lemon/goldberg_tarjan.h
author deba
Tue, 08 Apr 2008 15:16:16 +0000
changeset 2604 e4ec01f1a4cd
parent 2527 10f3b3286e63
permissions -rw-r--r--
Fix Makefile.am
     1 /* -*- C++ -*-
     2  *
     3  * This file is a part of LEMON, a generic C++ optimization library
     4  *
     5  * Copyright (C) 2003-2008
     6  * Egervary Jeno Kombinatorikus Optimalizalasi Kutatocsoport
     7  * (Egervary Research Group on Combinatorial Optimization, EGRES).
     8  *
     9  * Permission to use, modify and distribute this software is granted
    10  * provided that this copyright notice appears in all copies. For
    11  * precise terms see the accompanying LICENSE file.
    12  *
    13  * This software is provided "AS IS" with no warranty of any kind,
    14  * express or implied, and with no claim as to its suitability for any
    15  * purpose.
    16  *
    17  */
    18 
    19 #ifndef LEMON_GOLDBERG_TARJAN_H
    20 #define LEMON_GOLDBERG_TARJAN_H
    21 
    22 #include <vector>
    23 #include <queue>
    24 
    25 #include <lemon/error.h>
    26 #include <lemon/bits/invalid.h>
    27 #include <lemon/tolerance.h>
    28 #include <lemon/maps.h>
    29 #include <lemon/graph_utils.h>
    30 #include <lemon/dynamic_tree.h>
    31 #include <limits>
    32 
    33 /// \file
    34 /// \ingroup max_flow
    35 /// \brief Implementation of the preflow algorithm.
    36 
    37 namespace lemon {
    38 
    39   /// \brief Default traits class of GoldbergTarjan class.
    40   ///
    41   /// Default traits class of GoldbergTarjan class.
    42   /// \param _Graph Graph type.
    43   /// \param _CapacityMap Type of capacity map.
    44   template <typename _Graph, typename _CapacityMap>
    45   struct GoldbergTarjanDefaultTraits {
    46 
    47     /// \brief The graph type the algorithm runs on. 
    48     typedef _Graph Graph;
    49 
    50     /// \brief The type of the map that stores the edge capacities.
    51     ///
    52     /// The type of the map that stores the edge capacities.
    53     /// It must meet the \ref concepts::ReadMap "ReadMap" concept.
    54     typedef _CapacityMap CapacityMap;
    55 
    56     /// \brief The type of the length of the edges.
    57     typedef typename CapacityMap::Value Value;
    58 
    59     /// \brief The map type that stores the flow values.
    60     ///
    61     /// The map type that stores the flow values. 
    62     /// It must meet the \ref concepts::ReadWriteMap "ReadWriteMap" concept.
    63     typedef typename Graph::template EdgeMap<Value> FlowMap;
    64 
    65     /// \brief Instantiates a FlowMap.
    66     ///
    67     /// This function instantiates a \ref FlowMap. 
    68     /// \param graph The graph, to which we would like to define the flow map.
    69     static FlowMap* createFlowMap(const Graph& graph) {
    70       return new FlowMap(graph);
    71     }
    72 
    73     /// \brief The eleavator type used by GoldbergTarjan algorithm.
    74     /// 
    75     /// The elevator type used by GoldbergTarjan algorithm.
    76     ///
    77     /// \sa Elevator
    78     /// \sa LinkedElevator
    79     typedef LinkedElevator<Graph, typename Graph::Node> Elevator;
    80     
    81     /// \brief Instantiates an Elevator.
    82     ///
    83     /// This function instantiates a \ref Elevator. 
    84     /// \param graph The graph, to which we would like to define the elevator.
    85     /// \param max_level The maximum level of the elevator.
    86     static Elevator* createElevator(const Graph& graph, int max_level) {
    87       return new Elevator(graph, max_level);
    88     }
    89 
    90     /// \brief The tolerance used by the algorithm
    91     ///
    92     /// The tolerance used by the algorithm to handle inexact computation.
    93     typedef Tolerance<Value> Tolerance;
    94 
    95   };
    96 
    97   /// \ingroup max_flow
    98   /// \brief Goldberg-Tarjan algorithms class.
    99   ///
   100   /// This class provides an implementation of the \e GoldbergTarjan
   101   /// \e algorithm producing a flow of maximum value in a directed
   102   /// graph. The GoldbergTarjan algorithm is a theoretical improvement
   103   /// of the Goldberg's \ref Preflow "preflow" algorithm by using the \ref
   104   /// DynamicTree "dynamic tree" data structure of Sleator and Tarjan.
   105   /// 
   106   /// The original preflow algorithm with \e highest \e label
   107   /// heuristic has \f$O(n^2\sqrt{e})\f$ or with \e FIFO heuristic has
   108   /// \f$O(n^3)\f$ time complexity. The current algorithm improved
   109   /// this complexity to \f$O(nm\log(\frac{n^2}{m}))\f$. The algorithm
   110   /// builds limited size dynamic trees on the residual graph, which
   111   /// can be used to make multilevel push operations and gives a
   112   /// better bound on the number of non-saturating pushes.
   113   ///
   114   /// \param Graph The directed graph type the algorithm runs on.
   115   /// \param CapacityMap The capacity map type.
   116   /// \param _Traits Traits class to set various data types used by
   117   /// the algorithm.  The default traits class is \ref
   118   /// GoldbergTarjanDefaultTraits.  See \ref
   119   /// GoldbergTarjanDefaultTraits for the documentation of a
   120   /// Goldberg-Tarjan traits class.
   121   ///
   122   /// \author Tamas Hamori and Balazs Dezso
   123 #ifdef DOXYGEN
   124   template <typename _Graph, typename _CapacityMap, typename _Traits> 
   125 #else
   126   template <typename _Graph,
   127 	    typename _CapacityMap = typename _Graph::template EdgeMap<int>,
   128             typename _Traits = 
   129 	    GoldbergTarjanDefaultTraits<_Graph, _CapacityMap> >
   130 #endif
   131   class GoldbergTarjan {
   132   public:
   133 
   134     typedef _Traits Traits;
   135     typedef typename Traits::Graph Graph;
   136     typedef typename Traits::CapacityMap CapacityMap;
   137     typedef typename Traits::Value Value; 
   138 
   139     typedef typename Traits::FlowMap FlowMap;
   140     typedef typename Traits::Elevator Elevator;
   141     typedef typename Traits::Tolerance Tolerance;
   142     
   143   protected:
   144 
   145     GRAPH_TYPEDEFS(typename Graph);
   146 
   147     typedef typename Graph::template NodeMap<Node> NodeNodeMap;
   148     typedef typename Graph::template NodeMap<int> IntNodeMap;
   149 
   150     typedef typename Graph::template NodeMap<Edge> EdgeNodeMap;
   151     typedef typename Graph::template EdgeMap<Edge> EdgeEdgeMap;
   152 
   153     typedef typename std::vector<Node> VecNode;
   154  
   155     typedef DynamicTree<Value,IntNodeMap,Tolerance> DynTree;
   156 
   157     const Graph& _graph;
   158     const CapacityMap* _capacity;
   159     int _node_num; //the number of nodes of G
   160 
   161     Node _source;
   162     Node _target;
   163 
   164     FlowMap* _flow;
   165     bool _local_flow;
   166 
   167     Elevator* _level;
   168     bool _local_level;
   169 
   170     typedef typename Graph::template NodeMap<Value> ExcessMap;
   171     ExcessMap* _excess;
   172     
   173     Tolerance _tolerance;
   174 
   175     bool _phase;
   176 
   177     // constant for treesize
   178     static const double _tree_bound = 2;
   179     int _max_tree_size;
   180     
   181     //tags for the dynamic tree
   182     DynTree *_dt; 
   183     //datastructure of dyanamic tree
   184     IntNodeMap *_dt_index;
   185     //datastrucure for solution of the communication between the two class
   186     EdgeNodeMap *_dt_edges; 
   187     //nodeMap for storing the outgoing edge from the node in the tree  
   188 
   189     //max of the type Value
   190     const Value _max_value;
   191             
   192   public:
   193 
   194     typedef GoldbergTarjan Create;
   195 
   196     ///\name Named template parameters
   197 
   198     ///@{
   199 
   200     template <typename _FlowMap>
   201     struct DefFlowMapTraits : public Traits {
   202       typedef _FlowMap FlowMap;
   203       static FlowMap *createFlowMap(const Graph&) {
   204 	throw UninitializedParameter();
   205       }
   206     };
   207 
   208     /// \brief \ref named-templ-param "Named parameter" for setting
   209     /// FlowMap type
   210     ///
   211     /// \ref named-templ-param "Named parameter" for setting FlowMap
   212     /// type
   213     template <typename _FlowMap>
   214     struct DefFlowMap 
   215       : public GoldbergTarjan<Graph, CapacityMap, 
   216 			      DefFlowMapTraits<_FlowMap> > {
   217       typedef GoldbergTarjan<Graph, CapacityMap, 
   218 			     DefFlowMapTraits<_FlowMap> > Create;
   219     };
   220 
   221     template <typename _Elevator>
   222     struct DefElevatorTraits : public Traits {
   223       typedef _Elevator Elevator;
   224       static Elevator *createElevator(const Graph&, int) {
   225 	throw UninitializedParameter();
   226       }
   227     };
   228 
   229     /// \brief \ref named-templ-param "Named parameter" for setting
   230     /// Elevator type
   231     ///
   232     /// \ref named-templ-param "Named parameter" for setting Elevator
   233     /// type
   234     template <typename _Elevator>
   235     struct DefElevator 
   236       : public GoldbergTarjan<Graph, CapacityMap, 
   237 			      DefElevatorTraits<_Elevator> > {
   238       typedef GoldbergTarjan<Graph, CapacityMap, 
   239 			     DefElevatorTraits<_Elevator> > Create;
   240     };
   241 
   242     template <typename _Elevator>
   243     struct DefStandardElevatorTraits : public Traits {
   244       typedef _Elevator Elevator;
   245       static Elevator *createElevator(const Graph& graph, int max_level) {
   246 	return new Elevator(graph, max_level);
   247       }
   248     };
   249 
   250     /// \brief \ref named-templ-param "Named parameter" for setting
   251     /// Elevator type
   252     ///
   253     /// \ref named-templ-param "Named parameter" for setting Elevator
   254     /// type. The Elevator should be standard constructor interface, ie.
   255     /// the graph and the maximum level should be passed to it.
   256     template <typename _Elevator>
   257     struct DefStandardElevator 
   258       : public GoldbergTarjan<Graph, CapacityMap, 
   259 			      DefStandardElevatorTraits<_Elevator> > {
   260       typedef GoldbergTarjan<Graph, CapacityMap, 
   261 			     DefStandardElevatorTraits<_Elevator> > Create;
   262     };    
   263 
   264 
   265     ///\ref Exception for the case when s=t.
   266 
   267     ///\ref Exception for the case when the source equals the target.
   268     class InvalidArgument : public lemon::LogicError {
   269     public:
   270       virtual const char* what() const throw() {
   271 	return "lemon::GoldbergTarjan::InvalidArgument";
   272       }
   273     };
   274 
   275   protected:
   276     
   277     GoldbergTarjan() {}
   278 
   279   public:
   280 
   281     /// \brief The constructor of the class.
   282     ///
   283     /// The constructor of the class. 
   284     /// \param graph The directed graph the algorithm runs on. 
   285     /// \param capacity The capacity of the edges. 
   286     /// \param source The source node.
   287     /// \param target The target node.
   288     /// Except the graph, all of these parameters can be reset by
   289     /// calling \ref source, \ref target, \ref capacityMap and \ref
   290     /// flowMap, resp.
   291     GoldbergTarjan(const Graph& graph, const CapacityMap& capacity,
   292 		   Node source, Node target) 
   293       : _graph(graph), _capacity(&capacity), 
   294 	_node_num(), _source(source), _target(target),
   295 	_flow(0), _local_flow(false),
   296 	_level(0), _local_level(false),
   297 	_excess(0), _tolerance(), 
   298 	_phase(), _max_tree_size(),
   299 	_dt(0), _dt_index(0), _dt_edges(0), 
   300 	_max_value(std::numeric_limits<Value>::max()) { 
   301       if (_source == _target) throw InvalidArgument();
   302     }
   303 
   304     /// \brief Destrcutor.
   305     ///
   306     /// Destructor.
   307     ~GoldbergTarjan() {
   308       destroyStructures();
   309     }
   310     
   311     /// \brief Sets the capacity map.
   312     ///
   313     /// Sets the capacity map.
   314     /// \return \c (*this)
   315     GoldbergTarjan& capacityMap(const CapacityMap& map) {
   316       _capacity = &map;
   317       return *this;
   318     }
   319 
   320     /// \brief Sets the flow map.
   321     ///
   322     /// Sets the flow map.
   323     /// \return \c (*this)
   324     GoldbergTarjan& flowMap(FlowMap& map) {
   325       if (_local_flow) {
   326 	delete _flow;
   327 	_local_flow = false;
   328       }
   329       _flow = &map;
   330       return *this;
   331     }
   332 
   333     /// \brief Returns the flow map.
   334     ///
   335     /// \return The flow map.
   336     const FlowMap& flowMap() {
   337       return *_flow;
   338     }
   339 
   340     /// \brief Sets the elevator.
   341     ///
   342     /// Sets the elevator.
   343     /// \return \c (*this)
   344     GoldbergTarjan& elevator(Elevator& elevator) {
   345       if (_local_level) {
   346 	delete _level;
   347 	_local_level = false;
   348       }
   349       _level = &elevator;
   350       return *this;
   351     }
   352 
   353     /// \brief Returns the elevator.
   354     ///
   355     /// \return The elevator.
   356     const Elevator& elevator() {
   357       return *_level;
   358     }
   359 
   360     /// \brief Sets the source node.
   361     ///
   362     /// Sets the source node.
   363     /// \return \c (*this)
   364     GoldbergTarjan& source(const Node& node) {
   365       _source = node;
   366       return *this;
   367     }
   368 
   369     /// \brief Sets the target node.
   370     ///
   371     /// Sets the target node.
   372     /// \return \c (*this)
   373     GoldbergTarjan& target(const Node& node) {
   374       _target = node;
   375       return *this;
   376     }
   377  
   378     /// \brief Sets the tolerance used by algorithm.
   379     ///
   380     /// Sets the tolerance used by algorithm.
   381     GoldbergTarjan& tolerance(const Tolerance& tolerance) const {
   382       _tolerance = tolerance;
   383       if (_dt) {
   384 	_dt->tolerance(_tolerance);
   385       }
   386       return *this;
   387     } 
   388 
   389     /// \brief Returns the tolerance used by algorithm.
   390     ///
   391     /// Returns the tolerance used by algorithm.
   392     const Tolerance& tolerance() const {
   393       return tolerance;
   394     } 
   395 
   396          
   397   private:
   398     
   399     void createStructures() {
   400       _node_num = countNodes(_graph);
   401 
   402       _max_tree_size = int((double(_node_num) * double(_node_num)) / 
   403 			   double(countEdges(_graph)));
   404 
   405       if (!_flow) {
   406 	_flow = Traits::createFlowMap(_graph);
   407 	_local_flow = true;
   408       }
   409       if (!_level) {
   410 	_level = Traits::createElevator(_graph, _node_num);
   411 	_local_level = true;
   412       }
   413       if (!_excess) {
   414 	_excess = new ExcessMap(_graph);
   415       }
   416       if (!_dt_index && !_dt) {
   417 	_dt_index = new IntNodeMap(_graph);
   418 	_dt = new DynTree(*_dt_index, _tolerance);
   419       }
   420       if (!_dt_edges) {
   421 	_dt_edges = new EdgeNodeMap(_graph);
   422       }
   423     }
   424 
   425     void destroyStructures() {
   426       if (_local_flow) {
   427 	delete _flow;
   428       }
   429       if (_local_level) {
   430 	delete _level;
   431       }
   432       if (_excess) {
   433 	delete _excess;
   434       }
   435       if (_dt) {
   436 	delete _dt;
   437       }
   438       if (_dt_index) {
   439 	delete _dt_index;
   440       }
   441       if (_dt_edges) {
   442 	delete _dt_edges;
   443       }
   444     }
   445 
   446     bool sendOut(Node n, Value& excess) {
   447 
   448       Node w = _dt->findRoot(n);
   449       
   450       while (w != n) {
   451       
   452 	Value rem;      
   453 	Node u = _dt->findCost(n, rem);
   454 
   455         if (_tolerance.positive(rem) && !_level->active(w) && w != _target) {
   456 	  _level->activate(w);
   457         }
   458 
   459         if (!_tolerance.less(rem, excess)) {
   460 	  _dt->addCost(n, - excess);
   461 	  _excess->set(w, (*_excess)[w] + excess);
   462 	  excess = 0;
   463 	  return true;
   464 	}
   465 	
   466 	
   467         _dt->addCost(n, - rem);
   468 
   469         excess -= rem;
   470         _excess->set(w, (*_excess)[w] + rem);
   471         
   472 	_dt->cut(u);
   473 	_dt->addCost(u, _max_value);
   474 	  
   475 	Edge e = (*_dt_edges)[u];
   476 	_dt_edges->set(u, INVALID);
   477         
   478 	if (u == _graph.source(e)) {
   479 	  _flow->set(e, (*_capacity)[e]);
   480 	} else {
   481 	  _flow->set(e, 0);
   482 	}           
   483 
   484         w = _dt->findRoot(n);
   485       }      
   486       return false;
   487     }
   488 
   489     bool sendIn(Node n, Value& excess) {
   490 
   491       Node w = _dt->findRoot(n);
   492       
   493       while (w != n) {
   494       
   495 	Value rem;      
   496 	Node u = _dt->findCost(n, rem);
   497 
   498         if (_tolerance.positive(rem) && !_level->active(w) && w != _source) {
   499 	  _level->activate(w);
   500         }
   501 
   502         if (!_tolerance.less(rem, excess)) {
   503 	  _dt->addCost(n, - excess);
   504 	  _excess->set(w, (*_excess)[w] + excess);
   505 	  excess = 0;
   506 	  return true;
   507 	}
   508 	
   509 	
   510         _dt->addCost(n, - rem);
   511 
   512         excess -= rem;
   513         _excess->set(w, (*_excess)[w] + rem);
   514         
   515 	_dt->cut(u);
   516 	_dt->addCost(u, _max_value);
   517 	  
   518 	Edge e = (*_dt_edges)[u];
   519 	_dt_edges->set(u, INVALID);
   520         
   521 	if (u == _graph.source(e)) {
   522 	  _flow->set(e, (*_capacity)[e]);
   523 	} else {
   524 	  _flow->set(e, 0);
   525 	}           
   526 
   527         w = _dt->findRoot(n);
   528       }      
   529       return false;
   530     }
   531 
   532     void cutChildren(Node n) {
   533     
   534       for (OutEdgeIt e(_graph, n); e != INVALID; ++e) {
   535         
   536         Node v = _graph.target(e);
   537         
   538         if ((*_dt_edges)[v] != INVALID && (*_dt_edges)[v] == e) {
   539           _dt->cut(v);
   540           _dt_edges->set(v, INVALID);
   541 	  Value rem;
   542           _dt->findCost(v, rem);
   543           _dt->addCost(v, - rem);
   544           _dt->addCost(v, _max_value);
   545           _flow->set(e, rem);
   546 
   547         }      
   548       }
   549 
   550       for (InEdgeIt e(_graph, n); e != INVALID; ++e) {
   551         
   552         Node v = _graph.source(e);
   553         
   554         if ((*_dt_edges)[v] != INVALID && (*_dt_edges)[v] == e) {
   555           _dt->cut(v);
   556           _dt_edges->set(v, INVALID);
   557 	  Value rem;
   558           _dt->findCost(v, rem);
   559           _dt->addCost(v, - rem);
   560           _dt->addCost(v, _max_value);
   561           _flow->set(e, (*_capacity)[e] - rem);
   562 
   563         }      
   564       }
   565     }
   566 
   567     void extractTrees() {
   568       for (NodeIt n(_graph); n != INVALID; ++n) {
   569 	
   570 	Node w = _dt->findRoot(n);
   571       
   572 	while (w != n) {
   573       
   574 	  Value rem;      
   575 	  Node u = _dt->findCost(n, rem);
   576 
   577 	  _dt->cut(u);
   578 	  _dt->addCost(u, - rem);
   579 	  _dt->addCost(u, _max_value);
   580 	  
   581 	  Edge e = (*_dt_edges)[u];
   582 	  _dt_edges->set(u, INVALID);
   583 	  
   584 	  if (u == _graph.source(e)) {
   585 	    _flow->set(e, (*_capacity)[e] - rem);
   586 	  } else {
   587 	    _flow->set(e, rem);
   588 	  }
   589 	  
   590 	  w = _dt->findRoot(n);
   591 	}      
   592       }
   593     }
   594 
   595   public:    
   596 
   597     /// \name Execution control The simplest way to execute the
   598     /// algorithm is to use one of the member functions called \c
   599     /// run().  
   600     /// \n
   601     /// If you need more control on initial solution or
   602     /// execution then you have to call one \ref init() function and then
   603     /// the startFirstPhase() and if you need the startSecondPhase().  
   604     
   605     ///@{
   606 
   607     /// \brief Initializes the internal data structures.
   608     ///
   609     /// Initializes the internal data structures.
   610     ///
   611     void init() {
   612       createStructures();
   613 
   614       for (NodeIt n(_graph); n != INVALID; ++n) {
   615 	_excess->set(n, 0);
   616       }
   617 
   618       for (EdgeIt e(_graph); e != INVALID; ++e) {
   619 	_flow->set(e, 0);
   620       }
   621 
   622       _dt->clear();
   623       for (NodeIt v(_graph); v != INVALID; ++v) {
   624         (*_dt_edges)[v] = INVALID;
   625         _dt->makeTree(v);
   626         _dt->addCost(v, _max_value);
   627       }
   628 
   629       typename Graph::template NodeMap<bool> reached(_graph, false);
   630 
   631       _level->initStart();
   632       _level->initAddItem(_target);
   633 
   634       std::vector<Node> queue;
   635       reached.set(_source, true);
   636 
   637       queue.push_back(_target);
   638       reached.set(_target, true);
   639       while (!queue.empty()) {
   640 	_level->initNewLevel();
   641 	std::vector<Node> nqueue;
   642 	for (int i = 0; i < int(queue.size()); ++i) {
   643 	  Node n = queue[i];
   644 	  for (InEdgeIt e(_graph, n); e != INVALID; ++e) {
   645 	    Node u = _graph.source(e);
   646 	    if (!reached[u] && _tolerance.positive((*_capacity)[e])) {
   647 	      reached.set(u, true);
   648 	      _level->initAddItem(u);
   649 	      nqueue.push_back(u);
   650 	    }
   651 	  }
   652 	}
   653 	queue.swap(nqueue);
   654       }
   655       _level->initFinish();
   656 
   657       for (OutEdgeIt e(_graph, _source); e != INVALID; ++e) {
   658 	if (_tolerance.positive((*_capacity)[e])) {
   659 	  Node u = _graph.target(e);
   660 	  if ((*_level)[u] == _level->maxLevel()) continue;
   661 	  _flow->set(e, (*_capacity)[e]);
   662 	  _excess->set(u, (*_excess)[u] + (*_capacity)[e]);
   663 	  if (u != _target && !_level->active(u)) {
   664 	    _level->activate(u);
   665 	  }
   666 	}
   667       }
   668     }
   669 
   670     /// \brief Starts the first phase of the preflow algorithm.
   671     ///
   672     /// The preflow algorithm consists of two phases, this method runs
   673     /// the first phase. After the first phase the maximum flow value
   674     /// and a minimum value cut can already be computed, although a
   675     /// maximum flow is not yet obtained. So after calling this method
   676     /// \ref flowValue() returns the value of a maximum flow and \ref
   677     /// minCut() returns a minimum cut.     
   678     /// \pre One of the \ref init() functions should be called. 
   679     void startFirstPhase() {
   680       _phase = true;
   681       Node n;
   682 
   683       while ((n = _level->highestActive()) != INVALID) {
   684 	Value excess = (*_excess)[n];
   685 	int level = _level->highestActiveLevel();
   686 	int new_level = _level->maxLevel();
   687 
   688 	if (_dt->findRoot(n) != n) {
   689 	  if (sendOut(n, excess)) goto no_more_push;
   690 	}
   691 	
   692 	for (OutEdgeIt e(_graph, n); e != INVALID; ++e) {
   693 	  Value rem = (*_capacity)[e] - (*_flow)[e];
   694 	  Node v = _graph.target(e);
   695 	  
   696 	  if (!_tolerance.positive(rem) && (*_dt_edges)[v] != e) continue;
   697 
   698 	  if ((*_level)[v] < level) {
   699 	    
   700 	    if (_dt->findSize(n) + _dt->findSize(v) < 
   701 		_tree_bound * _max_tree_size) {
   702 	      _dt->addCost(n, -_max_value);
   703 	      _dt->addCost(n, rem);
   704 	      _dt->link(n, v);
   705 	      _dt_edges->set(n, e);
   706 	      if (sendOut(n, excess)) goto no_more_push;
   707 	    } else {
   708 	      if (!_level->active(v) && v != _target) {
   709 		_level->activate(v);
   710 	      }
   711 	      if (!_tolerance.less(rem, excess)) {
   712 		_flow->set(e, (*_flow)[e] + excess);
   713 		_excess->set(v, (*_excess)[v] + excess);
   714 		excess = 0;		  
   715 		goto no_more_push;
   716 	      } else {
   717 		excess -= rem;
   718 		_excess->set(v, (*_excess)[v] + rem);
   719 		_flow->set(e, (*_capacity)[e]);
   720 	      }		
   721 	    }
   722 	  } else if (new_level > (*_level)[v]) {
   723 	    new_level = (*_level)[v];
   724 	  }
   725 	}
   726 
   727 	for (InEdgeIt e(_graph, n); e != INVALID; ++e) {
   728 	  Value rem = (*_flow)[e];
   729 	  Node v = _graph.source(e);
   730 	  if (!_tolerance.positive(rem) && (*_dt_edges)[v] != e) continue;
   731 
   732 	  if ((*_level)[v] < level) {
   733 	    
   734 	    if (_dt->findSize(n) + _dt->findSize(v) < 
   735 		_tree_bound * _max_tree_size) {
   736 	      _dt->addCost(n, - _max_value);
   737 	      _dt->addCost(n, rem);
   738 	      _dt->link(n, v);
   739 	      _dt_edges->set(n, e);
   740 	      if (sendOut(n, excess)) goto no_more_push;
   741 	    } else {
   742 	      if (!_level->active(v) && v != _target) {
   743 		_level->activate(v);
   744 	      }
   745 	      if (!_tolerance.less(rem, excess)) {
   746 		_flow->set(e, (*_flow)[e] - excess);
   747 		_excess->set(v, (*_excess)[v] + excess);
   748 		excess = 0;		  
   749 		goto no_more_push;
   750 	      } else {
   751 		excess -= rem;
   752 		_excess->set(v, (*_excess)[v] + rem);
   753 		_flow->set(e, 0);
   754 	      }		
   755 	    }
   756 	  } else if (new_level > (*_level)[v]) {
   757 	    new_level = (*_level)[v];
   758 	  }
   759 	}
   760 		
   761       no_more_push:
   762 
   763 	_excess->set(n, excess);
   764 	
   765 	if (excess != 0) {
   766 	  cutChildren(n);
   767 	  if (new_level + 1 < _level->maxLevel()) {
   768 	    _level->liftHighestActive(new_level + 1);
   769 	  } else {
   770 	    _level->liftHighestActiveToTop();
   771 	  }
   772 	  if (_level->emptyLevel(level)) {
   773 	    _level->liftToTop(level);
   774 	  }
   775 	} else {
   776 	  _level->deactivate(n);
   777 	}	
   778       }
   779       extractTrees();
   780     }
   781 
   782     /// \brief Starts the second phase of the preflow algorithm.
   783     ///
   784     /// The preflow algorithm consists of two phases, this method runs
   785     /// the second phase. After calling \ref init() and \ref
   786     /// startFirstPhase() and then \ref startSecondPhase(), \ref
   787     /// flowMap() return a maximum flow, \ref flowValue() returns the
   788     /// value of a maximum flow, \ref minCut() returns a minimum cut
   789     /// \pre The \ref init() and startFirstPhase() functions should be
   790     /// called before.
   791     void startSecondPhase() {
   792       _phase = false;
   793       
   794       typename Graph::template NodeMap<bool> reached(_graph, false);
   795       for (NodeIt n(_graph); n != INVALID; ++n) {
   796 	reached.set(n, (*_level)[n] < _level->maxLevel());
   797       }
   798 
   799       _level->initStart();
   800       _level->initAddItem(_source);
   801  
   802       std::vector<Node> queue;
   803       queue.push_back(_source);
   804       reached.set(_source, true);
   805 
   806       while (!queue.empty()) {
   807 	_level->initNewLevel();
   808 	std::vector<Node> nqueue;
   809 	for (int i = 0; i < int(queue.size()); ++i) {
   810 	  Node n = queue[i];
   811 	  for (OutEdgeIt e(_graph, n); e != INVALID; ++e) {
   812 	    Node v = _graph.target(e);
   813 	    if (!reached[v] && _tolerance.positive((*_flow)[e])) {
   814 	      reached.set(v, true);
   815 	      _level->initAddItem(v);
   816 	      nqueue.push_back(v);
   817 	    }
   818 	  }
   819 	  for (InEdgeIt e(_graph, n); e != INVALID; ++e) {
   820 	    Node u = _graph.source(e);
   821 	    if (!reached[u] && 
   822 		_tolerance.positive((*_capacity)[e] - (*_flow)[e])) {
   823 	      reached.set(u, true);
   824 	      _level->initAddItem(u);
   825 	      nqueue.push_back(u);
   826 	    }
   827 	  }
   828 	}
   829 	queue.swap(nqueue);
   830       }
   831       _level->initFinish();
   832 
   833       for (NodeIt n(_graph); n != INVALID; ++n) {
   834 	if (!reached[n]) {
   835 	  _level->markToBottom(n);
   836 	} else if ((*_excess)[n] > 0 && _target != n) {
   837 	  _level->activate(n);
   838 	}
   839       }
   840 
   841       Node n;
   842 
   843       while ((n = _level->highestActive()) != INVALID) {
   844 	Value excess = (*_excess)[n];
   845 	int level = _level->highestActiveLevel();
   846 	int new_level = _level->maxLevel();
   847 
   848 	if (_dt->findRoot(n) != n) {
   849 	  if (sendIn(n, excess)) goto no_more_push;
   850 	}
   851 	
   852 	for (OutEdgeIt e(_graph, n); e != INVALID; ++e) {
   853 	  Value rem = (*_capacity)[e] - (*_flow)[e];
   854 	  Node v = _graph.target(e);
   855 	  
   856 	  if (!_tolerance.positive(rem) && (*_dt_edges)[v] != e) continue;
   857 
   858 	  if ((*_level)[v] < level) {
   859 	    
   860 	    if (_dt->findSize(n) + _dt->findSize(v) < 
   861 		_tree_bound * _max_tree_size) {
   862 	      _dt->addCost(n, -_max_value);
   863 	      _dt->addCost(n, rem);
   864 	      _dt->link(n, v);
   865 	      _dt_edges->set(n, e);
   866 	      if (sendIn(n, excess)) goto no_more_push;
   867 	    } else {
   868 	      if (!_level->active(v) && v != _source) {
   869 		_level->activate(v);
   870 	      }
   871 	      if (!_tolerance.less(rem, excess)) {
   872 		_flow->set(e, (*_flow)[e] + excess);
   873 		_excess->set(v, (*_excess)[v] + excess);
   874 		excess = 0;		  
   875 		goto no_more_push;
   876 	      } else {
   877 		excess -= rem;
   878 		_excess->set(v, (*_excess)[v] + rem);
   879 		_flow->set(e, (*_capacity)[e]);
   880 	      }		
   881 	    }
   882 	  } else if (new_level > (*_level)[v]) {
   883 	    new_level = (*_level)[v];
   884 	  }
   885 	}
   886 
   887 	for (InEdgeIt e(_graph, n); e != INVALID; ++e) {
   888 	  Value rem = (*_flow)[e];
   889 	  Node v = _graph.source(e);
   890 	  if (!_tolerance.positive(rem) && (*_dt_edges)[v] != e) continue;
   891 
   892 	  if ((*_level)[v] < level) {
   893 	    
   894 	    if (_dt->findSize(n) + _dt->findSize(v) < 
   895 		_tree_bound * _max_tree_size) {
   896 	      _dt->addCost(n, - _max_value);
   897 	      _dt->addCost(n, rem);
   898 	      _dt->link(n, v);
   899 	      _dt_edges->set(n, e);
   900 	      if (sendIn(n, excess)) goto no_more_push;
   901 	    } else {
   902 	      if (!_level->active(v) && v != _source) {
   903 		_level->activate(v);
   904 	      }
   905 	      if (!_tolerance.less(rem, excess)) {
   906 		_flow->set(e, (*_flow)[e] - excess);
   907 		_excess->set(v, (*_excess)[v] + excess);
   908 		excess = 0;		  
   909 		goto no_more_push;
   910 	      } else {
   911 		excess -= rem;
   912 		_excess->set(v, (*_excess)[v] + rem);
   913 		_flow->set(e, 0);
   914 	      }		
   915 	    }
   916 	  } else if (new_level > (*_level)[v]) {
   917 	    new_level = (*_level)[v];
   918 	  }
   919 	}
   920 		
   921       no_more_push:
   922 
   923 	_excess->set(n, excess);
   924 	
   925 	if (excess != 0) {
   926 	  cutChildren(n);
   927 	  if (new_level + 1 < _level->maxLevel()) {
   928 	    _level->liftHighestActive(new_level + 1);
   929 	  } else {
   930 	    _level->liftHighestActiveToTop();
   931 	  }
   932 	  if (_level->emptyLevel(level)) {
   933 	    _level->liftToTop(level);
   934 	  }
   935 	} else {
   936 	  _level->deactivate(n);
   937 	}	
   938       }
   939       extractTrees();
   940     }
   941 
   942     /// \brief Runs the Goldberg-Tarjan algorithm.  
   943     ///
   944     /// Runs the Goldberg-Tarjan algorithm.
   945     /// \note pf.run() is just a shortcut of the following code.
   946     /// \code
   947     ///   pf.init();
   948     ///   pf.startFirstPhase();
   949     ///   pf.startSecondPhase();
   950     /// \endcode
   951     void run() {
   952       init();
   953       startFirstPhase();
   954       startSecondPhase();
   955     }
   956 
   957     /// \brief Runs the Goldberg-Tarjan algorithm to compute the minimum cut.  
   958     ///
   959     /// Runs the Goldberg-Tarjan algorithm to compute the minimum cut.
   960     /// \note pf.runMinCut() is just a shortcut of the following code.
   961     /// \code
   962     ///   pf.init();
   963     ///   pf.startFirstPhase();
   964     /// \endcode
   965     void runMinCut() {
   966       init();
   967       startFirstPhase();
   968     }
   969 
   970     /// @}
   971 
   972     /// \name Query Functions 
   973     /// The result of the Goldberg-Tarjan algorithm can be obtained
   974     /// using these functions.
   975     /// \n
   976     /// Before the use of these functions, either run() or start() must
   977     /// be called.
   978     
   979     ///@{
   980 
   981     /// \brief Returns the value of the maximum flow.
   982     ///
   983     /// Returns the value of the maximum flow by returning the excess
   984     /// of the target node \c t. This value equals to the value of
   985     /// the maximum flow already after the first phase.
   986     Value flowValue() const {
   987       return (*_excess)[_target];
   988     }
   989 
   990     /// \brief Returns true when the node is on the source side of minimum cut.
   991     ///
   992     /// Returns true when the node is on the source side of minimum
   993     /// cut. This method can be called both after running \ref
   994     /// startFirstPhase() and \ref startSecondPhase().
   995     bool minCut(const Node& node) const {
   996       return ((*_level)[node] == _level->maxLevel()) == _phase;
   997     }
   998  
   999     /// \brief Returns a minimum value cut.
  1000     ///
  1001     /// Sets the \c cutMap to the characteristic vector of a minimum value
  1002     /// cut. This method can be called both after running \ref
  1003     /// startFirstPhase() and \ref startSecondPhase(). The result after second
  1004     /// phase could be changed slightly if inexact computation is used.
  1005     /// \pre The \c cutMap should be a bool-valued node-map.
  1006     template <typename CutMap>
  1007     void minCutMap(CutMap& cutMap) const {
  1008       for (NodeIt n(_graph); n != INVALID; ++n) {
  1009 	cutMap.set(n, minCut(n));
  1010       }
  1011     }
  1012 
  1013     /// \brief Returns the flow on the edge.
  1014     ///
  1015     /// Sets the \c flowMap to the flow on the edges. This method can
  1016     /// be called after the second phase of algorithm.
  1017     Value flow(const Edge& edge) const {
  1018       return (*_flow)[edge];
  1019     }
  1020 
  1021     /// @}
  1022 
  1023   }; 
  1024   
  1025 } //namespace lemon
  1026 
  1027 #endif