P

```
package com.thealgorithms.datastructures.graphs;
import java.util.ArrayList;
import java.util.List;
import java.util.Stack;
/**
* Java program that implements Tarjan's Algorithm.
* @author Shivanagouda S A (https://github.com/shivu2002a)
*
*/
/**
* Tarjan's algorithm is a linear time algorithm to find the strongly connected components of a
directed graph, which, from here onwards will be referred as SCC.
* A graph is said to be strongly connected if every vertex is reachable from every other vertex.
The SCCs of a directed graph form a partition into subgraphs that are themselves strongly
connected. Single node is always a SCC.
* Example:
0 --------> 1 -------> 3 --------> 4
^ /
| /
| /
| /
| /
| /
| /
| /
| /
| /
|V
2
For the above graph, the SCC list goes as follows:
1, 2, 0
3
4
We can also see that order of the nodes in an SCC doesn't matter since they are in cycle.
{@summary}
Tarjan's Algorithm:
* DFS search produces a DFS tree
* Strongly Connected Components form subtrees of the DFS tree.
* If we can find the head of these subtrees, we can get all the nodes in that subtree (including
the head) and that will be one SCC.
* There is no back edge from one SCC to another (here can be cross edges, but they will not be
used).
* Kosaraju Algorithm aims at doing the same but uses two DFS traversalse whereas Tarjan’s
algorithm does the same in a single DFS, which leads to much lower constant factors in the latter.
*/
public class TarjansAlgorithm {
// Timer for tracking lowtime and insertion time
private int Time;
private List<List<Integer>> SCClist = new ArrayList<List<Integer>>();
public List<List<Integer>> stronglyConnectedComponents(int V, List<List<Integer>> graph) {
// Initially all vertices as unvisited, insertion and low time are undefined
// insertionTime:Time when a node is visited 1st time while DFS traversal
// lowTime: indicates the earliest visited vertex (the vertex with minimum insertion time)
// that can be reached from a subtree rooted with a particular node.
int[] lowTime = new int[V];
int[] insertionTime = new int[V];
for (int i = 0; i < V; i++) {
insertionTime[i] = -1;
lowTime[i] = -1;
}
// To check if element is present in stack
boolean[] isInStack = new boolean[V];
// Store nodes during DFS
Stack<Integer> st = new Stack<Integer>();
for (int i = 0; i < V; i++) {
if (insertionTime[i] == -1) stronglyConnCompsUtil(i, lowTime, insertionTime, isInStack, st, graph);
}
return SCClist;
}
private void stronglyConnCompsUtil(int u, int[] lowTime, int[] insertionTime, boolean[] isInStack, Stack<Integer> st, List<List<Integer>> graph) {
// Initialize insertion time and lowTime value of current node
insertionTime[u] = Time;
lowTime[u] = Time;
Time += 1;
// Push current node into stack
isInStack[u] = true;
st.push(u);
// Go through all vertices adjacent to this
for (Integer vertex : graph.get(u)) {
// If the adjacent node is unvisited, do DFS
if (insertionTime[vertex] == -1) {
stronglyConnCompsUtil(vertex, lowTime, insertionTime, isInStack, st, graph);
// update lowTime for the current node comparing lowtime of adj node
lowTime[u] = Math.min(lowTime[u], lowTime[vertex]);
} else if (isInStack[vertex]) {
// If adj node is in stack, update low
lowTime[u] = Math.min(lowTime[u], insertionTime[vertex]);
}
}
// If lowtime and insertion time are same, current node is the head of an SCC
// head node found, get all the nodes in this SCC
if (lowTime[u] == insertionTime[u]) {
int w = -1;
var scc = new ArrayList<Integer>();
// Stack has all the nodes of the current SCC
while (w != u) {
w = st.pop();
scc.add(w);
isInStack[w] = false;
}
SCClist.add(scc);
}
}
}
```