Max-flow min-cut Overview of the Max-flow problem with sample code and example problem. Georgi Stoyanov Student at Sofia University http://backtrack-it.blogspot.com Table of Contents 1. Definition of the problem 2. Where does it occur? 3. Max-flow min-cut theorem 4. Example 5. Max-flow algorithm 6. Run-time estimation 7. Questions 2 Definition of the problem Definition of the problem Maximum flow problems Finding feasible flow Through a single -source, -sink flow network Flow is maximum Many problems solved by Max-flow The problem is often present at algorithmic competitions The Max-flow algorithm Additional definitions Edge capacity – maximum flow that can go through the edge Residual edge capacity – maximum flow that can pass after a certain amount has passed residualCapacity = edgeCapacity – alreadyPassedFlow Augmented path – path starting from source to sink Only edges with residual capacity above zero 5 Where does it occur? Where does it occur? In any kind of network with certain capacity Network of pipes – how much water can pass through the pipe network per unit of time? 7 Where does it occur? Electricity network – how much electricity can go through the grid? 8 Where does it occur? The internet network – how much traffic can go through a local network or the internet? 9 Where does it occur? In other problems Matching problem Group of N guys and M girls Every girl/guy likes a certain amount of people from the other group What is the maximum number of couples, with people who like each other? 10 Where does it occur? Converting the matching problem to a max-flow problem: We add an edge with capacity one for every couple that is acceptable We add two bonus nodes – source and sink We connect the source with the first group and the second group with the sink 11 Max-flow min-cut theorem Max-flow min-cut theorem The max-flow min-cut theorem states that in a flow network, the maximum amount of flow passing from the source to the sink is equal to the minimum capacity that when removed in a specific way from the network causes the situation that no flow can pass from the source to the sink. 13 Example Example Example min( cf(A,D), cf(D,E), cf(E,G)) = min( 3 – 0, 2 – 0, 1 – 0) = min( 3, 2, 1) = 1 maxFlow = maxFlow + 1 = 1 15 Example Example min( cf(A,D), cf(D,F), cf(F,G)) = min( 3 – 1, 6 – 0, 9 – 0) = min( 2, 6, 9) = 2 maxFlow = maxFlow + 2 = 3 16 Example Example min( cf(A,B), cf(B,C), cf(C,D), cf(D,F), cf(F,G)) = min( 3 – 0, 4 – 0, 1 – 0, 6 – 2, 9 - 2) = min( 3, 4, 1, 4, 7) = 1 maxFlow = maxFlow + 1 = 4 17 Example The flow in the previous slide is not optimal! Reverting some of the flow through a different path will achieve the optimal answer To do that for each directed edge (u, v) we will add an imaginary reverse edge (v, u) The new edge shall be used only if a certain amount of flow has already passed through the original edge! 18 Example Example min( cf(A,B), cf(B,C), cf(C,E), cf(E,D), cf(D,F), cf(F,g) ) = min( 3 – 1, 4 – 1, 2 – 0, 0 – -1, 6 – 3, 9 - 3) = min( 2, 3, 2, 1, 3, 6 ) = 1 maxFlow = maxFlow + 1 = 5 (which is the final answer) 19 The Max-flow algorithm The Max-flow algorithm The Edmonds-Karp algorithm Uses a graph structure Uses matrix of the capacities Uses matrix for the passed flow 21 The Max-flow algorithm The Edmonds-Karp algorithm Uses breadth-first search on each iteration to find a path from the source to the sink Uses parent table to store the path Uses path capacity table to store the value of the maximum flow to a node in the path 22 The Max-flow algorithm initialization #include<cstdio> #include<queue> #include<cstring> #include<vector> #include<iostream> #define MAX_NODES 100 // the maximum number of nodes in the graph #define INF 2147483646 // represents infity #define UNINITIALIZED -1 // value for node with no parent using namespace std; // represents the capacities of the edges int capacities[MAX_NODES][MAX_NODES]; // shows how much flow has passed through an edge int flowPassed[MAX_NODES][MAX_NODES]; // represents the graph. The graph must contain the negative edges too! vector<int> graph[MAX_NODES]; //shows the parents of the nodes of the path built by the BFS int parentsList[MAX_NODES]; //shows the maximum flow to a node in the path built by the BFS int currentPathCapacity[MAX_NODES]; 23 The Max-flow algorithm - core The “heart” of the algorithm: int edmondsKarp(int startNode, int endNode) { int maxFlow=0; while(true) { int flow=bfs(startNode, endNode); if(flow==0) break; maxFlow +=flow; int currentNode=endNode; while(currentNode != startNode) { int previousNode = parentsList[currentNode]; flowPassed[previousNode][currentNode] += flow; flowPassed[currentNode][previousNode] -= flow; currentNode=previousNode; } } return maxFlow; } 24 The Max-flow algorithm – Breadth-first search Breadth-first search int bfs(int startNode, int endNode) { memset(parentsList, UNINITIALIZED, sizeof(parentsList)); memset(currentPathCapacity, 0, sizeof(currentPathCapacity)); queue<int> q; q.push(startNode); parentsList[startNode]=-2; currentPathCapacity[startNode]=INF; . . . 25 The Max-flow algorithm – Breadth-first search ... while(!q.empty()) { int currentNode = q.front(); q.pop(); for(int i=0; i<graph[currentNode].size(); i++) { int to = graph[currentNode][i]; if(parentsList[to] == UNINITIALIZED && capacities[currentNode][to] - flowPassed[currentNode][to] > 0) { parentsList[to] = currentNode; currentPathCapacity[to] = min(currentPathCapacity[currentNode], capacities[currentNode][to] - flowPassed[currentNode][to]); if(to == endNode) return currentPathCapacity[endNode]; q.push(to); } } } return 0; } 26 Run-time estimation Breaking down the algorithm: The BFS will cost O(E) operations to find a path on each iteration We will have total O(VE) path augmentations (proved with Theorem and Lemmas) This gives us total run-time of O(VE*E) 27 Run-time estimation There are other algorithms that can run in O(V³) time but are far more complicated to implement ! Note - this algorithm can also run in O(V³) time for sparse graphs 28 The Max-flow algorithm Perks of using the Edmonds-Karp algorithm Runs relatively fast in sparse graphs Represents a refined version of the FordFulkerson algorithm Unlike the Ford-Fulkerson algorithm, this will always terminate It is relatively simple to implement 29 Summary Many problems can be transformed to a max-flow problem. So keep your eyes open! The Edmonds-Karp algorithm is: fairly fast for sparse graphs – O(V³) easy to implement runs in O(VE²) time 30 Summary Don’t forget to add the reverse edges to your graph! The algorithm Looks for augmenting path from source to sink on each iteration Maximum flow == smallest residual capacity of an edge in that path 31 Resources Video lectures (in kind of English) http://www.youtube.com/watch?v=J0wzih3_5Wo http://en.wikipedia.org/wiki/Maximum_flow_problem http://en.wikipedia.org/wiki/Edmonds%E2%80%93Kar p_algorithm http://en.wikipedia.org/wiki/Matching_(graph_theory) Nakov’s book: Programming = ++Algorithms; 32 Combinatorics курсове и уроци по програмиране, уеб дизайн – безплатно курсове и уроци по програмиране – Телерик академия уроци по програмиране и уеб дизайн за ученици програмиране за деца – безплатни курсове и уроци безплатен SEO курс - оптимизация за търсачки курсове и уроци по програмиране, книги – безплатно от Наков уроци по уеб дизайн, HTML, CSS, JavaScript, Photoshop free C# book, безплатна книга C#, книга Java, книга C# безплатен курс "Качествен програмен код" безплатен курс "Разработка на софтуер в cloud среда" BG Coder - онлайн състезателна система - online judge форум програмиране, форум уеб дизайн ASP.NET курс - уеб програмиране, бази данни, C#, .NET, ASP.NET ASP.NET MVC курс – HTML, SQL, C#, .NET, ASP.NET MVC алго академия – състезателно програмиране, състезания курс мобилни приложения с iPhone, Android, WP7, PhoneGap Дончо Минков - сайт за програмиране Николай Костов - блог за програмиране C# курс, програмиране, безплатно http://algoacademy.telerik.com Free Trainings @ Telerik Academy “C# Programming @ Telerik Academy Telerik Software Academy academy.telerik.com Telerik Academy @ Facebook csharpfundamentals.telerik.com facebook.com/TelerikAcademy Telerik Software Academy Forums forums.academy.telerik.com