Complex Networks: Lecture 3b: Introduction to Graph Theory
EE 6605
Instructor: G Ron Chen
Most pictures on this ppt were taken from un-copyrighted websites on the web with thanks
Trail: A walk in which all edges are distinct.
Review
Walk: A finite sequence of edges, one after another, in the form of v v ,v v ,…,v v where N(G) v ,v ,…,v are nodes.
1223 n1n 12n
12n
A walk is denoted by v v v and the number of
edges in a walk is called its length.
Path: A trail in which all nodes are distinct, except perhaps v v which is called a closed path, often
1n
called a circuit (or a cycle, a loop).
Tree: A graph with no circuits.
Eulerian Graphs
Eulerian graph: If it has a closed trail that traverses every edge once and once only
There may be more than one such trail, each of which is called an Eulerian trail (or, Euler Circuit)
Semi-Eulerian graph: If it has a trail, need not be closed, that traverses each edge once and once only
Main Results
Theorem: A connected graph is Eulerian if and only if the every node in the graph has an even degree.
Proof: If a note is a starting point, then the trail has to return to it:
If a note has a trail enters it, then the trail has to leave it:
Corollary: The seven-bridge problem of Königsburg has no solutions.
Proof: It is not Eulerian.
Example:
Algorithm to find Eulerian Graphs
Corollary (Fleury Algorithm) In any given Eulerian graph G, an Eulerian trail can be found by the following procedure:
Start from any node in G. Walk along the edges of G in an arbitrary manner, subject to the following rules:
erase the edges as they are traversed;
erase the resulting isolated nodes;
walk through a bridge only if there are no other alternatives.
Example:
Example
(path usually not unique)
Sir William Rowan Hamilton (1805-1865)
In 1856, the English mathematician William R. Hamilton studied the world navigation problem and considered a map with 20 nodes representing cities connected by sailing routes, as depicted by the following figure. He wanted to find out if one can traverse through every city once and once only, and finally return to the starting city. His study eventually led to the establishment of the now- famous Hamiltonian graph theory.
One solution
Solution not unique
Another Example
Hamiltonian Graphs
Hamiltonian graph: If it has a closed trail that traverses each node once and once only
A trivial case is a single node and a nontrivial Hamiltonian graph must be a circuit, called a Hamiltonian circuit
Semi-Hamiltonian graph: If it has a trail, need not be closed, that traverses each node once and once only
(3)
Some Results
k(v) k(u) N vvv
123
k(v )k(v )3 v v
v 1313i
vi
| v1 v2 v3
k(v ) k(v ) 3 v v 1i 1i
k(v ) k(v ) 3
1i 1i
… vi ||
v1 v2 v3
For another proof, see Wiki
v v
Corollary:LetGbeasimplegraphwithN(3) nodes.
If every pair of nodes has degree , then G is
Hamiltonian. [Note: “every pair” contains “every non-adjacent pair”]
The above conditions are sufficient but not necessary
For example: N = 20
k(v) k(u) N k(v) N / 2
* There is no “if and only if” condition for Hamiltonian graphs so far
Travelling Salesman Problem
Travelling Salesman Problem –
Given a set of cities and the distances between each pair of cities, what is the shortest possible route that visits every city once and once only, and then returns to the starting city?
It is called an optimal Hamiltonian graph problem, which is NP-hard
A special case: The given cities are fully connected Fully-connected graphs are called complete graphs
Travelling Salesman Problem: On complete graphs Nearest-Neighbor Algorithm: To find a near-optimal solution
Start with an arbitrary node, v0. Find one adjacent edge with the smallest weight, which connects to node v1, so as to have v0 – v1
Start with v1, repeat the same process, … until having v0 -v1 -…-vn
If vn still has nearest neighbor(s) not traversed, then continue; If vn does not have anymore nearest neighbor not traversed,
then connect vn back to v0 Solution: v0 -v1 -…-vn -v0
Travelling Salesman Problem: A complete graph Example:
A
9 12
E 9 B10
7
13
8
5 6
D 11 C
Start from A
A–D–B–C–E–A
Total path length = 7 + 6 + 8 + 5 + 9 = 35
= Is this a Hamiltonian graph ?
Knight’s Tour Game
Can the knight visit every block, once and once only, and finally return to the starting point?
5 X 5 has open solution (semi-Hamiltonian) but no closed solution (Hamiltonian)
8 X 8 – Yes, it always has a closed solution Namely, it is Hamiltonian (proved by Euler in 1759)
There are 26,534,728,821,064 Hamiltonian paths
8 X 8 – Yes, it is Hamiltonian This shows one solution
Theorem (Cull and de Curtins,1978) On any rectangular board whose smaller dimension is at least 5, there exists a (possibly open) knight’s tour (= semi-Hamiltonian)
mn a closed knight’s tour is always possible (= Hamiltonian),
Theorem (Schwenk, 1991) For any m × n board with
except for the following few cases: (1) m and n are both odd
(2) m = 1, 2, or 4
(3) m = 3 and n = 4, 6, or 8
Note:
There is always a solution, which does not mean that starting anywhere will be a solution (Reason: 8X8 is a square, but the walking step of a knight is not a square)
Example
Wiki
24 x 24
Shortest Path Length Problem
Q: What is the shortest path length from A to L ?
B2D3G5J
3
45
15 A9H9
E
2
63 1
2
L
6
C 9 F2I2K
Solving the Shortest Path Length Problem Dijkstra’s Algorithm
Dijkstra’s algorithm is a greedy algorithm (which needs to explore all nodes and all edges).
Dijkstra’s algorithm is adopted by edge-state routing protocols, OSPF (Open Shortest Path First), IS-IS (Intermediate System to Intermediate System), etc.
Dijkstra’s algorithm cannot handle negative edge weights (for which Bellman–Ford algorithm works)
Edsger Wybe Dijkstra (1930-2002)
Solving the Shortest Path Length Problem
Dijkstra’s Algorithm: Moving from A toward L and associating each intermediate node V with a number l(V ) that is equal to the shortest path length from A to V.
Example: From A to J, one has l(J ) l(G) 5 or l(J ) l(H ) 5
whichever is shorter. B2D3G5J
3
A
4155
E 9H9
2
663 1
L
2
C 9 F2I2K
Solving the Shortest Path Length Problem
Starting from A with l(A) 0 , moving from A toward L step by step, one has l(B) l(A) 3 3 l(E) l(A) 9 9 l(C) l(A) 2 2
Therefore, node B is chosen, with l(B) 3 , node C is chosen, with l(C) 2 and node E is chosen with l(E) 9
B2D3G5J
3
A
4155
E 9H9
2 613
L
2
C 9 F2I2K
6
Solving the Shortest Path Length Problem Looking at the nodes adjacent to B, one has
l(D) l(B) 2 5
Looking at the nodes adjacent to C, one has
l(E) l(B) 4 7 l(F) l(C) 9 11
l(E) l(C) 6 8
So, l(D) 5 is uniquely determined.
But l(E)7,or8or9.Hence,nodeEischosenwithl(E)l(B)47
so that
l(H ) l(E) 2 9
F is chosen with l(F) 11
But also l(F) l(H) 1 so F is chosen with B2D3G5J
3
A
l ( F ) l ( H ) 1 10 4155
E 9H9
26 613
L
2
C 9 F2I2K
Solving the Shortest Path Length Problem
Continuing this way, one finally obtains
l(A) 0 l(B) 3 l(C) 2 l(D) 5 l(E) 7 l(F) 10
l(L) 17
Result: total shortest path length = 17 from A to L :
l(J) 13 l(K) 14 B2D3G5J
l(G)8 l(H)9 l(I)12
3 A9H9
4155
E
2
663 1
L
2
C 9 F2I2K
The above provides all shortest lengths from A to any other nodes Solutions may not be unique, but all will be shortest
Chinese Postman Problem
The problem is for a postman to deliver letters in such a way that he passes every street at least once and finally returns to the starting point (the post office), traversing a shortest possible total path-length.
First solution given by Prof Mei-ko Kwan (1934—) (管梅谷 from Shanghai)
Chinese Postman Problem
Ideas:
To have a solution, the graph must be Eulerian
To make the graph be Eulerian, some edges need to be added to nodes with odd degrees
The postman must traverse on every added edge (otherwise, they do not need to be added).
The choice with the least total length of added edges provides an optimal solution.
Chinese Postman Problem
Example:
There are 3 options to make all node degrees become even:
(1) add edges AC and EG (2) add edges AEG and CE (3) add edges AE and CG
Note:
Adding an edge does not mean to build a street, but just to walk through it twice
Chinese Postman Problem
Let T be the total time for the postman to walk through all original streets (T = 97 mins is not important in this analysis)
(1) T + (AC + EG) = T + 17 mins
(2) T+(AEG+CE) = T+20mins
(3) T + (AE + CG) = T + 16 mins optimal solution
Kuan Mei-Ko (2016)
BREAK
10 minutes
Directed Graphs
Digraph (= directed graph): edge (u,v): uv v
u
simple digraph
Underlying graph:
non-simple graph
Two digraphs are isomorphic if their underlying graphs are isomorphic
Degrees
In-degree: d_in
Out-degree: d_out
Total degree: d = d_in + d_out Balanced, if d_in = d_out
The following can be similarly defined:
Directed walk
Directed trail
Directed path
Directed circuit (cycle) Directed tree
Directed Eulerian graph Directed Hamiltonian graph
Connectivity
A digraph is connected if its underlying graph is connected. A digraph is strongly connected if there is a directed path
from any node to any other node.
1 2
4
3 1
connected but not strongly connected
4
3
strongly connected
2
Some Results
Theorem: Let G be a connected digraph. If all its nodes have d_in = 1, regardless of their d_out, then G has one and only one directed circuit.
Proof.
Corollary: Let G be a connected digraph. If all its nodes have d_out = 1, regardless of their d_in, then G has one and only one directed circuit.
Theorem: A connected (sub)digraph is a directed circuit if and only if all its nodes satisfy d_in = d_out.
Corollary: A connected digraph contains a directed circuit if and only if it has a subgraph with all nodes satisfying d_in = d_out.
1 2
0 1 0 0 0 0 1 0
Adjacency Matrix
4
3
Ad [aij]1 1 0 0 0 0 1 0
(asymmetric)
aij –fromitoj
Underlying network A is not well-defined due to multiple edges
More Results
Directed Eulerian graphs
A directed graph is said to be balanced if for all nodes: d_in = d_out
Theorem: A strongly connected digraph is directed Eulerian if and only if it is balanced.
Proof. If a node is not the end, then after walking into it you must walk out from it, which makes d_in = d_out (may be repeated). Continue, until returning to the starting node.
More Results
Directed Hamiltonian graphs
Theorem: Let G be a strongly connected digraph
with N nodes. If all nodes satisfy both dinN/2 and doutN/2
Then, G is directed Hamiltonian (sufficient condition)
Density of Graph Recall undirected graph:
directed graph:
Orientable Graphs
A graph is orientable if its edges can be directed such that the resulting digraph is strongly connected.
not orientable
importance of bridges importance of circuits
Some More Results
Theorem: A connected graph is orientable if and only if every edge of the graph is contained in a circuit.
Proof.
Necessity – G is orientableevery edge can be directed connecting from any node to any other node
Sufficiency – Choose any circuit C and direct it cyclically.
If C = G , done.
If C G , then consider any adjacent edge e. By assumption, e is in some circuit eC*G Since C* is a circuit, there is a path to come back to e . One can then direct this path cyclically.
e
C*
Repeat the above for all edges, then done.
C
Directed and Weighted Networks
Applications: coding, high-security computer networks (data diode), Industrial control systems, …
Shortest Path: For the following directed graph, find the shortest path from Node 1 to Node 8
12263 1 10 5 9
34756
5234 64788
Path: X={1}, p1=0 p1=0
12263 1 10 5 9
34756
5234 64788
min {c12,c14,c16}=min {0+2,0+1,0+3}=min {2,1,3}=1 X={1,4}, p4=1
p4=1
X={1,4}
p1=0 p2=2
12263 1 10 5 9
34756
5234 64788
min {c12,c16,c42,c47}=min {0+2,0+3,1+10,1+2}=min {2,3,11,3}=2 X={1,2,4}, p2=2
p4=1
X={1,2,4}
p1=0 p2=2
12263 1 10 5 9
34756
5234 64788
p6=3
min {c16,c23,c25,c47}=min {0+3,2+6,2+5,1+2}=min {3,8,7,3}=3 X={1,2,4,6}, p6=3
p4=1
X={1,2,4,6}
p1=0 p2=2
12263 1 10 5 9
34756
5234 64788
p6=3 p7=3
min {c23,c25,c47,c67}=min {2+6,2+5,1+2,3+4}=min {8,7,3,7}=3 X={1,2,4,6,7}, p7=3
p4=1
X={1,2,4,6,7}
p1=0 p2=2
12263 1 10 5 9
p4=1 p5=6
34756
5234 64788
p6=3 p7=3
min {c23,c25,c75,c78}=min {2+6,2+5,3+3,3+8}=min {8,7,6,11}=6 X={1,2,4,5,6,7}, p5=6
X={1,2,4,6,7}
p1=0 p2=2
p3=8
12263 1 10 5 9
p4=1 p5=6
34756
5234 64788
p6=3 p7=3
min {c23,c53,c58,c78}=min {2+6,6+9,6+4,3+8}=min {8,15,10,11}=8 X={1,2,3,4,5,6,7}, p3=8
X={1,2,3,4,6,7}
p1=0 p2=2
p3=8
12263 1 10 5 9
p4=1 p5=6
34756
5234 64788
p6=3 p7=3 p8=10
min {c38,c58,c78}=min {8+6,6+4,3+7}=min {14,10,11}=10 X={1,2,3,4,5,6,7,8}, p8=10
X={1,2,3,4,6,7,8}
p1=0 p2=2 p3=8
12263 1 10 5 9
p4=1 p5=6
34756
5234
64788
p6=3 p7=3 p8=10
X={1,2,3,4,6,7,8}
p1=0 p2=2 p3=8
12263 1 10 5 9
p4=1 p5=6
34756
5234 64788
p6=3 p7=3 p8=10
Upper-left corner: Compare c12 = 2, c14 = 1, c16 = 3 Keep c14 = 1
X={1,2,3,4,6,7,8}
p1=0 p2=2 p3=8
12263 1 10 5 9
p4=1 p5=6
34756
5234 64788
p6=3 p7=3 p8=10
Middle part: Compare c23 = 6, c47 = 2 Keep c47 = 2
X={1,2,3,4,6,7,8}
p1=0 p2=2 p3=8
12263 1 10 5 9
p4=1 p5=6
34756
5234 64788
p6=3 p7=3 p8=10
Middle part: Compare c23 = 6, c47 = 2 Keepc47 =2 Therest: c75 =3,c58 =4
X={1,2,3,4,6,7,8}
p1=0 p2=2 p3=8
12263 1 10 5 9
p4=1 p5=6
34756
5234 64788
p6=3 p7=3 p8=10
Solution:
The shortest path from Node 1 to Node 8 is 10, via {1,4,7,5,8}
Longest Path or Maximum Flows Consider the following roadmap, where weights are profit values.
Starting from A, follow the directed road to traverse to F, not allowed to repeat any road, finally arrive at F while making the maximum profits.
B
6
A
10
7
D
E
11
5
18
4
10
1
C
F
Ford–Fulkerson Algorithm
l(A) = 0
l(B) = l(A) + 6 = 6
l(C) = l(A) + 10 = 10l(B) = l(C) + 4 = 14l(E) = l(B) + 7 = 21l(F) = l(E) + 18 = 39 l(D) = l(C) + 1 = 11l(E) = l(D) + 11 = 22l(F) = l(E) + 18 = 40
l(F) = l(C) + 5 = 15
l(D) = l(A) + 10 = 10l(E) = l(D) +11 = 21l(F) = l(E) + 18 = 39
B
6
A
10
7
D
E
11
5
18
4
10
1
C
F
l(A) = 0
l(B) = l(A) + 6 = 6
B
6
A
10
7
D
E
11
5
18
Optimal Solution
l(C) = l(A) + 10 = 10l(B) = l(C) + 4 = 14l(E) = l(B) + 7 = 21l(F) = l(E) + 18 = 39 l(D) = l(C) + 1 = 11l(E) = l(D) + 11 = 22l(F) = l(E) + 18 = 40
l(F) = l(C) + 5 = 15
l(D) = l(A) + 10 = 10l(E) = l(D) +11 = 21l(F) = l(E) + 18 = 39
4
10
1
C
F
Matching
A matching in a graph G is a set of non-adjacent edges Examples: all sets of red-colored edges in the following figures are
matchings (in fact, they are also maximal matchings):
A maximal matching is one on which no more edge can be added. A maximum matching, furthermore, is a maximal matching that
contains the largest possible number of edges.
For example: Matchings in figures (a), (b), (c) and (f), are maximum, but those in (d) and (e) are only maximal but not maximum.
Undirected Graphs
Matched Unmatched node node
Both maximal and maximum matchings in a graph may not be unique. This is clear by inspecting the above figures.
A node is matched if it is an end-point of an edge in a matching; otherwise, the node is unmatched.
A perfect matching is a matching which matches all nodes of the graph. Figure (b) is a perfect matching, while the others are not.
Every perfect matching is always maximum (hence, also maximal)
Matching in Directed Graphs
Matching: a set of directed edges without common heads and tails
Matched node: the head node of a matching edge
A perfect matching is a maximum matching where all nodes are matched nodes
Matching Paths
Alternating path is a path whose edges are alternately in and out of the matching (for directed graphs, also change its direction to opposite)
Example:
At top, there is a matching {BC, DE} At bottom, it is an alternating path
Note: at top, nodes A, F are unmatched; at bottom, nodes A, F become matched. In such a case, the bottom alternating path is called an augmented path of the above path (next page)
Augmenting path is an alternating path between two unmatched nodes
Example:
Consider the graph on the left-hand side, which has a matching {AD, BE}. Here, C and F are unmatched nodes.
For path C-E-B-F, which goes from unmatched node C to unmatched node F, its augmenting path is the one on the right-hand side, “C-E-B-F”, in which dashed lines became solid lines and solid lines became dashed lines.
Augmenting path algorithm (to find maximum matching) This algorithm works only for bipartite graphs
Start with an empty matching
Rule: If there is an augmenting path, do an augmentation
Procedure: Consider the directed bipartite graph (a)
Step 1) Start from node A, search for a path (here, it can go to either D [or E, which is not chosen here], because of the given directions); then augment it, which yields graph (b) [since it is directed, remember to change its direction after augmentation]
(a) (b)
0 solid edge 1 solid edge
Step 2) Start from node B, search for a path, which can go through B-D-A-E [or stop only at E or F]; augment it, which yields graph (c)
(b) (c)
1 solid edge 2 solid edges
Step 3) Start from node C, search for a path, which can only go to F: C-E-A-D-B-F; then augment it, so as to obtain graph (d).
(c) (d)
2 solid edges 3 solid edges
Summary
……
(a) (b) (c) (d)
The search has completed on one side (top) of the bipartile graph, stop Finally, change all the edge directions to the opposite
Solution:
(maximum matching)
Solution may not be unique
Theorem:
A matching is maximum if and only if there is no augmenting path
Maximum matching is useful for network control
(To be studied in Lecture-9)
End