Karp's Minimum Mean Cycle Algorithm

Free book on Binary Tree

Get this book -> Problems on Array: For Interviews and Competitive Programming

In this article, we have presented Karp's Minimum Mean Cycle Algorithm along with C++ Implementation. Karp's theorem and Complexity Analysis.

Table of contents:

  1. Problem statement with example
  2. Theorem given by Karp
  3. Karp's Minimum Mean Cycle Algorithm
  4. Implementation
  5. Complexity Analysis
  6. References

Problem statement with example

In 1978 , Richard M. Karp gave a characterization of Minimum Mean Cycle. Given a strongly connected digraph, G(V,E) with n vertices consisting of non-negative weights. Mean Weight of a directed cycle defined as summation of edge weights of all edges present in a cycles over number of edges present in a cycle. Our task is to find Minimum Mean Cycle over all directed cycles present in a digraph.

Example :


In a given directed graph, 3 directed cycles are possible and their following mean weight are as shown :

Cycles Path Mean Weight
C1 v1->v2-v4->v1 (1+2+0)/3
C2 v1->v2->v5->v1 (1+2+2)/3
C3 v1->v2->v3->v5->v1 (1+3+1+2)/4
Our task is to find a directed cycle with minimum mean weight over all directed cycles. For a given graph,it comes out to be for Cycle C1 i.e 1.

Theorem given by Karp

Let s be an arbitrary start vertex of G and λ* be Minimum Mean Weight over all all directed cycles.For each vertex v ∈ V and non-negative integer k , let Fk(v) be the minimum weight of any edge progression of length exactly k from s to v, of ∞ if no such edge progression exists. λ* can be defined as


It can be observed that if each edge weight is reduced by constant c then λ* is also reduced by c. RHS of above equation is also reduced by c as F(v) is reduced by c and Fk(v) reduced by k*c. Let translation constant be such that λ* becomes zero.

If λ*= 0 then (Fn(v) - Fk(v))/(n-k) = 0 for some value of v and k. It is known that graph doesn't consists of negative edges and there exists a cycle of weight zero . Hence , we can say that there is an edge progression of edge length k less than n ,from vertex s to v having minimum-mean weight,Fn(v) >= min0<=k<n Fk(v). It implies that max0<=k<n(Fn(v) - Fk(v)) >= 0 . On dividing n-k on both sides, it can be deduced to max0<=k<n(Fn(v) - Fk(v))/(n-k)>= 0.
Above condition λ*= 0 holds if and only if Fn(v) = min0<=k<n Fk(v) for some vertex v which can be proved simply by assuming cycle C and repeating over and over until its path length becomes n and k.

For more details, refer to [1].

Karp's Minimum Mean Cycle Algorithm


|E| = Number of edges present in a graph
|V| = n, Number of Vertices in a graph


  1. Choose an arbitrary starting vertex s.
  2. Compute the shortest path from vertex s to v having edge length k, where 0 < k <= |V|.
  3. Store the computed value in step-2 in an 2-D array say 'dp' ,where dp[k][v] tells shortest length from vertex s to v having edge length k.
  4. For every vertex v , calculate the Mean Weight using Karp's theorem, max((dp[n][v] - dp[k][v])/(n-k)) where 0<= k <= n-1
  5. Store the results computed in step-4 in an 1D-array say 'mean'.
  6. Minimum Mean is the minimum value in the array 'mean'.

How to compute the shortest path?

Dynamic programming can be implemented to find shortest path from vertex s to vertex v containing k edges contains optimal sub-structure and overlapping sub-problems.
 Recursive Solution 
 dp[k][v] = min(dp[k][v], dp[k-1][u] + weight(u,v))
For each k in range(0,n), we compute above relation for every vertex v in range(0,n) and store it in dp[k][v]. Here, W(u,v) is a weight of direct edge between vertex u and v. 


Implementation in C++:

using namespace std;

//represents number of edges
int N;

//represents an edge
typedef struct{
	int from,weight;

vector<edge> edges[N];

void addedge(int from,int to,int weight){

//computes the shortest Path 
void shortestpath(int dp[][N]){

	//intialize all distances to -1
	for(int i=0; i<=N; i++){
		for(int j=0; j<N; j++)
			dp[i][j]= -1;

	//starting vertex to be first vertex 0
	dp[0][0]= 0;

	//for all path length
	for(int k=1; k<=N; k++){

		//for every vertex
		for(int v=0; v<N; v++){
			//direct edges (u,v)
			for(int u=0; u<edges[v].size(); u++){
				if(dp[k-1][edges[u].from] == -1)

				int weight_u_to_v=edges[v][u].weight;

					dp[k][v]= dp[k-1][edges[u].from] + weight_u_to_v;
					dp[k][v] = min(dp[k][v], dp[k-1][edges[u].from] + weight_u_to_v);


//computes minimum mean weight of cycle in graph
double minMeanWeight(){

	int dp[N+1][N];


	//stores mean values
	double mean[N];
	for(int i=0;i<N;i++)

	//Applying Karp's Theorem
	for(int i=0;i<N;i++){


		for(int k=0;k<N;k++){
			if(dp[k][i] != -1)
				mean[i] = max(mean[i],((double)dp[N][i]-dp[k][i])/(N-k));

	//computes minimum value from mean[]
	double minWeight = mean[0];
	for(int i=0;i<N;i++){
		if(mean[i] != -1)
			minWeight = min(minWeight,mean[i]);
	//if not found, returns -1
	return minWeight;
int main(){

	return 0;



A Table could be generated for following approach, where each entry (i,j) represents a value stored in dp[i][j], 0<=i, j<|V|.
(k,v) 0 1 2 3 4 max((dp[n][v] - dp[k][v])/(n-k))
v1 0 -1 -1 20 14 7/2
v2 -1 1 -1 12 6 5/3
v3 -1 10 4 -1 15 11/2
v4 -1 -1 12 6 -1 -1



Complexity Analysis

  • Time Complexity : O(|V| * |E|), for computation of shortest path using dynamic programming.
  • Auxiliary Space Complexity: O(|V| * |E| + |V|), for DP matrix and storing mean weights of every path length

With this article at OpenGenus, you must have a strong idea of Karp's Minimum Mean Cycle Algorithm.


  1. R. M. Karp, A characterization of the minimum cycle mean in a digraph,Discrete mathematics 23 (3) (1978) 309–311
  2. Introduction to Algorithms - Problem Set 9 Solutions by Massachusetts Institute of Technology courses.csail.mit.edu/6.046/fall01/handouts/ps9sol.pdf