流网络(Flow Networks)指的是一个有向图 G = (V, E),其中每条边 (u, v) ∈ E 均有一非负容量 c(u, v) ≥ 0。如果 (u, v) ∉ E 则可以规定 c(u, v) = 0。流网络中有两个特殊的顶点:源点 s (source)和汇点 t(sink)。为方便起见,假定每个顶点均处于从源点到汇点的某条路径上,就是说,对每个顶点 v ∈ E,存在一条路径 s --> v --> t。因此,图 G 为连通图,且 |E| ≥ |V| - 1。
下图展示了一个流网络实例。
设 G = (V, E) 是一个流网络,其容量函数为 c。设 s 为网络的源点,t 为汇点。G 的流的一个实值函数 f:V×V → R,且满足下列三个性质:
- 容量限制(Capacity Constraint):对所有顶点对 u, v ∈ V,要求 f(u, v) ≤ c(u, v)。
- 反对称性(Skew Symmetry):对所有顶点对 u, v ∈ V,要求 f(u, v) = - f(v, u)。
- 流守恒性(Flow Conservation):对所有顶点对 u ∈ V - {s, t},要求 Σv∈Vf(u, v) = 0。
f(u, v) 称为从顶点 u 到顶点 v 的流,流的值定义为:|f| =Σv∈Vf(s, v),即从源点 s 出发的总流。
最大流问题(Maximum-flow problem)中,给出源点 s 和汇点 t 的流网络 G,希望找出从 s 到 t 的最大值流。
满足流网络的性质的实际上定义了问题的限制:
- 经过边的流不能超过边的容量;
- 除了源点 s 和汇点 t,对于其它所有顶点,流入量与流出量要相等;
上面的图中描述的流网络可简化为下图,其中源点 s = 0,汇点 t = 5。
上图的最大流为 23,流向如下图所示。
Ford-Fulkerson 算法是一种解决最大流的方法,其依赖于三种重要思想:
- 残留网络(Residual networks)
- 增广路径(Augmenting paths)
- 割(Cut)
这些思想是最大流最小割定理的精髓,该定理用流网络的割来描述最大流的值。
最大流最小割定理
如果 f 是具有源点 s 和汇点 t 的流网络 G = (V, E) 中的一个流,则下列条件是等价的:
- f 是 G 的一个最大流。
- 残留网络 Gf 不包含增广路径。
- 对 G 的某个割 (S, T),有 |f| = c(S, T)。
Ford-Fulkerson 算法是一种迭代方法。开始时,对所有 u, v ∈ V 有 f(u, v) = 0,即初始状态时流的值为 0。在每次迭代中,可通过寻找一条增广路径来增加流值。增广路径可以看做是从源点 s 到汇点 t 之间的一条路径,沿该路径可以压入更多的流,从而增加流的值。反复进行这一过程,直至增广路径都被找出为止。最大流最小割定理将说明在算法终止时,这一过程可产生出最大流。
1 FORD-FULKERSON-METHOD(G, s, t) 2 initialize flow f to 0 3 while there exists an augmenting path p 4 do augment flow f along p 5 return f
上述伪码实现的时间复杂度为 O(max_flow * E)。
C# 代码实现如下:
1 using System; 2 using System.Collections.Generic; 3 using System.Linq; 4 5 namespace GraphAlgorithmTesting 6 { 7 class Program 8 { 9 static void Main(string[] args) 10 { 11 Graph g = new Graph(6); 12 g.AddEdge(0, 1, 16); 13 g.AddEdge(0, 2, 13); 14 g.AddEdge(1, 2, 10); 15 g.AddEdge(1, 3, 12); 16 g.AddEdge(2, 1, 4); 17 g.AddEdge(2, 4, 14); 18 g.AddEdge(3, 2, 9); 19 g.AddEdge(3, 5, 20); 20 g.AddEdge(4, 3, 7); 21 g.AddEdge(4, 5, 4); 22 23 Console.WriteLine(); 24 Console.WriteLine("Graph Vertex Count : {0}", g.VertexCount); 25 Console.WriteLine("Graph Edge Count : {0}", g.EdgeCount); 26 Console.WriteLine(); 27 28 int maxFlow = g.FordFulkerson(0, 5); 29 Console.WriteLine("The Max Flow is : {0}", maxFlow); 30 31 Console.ReadKey(); 32 } 33 34 class Edge 35 { 36 public Edge(int begin, int end, int weight) 37 { 38 this.Begin = begin; 39 this.End = end; 40 this.Weight = weight; 41 } 42 43 public int Begin { get; private set; } 44 public int End { get; private set; } 45 public int Weight { get; private set; } 46 47 public override string ToString() 48 { 49 return string.Format( 50 "Begin[{0}], End[{1}], Weight[{2}]", 51 Begin, End, Weight); 52 } 53 } 54 55 class Graph 56 { 57 private Dictionary<int, List<Edge>> _adjacentEdges 58 = new Dictionary<int, List<Edge>>(); 59 60 public Graph(int vertexCount) 61 { 62 this.VertexCount = vertexCount; 63 } 64 65 public int VertexCount { get; private set; } 66 67 public IEnumerable<int> Vertices 68 { 69 get 70 { 71 return _adjacentEdges.Keys; 72 } 73 } 74 75 public IEnumerable<Edge> Edges 76 { 77 get 78 { 79 return _adjacentEdges.Values.SelectMany(e => e); 80 } 81 } 82 83 public int EdgeCount 84 { 85 get 86 { 87 return this.Edges.Count(); 88 } 89 } 90 91 public void AddEdge(int begin, int end, int weight) 92 { 93 if (!_adjacentEdges.ContainsKey(begin)) 94 { 95 var edges = new List<Edge>(); 96 _adjacentEdges.Add(begin, edges); 97 } 98 99 _adjacentEdges[begin].Add(new Edge(begin, end, weight)); 100 } 101 102 public int FordFulkerson(int s, int t) 103 { 104 int u, v; 105 106 // Create a residual graph and fill the residual graph with 107 // given capacities in the original graph as residual capacities 108 // in residual graph 109 int[,] residual = new int[VertexCount, VertexCount]; 110 111 // Residual graph where rGraph[i,j] indicates 112 // residual capacity of edge from i to j (if there 113 // is an edge. If rGraph[i,j] is 0, then there is not) 114 for (u = 0; u < VertexCount; u++) 115 for (v = 0; v < VertexCount; v++) 116 residual[u, v] = 0; 117 foreach (var edge in this.Edges) 118 { 119 residual[edge.Begin, edge.End] = edge.Weight; 120 } 121 122 // This array is filled by BFS and to store path 123 int[] parent = new int[VertexCount]; 124 125 // There is no flow initially 126 int maxFlow = 0; 127 128 // Augment the flow while there is path from source to sink 129 while (BFS(residual, s, t, parent)) 130 { 131 // Find minimum residual capacity of the edhes along the 132 // path filled by BFS. Or we can say find the maximum flow 133 // through the path found. 134 int pathFlow = int.MaxValue; 135 for (v = t; v != s; v = parent[v]) 136 { 137 u = parent[v]; 138 pathFlow = pathFlow < residual[u, v] 139 ? pathFlow : residual[u, v]; 140 } 141 142 // update residual capacities of the edges and reverse edges 143 // along the path 144 for (v = t; v != s; v = parent[v]) 145 { 146 u = parent[v]; 147 residual[u, v] -= pathFlow; 148 residual[v, u] += pathFlow; 149 } 150 151 // Add path flow to overall flow 152 maxFlow += pathFlow; 153 } 154 155 // Return the overall flow 156 return maxFlow; 157 } 158 159 // Returns true if there is a path from source 's' to sink 't' in 160 // residual graph. Also fills parent[] to store the path. 161 private bool BFS(int[,] residual, int s, int t, int[] parent) 162 { 163 bool[] visited = new bool[VertexCount]; 164 for (int i = 0; i < visited.Length; i++) 165 { 166 visited[i] = false; 167 } 168 169 Queue<int> q = new Queue<int>(); 170 171 visited[s] = true; 172 q.Enqueue(s); 173 parent[s] = -1; 174 175 // standard BFS loop 176 while (q.Count > 0) 177 { 178 int u = q.Dequeue(); 179 180 for (int v = 0; v < VertexCount; v++) 181 { 182 if (!visited[v] 183 && residual[u, v] > 0) 184 { 185 q.Enqueue(v); 186 visited[v] = true; 187 parent[v] = u; 188 } 189 } 190 } 191 192 // If we reached sink in BFS starting from source, 193 // then return true, else false 194 return visited[t] == true; 195 } 196 } 197 } 198 }
运行结果如下:
参考资料
- 广度优先搜索
- 深度优先搜索
- Breadth First Traversal for a Graph
- Depth First Traversal for a Graph
- Dijkstra 单源最短路径算法
- Bellman-Ford 单源最短路径算法
- Bellman–Ford algorithm
- Introduction To Algorithm
- Floyd-Warshall's algorithm
- Bellman-Ford algorithm for single-source shortest paths
- Dynamic Programming | Set 23 (Bellman–Ford Algorithm)
- Dynamic Programming | Set 16 (Floyd Warshall Algorithm)
- Johnson’s algorithm for All-pairs shortest paths
- Floyd-Warshall's algorithm
- 最短路径算法--Dijkstra算法,Bellmanford算法,Floyd算法,Johnson算法
- QuickGraph, Graph Data Structures And Algorithms for .NET
- CHAPTER 26: ALL-PAIRS SHORTEST PATHS