INF280

Graph Traversals & Paths

Florian Brandner
February 27, 2018

1/21
Contents

Introduction

Simple Traversals
Depth-First Search
Breadth-First Search

Finding Paths
Dijkstra
Bellman-Ford
Floyd-Warshall
Improvements

Eulerian Circuits

2/21
Introduction

You all know graphs:

• Set of nodes N
• Set of edges E ⊆ N × N
• Edges can be undirected or directed, i.e., (a, b) 6= (b, a)

A B

N {A, B, C}
E {(A, B), (A, C), (B, C)}
C

3/21
Data Structures

Several options to represent graphs:

• Adjacency matrix:
• bool G[MAXN][MAXN];
• G[x][y] is true if an edge between node x and y exists
• Replace bool by int to represent weighted edges
• Adjacency list:
• vector<int> Adj[MAXN];
• y is in Adj[x] if an edge between node x and y exists
• Pairs to represent weights
• Edge list:
• vector<pair<int, int> > Edges;
• Edges contains a pair of nodes if an edge exists between them
• Nodes and edges may also be custom structs or classes
4/21
Contents

Introduction

Simple Traversals
Depth-First Search
Breadth-First Search

Finding Paths
Dijkstra
Bellman-Ford
Floyd-Warshall
Improvements

Eulerian Circuits

5/21
Depth-First Search

Visit each node in the graph once:

• Recursive implementation below

• Manage stack yourself for iterative version
• First visit child nodes then siblings
bool Visited[MAXN] = {};
void DFS(int u) {
if (Visited[u])
return;
Visited[u] = true;
// maybe do something with u (pre-order) ...
for (auto v : Adj[u])
DFS(v);
// or do something here (post-order)
}

6/21
Applications of DFS

• Determine a topological order of nodes

• Only works for graphs without cycles (i.e., x → y → z → x)
• Add visited node at the head of an ordered list
(at the end of DFS: ordering.push_front(u))
• Detect if a cycle exists:
• Check if the currently visited node is on the stack
A) Use three states for Visited array:
UNVISITED, ONSTACK, VISITED
B) Explicitly search in the stack of the iterative algorithm

• Examples: https://visualgo.net/dfsbfs

7/21
Breadth-First Search

Visit each node in the graph once:

• Similar to DFS, but replaces stack by queue
queue<int> Q;
bool Visited[MAXN] = {};
void BFS(int root) {
Q.push(root);
while (!Q.empty()) {
int u = Q.front();
Q.pop();
if (Visited[u])
continue;
Visited[u] = true;
for (auto v : Adj[u])
Q.push(v); // usually do something with v
}
}
8/21
Applications of BFS

• Shortest path search

• Stop processing when a given node d was found
• Minimizes number of hops, i.e., all edges have same weight
• Generalization follows next
• Examples: https://visualgo.net/dfsbfs

9/21
Contents

Introduction

Simple Traversals
Depth-First Search
Breadth-First Search

Finding Paths
Dijkstra
Bellman-Ford
Floyd-Warshall
Improvements

Eulerian Circuits

10/21
Dijkstra

BFS can only be used if edge weights are uniform

• Dijkstra’s algorithm generalizes this

• Constraint: all edges need to have non-negative weights
• Use a priority queue to choose which node to examine next
• Would require a function to update the priority of an element
• Would need to update order in the priority queue
• We’ll use the standard priority_queue of STL
• Simply insert a new element in the queue (no update)
• Ok since priorities decrease monotonically
• This slightly diverges from Dijkstra’s algorithm

• May revisit nodes several times

11/21
Dijkstra’s Algorithm
unsigned int Dist[MAXN];
typedef pair<unsigned int, int> WeightNode; // weight goes first
priority_queue<WeightNode, std::vector<WeightNode>,
std::greater<WeightNode> > Q;
void Dijkstra(int root) {
fill_n(Dist, MAXN, MAXLEN);
Dist[root] = 0;
Q.push(make_pair(0, root));
while (!Q.empty()) {
int u = Q.top().second; // get node with least priority
Q.pop();
for (auto tmp : Adj[u]) {
int v = tmp.second;
unsigned int weight = tmp.first;
if (Dist[v] > Dist[u] + weight) { // shorter path found?
Dist[v] = Dist[u] + weight;
Q.push(make_pair(Dist[v], v)); // simply push, no update here
}
}
} https://visualgo.net/sssp
}
12/21
Bellman-Ford

• Dijkstra’s algorithm is limited to non-negative edge weights

• Bellman-Ford extends this to general edge weights
• Constraint: no cycle with negative total costs
• May again revisit nodes several times

13/21
Bellman-Ford Algorithm

unsigned int Dist[MAXN];

void BellmanFord(int root) {
fill_n(Dist, MAXN, MAXLEN);
Dist[root] = 0;
for(int u=0; u < N - 1; u++) { // just iterate N - 1 times
for (auto tmp : Adj[u]) {
int v = tmp.second; // same as Dijkstra before
unsigned int weight = tmp.first;
Dist[v] = min(Dist[v], Dist[u] + weight);
}
}
} https://visualgo.net/sssp

14/21
Floyd-Warshall

• Dijkstra and Bellman-Ford compute shortest paths

• From a single source (root)
• To all other (reachable) nodes
• This is known as: single-source shortest path problem
• Floyd-Warshall extends this to compute the shortest paths
between all pairs of nodes
• This is known as: all-pairs shortest path problem

15/21
Floyd-Warshall Algorithm

int Dist[MAXN][MAXN];
void FloydWarshall() {
fill_n((int*)Dist, MAXN*MAXN, MAXLEN);
for(int u=0; u < N; u++) {
Dist[u][u] = 0;
for (auto tmp : Adj[u])
Dist[u][tmp.second] = tmp.first;
}
for(int k=0; k < N; k++) // check sub-path combinations
for(int i=0; i < N; i++)
for(int j=0; j < N; j++) // concatenate paths
Dist[i][j] = min(Dist[i][j], Dist[i][k] + Dist[k][j]);
}

16/21
Keeping track of the path

We only considered the length of the path so far:

• All of the above algorithms can track the actual shortest path
• This allows to print each edge/node along the path
• Basic idea:
• Introduce an array int Predecessor[MAXN]
(Use two-dimensional array for Floyd-Warshall)
• Updated whenever Dist[v] changes
• Simply set to the new predecessor u

17/21
Heuristics – A* Search

Heuristics may speed-up the path search

• Bellman-Ford and Floyd-Warshall equally explore all possibilities

• Dijkstra prefers nodes with shorter distance
• Basic idea behind A* Search:
• Extension to Dijkstra’s algorithm
• Introduce an estimation of the remaining distance
• Prefer nodes with minimal estimated remaining distance
• Advantages
• May converge faster than Dijkstra
• Can be used to compute approximate solutions
(trading speed for precision)

18/21
Contents

Introduction

Simple Traversals
Depth-First Search
Breadth-First Search

Finding Paths
Dijkstra
Bellman-Ford
Floyd-Warshall
Improvements

Eulerian Circuits

19/21
Eulerian Circuits

We study undirected graphs and assume they are connected:

• Eulerian path:
Use every edge of a graph exactly once. Start and end may differ
• Eulerian circuit:
Use every edge exactly once. Start and end at the same node

• Conditions to find Eulerian path:

• All nodes have even degree or
• Precisely two nodes have odd degree
• For Eulerian circuit, all nodes must have even degree

20/21
Hierholzer’s Algorithm for Eulerian Paths (assuming they exist)
set<int> Adj[MAXN]; vector<int> Circuit;
void Hierholzer() {
int v = 0; // find node with odd degree, else start with node 0
for (int u=0; u < N && v == 0; u++)
if (Adj[u].size() & 1)
v = u; // node with odd degree
stack<int> Stack;
Stack.push(v);
while (!Stack.empty()) {
if (!Adj[v].empty()) { // follow edges until stuck
Stack.push(v);
int tmp = *Adj[v].begin();
Adj[v].erase(tmp); // remove edge, modifying graph
Adj[tmp].erase(v);
v = tmp;
} else { // got stuck: stack contains a circuit
Circuit.push_back(v); // append node at the end of circuit
v = Stack.top(); // backtrack using stack, find larger circuit
Stack.pop();
}
https://www-m9.ma.tum.de/graph-algorithms/hierholzer/index_en.html
}
21/21
}