Institute for Design and Control of Mechatronical Systems
Introduction aux Systèmes Collaboratifs Multi-Agents
UPJV, Département EEA
Fabio MORBIDI
Laboratoire MIS
Équipe Perception Robotique E-mail: [email protected]
Année Universitaire 2016/2017
Jeudi 13h30-16h30, Salle 8
M1 EEAII - Découverte de la Recherche (ViRob)
Graph theory: introduction
Graphs provide natural abstractions for how information is shared between agents in a network
The graph-based abstraction contains high-level descriptions of the network topology in terms of objects referred to as vertices and edges
In the next slides we will see:
Basic notions of:
Algebraic graph theory (adjacency, incidence, Laplacian matrices) Spectral graph theory (spectrum of the Laplacian)
i.e. = id est (latin) = à savoir, c’est à dire e.g. = exempli gratia (latin) = par exemple Remark:
Institute for Design and Control of Mechatronical Systems
Harald Kirchsteiger 2011/04
Graph theory
A finite, undirected, simple graph, or graph for short, is built upon a finite set:
3
The graph is then formally defined as the pair :
the vertex (or node) set.
We also define the edge set:
It consists of elements of the form or such that
Fabio Morbidi
Graph theory
When an edge exists between vertices and , they are called adjacent and denoted . In this case is called incident with vertices Example
The neighborhood of vertex is the set i.e., the set of all vertices adjacent to . In the example:
Institute for Design and Control of Mechatronical Systems
Harald Kirchsteiger 2011/04
Graph theory
A path of length
m
in is given by a sequence of distinct verticessuch that for , the vertices and are adjacent.
5
When the vertices of the path are distinct except for its end vertices, the path is called a cycle.
A graph is called connected if for every pair of vertices in , there is a path that has them as end vertices. Otherwise the graph is called
disconnected (e.g. the graph below is connected).
Fabio Morbidi
Cycle
1
2
3 5 4
Petite historique
The problem of the “7 bridges of Königsberg“ : first studied in graph theory Find a walk through the city that crosses each bridge once and only once
(i.e. find an Eulerian cycle in a graph)
Negative answer by Leonhard Euler in 1735
An Eulerian cycle exists if the graph is connected and has exactly zero or two nodes of odd degree (in the problem of Königsberg, 4 nodes have odd degree)
Leonhard Euler
Graph theory
(1707-1783)
Institute for Design and Control of Mechatronical Systems
Harald Kirchsteiger 2011/04 7
: Complete or fully connected graph Each vertex is adjacent to every other vertex Example:
Fabio Morbidi
(7 vertices)
Standard graphs
Standard graphs
: Path graph
where if and only if Example:
:
n
-cyclewhere if and only if Example:
Note:
if is an integer multiple of n (i.e. and are congruent modulo n)
Institute for Design and Control of Mechatronical Systems
Harald Kirchsteiger 2011/04
Standard graphs
9
: Star graph
where if and only if
Examples:
Fabio Morbidi
Generalizations of the notion of graph
When the edges in a graph are given directions, the resulting interconnection is no longer considered an undirected graph.
A directed graph (or digraph), denoted by can be obtained in two ways:
1. Drop the requirement that the edge set
E
contains unordered pairs of verticesIf the ordered pair then is said to be the tail (where the arrow starts) of the edge, while is its head (where the arrow ends) 2. Associate an orientation to the unordered edge set Directed graphs (digraphs)
Orientation +1 Orientation -1
Institute for Design and Control of Mechatronical Systems
Harald Kirchsteiger 2011/04
Generalizations of the notion of graph
11
Directed graphs (digraphs) Example:
or
Fabio Morbidi
Generalizations of the notion of graph
Directed graphs (digraphs)
The notions of adjacency, neighborhood, and connectness can be extended in the context of digraphs.
For example, a directed path of length m in is given by the sequence of distinct vertices:
such that for the vertices
A digraph is called:
a) Strongly connected: if for every pair of vertices there is a directed path between them
b) Weakly connected: if it is connected when viewed as a graph, i.e.
a “disoriented“ graph
For example, the graph in the previous slide is weakly connected but not strongly connected
Institute for Design and Control of Mechatronical Systems
Harald Kirchsteiger 2011/04
Graphs and matrices
13
Graphs admit a representation in terms of matrices, beside a graphical representation in terms of vertices and edges.
For an undirected graph , the degree of a given vertex, , is the cardinality of the neighborhood set , i.e., the number of vertices that are adjacent to vertex in .
Example:
Fabio Morbidi
Degree and adjacency matrix
The degree matrix of a graph is the
n × n
diagonal matrix containing the vertex-degree of on the diagonal:The adjacency matrix is the symmetric
n × n
matrix encoding the adjacency relationships in the graph :Institute for Design and Control of Mechatronical Systems
Harald Kirchsteiger 2011/04
Degree and adjacency matrix
15
Example:
Symmetric matrix
Fabio Morbidi
,
Exercise:
Compute and for
1
2 3 4
5
6
1
2 3 4
5
6
6
×
6 identity matrixDegree and adjacency matrix
Institute for Design and Control of Mechatronical Systems
Harald Kirchsteiger 2011/04
Incidence matrix
17
Remark:
captures not only the adjacency relationships, but also the orientation that the graph now enjoys
Under the assumptions that labels have been associated with edges in a graph whose edges have been arbitrarily oriented, the
n × m
incidence matrix is defined as (n
= number of vertices;m
= number of edges):Fabio Morbidi
Example:
Vertices
Edges
Incidence matrix
Institute for Design and Control of Mechatronical Systems
Harald Kirchsteiger 2011/04
Incidence matrix
19
Remarks:
has a column sum equal to zero.
The incidence matrix of a digraph can be defined analogously by skipping the pre-orientation that is needed for graphs.
• The incidence matrix is denoted in this case by
Fabio Morbidi
Graph Laplacian
The graph Laplacian associated with an undirected graph is:
From this definition, it follows that for all graphs, the rows of the Laplacian sum zero
Example:
Pierre-S. Laplace
where
(1749-1827)
Institute for Design and Control of Mechatronical Systems
Harald Kirchsteiger 2011/04
Graph Laplacian: alternative definition
21
Given an arbitrary orientation to the edge set , the graph Laplacian of can be alternatively defined as:
This definition reveals that is a:
• Symmetric matrix (i.e. )
• Positive semidefinite matrix (i.e. )
The two definitions that we have given are equivalent and since no notion of orientation is needed in the first one, the graph Laplacian is orientation independent
Fabio Morbidi
Weighted graph Laplacian
Given a weighted graph the weighted graph Laplacian is defined as:
where
Institute for Design and Control of Mechatronical Systems
Harald Kirchsteiger 2011/04
Weighted Laplacian for digraphs
23
Let be a weighted digraph.
For the adjacency matrix, we let:
and for the diagonal degree matrix , we set:
where
is the weighted in-degree of vertex .
Fabio Morbidi
The corresponding (in-degree) weighted Laplacian is defined as:
Weighted Laplacian for digraphs
Example:
5 2
7
Not symmetric !!
Institute for Design and Control of Mechatronical Systems
Harald Kirchsteiger 2011/04
Weighted Laplacian for digraphs
25
Exercise:
3 1
4
Compute , and for the digraph reported below:
2
Fabio Morbidi
Algebraic and spectral graph theory
Algebraic graph theory associates algebraic objects (e.g., the degree, adjacency, incidence and Laplacian matrices) to graphs Spectral graph theory studies the eigenvalues associated to the
adjacency and Laplacian matrices
Definition
A nonzero vector is a eigenvector of a matrix if and only if there exists a scalar such that:
where is called eigenvalue associated to .
We find the eigenvalues and eigenvectors of
by solving the equation:
Recall that ….
Institute for Design and Control of Mechatronical Systems
Harald Kirchsteiger 2011/04
Algebraic and spectral graph theory
27
Consider the graph Laplacian . This matrix is symmetric and positive semidefinite, hence its
real eigenvalues can be ordered as:
Theorem:
The graph is connected if and only if
Fabio Morbidi
Spectrum of the Laplacian matrix
• To find the Laplacian spectrum of an arbitrary graph is not trivial
• For some special graphs we can easily compute the eigenvalues (and the associated eigenvectors)
Since the spectrum of is that of shifted by
n
. Since the spectrum of the matrix isthe Laplacian spectrum of is:
Complete graph
n
-cycleThe Laplacian spectrum of is:
Institute for Design and Control of Mechatronical Systems
Harald Kirchsteiger 2011/04
Algebraic and spectral graph theory
29
is the second smallest eigenvalue of the Laplacian is called Fiedler value and the associated eigenvector is called Fiedler vector
The Fiedler value is important not only as a measure of robustness or level of connectedness of a graph, but also for the convergence properties of a collection of distributed coordination algorithms:
Remark:
Miroslav Fiedler
Fabio Morbidi
Consensus protocols
(1926 – 2015)