lemon/network_simplex.h
author Alpar Juttner <alpar@cs.elte.hu>
Tue, 05 Jun 2012 12:06:28 +0200
changeset 1054 c40a9d94442d
parent 922 9312d6c89d02
child 1004 d59484d5fc1f
permissions -rw-r--r--
New features in time_measure.h (#442)
     1 /* -*- mode: C++; indent-tabs-mode: nil; -*-
     2  *
     3  * This file is a part of LEMON, a generic C++ optimization library.
     4  *
     5  * Copyright (C) 2003-2010
     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_NETWORK_SIMPLEX_H
    20 #define LEMON_NETWORK_SIMPLEX_H
    21 
    22 /// \ingroup min_cost_flow_algs
    23 ///
    24 /// \file
    25 /// \brief Network Simplex algorithm for finding a minimum cost flow.
    26 
    27 #include <vector>
    28 #include <limits>
    29 #include <algorithm>
    30 
    31 #include <lemon/core.h>
    32 #include <lemon/math.h>
    33 
    34 namespace lemon {
    35 
    36   /// \addtogroup min_cost_flow_algs
    37   /// @{
    38 
    39   /// \brief Implementation of the primal Network Simplex algorithm
    40   /// for finding a \ref min_cost_flow "minimum cost flow".
    41   ///
    42   /// \ref NetworkSimplex implements the primal Network Simplex algorithm
    43   /// for finding a \ref min_cost_flow "minimum cost flow"
    44   /// \ref amo93networkflows, \ref dantzig63linearprog,
    45   /// \ref kellyoneill91netsimplex.
    46   /// This algorithm is a highly efficient specialized version of the
    47   /// linear programming simplex method directly for the minimum cost
    48   /// flow problem.
    49   ///
    50   /// In general, \ref NetworkSimplex and \ref CostScaling are the fastest
    51   /// implementations available in LEMON for this problem.
    52   /// Furthermore, this class supports both directions of the supply/demand
    53   /// inequality constraints. For more information, see \ref SupplyType.
    54   ///
    55   /// Most of the parameters of the problem (except for the digraph)
    56   /// can be given using separate functions, and the algorithm can be
    57   /// executed using the \ref run() function. If some parameters are not
    58   /// specified, then default values will be used.
    59   ///
    60   /// \tparam GR The digraph type the algorithm runs on.
    61   /// \tparam V The number type used for flow amounts, capacity bounds
    62   /// and supply values in the algorithm. By default, it is \c int.
    63   /// \tparam C The number type used for costs and potentials in the
    64   /// algorithm. By default, it is the same as \c V.
    65   ///
    66   /// \warning Both \c V and \c C must be signed number types.
    67   /// \warning All input data (capacities, supply values, and costs) must
    68   /// be integer.
    69   ///
    70   /// \note %NetworkSimplex provides five different pivot rule
    71   /// implementations, from which the most efficient one is used
    72   /// by default. For more information, see \ref PivotRule.
    73   template <typename GR, typename V = int, typename C = V>
    74   class NetworkSimplex
    75   {
    76   public:
    77 
    78     /// The type of the flow amounts, capacity bounds and supply values
    79     typedef V Value;
    80     /// The type of the arc costs
    81     typedef C Cost;
    82 
    83   public:
    84 
    85     /// \brief Problem type constants for the \c run() function.
    86     ///
    87     /// Enum type containing the problem type constants that can be
    88     /// returned by the \ref run() function of the algorithm.
    89     enum ProblemType {
    90       /// The problem has no feasible solution (flow).
    91       INFEASIBLE,
    92       /// The problem has optimal solution (i.e. it is feasible and
    93       /// bounded), and the algorithm has found optimal flow and node
    94       /// potentials (primal and dual solutions).
    95       OPTIMAL,
    96       /// The objective function of the problem is unbounded, i.e.
    97       /// there is a directed cycle having negative total cost and
    98       /// infinite upper bound.
    99       UNBOUNDED
   100     };
   101 
   102     /// \brief Constants for selecting the type of the supply constraints.
   103     ///
   104     /// Enum type containing constants for selecting the supply type,
   105     /// i.e. the direction of the inequalities in the supply/demand
   106     /// constraints of the \ref min_cost_flow "minimum cost flow problem".
   107     ///
   108     /// The default supply type is \c GEQ, the \c LEQ type can be
   109     /// selected using \ref supplyType().
   110     /// The equality form is a special case of both supply types.
   111     enum SupplyType {
   112       /// This option means that there are <em>"greater or equal"</em>
   113       /// supply/demand constraints in the definition of the problem.
   114       GEQ,
   115       /// This option means that there are <em>"less or equal"</em>
   116       /// supply/demand constraints in the definition of the problem.
   117       LEQ
   118     };
   119 
   120     /// \brief Constants for selecting the pivot rule.
   121     ///
   122     /// Enum type containing constants for selecting the pivot rule for
   123     /// the \ref run() function.
   124     ///
   125     /// \ref NetworkSimplex provides five different implementations for
   126     /// the pivot strategy that significantly affects the running time
   127     /// of the algorithm.
   128     /// According to experimental tests conducted on various problem
   129     /// instances, \ref BLOCK_SEARCH "Block Search" and
   130     /// \ref ALTERING_LIST "Altering Candidate List" rules turned out
   131     /// to be the most efficient.
   132     /// Since \ref BLOCK_SEARCH "Block Search" is a simpler strategy that
   133     /// seemed to be slightly more robust, it is used by default.
   134     /// However, another pivot rule can easily be selected using the
   135     /// \ref run() function with the proper parameter.
   136     enum PivotRule {
   137 
   138       /// The \e First \e Eligible pivot rule.
   139       /// The next eligible arc is selected in a wraparound fashion
   140       /// in every iteration.
   141       FIRST_ELIGIBLE,
   142 
   143       /// The \e Best \e Eligible pivot rule.
   144       /// The best eligible arc is selected in every iteration.
   145       BEST_ELIGIBLE,
   146 
   147       /// The \e Block \e Search pivot rule.
   148       /// A specified number of arcs are examined in every iteration
   149       /// in a wraparound fashion and the best eligible arc is selected
   150       /// from this block.
   151       BLOCK_SEARCH,
   152 
   153       /// The \e Candidate \e List pivot rule.
   154       /// In a major iteration a candidate list is built from eligible arcs
   155       /// in a wraparound fashion and in the following minor iterations
   156       /// the best eligible arc is selected from this list.
   157       CANDIDATE_LIST,
   158 
   159       /// The \e Altering \e Candidate \e List pivot rule.
   160       /// It is a modified version of the Candidate List method.
   161       /// It keeps only a few of the best eligible arcs from the former
   162       /// candidate list and extends this list in every iteration.
   163       ALTERING_LIST
   164     };
   165 
   166   private:
   167 
   168     TEMPLATE_DIGRAPH_TYPEDEFS(GR);
   169 
   170     typedef std::vector<int> IntVector;
   171     typedef std::vector<Value> ValueVector;
   172     typedef std::vector<Cost> CostVector;
   173     typedef std::vector<signed char> CharVector;
   174     // Note: vector<signed char> is used instead of vector<ArcState> and
   175     // vector<ArcDirection> for efficiency reasons
   176 
   177     // State constants for arcs
   178     enum ArcState {
   179       STATE_UPPER = -1,
   180       STATE_TREE  =  0,
   181       STATE_LOWER =  1
   182     };
   183 
   184     // Direction constants for tree arcs
   185     enum ArcDirection {
   186       DIR_DOWN = -1,
   187       DIR_UP   =  1
   188     };
   189 
   190   private:
   191 
   192     // Data related to the underlying digraph
   193     const GR &_graph;
   194     int _node_num;
   195     int _arc_num;
   196     int _all_arc_num;
   197     int _search_arc_num;
   198 
   199     // Parameters of the problem
   200     bool _have_lower;
   201     SupplyType _stype;
   202     Value _sum_supply;
   203 
   204     // Data structures for storing the digraph
   205     IntNodeMap _node_id;
   206     IntArcMap _arc_id;
   207     IntVector _source;
   208     IntVector _target;
   209     bool _arc_mixing;
   210 
   211     // Node and arc data
   212     ValueVector _lower;
   213     ValueVector _upper;
   214     ValueVector _cap;
   215     CostVector _cost;
   216     ValueVector _supply;
   217     ValueVector _flow;
   218     CostVector _pi;
   219 
   220     // Data for storing the spanning tree structure
   221     IntVector _parent;
   222     IntVector _pred;
   223     IntVector _thread;
   224     IntVector _rev_thread;
   225     IntVector _succ_num;
   226     IntVector _last_succ;
   227     CharVector _pred_dir;
   228     CharVector _state;
   229     IntVector _dirty_revs;
   230     int _root;
   231 
   232     // Temporary data used in the current pivot iteration
   233     int in_arc, join, u_in, v_in, u_out, v_out;
   234     Value delta;
   235 
   236     const Value MAX;
   237 
   238   public:
   239 
   240     /// \brief Constant for infinite upper bounds (capacities).
   241     ///
   242     /// Constant for infinite upper bounds (capacities).
   243     /// It is \c std::numeric_limits<Value>::infinity() if available,
   244     /// \c std::numeric_limits<Value>::max() otherwise.
   245     const Value INF;
   246 
   247   private:
   248 
   249     // Implementation of the First Eligible pivot rule
   250     class FirstEligiblePivotRule
   251     {
   252     private:
   253 
   254       // References to the NetworkSimplex class
   255       const IntVector  &_source;
   256       const IntVector  &_target;
   257       const CostVector &_cost;
   258       const CharVector &_state;
   259       const CostVector &_pi;
   260       int &_in_arc;
   261       int _search_arc_num;
   262 
   263       // Pivot rule data
   264       int _next_arc;
   265 
   266     public:
   267 
   268       // Constructor
   269       FirstEligiblePivotRule(NetworkSimplex &ns) :
   270         _source(ns._source), _target(ns._target),
   271         _cost(ns._cost), _state(ns._state), _pi(ns._pi),
   272         _in_arc(ns.in_arc), _search_arc_num(ns._search_arc_num),
   273         _next_arc(0)
   274       {}
   275 
   276       // Find next entering arc
   277       bool findEnteringArc() {
   278         Cost c;
   279         for (int e = _next_arc; e != _search_arc_num; ++e) {
   280           c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
   281           if (c < 0) {
   282             _in_arc = e;
   283             _next_arc = e + 1;
   284             return true;
   285           }
   286         }
   287         for (int e = 0; e != _next_arc; ++e) {
   288           c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
   289           if (c < 0) {
   290             _in_arc = e;
   291             _next_arc = e + 1;
   292             return true;
   293           }
   294         }
   295         return false;
   296       }
   297 
   298     }; //class FirstEligiblePivotRule
   299 
   300 
   301     // Implementation of the Best Eligible pivot rule
   302     class BestEligiblePivotRule
   303     {
   304     private:
   305 
   306       // References to the NetworkSimplex class
   307       const IntVector  &_source;
   308       const IntVector  &_target;
   309       const CostVector &_cost;
   310       const CharVector &_state;
   311       const CostVector &_pi;
   312       int &_in_arc;
   313       int _search_arc_num;
   314 
   315     public:
   316 
   317       // Constructor
   318       BestEligiblePivotRule(NetworkSimplex &ns) :
   319         _source(ns._source), _target(ns._target),
   320         _cost(ns._cost), _state(ns._state), _pi(ns._pi),
   321         _in_arc(ns.in_arc), _search_arc_num(ns._search_arc_num)
   322       {}
   323 
   324       // Find next entering arc
   325       bool findEnteringArc() {
   326         Cost c, min = 0;
   327         for (int e = 0; e != _search_arc_num; ++e) {
   328           c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
   329           if (c < min) {
   330             min = c;
   331             _in_arc = e;
   332           }
   333         }
   334         return min < 0;
   335       }
   336 
   337     }; //class BestEligiblePivotRule
   338 
   339 
   340     // Implementation of the Block Search pivot rule
   341     class BlockSearchPivotRule
   342     {
   343     private:
   344 
   345       // References to the NetworkSimplex class
   346       const IntVector  &_source;
   347       const IntVector  &_target;
   348       const CostVector &_cost;
   349       const CharVector &_state;
   350       const CostVector &_pi;
   351       int &_in_arc;
   352       int _search_arc_num;
   353 
   354       // Pivot rule data
   355       int _block_size;
   356       int _next_arc;
   357 
   358     public:
   359 
   360       // Constructor
   361       BlockSearchPivotRule(NetworkSimplex &ns) :
   362         _source(ns._source), _target(ns._target),
   363         _cost(ns._cost), _state(ns._state), _pi(ns._pi),
   364         _in_arc(ns.in_arc), _search_arc_num(ns._search_arc_num),
   365         _next_arc(0)
   366       {
   367         // The main parameters of the pivot rule
   368         const double BLOCK_SIZE_FACTOR = 1.0;
   369         const int MIN_BLOCK_SIZE = 10;
   370 
   371         _block_size = std::max( int(BLOCK_SIZE_FACTOR *
   372                                     std::sqrt(double(_search_arc_num))),
   373                                 MIN_BLOCK_SIZE );
   374       }
   375 
   376       // Find next entering arc
   377       bool findEnteringArc() {
   378         Cost c, min = 0;
   379         int cnt = _block_size;
   380         int e;
   381         for (e = _next_arc; e != _search_arc_num; ++e) {
   382           c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
   383           if (c < min) {
   384             min = c;
   385             _in_arc = e;
   386           }
   387           if (--cnt == 0) {
   388             if (min < 0) goto search_end;
   389             cnt = _block_size;
   390           }
   391         }
   392         for (e = 0; e != _next_arc; ++e) {
   393           c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
   394           if (c < min) {
   395             min = c;
   396             _in_arc = e;
   397           }
   398           if (--cnt == 0) {
   399             if (min < 0) goto search_end;
   400             cnt = _block_size;
   401           }
   402         }
   403         if (min >= 0) return false;
   404 
   405       search_end:
   406         _next_arc = e;
   407         return true;
   408       }
   409 
   410     }; //class BlockSearchPivotRule
   411 
   412 
   413     // Implementation of the Candidate List pivot rule
   414     class CandidateListPivotRule
   415     {
   416     private:
   417 
   418       // References to the NetworkSimplex class
   419       const IntVector  &_source;
   420       const IntVector  &_target;
   421       const CostVector &_cost;
   422       const CharVector &_state;
   423       const CostVector &_pi;
   424       int &_in_arc;
   425       int _search_arc_num;
   426 
   427       // Pivot rule data
   428       IntVector _candidates;
   429       int _list_length, _minor_limit;
   430       int _curr_length, _minor_count;
   431       int _next_arc;
   432 
   433     public:
   434 
   435       /// Constructor
   436       CandidateListPivotRule(NetworkSimplex &ns) :
   437         _source(ns._source), _target(ns._target),
   438         _cost(ns._cost), _state(ns._state), _pi(ns._pi),
   439         _in_arc(ns.in_arc), _search_arc_num(ns._search_arc_num),
   440         _next_arc(0)
   441       {
   442         // The main parameters of the pivot rule
   443         const double LIST_LENGTH_FACTOR = 0.25;
   444         const int MIN_LIST_LENGTH = 10;
   445         const double MINOR_LIMIT_FACTOR = 0.1;
   446         const int MIN_MINOR_LIMIT = 3;
   447 
   448         _list_length = std::max( int(LIST_LENGTH_FACTOR *
   449                                      std::sqrt(double(_search_arc_num))),
   450                                  MIN_LIST_LENGTH );
   451         _minor_limit = std::max( int(MINOR_LIMIT_FACTOR * _list_length),
   452                                  MIN_MINOR_LIMIT );
   453         _curr_length = _minor_count = 0;
   454         _candidates.resize(_list_length);
   455       }
   456 
   457       /// Find next entering arc
   458       bool findEnteringArc() {
   459         Cost min, c;
   460         int e;
   461         if (_curr_length > 0 && _minor_count < _minor_limit) {
   462           // Minor iteration: select the best eligible arc from the
   463           // current candidate list
   464           ++_minor_count;
   465           min = 0;
   466           for (int i = 0; i < _curr_length; ++i) {
   467             e = _candidates[i];
   468             c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
   469             if (c < min) {
   470               min = c;
   471               _in_arc = e;
   472             }
   473             else if (c >= 0) {
   474               _candidates[i--] = _candidates[--_curr_length];
   475             }
   476           }
   477           if (min < 0) return true;
   478         }
   479 
   480         // Major iteration: build a new candidate list
   481         min = 0;
   482         _curr_length = 0;
   483         for (e = _next_arc; e != _search_arc_num; ++e) {
   484           c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
   485           if (c < 0) {
   486             _candidates[_curr_length++] = e;
   487             if (c < min) {
   488               min = c;
   489               _in_arc = e;
   490             }
   491             if (_curr_length == _list_length) goto search_end;
   492           }
   493         }
   494         for (e = 0; e != _next_arc; ++e) {
   495           c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
   496           if (c < 0) {
   497             _candidates[_curr_length++] = e;
   498             if (c < min) {
   499               min = c;
   500               _in_arc = e;
   501             }
   502             if (_curr_length == _list_length) goto search_end;
   503           }
   504         }
   505         if (_curr_length == 0) return false;
   506 
   507       search_end:
   508         _minor_count = 1;
   509         _next_arc = e;
   510         return true;
   511       }
   512 
   513     }; //class CandidateListPivotRule
   514 
   515 
   516     // Implementation of the Altering Candidate List pivot rule
   517     class AlteringListPivotRule
   518     {
   519     private:
   520 
   521       // References to the NetworkSimplex class
   522       const IntVector  &_source;
   523       const IntVector  &_target;
   524       const CostVector &_cost;
   525       const CharVector &_state;
   526       const CostVector &_pi;
   527       int &_in_arc;
   528       int _search_arc_num;
   529 
   530       // Pivot rule data
   531       int _block_size, _head_length, _curr_length;
   532       int _next_arc;
   533       IntVector _candidates;
   534       CostVector _cand_cost;
   535 
   536       // Functor class to compare arcs during sort of the candidate list
   537       class SortFunc
   538       {
   539       private:
   540         const CostVector &_map;
   541       public:
   542         SortFunc(const CostVector &map) : _map(map) {}
   543         bool operator()(int left, int right) {
   544           return _map[left] < _map[right];
   545         }
   546       };
   547 
   548       SortFunc _sort_func;
   549 
   550     public:
   551 
   552       // Constructor
   553       AlteringListPivotRule(NetworkSimplex &ns) :
   554         _source(ns._source), _target(ns._target),
   555         _cost(ns._cost), _state(ns._state), _pi(ns._pi),
   556         _in_arc(ns.in_arc), _search_arc_num(ns._search_arc_num),
   557         _next_arc(0), _cand_cost(ns._search_arc_num), _sort_func(_cand_cost)
   558       {
   559         // The main parameters of the pivot rule
   560         const double BLOCK_SIZE_FACTOR = 1.0;
   561         const int MIN_BLOCK_SIZE = 10;
   562         const double HEAD_LENGTH_FACTOR = 0.01;
   563         const int MIN_HEAD_LENGTH = 3;
   564 
   565         _block_size = std::max( int(BLOCK_SIZE_FACTOR *
   566                                     std::sqrt(double(_search_arc_num))),
   567                                 MIN_BLOCK_SIZE );
   568         _head_length = std::max( int(HEAD_LENGTH_FACTOR * _block_size),
   569                                  MIN_HEAD_LENGTH );
   570         _candidates.resize(_head_length + _block_size);
   571         _curr_length = 0;
   572       }
   573 
   574       // Find next entering arc
   575       bool findEnteringArc() {
   576         // Check the current candidate list
   577         int e;
   578         Cost c;
   579         for (int i = 0; i != _curr_length; ++i) {
   580           e = _candidates[i];
   581           c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
   582           if (c < 0) {
   583             _cand_cost[e] = c;
   584           } else {
   585             _candidates[i--] = _candidates[--_curr_length];
   586           }
   587         }
   588 
   589         // Extend the list
   590         int cnt = _block_size;
   591         int limit = _head_length;
   592 
   593         for (e = _next_arc; e != _search_arc_num; ++e) {
   594           c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
   595           if (c < 0) {
   596             _cand_cost[e] = c;
   597             _candidates[_curr_length++] = e;
   598           }
   599           if (--cnt == 0) {
   600             if (_curr_length > limit) goto search_end;
   601             limit = 0;
   602             cnt = _block_size;
   603           }
   604         }
   605         for (e = 0; e != _next_arc; ++e) {
   606           c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
   607           if (c < 0) {
   608             _cand_cost[e] = c;
   609             _candidates[_curr_length++] = e;
   610           }
   611           if (--cnt == 0) {
   612             if (_curr_length > limit) goto search_end;
   613             limit = 0;
   614             cnt = _block_size;
   615           }
   616         }
   617         if (_curr_length == 0) return false;
   618 
   619       search_end:
   620 
   621         // Perform partial sort operation on the candidate list
   622         int new_length = std::min(_head_length + 1, _curr_length);
   623         std::partial_sort(_candidates.begin(), _candidates.begin() + new_length,
   624                           _candidates.begin() + _curr_length, _sort_func);
   625 
   626         // Select the entering arc and remove it from the list
   627         _in_arc = _candidates[0];
   628         _next_arc = e;
   629         _candidates[0] = _candidates[new_length - 1];
   630         _curr_length = new_length - 1;
   631         return true;
   632       }
   633 
   634     }; //class AlteringListPivotRule
   635 
   636   public:
   637 
   638     /// \brief Constructor.
   639     ///
   640     /// The constructor of the class.
   641     ///
   642     /// \param graph The digraph the algorithm runs on.
   643     /// \param arc_mixing Indicate if the arcs will be stored in a
   644     /// mixed order in the internal data structure.
   645     /// In general, it leads to similar performance as using the original
   646     /// arc order, but it makes the algorithm more robust and in special
   647     /// cases, even significantly faster. Therefore, it is enabled by default.
   648     NetworkSimplex(const GR& graph, bool arc_mixing = true) :
   649       _graph(graph), _node_id(graph), _arc_id(graph),
   650       _arc_mixing(arc_mixing),
   651       MAX(std::numeric_limits<Value>::max()),
   652       INF(std::numeric_limits<Value>::has_infinity ?
   653           std::numeric_limits<Value>::infinity() : MAX)
   654     {
   655       // Check the number types
   656       LEMON_ASSERT(std::numeric_limits<Value>::is_signed,
   657         "The flow type of NetworkSimplex must be signed");
   658       LEMON_ASSERT(std::numeric_limits<Cost>::is_signed,
   659         "The cost type of NetworkSimplex must be signed");
   660 
   661       // Reset data structures
   662       reset();
   663     }
   664 
   665     /// \name Parameters
   666     /// The parameters of the algorithm can be specified using these
   667     /// functions.
   668 
   669     /// @{
   670 
   671     /// \brief Set the lower bounds on the arcs.
   672     ///
   673     /// This function sets the lower bounds on the arcs.
   674     /// If it is not used before calling \ref run(), the lower bounds
   675     /// will be set to zero on all arcs.
   676     ///
   677     /// \param map An arc map storing the lower bounds.
   678     /// Its \c Value type must be convertible to the \c Value type
   679     /// of the algorithm.
   680     ///
   681     /// \return <tt>(*this)</tt>
   682     template <typename LowerMap>
   683     NetworkSimplex& lowerMap(const LowerMap& map) {
   684       _have_lower = true;
   685       for (ArcIt a(_graph); a != INVALID; ++a) {
   686         _lower[_arc_id[a]] = map[a];
   687       }
   688       return *this;
   689     }
   690 
   691     /// \brief Set the upper bounds (capacities) on the arcs.
   692     ///
   693     /// This function sets the upper bounds (capacities) on the arcs.
   694     /// If it is not used before calling \ref run(), the upper bounds
   695     /// will be set to \ref INF on all arcs (i.e. the flow value will be
   696     /// unbounded from above).
   697     ///
   698     /// \param map An arc map storing the upper bounds.
   699     /// Its \c Value type must be convertible to the \c Value type
   700     /// of the algorithm.
   701     ///
   702     /// \return <tt>(*this)</tt>
   703     template<typename UpperMap>
   704     NetworkSimplex& upperMap(const UpperMap& map) {
   705       for (ArcIt a(_graph); a != INVALID; ++a) {
   706         _upper[_arc_id[a]] = map[a];
   707       }
   708       return *this;
   709     }
   710 
   711     /// \brief Set the costs of the arcs.
   712     ///
   713     /// This function sets the costs of the arcs.
   714     /// If it is not used before calling \ref run(), the costs
   715     /// will be set to \c 1 on all arcs.
   716     ///
   717     /// \param map An arc map storing the costs.
   718     /// Its \c Value type must be convertible to the \c Cost type
   719     /// of the algorithm.
   720     ///
   721     /// \return <tt>(*this)</tt>
   722     template<typename CostMap>
   723     NetworkSimplex& costMap(const CostMap& map) {
   724       for (ArcIt a(_graph); a != INVALID; ++a) {
   725         _cost[_arc_id[a]] = map[a];
   726       }
   727       return *this;
   728     }
   729 
   730     /// \brief Set the supply values of the nodes.
   731     ///
   732     /// This function sets the supply values of the nodes.
   733     /// If neither this function nor \ref stSupply() is used before
   734     /// calling \ref run(), the supply of each node will be set to zero.
   735     ///
   736     /// \param map A node map storing the supply values.
   737     /// Its \c Value type must be convertible to the \c Value type
   738     /// of the algorithm.
   739     ///
   740     /// \return <tt>(*this)</tt>
   741     ///
   742     /// \sa supplyType()
   743     template<typename SupplyMap>
   744     NetworkSimplex& supplyMap(const SupplyMap& map) {
   745       for (NodeIt n(_graph); n != INVALID; ++n) {
   746         _supply[_node_id[n]] = map[n];
   747       }
   748       return *this;
   749     }
   750 
   751     /// \brief Set single source and target nodes and a supply value.
   752     ///
   753     /// This function sets a single source node and a single target node
   754     /// and the required flow value.
   755     /// If neither this function nor \ref supplyMap() is used before
   756     /// calling \ref run(), the supply of each node will be set to zero.
   757     ///
   758     /// Using this function has the same effect as using \ref supplyMap()
   759     /// with a map in which \c k is assigned to \c s, \c -k is
   760     /// assigned to \c t and all other nodes have zero supply value.
   761     ///
   762     /// \param s The source node.
   763     /// \param t The target node.
   764     /// \param k The required amount of flow from node \c s to node \c t
   765     /// (i.e. the supply of \c s and the demand of \c t).
   766     ///
   767     /// \return <tt>(*this)</tt>
   768     NetworkSimplex& stSupply(const Node& s, const Node& t, Value k) {
   769       for (int i = 0; i != _node_num; ++i) {
   770         _supply[i] = 0;
   771       }
   772       _supply[_node_id[s]] =  k;
   773       _supply[_node_id[t]] = -k;
   774       return *this;
   775     }
   776 
   777     /// \brief Set the type of the supply constraints.
   778     ///
   779     /// This function sets the type of the supply/demand constraints.
   780     /// If it is not used before calling \ref run(), the \ref GEQ supply
   781     /// type will be used.
   782     ///
   783     /// For more information, see \ref SupplyType.
   784     ///
   785     /// \return <tt>(*this)</tt>
   786     NetworkSimplex& supplyType(SupplyType supply_type) {
   787       _stype = supply_type;
   788       return *this;
   789     }
   790 
   791     /// @}
   792 
   793     /// \name Execution Control
   794     /// The algorithm can be executed using \ref run().
   795 
   796     /// @{
   797 
   798     /// \brief Run the algorithm.
   799     ///
   800     /// This function runs the algorithm.
   801     /// The paramters can be specified using functions \ref lowerMap(),
   802     /// \ref upperMap(), \ref costMap(), \ref supplyMap(), \ref stSupply(),
   803     /// \ref supplyType().
   804     /// For example,
   805     /// \code
   806     ///   NetworkSimplex<ListDigraph> ns(graph);
   807     ///   ns.lowerMap(lower).upperMap(upper).costMap(cost)
   808     ///     .supplyMap(sup).run();
   809     /// \endcode
   810     ///
   811     /// This function can be called more than once. All the given parameters
   812     /// are kept for the next call, unless \ref resetParams() or \ref reset()
   813     /// is used, thus only the modified parameters have to be set again.
   814     /// If the underlying digraph was also modified after the construction
   815     /// of the class (or the last \ref reset() call), then the \ref reset()
   816     /// function must be called.
   817     ///
   818     /// \param pivot_rule The pivot rule that will be used during the
   819     /// algorithm. For more information, see \ref PivotRule.
   820     ///
   821     /// \return \c INFEASIBLE if no feasible flow exists,
   822     /// \n \c OPTIMAL if the problem has optimal solution
   823     /// (i.e. it is feasible and bounded), and the algorithm has found
   824     /// optimal flow and node potentials (primal and dual solutions),
   825     /// \n \c UNBOUNDED if the objective function of the problem is
   826     /// unbounded, i.e. there is a directed cycle having negative total
   827     /// cost and infinite upper bound.
   828     ///
   829     /// \see ProblemType, PivotRule
   830     /// \see resetParams(), reset()
   831     ProblemType run(PivotRule pivot_rule = BLOCK_SEARCH) {
   832       if (!init()) return INFEASIBLE;
   833       return start(pivot_rule);
   834     }
   835 
   836     /// \brief Reset all the parameters that have been given before.
   837     ///
   838     /// This function resets all the paramaters that have been given
   839     /// before using functions \ref lowerMap(), \ref upperMap(),
   840     /// \ref costMap(), \ref supplyMap(), \ref stSupply(), \ref supplyType().
   841     ///
   842     /// It is useful for multiple \ref run() calls. Basically, all the given
   843     /// parameters are kept for the next \ref run() call, unless
   844     /// \ref resetParams() or \ref reset() is used.
   845     /// If the underlying digraph was also modified after the construction
   846     /// of the class or the last \ref reset() call, then the \ref reset()
   847     /// function must be used, otherwise \ref resetParams() is sufficient.
   848     ///
   849     /// For example,
   850     /// \code
   851     ///   NetworkSimplex<ListDigraph> ns(graph);
   852     ///
   853     ///   // First run
   854     ///   ns.lowerMap(lower).upperMap(upper).costMap(cost)
   855     ///     .supplyMap(sup).run();
   856     ///
   857     ///   // Run again with modified cost map (resetParams() is not called,
   858     ///   // so only the cost map have to be set again)
   859     ///   cost[e] += 100;
   860     ///   ns.costMap(cost).run();
   861     ///
   862     ///   // Run again from scratch using resetParams()
   863     ///   // (the lower bounds will be set to zero on all arcs)
   864     ///   ns.resetParams();
   865     ///   ns.upperMap(capacity).costMap(cost)
   866     ///     .supplyMap(sup).run();
   867     /// \endcode
   868     ///
   869     /// \return <tt>(*this)</tt>
   870     ///
   871     /// \see reset(), run()
   872     NetworkSimplex& resetParams() {
   873       for (int i = 0; i != _node_num; ++i) {
   874         _supply[i] = 0;
   875       }
   876       for (int i = 0; i != _arc_num; ++i) {
   877         _lower[i] = 0;
   878         _upper[i] = INF;
   879         _cost[i] = 1;
   880       }
   881       _have_lower = false;
   882       _stype = GEQ;
   883       return *this;
   884     }
   885 
   886     /// \brief Reset the internal data structures and all the parameters
   887     /// that have been given before.
   888     ///
   889     /// This function resets the internal data structures and all the
   890     /// paramaters that have been given before using functions \ref lowerMap(),
   891     /// \ref upperMap(), \ref costMap(), \ref supplyMap(), \ref stSupply(),
   892     /// \ref supplyType().
   893     ///
   894     /// It is useful for multiple \ref run() calls. Basically, all the given
   895     /// parameters are kept for the next \ref run() call, unless
   896     /// \ref resetParams() or \ref reset() is used.
   897     /// If the underlying digraph was also modified after the construction
   898     /// of the class or the last \ref reset() call, then the \ref reset()
   899     /// function must be used, otherwise \ref resetParams() is sufficient.
   900     ///
   901     /// See \ref resetParams() for examples.
   902     ///
   903     /// \return <tt>(*this)</tt>
   904     ///
   905     /// \see resetParams(), run()
   906     NetworkSimplex& reset() {
   907       // Resize vectors
   908       _node_num = countNodes(_graph);
   909       _arc_num = countArcs(_graph);
   910       int all_node_num = _node_num + 1;
   911       int max_arc_num = _arc_num + 2 * _node_num;
   912 
   913       _source.resize(max_arc_num);
   914       _target.resize(max_arc_num);
   915 
   916       _lower.resize(_arc_num);
   917       _upper.resize(_arc_num);
   918       _cap.resize(max_arc_num);
   919       _cost.resize(max_arc_num);
   920       _supply.resize(all_node_num);
   921       _flow.resize(max_arc_num);
   922       _pi.resize(all_node_num);
   923 
   924       _parent.resize(all_node_num);
   925       _pred.resize(all_node_num);
   926       _pred_dir.resize(all_node_num);
   927       _thread.resize(all_node_num);
   928       _rev_thread.resize(all_node_num);
   929       _succ_num.resize(all_node_num);
   930       _last_succ.resize(all_node_num);
   931       _state.resize(max_arc_num);
   932 
   933       // Copy the graph
   934       int i = 0;
   935       for (NodeIt n(_graph); n != INVALID; ++n, ++i) {
   936         _node_id[n] = i;
   937       }
   938       if (_arc_mixing) {
   939         // Store the arcs in a mixed order
   940         const int skip = std::max(_arc_num / _node_num, 3);
   941         int i = 0, j = 0;
   942         for (ArcIt a(_graph); a != INVALID; ++a) {
   943           _arc_id[a] = i;
   944           _source[i] = _node_id[_graph.source(a)];
   945           _target[i] = _node_id[_graph.target(a)];
   946           if ((i += skip) >= _arc_num) i = ++j;
   947         }
   948       } else {
   949         // Store the arcs in the original order
   950         int i = 0;
   951         for (ArcIt a(_graph); a != INVALID; ++a, ++i) {
   952           _arc_id[a] = i;
   953           _source[i] = _node_id[_graph.source(a)];
   954           _target[i] = _node_id[_graph.target(a)];
   955         }
   956       }
   957 
   958       // Reset parameters
   959       resetParams();
   960       return *this;
   961     }
   962 
   963     /// @}
   964 
   965     /// \name Query Functions
   966     /// The results of the algorithm can be obtained using these
   967     /// functions.\n
   968     /// The \ref run() function must be called before using them.
   969 
   970     /// @{
   971 
   972     /// \brief Return the total cost of the found flow.
   973     ///
   974     /// This function returns the total cost of the found flow.
   975     /// Its complexity is O(e).
   976     ///
   977     /// \note The return type of the function can be specified as a
   978     /// template parameter. For example,
   979     /// \code
   980     ///   ns.totalCost<double>();
   981     /// \endcode
   982     /// It is useful if the total cost cannot be stored in the \c Cost
   983     /// type of the algorithm, which is the default return type of the
   984     /// function.
   985     ///
   986     /// \pre \ref run() must be called before using this function.
   987     template <typename Number>
   988     Number totalCost() const {
   989       Number c = 0;
   990       for (ArcIt a(_graph); a != INVALID; ++a) {
   991         int i = _arc_id[a];
   992         c += Number(_flow[i]) * Number(_cost[i]);
   993       }
   994       return c;
   995     }
   996 
   997 #ifndef DOXYGEN
   998     Cost totalCost() const {
   999       return totalCost<Cost>();
  1000     }
  1001 #endif
  1002 
  1003     /// \brief Return the flow on the given arc.
  1004     ///
  1005     /// This function returns the flow on the given arc.
  1006     ///
  1007     /// \pre \ref run() must be called before using this function.
  1008     Value flow(const Arc& a) const {
  1009       return _flow[_arc_id[a]];
  1010     }
  1011 
  1012     /// \brief Return the flow map (the primal solution).
  1013     ///
  1014     /// This function copies the flow value on each arc into the given
  1015     /// map. The \c Value type of the algorithm must be convertible to
  1016     /// the \c Value type of the map.
  1017     ///
  1018     /// \pre \ref run() must be called before using this function.
  1019     template <typename FlowMap>
  1020     void flowMap(FlowMap &map) const {
  1021       for (ArcIt a(_graph); a != INVALID; ++a) {
  1022         map.set(a, _flow[_arc_id[a]]);
  1023       }
  1024     }
  1025 
  1026     /// \brief Return the potential (dual value) of the given node.
  1027     ///
  1028     /// This function returns the potential (dual value) of the
  1029     /// given node.
  1030     ///
  1031     /// \pre \ref run() must be called before using this function.
  1032     Cost potential(const Node& n) const {
  1033       return _pi[_node_id[n]];
  1034     }
  1035 
  1036     /// \brief Return the potential map (the dual solution).
  1037     ///
  1038     /// This function copies the potential (dual value) of each node
  1039     /// into the given map.
  1040     /// The \c Cost type of the algorithm must be convertible to the
  1041     /// \c Value type of the map.
  1042     ///
  1043     /// \pre \ref run() must be called before using this function.
  1044     template <typename PotentialMap>
  1045     void potentialMap(PotentialMap &map) const {
  1046       for (NodeIt n(_graph); n != INVALID; ++n) {
  1047         map.set(n, _pi[_node_id[n]]);
  1048       }
  1049     }
  1050 
  1051     /// @}
  1052 
  1053   private:
  1054 
  1055     // Initialize internal data structures
  1056     bool init() {
  1057       if (_node_num == 0) return false;
  1058 
  1059       // Check the sum of supply values
  1060       _sum_supply = 0;
  1061       for (int i = 0; i != _node_num; ++i) {
  1062         _sum_supply += _supply[i];
  1063       }
  1064       if ( !((_stype == GEQ && _sum_supply <= 0) ||
  1065              (_stype == LEQ && _sum_supply >= 0)) ) return false;
  1066 
  1067       // Remove non-zero lower bounds
  1068       if (_have_lower) {
  1069         for (int i = 0; i != _arc_num; ++i) {
  1070           Value c = _lower[i];
  1071           if (c >= 0) {
  1072             _cap[i] = _upper[i] < MAX ? _upper[i] - c : INF;
  1073           } else {
  1074             _cap[i] = _upper[i] < MAX + c ? _upper[i] - c : INF;
  1075           }
  1076           _supply[_source[i]] -= c;
  1077           _supply[_target[i]] += c;
  1078         }
  1079       } else {
  1080         for (int i = 0; i != _arc_num; ++i) {
  1081           _cap[i] = _upper[i];
  1082         }
  1083       }
  1084 
  1085       // Initialize artifical cost
  1086       Cost ART_COST;
  1087       if (std::numeric_limits<Cost>::is_exact) {
  1088         ART_COST = std::numeric_limits<Cost>::max() / 2 + 1;
  1089       } else {
  1090         ART_COST = 0;
  1091         for (int i = 0; i != _arc_num; ++i) {
  1092           if (_cost[i] > ART_COST) ART_COST = _cost[i];
  1093         }
  1094         ART_COST = (ART_COST + 1) * _node_num;
  1095       }
  1096 
  1097       // Initialize arc maps
  1098       for (int i = 0; i != _arc_num; ++i) {
  1099         _flow[i] = 0;
  1100         _state[i] = STATE_LOWER;
  1101       }
  1102 
  1103       // Set data for the artificial root node
  1104       _root = _node_num;
  1105       _parent[_root] = -1;
  1106       _pred[_root] = -1;
  1107       _thread[_root] = 0;
  1108       _rev_thread[0] = _root;
  1109       _succ_num[_root] = _node_num + 1;
  1110       _last_succ[_root] = _root - 1;
  1111       _supply[_root] = -_sum_supply;
  1112       _pi[_root] = 0;
  1113 
  1114       // Add artificial arcs and initialize the spanning tree data structure
  1115       if (_sum_supply == 0) {
  1116         // EQ supply constraints
  1117         _search_arc_num = _arc_num;
  1118         _all_arc_num = _arc_num + _node_num;
  1119         for (int u = 0, e = _arc_num; u != _node_num; ++u, ++e) {
  1120           _parent[u] = _root;
  1121           _pred[u] = e;
  1122           _thread[u] = u + 1;
  1123           _rev_thread[u + 1] = u;
  1124           _succ_num[u] = 1;
  1125           _last_succ[u] = u;
  1126           _cap[e] = INF;
  1127           _state[e] = STATE_TREE;
  1128           if (_supply[u] >= 0) {
  1129             _pred_dir[u] = DIR_UP;
  1130             _pi[u] = 0;
  1131             _source[e] = u;
  1132             _target[e] = _root;
  1133             _flow[e] = _supply[u];
  1134             _cost[e] = 0;
  1135           } else {
  1136             _pred_dir[u] = DIR_DOWN;
  1137             _pi[u] = ART_COST;
  1138             _source[e] = _root;
  1139             _target[e] = u;
  1140             _flow[e] = -_supply[u];
  1141             _cost[e] = ART_COST;
  1142           }
  1143         }
  1144       }
  1145       else if (_sum_supply > 0) {
  1146         // LEQ supply constraints
  1147         _search_arc_num = _arc_num + _node_num;
  1148         int f = _arc_num + _node_num;
  1149         for (int u = 0, e = _arc_num; u != _node_num; ++u, ++e) {
  1150           _parent[u] = _root;
  1151           _thread[u] = u + 1;
  1152           _rev_thread[u + 1] = u;
  1153           _succ_num[u] = 1;
  1154           _last_succ[u] = u;
  1155           if (_supply[u] >= 0) {
  1156             _pred_dir[u] = DIR_UP;
  1157             _pi[u] = 0;
  1158             _pred[u] = e;
  1159             _source[e] = u;
  1160             _target[e] = _root;
  1161             _cap[e] = INF;
  1162             _flow[e] = _supply[u];
  1163             _cost[e] = 0;
  1164             _state[e] = STATE_TREE;
  1165           } else {
  1166             _pred_dir[u] = DIR_DOWN;
  1167             _pi[u] = ART_COST;
  1168             _pred[u] = f;
  1169             _source[f] = _root;
  1170             _target[f] = u;
  1171             _cap[f] = INF;
  1172             _flow[f] = -_supply[u];
  1173             _cost[f] = ART_COST;
  1174             _state[f] = STATE_TREE;
  1175             _source[e] = u;
  1176             _target[e] = _root;
  1177             _cap[e] = INF;
  1178             _flow[e] = 0;
  1179             _cost[e] = 0;
  1180             _state[e] = STATE_LOWER;
  1181             ++f;
  1182           }
  1183         }
  1184         _all_arc_num = f;
  1185       }
  1186       else {
  1187         // GEQ supply constraints
  1188         _search_arc_num = _arc_num + _node_num;
  1189         int f = _arc_num + _node_num;
  1190         for (int u = 0, e = _arc_num; u != _node_num; ++u, ++e) {
  1191           _parent[u] = _root;
  1192           _thread[u] = u + 1;
  1193           _rev_thread[u + 1] = u;
  1194           _succ_num[u] = 1;
  1195           _last_succ[u] = u;
  1196           if (_supply[u] <= 0) {
  1197             _pred_dir[u] = DIR_DOWN;
  1198             _pi[u] = 0;
  1199             _pred[u] = e;
  1200             _source[e] = _root;
  1201             _target[e] = u;
  1202             _cap[e] = INF;
  1203             _flow[e] = -_supply[u];
  1204             _cost[e] = 0;
  1205             _state[e] = STATE_TREE;
  1206           } else {
  1207             _pred_dir[u] = DIR_UP;
  1208             _pi[u] = -ART_COST;
  1209             _pred[u] = f;
  1210             _source[f] = u;
  1211             _target[f] = _root;
  1212             _cap[f] = INF;
  1213             _flow[f] = _supply[u];
  1214             _state[f] = STATE_TREE;
  1215             _cost[f] = ART_COST;
  1216             _source[e] = _root;
  1217             _target[e] = u;
  1218             _cap[e] = INF;
  1219             _flow[e] = 0;
  1220             _cost[e] = 0;
  1221             _state[e] = STATE_LOWER;
  1222             ++f;
  1223           }
  1224         }
  1225         _all_arc_num = f;
  1226       }
  1227 
  1228       return true;
  1229     }
  1230 
  1231     // Find the join node
  1232     void findJoinNode() {
  1233       int u = _source[in_arc];
  1234       int v = _target[in_arc];
  1235       while (u != v) {
  1236         if (_succ_num[u] < _succ_num[v]) {
  1237           u = _parent[u];
  1238         } else {
  1239           v = _parent[v];
  1240         }
  1241       }
  1242       join = u;
  1243     }
  1244 
  1245     // Find the leaving arc of the cycle and returns true if the
  1246     // leaving arc is not the same as the entering arc
  1247     bool findLeavingArc() {
  1248       // Initialize first and second nodes according to the direction
  1249       // of the cycle
  1250       int first, second;
  1251       if (_state[in_arc] == STATE_LOWER) {
  1252         first  = _source[in_arc];
  1253         second = _target[in_arc];
  1254       } else {
  1255         first  = _target[in_arc];
  1256         second = _source[in_arc];
  1257       }
  1258       delta = _cap[in_arc];
  1259       int result = 0;
  1260       Value c, d;
  1261       int e;
  1262 
  1263       // Search the cycle form the first node to the join node
  1264       for (int u = first; u != join; u = _parent[u]) {
  1265         e = _pred[u];
  1266         d = _flow[e];
  1267         if (_pred_dir[u] == DIR_DOWN) {
  1268           c = _cap[e];
  1269           d = c >= MAX ? INF : c - d;
  1270         }
  1271         if (d < delta) {
  1272           delta = d;
  1273           u_out = u;
  1274           result = 1;
  1275         }
  1276       }
  1277 
  1278       // Search the cycle form the second node to the join node
  1279       for (int u = second; u != join; u = _parent[u]) {
  1280         e = _pred[u];
  1281         d = _flow[e];
  1282         if (_pred_dir[u] == DIR_UP) {
  1283           c = _cap[e];
  1284           d = c >= MAX ? INF : c - d;
  1285         }
  1286         if (d <= delta) {
  1287           delta = d;
  1288           u_out = u;
  1289           result = 2;
  1290         }
  1291       }
  1292 
  1293       if (result == 1) {
  1294         u_in = first;
  1295         v_in = second;
  1296       } else {
  1297         u_in = second;
  1298         v_in = first;
  1299       }
  1300       return result != 0;
  1301     }
  1302 
  1303     // Change _flow and _state vectors
  1304     void changeFlow(bool change) {
  1305       // Augment along the cycle
  1306       if (delta > 0) {
  1307         Value val = _state[in_arc] * delta;
  1308         _flow[in_arc] += val;
  1309         for (int u = _source[in_arc]; u != join; u = _parent[u]) {
  1310           _flow[_pred[u]] -= _pred_dir[u] * val;
  1311         }
  1312         for (int u = _target[in_arc]; u != join; u = _parent[u]) {
  1313           _flow[_pred[u]] += _pred_dir[u] * val;
  1314         }
  1315       }
  1316       // Update the state of the entering and leaving arcs
  1317       if (change) {
  1318         _state[in_arc] = STATE_TREE;
  1319         _state[_pred[u_out]] =
  1320           (_flow[_pred[u_out]] == 0) ? STATE_LOWER : STATE_UPPER;
  1321       } else {
  1322         _state[in_arc] = -_state[in_arc];
  1323       }
  1324     }
  1325 
  1326     // Update the tree structure
  1327     void updateTreeStructure() {
  1328       int old_rev_thread = _rev_thread[u_out];
  1329       int old_succ_num = _succ_num[u_out];
  1330       int old_last_succ = _last_succ[u_out];
  1331       v_out = _parent[u_out];
  1332 
  1333       // Check if u_in and u_out coincide
  1334       if (u_in == u_out) {
  1335         // Update _parent, _pred, _pred_dir
  1336         _parent[u_in] = v_in;
  1337         _pred[u_in] = in_arc;
  1338         _pred_dir[u_in] = u_in == _source[in_arc] ? DIR_UP : DIR_DOWN;
  1339 
  1340         // Update _thread and _rev_thread
  1341         if (_thread[v_in] != u_out) {
  1342           int after = _thread[old_last_succ];
  1343           _thread[old_rev_thread] = after;
  1344           _rev_thread[after] = old_rev_thread;
  1345           after = _thread[v_in];
  1346           _thread[v_in] = u_out;
  1347           _rev_thread[u_out] = v_in;
  1348           _thread[old_last_succ] = after;
  1349           _rev_thread[after] = old_last_succ;
  1350         }
  1351       } else {
  1352         // Handle the case when old_rev_thread equals to v_in
  1353         // (it also means that join and v_out coincide)
  1354         int thread_continue = old_rev_thread == v_in ?
  1355           _thread[old_last_succ] : _thread[v_in];
  1356 
  1357         // Update _thread and _parent along the stem nodes (i.e. the nodes
  1358         // between u_in and u_out, whose parent have to be changed)
  1359         int stem = u_in;              // the current stem node
  1360         int par_stem = v_in;          // the new parent of stem
  1361         int next_stem;                // the next stem node
  1362         int last = _last_succ[u_in];  // the last successor of stem
  1363         int before, after = _thread[last];
  1364         _thread[v_in] = u_in;
  1365         _dirty_revs.clear();
  1366         _dirty_revs.push_back(v_in);
  1367         while (stem != u_out) {
  1368           // Insert the next stem node into the thread list
  1369           next_stem = _parent[stem];
  1370           _thread[last] = next_stem;
  1371           _dirty_revs.push_back(last);
  1372 
  1373           // Remove the subtree of stem from the thread list
  1374           before = _rev_thread[stem];
  1375           _thread[before] = after;
  1376           _rev_thread[after] = before;
  1377 
  1378           // Change the parent node and shift stem nodes
  1379           _parent[stem] = par_stem;
  1380           par_stem = stem;
  1381           stem = next_stem;
  1382 
  1383           // Update last and after
  1384           last = _last_succ[stem] == _last_succ[par_stem] ?
  1385             _rev_thread[par_stem] : _last_succ[stem];
  1386           after = _thread[last];
  1387         }
  1388         _parent[u_out] = par_stem;
  1389         _thread[last] = thread_continue;
  1390         _rev_thread[thread_continue] = last;
  1391         _last_succ[u_out] = last;
  1392 
  1393         // Remove the subtree of u_out from the thread list except for
  1394         // the case when old_rev_thread equals to v_in
  1395         if (old_rev_thread != v_in) {
  1396           _thread[old_rev_thread] = after;
  1397           _rev_thread[after] = old_rev_thread;
  1398         }
  1399 
  1400         // Update _rev_thread using the new _thread values
  1401         for (int i = 0; i != int(_dirty_revs.size()); ++i) {
  1402           int u = _dirty_revs[i];
  1403           _rev_thread[_thread[u]] = u;
  1404         }
  1405 
  1406         // Update _pred, _pred_dir, _last_succ and _succ_num for the
  1407         // stem nodes from u_out to u_in
  1408         int tmp_sc = 0, tmp_ls = _last_succ[u_out];
  1409         for (int u = u_out, p = _parent[u]; u != u_in; u = p, p = _parent[u]) {
  1410           _pred[u] = _pred[p];
  1411           _pred_dir[u] = -_pred_dir[p];
  1412           tmp_sc += _succ_num[u] - _succ_num[p];
  1413           _succ_num[u] = tmp_sc;
  1414           _last_succ[p] = tmp_ls;
  1415         }
  1416         _pred[u_in] = in_arc;
  1417         _pred_dir[u_in] = u_in == _source[in_arc] ? DIR_UP : DIR_DOWN;
  1418         _succ_num[u_in] = old_succ_num;
  1419       }
  1420 
  1421       // Update _last_succ from v_in towards the root
  1422       int up_limit_out = _last_succ[join] == v_in ? join : -1;
  1423       int last_succ_out = _last_succ[u_out];
  1424       for (int u = v_in; u != -1 && _last_succ[u] == v_in; u = _parent[u]) {
  1425         _last_succ[u] = last_succ_out;
  1426       }
  1427 
  1428       // Update _last_succ from v_out towards the root
  1429       if (join != old_rev_thread && v_in != old_rev_thread) {
  1430         for (int u = v_out; u != up_limit_out && _last_succ[u] == old_last_succ;
  1431              u = _parent[u]) {
  1432           _last_succ[u] = old_rev_thread;
  1433         }
  1434       }
  1435       else if (last_succ_out != old_last_succ) {
  1436         for (int u = v_out; u != up_limit_out && _last_succ[u] == old_last_succ;
  1437              u = _parent[u]) {
  1438           _last_succ[u] = last_succ_out;
  1439         }
  1440       }
  1441 
  1442       // Update _succ_num from v_in to join
  1443       for (int u = v_in; u != join; u = _parent[u]) {
  1444         _succ_num[u] += old_succ_num;
  1445       }
  1446       // Update _succ_num from v_out to join
  1447       for (int u = v_out; u != join; u = _parent[u]) {
  1448         _succ_num[u] -= old_succ_num;
  1449       }
  1450     }
  1451 
  1452     // Update potentials in the subtree that has been moved
  1453     void updatePotential() {
  1454       Cost sigma = _pi[v_in] - _pi[u_in] -
  1455                    _pred_dir[u_in] * _cost[in_arc];
  1456       int end = _thread[_last_succ[u_in]];
  1457       for (int u = u_in; u != end; u = _thread[u]) {
  1458         _pi[u] += sigma;
  1459       }
  1460     }
  1461 
  1462     // Heuristic initial pivots
  1463     bool initialPivots() {
  1464       Value curr, total = 0;
  1465       std::vector<Node> supply_nodes, demand_nodes;
  1466       for (NodeIt u(_graph); u != INVALID; ++u) {
  1467         curr = _supply[_node_id[u]];
  1468         if (curr > 0) {
  1469           total += curr;
  1470           supply_nodes.push_back(u);
  1471         }
  1472         else if (curr < 0) {
  1473           demand_nodes.push_back(u);
  1474         }
  1475       }
  1476       if (_sum_supply > 0) total -= _sum_supply;
  1477       if (total <= 0) return true;
  1478 
  1479       IntVector arc_vector;
  1480       if (_sum_supply >= 0) {
  1481         if (supply_nodes.size() == 1 && demand_nodes.size() == 1) {
  1482           // Perform a reverse graph search from the sink to the source
  1483           typename GR::template NodeMap<bool> reached(_graph, false);
  1484           Node s = supply_nodes[0], t = demand_nodes[0];
  1485           std::vector<Node> stack;
  1486           reached[t] = true;
  1487           stack.push_back(t);
  1488           while (!stack.empty()) {
  1489             Node u, v = stack.back();
  1490             stack.pop_back();
  1491             if (v == s) break;
  1492             for (InArcIt a(_graph, v); a != INVALID; ++a) {
  1493               if (reached[u = _graph.source(a)]) continue;
  1494               int j = _arc_id[a];
  1495               if (_cap[j] >= total) {
  1496                 arc_vector.push_back(j);
  1497                 reached[u] = true;
  1498                 stack.push_back(u);
  1499               }
  1500             }
  1501           }
  1502         } else {
  1503           // Find the min. cost incomming arc for each demand node
  1504           for (int i = 0; i != int(demand_nodes.size()); ++i) {
  1505             Node v = demand_nodes[i];
  1506             Cost c, min_cost = std::numeric_limits<Cost>::max();
  1507             Arc min_arc = INVALID;
  1508             for (InArcIt a(_graph, v); a != INVALID; ++a) {
  1509               c = _cost[_arc_id[a]];
  1510               if (c < min_cost) {
  1511                 min_cost = c;
  1512                 min_arc = a;
  1513               }
  1514             }
  1515             if (min_arc != INVALID) {
  1516               arc_vector.push_back(_arc_id[min_arc]);
  1517             }
  1518           }
  1519         }
  1520       } else {
  1521         // Find the min. cost outgoing arc for each supply node
  1522         for (int i = 0; i != int(supply_nodes.size()); ++i) {
  1523           Node u = supply_nodes[i];
  1524           Cost c, min_cost = std::numeric_limits<Cost>::max();
  1525           Arc min_arc = INVALID;
  1526           for (OutArcIt a(_graph, u); a != INVALID; ++a) {
  1527             c = _cost[_arc_id[a]];
  1528             if (c < min_cost) {
  1529               min_cost = c;
  1530               min_arc = a;
  1531             }
  1532           }
  1533           if (min_arc != INVALID) {
  1534             arc_vector.push_back(_arc_id[min_arc]);
  1535           }
  1536         }
  1537       }
  1538 
  1539       // Perform heuristic initial pivots
  1540       for (int i = 0; i != int(arc_vector.size()); ++i) {
  1541         in_arc = arc_vector[i];
  1542         if (_state[in_arc] * (_cost[in_arc] + _pi[_source[in_arc]] -
  1543             _pi[_target[in_arc]]) >= 0) continue;
  1544         findJoinNode();
  1545         bool change = findLeavingArc();
  1546         if (delta >= MAX) return false;
  1547         changeFlow(change);
  1548         if (change) {
  1549           updateTreeStructure();
  1550           updatePotential();
  1551         }
  1552       }
  1553       return true;
  1554     }
  1555 
  1556     // Execute the algorithm
  1557     ProblemType start(PivotRule pivot_rule) {
  1558       // Select the pivot rule implementation
  1559       switch (pivot_rule) {
  1560         case FIRST_ELIGIBLE:
  1561           return start<FirstEligiblePivotRule>();
  1562         case BEST_ELIGIBLE:
  1563           return start<BestEligiblePivotRule>();
  1564         case BLOCK_SEARCH:
  1565           return start<BlockSearchPivotRule>();
  1566         case CANDIDATE_LIST:
  1567           return start<CandidateListPivotRule>();
  1568         case ALTERING_LIST:
  1569           return start<AlteringListPivotRule>();
  1570       }
  1571       return INFEASIBLE; // avoid warning
  1572     }
  1573 
  1574     template <typename PivotRuleImpl>
  1575     ProblemType start() {
  1576       PivotRuleImpl pivot(*this);
  1577 
  1578       // Perform heuristic initial pivots
  1579       if (!initialPivots()) return UNBOUNDED;
  1580 
  1581       // Execute the Network Simplex algorithm
  1582       while (pivot.findEnteringArc()) {
  1583         findJoinNode();
  1584         bool change = findLeavingArc();
  1585         if (delta >= MAX) return UNBOUNDED;
  1586         changeFlow(change);
  1587         if (change) {
  1588           updateTreeStructure();
  1589           updatePotential();
  1590         }
  1591       }
  1592 
  1593       // Check feasibility
  1594       for (int e = _search_arc_num; e != _all_arc_num; ++e) {
  1595         if (_flow[e] != 0) return INFEASIBLE;
  1596       }
  1597 
  1598       // Transform the solution and the supply map to the original form
  1599       if (_have_lower) {
  1600         for (int i = 0; i != _arc_num; ++i) {
  1601           Value c = _lower[i];
  1602           if (c != 0) {
  1603             _flow[i] += c;
  1604             _supply[_source[i]] += c;
  1605             _supply[_target[i]] -= c;
  1606           }
  1607         }
  1608       }
  1609 
  1610       // Shift potentials to meet the requirements of the GEQ/LEQ type
  1611       // optimality conditions
  1612       if (_sum_supply == 0) {
  1613         if (_stype == GEQ) {
  1614           Cost max_pot = -std::numeric_limits<Cost>::max();
  1615           for (int i = 0; i != _node_num; ++i) {
  1616             if (_pi[i] > max_pot) max_pot = _pi[i];
  1617           }
  1618           if (max_pot > 0) {
  1619             for (int i = 0; i != _node_num; ++i)
  1620               _pi[i] -= max_pot;
  1621           }
  1622         } else {
  1623           Cost min_pot = std::numeric_limits<Cost>::max();
  1624           for (int i = 0; i != _node_num; ++i) {
  1625             if (_pi[i] < min_pot) min_pot = _pi[i];
  1626           }
  1627           if (min_pot < 0) {
  1628             for (int i = 0; i != _node_num; ++i)
  1629               _pi[i] -= min_pot;
  1630           }
  1631         }
  1632       }
  1633 
  1634       return OPTIMAL;
  1635     }
  1636 
  1637   }; //class NetworkSimplex
  1638 
  1639   ///@}
  1640 
  1641 } //namespace lemon
  1642 
  1643 #endif //LEMON_NETWORK_SIMPLEX_H