cleared before populated.
- Returns:
- Lgraph
The line graph of G.
Notes
Graph, node, and edge data are not propagated to the new graph. For undirected graphs, the nodes in G must be sortable, otherwise the constructed line graph may not be correct.
Self-loops in undirected graphs
For an undirected graph
G
without multiple edges, each edge can be written as a set{u, v}
. Its line graphL
has the edges ofG
as its nodes. Ifx
andy
are two nodes inL
, then{x, y}
is an edge inL
if and only if the intersection ofx
andy
is nonempty. Thus, the set of all edges is determined by the set of all pairwise intersections of edges inG
.Trivially, every edge in G would have a nonzero intersection with itself, and so every node in
L
should have a self-loop. This is not so interesting, and the original context of line graphs was with simple graphs, which had no self-loops or multiple edges. The line graph was also meant to be a simple graph and thus, self-loops inL
are not part of the standard definition of a line graph. In a pairwise intersection matrix, this is analogous to excluding the diagonal entries from the line graph definition.Self-loops and multiple edges in
G
add nodes toL
in a natural way, and do not require any fundamental changes to the definition. It might be argued that the self-loops we excluded before should now be included. However, the self-loops are still â€śtrivialâ€ť in some sense and thus, are usually excluded.Self-loops in directed graphs
For a directed graph
G
without multiple edges, each edge can be written as a tuple(u, v)
. Its line graphL
has the edges ofG
as its nodes. Ifx
andy
are two nodes inL
, then(x, y)
is an edge inL
if and only if the tail ofx
matches the head ofy
, for example, ifx = (a, b)
andy = (b, c)
for some verticesa
,b
, andc
inG
.Due to the directed nature of the edges, it is no longer the case that every edge in
G
should have a self-loop inL
. Now, the only time self-loops arise is if a node inG
itself has a self-loop. So such self-loops are no longer â€śtrivialâ€ť but instead, represent essential features of the topology ofG
. For this reason, the historical development of line digraphs is such that self-loops are included. When the graphG
has multiple edges, once again only superficial changes are required to the definition.References
Harary, Frank, and Norman, Robert Z., â€śSome properties of line digraphsâ€ť, Rend. Circ. Mat. Palermo, II. Ser. 9 (1960), 161â€“168.
Hemminger, R. L.; Beineke, L. W. (1978), â€śLine graphs and line digraphsâ€ť, in Beineke, L. W.; Wilson, R. J., Selected Topics in Graph Theory, Academic Press Inc., pp. 271â€“305.
Examples
>>> G = nx.star_graph(3) >>> L = nx.line_graph(G) >>> print(sorted(map(sorted, L.edges()))) # makes a 3-clique, K3 [[(0, 1), (0, 2)], [(0, 1), (0, 3)], [(0, 2), (0, 3)]]
Edge attributes from
G
are not copied over as node attributes inL
, but attributes can be copied manually:>>> G = nx.path_graph(4) >>> G.add_edges_from((u, v, {"tot": u+v}) for u, v in G.edges) >>> G.edges(data=True) EdgeDataView([(0, 1, {'tot': 1}), (1, 2, {'tot': 3}), (2, 3, {'tot': 5})]) >>> H = nx.line_graph(G) >>> H.add_nodes_from((node, G.edges[node]) for node in H) >>> H.nodes(data=True) NodeDataView({(0, 1): {'tot': 1}, (2, 3): {'tot': 5}, (1, 2): {'tot': 3}})