1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN">
4 <META HTTP-EQUIV="CONTENT-TYPE" CONTENT="text/html; charset=iso-8859-15">
5 <TITLE>Boost Graph Library: Boykov-Kolmogorov Maximum Flow</TITLE>
6 <META NAME="GENERATOR" CONTENT="OpenOffice.org 2.0 (Linux)">
7 <META NAME="CREATED" CONTENT="20060820;17315200">
8 <META NAME="CHANGEDBY" CONTENT="Stephan Diederich">
9 <META NAME="CHANGED" CONTENT="20060820;23125100">
11 // Copyright (c) 2006 Stephan Diederich
13 // This documentation may be used under either of the following two licences:
15 // Permission is hereby granted, free of charge, to any person
16 // obtaining a copy of this software and associated documentation
17 // files (the "Software"), to deal in the Software without
18 // restriction, including without limitation the rights to use,
19 // copy, modify, merge, publish, distribute, sublicense, and/or
20 // sell copies of the Software, and to permit persons to whom the
21 // Software is furnished to do so, subject to the following
24 // The above copyright notice and this permission notice shall be
25 // included in all copies or substantial portions of the Software.
27 // THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
28 // EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
29 // OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
30 // NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
31 // HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
32 // WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
33 // FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
34 // OTHER DEALINGS IN THE SOFTWARE. OF SUCH DAMAGE.
38 // Distributed under the Boost Software License, Version 1.0.
39 // (See accompanying file LICENSE_1_0.txt or copy at
40 // http://www.boost.org/LICENSE_1_0.txt)
44 TD P { color: #000000 }
47 PRE { color: #000000 }
49 BLOCKQUOTE { color: #000000 }
50 A:link { color: #0000ee }
51 A:visited { color: #551a8b }
55 <BODY LANG="de-DE" TEXT="#000000" LINK="#0000ee" VLINK="#551a8b" BGCOLOR="#ffffff" DIR="LTR">
56 <P><IMG SRC="../../../boost.png" NAME="Grafik1" ALT="C++ Boost" ALIGN=BOTTOM WIDTH=277 HEIGHT=86 BORDER=0>
58 <H1><A NAME="sec:boykov_kolmogorov_max_flow"></A><TT>boykov_kolmogorov_max_flow</TT>
60 <PRE><I>// named parameter version</I>
61 template <class Graph, class P, class T, class R>
62 typename property_traits<typename property_map<Graph, edge_capacity_t>::const_type>::value_type
63 boykov_kolmogorov_max_flow(Graph& g,
64 typename graph_traits<Graph>::vertex_descriptor src,
65 typename graph_traits<Graph>::vertex_descriptor sink,
66 const bgl_named_params<P, T, R>& params = <I>all defaults</I>)
68 <I>// non-named parameter version</I>
69 template <class Graph, class CapacityEdgeMap, class ResidualCapacityEdgeMap, class ReverseEdgeMap,
70 class PredecessorMap, class ColorMap, class DistanceMap, class IndexMap>
71 typename property_traits<CapacityEdgeMap>::value_type
72 boykov_kolmogorov_max_flow(Graph& g,
74 ResidualCapacityEdgeMap res_cap,
75 ReverseEdgeMap rev_map,
76 PredecessorMap pre_map,
80 typename graph_traits <Graph>::vertex_descriptor src,
81 typename graph_traits <Graph >::vertex_descriptor sink)</PRE><P>
82 <FONT SIZE=3>Additional overloaded versions for non-named parameters
83 are provided (without DistanceMap/ColorMap/DistanceMap; for those
84 iterator_property_maps with the provided index map are used)</FONT></P>
85 <P>The <TT>boykov_kolmogorov_max_flow()</TT> function calculates the maximum
86 flow of a network. See Section <A HREF="graph_theory_review.html#sec:network-flow-algorithms">Network
87 Flow Algorithms</A> for a description of maximum flow. The calculated
88 maximum flow will be the return value of the function. The function
89 also calculates the flow values <I>f(u,v)</I> for all <I>(u,v)</I> in
90 <I>E</I>, which are returned in the form of the residual capacity
91 <I>r(u,v) = c(u,v) - f(u,v)</I>.
93 <P><B>Requirements:</B><BR>The directed graph <I>G=(V,E)</I> that
94 represents the network must include a reverse edge for every edge in
95 <I>E</I>. That is, the input graph should be <I>G<SUB>in</SUB> =
96 (V,{E U E<SUP>T</SUP>})</I>. The <TT>ReverseEdgeMap</TT> argument <TT>rev</TT>
97 must map each edge in the original graph to its reverse edge, that is
98 <I>(u,v) -> (v,u)</I> for all <I>(u,v)</I> in <I>E</I>.
101 <P>Remarks: While the push-relabel method states that each edge in <I>E<SUP>T</SUP></I>
102 has to have capacity of 0, the reverse edges for this algorithm ARE
103 allowed to carry capacities. If there are already reverse edges in
104 the input Graph <I><FONT FACE="Courier New, monospace">G</FONT></I>,
105 those can be used. This can halve the amount of edges and will
106 noticeably increase the performance.</P>
109 <B>Algorithm description:</B><BR>The Boykov-Kolmogorov max-flow (or often
110 BK max-flow) algorithm is a variety of the augmenting-path algorithm. Standard
111 augmenting path algorithms find shortest paths from source to sink vertex and
112 augment them by substracting the bottleneck capacity found on that path from the
113 residual capacities of each edge and adding it to the total flow. Additionally
114 the minimum capacity is added to the residual capacity of the reverse edges. If
115 no more paths in the residual-edge tree are found, the algorithm terminates.
116 Instead of finding a new shortest path from source to sink in the graph in each
117 iteration, the Boykov-Kolmogorov algorithm keeps the already found paths as
120 <P>The algorithm builds up two search trees, a source-tree and a
121 sink-tree. Each vertex has a label (stored in <I>ColorMap</I>) to
122 which tree it belongs and a status-flag if this vertex is active or
123 passive. In the beginning of the algorithm only the source and the
124 sink are colored (source==black, sink==white) and have active status.
125 All other vertices are colored gray. The algorithm consists of three
127 <P><I>grow-phase</I>: In this phase active vertices are allowed to
128 acquire neighbor vertices that are connected through an edge that has
129 a capacity-value greater than zero. Acquiring means that those vertices
130 become active and belong now to the search tree of the current
131 active vertex. If there are no more valid connections to neighbor
132 vertices, the current vertex becomes passive and the grow phase
133 continues with the next active vertex. The grow phase terminates if
134 there are no more active vertices left or a vertex discovers a vertex
135 from the other search tree through an unsaturated edge. In this case
136 a path from source to sink is found.</P>
137 <P><I>augment-phase</I>: This phase augments the path that was found
138 in the grow phase. First it finds the bottleneck capacity of the
139 found path, and then it updates the residual-capacity of the edges
140 from this path by substracting the bottleneck capacity from the
141 residual capacity. Furthermore the residual capacity of the reverse
142 edges are updated by adding the bottleneck capacity. This phase can
143 destroy the built up search trees, as it creates at least one
144 saturated edge. That means, that the search trees collapse to
145 forests, because a condition for the search trees is, that each
146 vertex in them has a valid (=non-saturated) connection to a terminal.</P>
147 <P><I>adoption-phase</I>: Here the search trees are reconstructed. A
148 simple solution would be to mark all vertices coming after the first
149 orphan in the found path free vertices (gray). A more sophisticated
150 solution is to give those orphans new parents: The neighbor vertices
151 are checked if they have a valid connection to the same terminal like
152 this vertex had (a path with unsaturated edges). If there is one,
153 this vertex becomes the new parent of the current orphan and this
154 forest is re-included into the search tree. If no new valid parent is
155 found, this vertex becomes a free vertex (marked gray), and it's
156 children become orphans. The adoption phase terminates if there are
158 <P><IMG SRC="figs/bk_max_flow.gif" NAME="Grafik2" ALIGN=LEFT WIDTH=827 HEIGHT=311 BORDER=0><BR CLEAR=LEFT><B>Details:</B></P>
160 <LI><P>Marking heuristics: A timestamp is stored for each vertex
161 which shows in which iteration of the algorithm the distance to the
162 corresponding terminal was calculated.
165 <LI><P>This distance is used and gets calculated in the
166 adoption-phase. In order to find a valid new parent for an orphan,
167 the possible parent is checked for a connection to the terminal to
168 which tree it belongs. If there is such a connection, the path is
169 tagged with the current time-stamp, and the distance value. If
170 another orphan has to find a parent and it comes across a vertex
171 with a current timestamp, this information is used.</P>
172 <LI><P>The distance is also used in the grow-phase. If a vertex
173 comes across another vertex of the same tree while searching for
174 new vertices, the other's distance is compared to its distance. If
175 it is smaller, that other vertex becomes the new parent of the
176 current. This can decrease the length of the search paths, and so
177 amount of adoptions.</P>
179 <LI><P>Ordering of orphans: As described above, the augment-phase
180 and the adoption phase can create orphans. The orphans the
181 augment-phase generates, are ordered according to their distance to
182 the terminals (smallest first). This combined with the
183 distance/timestamp heuristics results in the possibility for not
184 having to recheck terminal-connections too often. New orphans which
185 are generated in adoption phase are processed before orphans from
186 the main queue for the same reason.</P>
188 <P><BR><B>Implementation notes:</B></P>
189 <P>The algorithm is mainly implemented as described by Boykov and Kolmogorov in
190 [<a href="bibliography.html#boykov-kolmogorov04">69</a>]. An extended version
191 can be found in the PhD Thesis of Kolmogorov [<A HREF="bibliography.html#kolmogorov03">68</a>].
192 The following changes are made to improve performance:</P>
194 <LI>initialization: the algorithm first augments all paths from
195 source->sink and all paths from source->VERTEX->sink. This
196 improves especially graph-cuts used in image vision where nearly
197 each vertex has a source and sink connect. During this step, all
198 vertices that have an unsaturated connection from source are added
199 to the active vertex list and so the source is not.</LI>
200 <LI>active vertices: Boykov-Kolmogorov uses two lists for active nodes
201 and states that new active vertices are added to the rear of the
202 second. Fetching an active vertex is done from the beginning of the
203 first list. If the first list is empty, it is exchanged by the
204 second. This implementation uses just one list.</LI>
205 <LI>grow-phase: In the grow phase the first vertex in the
206 active-list is taken and all outgoing edges are checked if they are
207 unsaturated. This decreases performance for graphs with high-edge
208 density. This implementation stores the last accessed edge and
209 continues with it, if the first vertex in the active-list is the
210 same one as during the last grow-phase.</LI>
212 <H3>Where Defined</H3>
213 <P><TT><A HREF="../../../boost/graph/boykov_kolmogorov_max_flow.hpp">boost/graph/boykov_kolmogorov_max_flow.hpp</A></TT>
216 <P>IN: <TT>Graph& g</TT>
218 <BLOCKQUOTE>A directed graph. The graph's type must be a model of
219 <A HREF="VertexListGraph.html">Vertex List Graph</A>, <A HREF="EdgeListGraph.html">Edge
220 List Graph</A> and <A HREF="IncidenceGraph.html">Incidence Graph</A>.
221 For each edge <I>(u,v)</I> in the graph, the reverse edge <I>(v,u)</I>
222 must also be in the graph. Performance of the algorithm will be slightly
223 improved if the graph type also models <a href="AdjacencyMatrix.html">Adjacency
226 <P>IN: <TT>vertex_descriptor src</TT>
228 <BLOCKQUOTE>The source vertex for the flow network graph.
230 <P>IN: <TT>vertex_descriptor sink</TT>
232 <BLOCKQUOTE>The sink vertex for the flow network graph.
234 <H3>Named Parameters</H3>
235 <P>IN: <TT>edge_capacity(EdgeCapacityMap cap)</TT>
237 <BLOCKQUOTE>The edge capacity property map. The type must be a model
238 of a constant <A HREF="../../property_map/doc/LvaluePropertyMap.html">Lvalue
239 Property Map</A>. The key type of the map must be the graph's edge
240 descriptor type.<BR><B>Default:</B> <TT>get(edge_capacity, g)</TT>
242 <P>OUT: <TT>edge_residual_capacity(ResidualCapacityEdgeMap res)</TT>
244 <BLOCKQUOTE>The edge residual capacity property map. The type must be
245 a model of a mutable <A HREF="../../property_map/doc/LvaluePropertyMap.html">Lvalue
246 Property Map</A>. The key type of the map must be the graph's edge
247 descriptor type.<BR><B>Default:</B> <TT>get(edge_residual_capacity,
250 <P>IN: <TT>edge_reverse(ReverseEdgeMap rev)</TT>
252 <BLOCKQUOTE>An edge property map that maps every edge <I>(u,v)</I> in
253 the graph to the reverse edge <I>(v,u)</I>. The map must be a model
254 of constant <A HREF="../../property_map/doc/LvaluePropertyMap.html">Lvalue
255 Property Map</A>. The key type of the map must be the graph's edge
256 descriptor type.<BR><B>Default:</B> <TT>get(edge_reverse, g)</TT>
258 <P>UTIL: <TT>vertex_predecessor(PredecessorMap pre_map)</TT>
260 <BLOCKQUOTE>A vertex property map that stores the edge to the vertex'
261 predecessor. The map must be a model of mutable <A HREF="../../property_map/doc/LvaluePropertyMap.html">Lvalue
262 Property Map</A>. The key type of the map must be the graph's vertex
263 descriptor type.<BR><B>Default:</B> <TT>get(vertex_predecessor, g)</TT>
265 <P>OUT/UTIL: <TT>vertex_color(ColorMap color)</TT>
267 <BLOCKQUOTE>A vertex property map that stores a color for edge
268 vertex. If the color of a vertex after running the algorithm is black
269 the vertex belongs to the source tree else it belongs to the
270 sink-tree (used for minimum cuts). The map must be a model of mutable
271 <A HREF="../../property_map/doc/LvaluePropertyMap.html">Lvalue Property
272 Map</A>. The key type of the map must be the graph's vertex
273 descriptor type.<BR><B>Default:</B> <TT>get(vertex_color, g)</TT>
275 <P>UTIL: <TT>vertex_distance(DistanceMap dist)</TT>
277 <BLOCKQUOTE>A vertex property map that stores the distance to the
278 corresponding terminal. It's a utility-map for speeding up the
279 algorithm. The map must be a model of mutable <A HREF="../../property_map/doc/LvaluePropertyMap.html">Lvalue
280 Property Map</A>. The key type of the map must be the graph's vertex
281 descriptor type.<BR><B>Default:</B> <TT>get(vertex_distance, g)</TT>
283 <P>IN: <TT>vertex_index(VertexIndexMap index_map)</TT>
285 <BLOCKQUOTE>Maps each vertex of the graph to a unique integer in the
286 range <TT>[0, num_vertices(g))</TT>. The map must be a model of
287 constant <A HREF="../../property_map/doc/LvaluePropertyMap.html">LvaluePropertyMap</A>.
288 The key type of the map must be the graph's vertex descriptor
289 type.<BR><B>Default:</B> <TT>get(vertex_index, g)</TT>
292 <P>This reads an example maximum flow problem (a graph with edge
293 capacities) from a file in the DIMACS format (<TT><A HREF="../example/max_flow.dat">example/max_flow.dat</A></TT>).
294 The source for this example can be found in
295 <TT><A HREF="../example/boykov_kolmogorov-eg.cpp">example/boykov_kolmogorov-eg.cpp</A></TT>.
297 <PRE>#include <boost/config.hpp>
298 #include <iostream>
299 #include <string>
300 #include <boost/graph/adjacency_list.hpp>
301 #include <boost/graph/boykov_kolmogorov_max_flow.hpp>
302 #include <boost/graph/read_dimacs.hpp>
303 #include <boost/graph/graph_utility.hpp>
308 using namespace boost;
310 typedef adjacency_list_traits < vecS, vecS, directedS > Traits;
311 typedef adjacency_list < vecS, vecS, directedS,
312 property < vertex_name_t, std::string,
313 property < vertex_index_t, long,
314 property < vertex_color_t, boost::default_color_type,
315 property < vertex_distance_t, long,
316 property < vertex_predecessor_t, Traits::edge_descriptor > > > > >,
318 property < edge_capacity_t, long,
319 property < edge_residual_capacity_t, long,
320 property < edge_reverse_t, Traits::edge_descriptor > > > > Graph;
323 property_map < Graph, edge_capacity_t >::type
324 capacity = get(edge_capacity, g);
325 property_map < Graph, edge_residual_capacity_t >::type
326 residual_capacity = get(edge_residual_capacity, g);
327 property_map < Graph, edge_reverse_t >::type rev = get(edge_reverse, g);
328 Traits::vertex_descriptor s, t;
329 read_dimacs_max_flow(g, capacity, rev, s, t);
331 std::vector<default_color_type> color(num_vertices(g));
332 std::vector<long> distance(num_vertices(g));
333 long flow = boykov_kolmogorov_max_flow(g ,s, t);
335 std::cout << "c The total flow:" << std::endl;
336 std::cout << "s " << flow << std::endl << std::endl;
338 std::cout << "c flow values:" << std::endl;
339 graph_traits < Graph >::vertex_iterator u_iter, u_end;
340 graph_traits < Graph >::out_edge_iterator ei, e_end;
341 for (boost::tie(u_iter, u_end) = vertices(g); u_iter != u_end; ++u_iter)
342 for (boost::tie(ei, e_end) = out_edges(*u_iter, g); ei != e_end; ++ei)
343 if (capacity[*ei] > 0)
344 std::cout << "f " << *u_iter << " " << target(*ei, g) << " "
345 << (capacity[*ei] - residual_capacity[*ei]) << std::endl;
351 <PRE>c The total flow:
376 <P STYLE="margin-bottom: 0cm">
377 <TT><A HREF="edmonds_karp_max_flow.html">edmonds_karp_max_flow()</A></TT>,
378 <TT><A HREF="push_relabel_max_flow.html">push_relabel_max_flow()</A></TT>.
381 <TABLE CELLPADDING=2 CELLSPACING=2>
384 <P>Copyright © 2006</P>
387 <P>Stephan Diederich, University
388 Mannheim(<A HREF="mailto:diederich@ti.uni-manheim.de">diederich@ti.uni-manheim.de</A>)</P>