mirror of
https://github.com/dholerobin/Lecture_Notes.git
synced 2025-03-16 22:29:57 +00:00
264 lines
8.9 KiB
Markdown
264 lines
8.9 KiB
Markdown
## Prim's Algorithm
|
|
|
|
Like Kruskal's algorithm, Prim's algorithm is another algorithm to find the minimum spanning tree for the given undirected weighted graph.
|
|
|
|
Prim's algorithm is also a greedy algorithm, which is quite similar to Dijkstra's algorithm. If you are familar with Dijkstra's algorithm then you know that, we need to find a minimum distance vertex at each step, however in Prim's algorithm we need to find a minimum weight edge. Let's see the actual algorithm.
|
|
|
|
**Notes:**
|
|
- Here the term **tree** stands for an intermediate tree in the formation of the whole MST.
|
|
- **Explored edges** means the edges which are already found in the run of the algorithm.
|
|
- Here, we are assuming that the given undireted graph is connected.
|
|
|
|
## Algorithm
|
|
|
|
1. Select an arbitrary vertex say $V$ from the graph and start the algorithm from that vertex.
|
|
2. Add vertex $V$ in the tree.
|
|
3. Explore all the edges connected to vertex $V$.
|
|
4. Find the minimum weight edge from all the explored edges, which connects the tree to a vertex $U$ which is not yet added in the tree.
|
|
5. Set $V$ to $U$ and continue from step 2 until all $|V|$ vertices are in the tree.
|
|
|
|
**Visualization**
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
We will discuss two different approaches:
|
|
1. Adjacency List representation of graph
|
|
2. Adjacency Matrix representation of graph
|
|
|
|
Generally, we use Adjacency Matrix representation in case of Dense graph because it uses lesser space than the list representation, whereas we use Adjacency List representation in the case of sparse graph.
|
|
|
|
**Note:** **Minimum weight** represents the weight of a minimum weight edge observed so far in the algorithm, which is connected to a particular vertex.
|
|
|
|
## Sparse Graphs - Adjacency list representation
|
|
|
|
Here we are representing the graph using Adjacency list representation.
|
|
|
|
**Implementation Algorithm**
|
|
|
|
1. Initialize a boolean array which keeps track of, whether a vertex is added in the tree.
|
|
2. Initialize an array of integers say $\text{Minweight[]}$, where each entry of it shows the minimum weight, by $\infty$.
|
|
3. Start from any vertex say $A$. Mark the weight of the minimum weight edge to reach it as $0$.
|
|
4. Explore all of the edges connected with $A$ and update the minimum weights to reach the adjacent vertices, if the below condition is satisfied,
|
|
For an edge $A\to B$,
|
|
$\text{Minweight}[B] < \text{EdgeWeight}(A,B)$
|
|
Note that we are only looking for those adjacent vertices which are not already in the tree.
|
|
5. Find the minimum weight edge from all the explored edges and repeat from step $4$. Say that edge is $a - b$, then take $V$ as $b$ and repeat from step $4$.
|
|
6. If all the $|V|$ vertices are added in the tree, then stop the algorithm.
|
|
|
|
Can you tell, how we will do the step 5, which is to find the minimum weight edge from all the exlpored edges?
|
|
|
|
Here, we need to use some data structure which finds out the minimum weight edge from all the explored edges efficiently.
|
|
|
|
Do you know any of them?
|
|
|
|
We can use anyone of priority queue, fibonacci heap, binomial heap, balanced binary tree, etc.
|
|
|
|
**Note:** Below in the code, the parent array is used to retrieve the formed MST and **set** (STL container) is a kind of balanced binary search tree.
|
|
|
|
|
|
```c++
|
|
#include <bits/stdc++.h>
|
|
#define MAX_Weight 1000000000
|
|
using namespace std;
|
|
|
|
int main()
|
|
{
|
|
int no_vertices = 4;
|
|
|
|
vector<vector<pair<int,int>> > graph(no_vertices+1, vector<pair<int,int>>());
|
|
|
|
graph[1].push_back({2,6}); // A-B
|
|
graph[2].push_back({1,6});
|
|
graph[1].push_back({4,5}); // A-D
|
|
graph[4].push_back({1,5});
|
|
graph[2].push_back({3,3}); // B-C
|
|
graph[3].push_back({2,3});
|
|
graph[2].push_back({4,4}); // B-D
|
|
graph[4].push_back({2,4});
|
|
graph[3].push_back({4,2}); // C-D
|
|
graph[4].push_back({3,2});
|
|
|
|
// To track if the vertex is added in MST
|
|
vector<bool> inMST(no_vertices+1);
|
|
|
|
vector<int> minWeight(no_vertices+1, MAX_Weight),
|
|
parent(no_vertices+1);
|
|
|
|
// Minimum finding(logN) DS
|
|
set<pair<int,int>> Explored_edges;
|
|
|
|
Explored_edges.insert({0,1});
|
|
|
|
for(int i=2;i<no_vertices+1;i++)
|
|
Explored_edges.insert({MAX_Weight,i});
|
|
|
|
int VertinMST = 0, MSTcost = 0;
|
|
|
|
minWeight[1] = 0;
|
|
parent[1] = -1;
|
|
|
|
while(VertinMST < no_vertices)
|
|
{
|
|
// Vertex connected by Minimum weight edge
|
|
int vertex = Explored_edges.begin()->second;
|
|
|
|
MSTcost += minWeight[vertex];
|
|
|
|
inMST[vertex] = true;
|
|
VertinMST++;
|
|
|
|
Explored_edges.erase(Explored_edges.begin());
|
|
|
|
// Exploring the adjacent edges
|
|
for(auto i:graph[vertex])
|
|
{
|
|
// If we reach by lesser weighted edge then update
|
|
if(!inMST[i.first] && minWeight[i.first] > i.second)
|
|
{
|
|
// Previous larger weighted edge
|
|
Explored_edges.erase({minWeight[i.first],i.first});
|
|
|
|
minWeight[i.first] = i.second;
|
|
parent[i.first] = vertex;
|
|
|
|
// New smaller weighted edge
|
|
Explored_edges.insert({minWeight[i.first],i.first});
|
|
}
|
|
}
|
|
|
|
}
|
|
|
|
cout << MSTcost << endl;
|
|
|
|
cout << "Edges in MST:" << endl;
|
|
for(int i = 1; i <= no_vertices; i++)
|
|
{
|
|
if(parent[i] != -1)
|
|
cout << parent[i] << " " << i << endl;
|
|
}
|
|
|
|
return 0;
|
|
}
|
|
```
|
|
|
|
**Time Complexity**
|
|
|
|
1. We need $\mathcal{O}(log|V|)$ time to find the minimum weight edge.
|
|
2. We are doing the above step $\mathcal{O}(|E|+|V|)$ times, which is similar to BFS.
|
|
|
|
Overall time complexity: $\mathcal{O}((|E|+|V|)log|V|)$
|
|
|
|
## Dense Graphs - Adjacency matrix representation
|
|
|
|
Here, to loop over the adjacent vertices, we have to loop over all |V| entries of the adjacency matrix.
|
|
|
|
So, basically to update minimum weights we have to spend $O(|V|)$ time. And also to find the minimum weight edge we have to spend $O(|V|)$ time.
|
|
|
|
The implementation is much simpler than the previous one.
|
|
|
|
```c++
|
|
#include <bits/stdc++.h>
|
|
#define MAX_Weight 1000000000
|
|
using namespace std;
|
|
|
|
int main()
|
|
{
|
|
int no_vertices = 4;
|
|
|
|
int graph[no_vertices+1][no_vertices+1];
|
|
|
|
graph[1][2]=6; // A-B
|
|
graph[2][1]=6;
|
|
graph[1][4]=5; // A-D
|
|
graph[4][1]=5;
|
|
graph[2][3]=3; // B-C
|
|
graph[3][2]=3;
|
|
graph[2][4]=4; // B-D
|
|
graph[4][2]=4;
|
|
graph[3][4]=2; // C-D
|
|
graph[4][3]=2;
|
|
|
|
// To track if the vertex is added in MST
|
|
vector<bool> inMST(no_vertices + 1);
|
|
|
|
int VertinMST = 0, MSTcost = 0;
|
|
|
|
vector<int> minWeight(no_vertices + 1, MAX_Weight),
|
|
parent(no_vertices + 1);
|
|
|
|
minWeight[1] = 0;
|
|
parent[1] = -1;
|
|
|
|
for(int i=1; i<=no_vertices; i++)
|
|
{
|
|
int minvertex = 0, weight = MAX_Weight;
|
|
|
|
// Find the minimum weighted edge
|
|
for(int j=1; j<=no_vertices; j++)
|
|
if(!inMST[j] && (minvertex==0 || minWeight[j] < minWeight[minvertex]))
|
|
{
|
|
minvertex = j;
|
|
weight = minWeight[j];
|
|
}
|
|
|
|
inMST[minvertex] = true;
|
|
MSTcost += weight;
|
|
|
|
// Update the min weights
|
|
for(int j=1;j<=no_vertices;j++)
|
|
if(!inMST[j] && graph[minvertex][j] < minWeight[j])
|
|
{
|
|
minWeight[j] = graph[minvertex][j];
|
|
parent[j] = minvertex;
|
|
}
|
|
|
|
}
|
|
|
|
cout << MSTcost << endl;
|
|
|
|
cout << "Edges in MST:" << endl;
|
|
for(int i = 1; i <= no_vertices; i++)
|
|
{
|
|
if(parent[i] != -1)
|
|
cout << parent[i] << " " << i << endl;
|
|
}
|
|
|
|
return 0;
|
|
}
|
|
|
|
```
|
|
### Time Complexity
|
|
|
|
1. It takes $\mathcal{O}(|V|)$ time to find the minimum weight edge and also to update the minimum weights.
|
|
2. The outer loop runs $|V|$ times.
|
|
|
|
Overall time complexity: $\mathcal{O}(|V|^2)$
|
|
|
|
### Other algorithms to find MST
|
|
|
|
1. Kruskal's Algorithm
|
|
2. Boruvka's Algorithm
|
|
|
|
|