lemon/network_simplex.h
author Alpar Juttner <alpar@cs.elte.hu>
Thu, 20 Aug 2009 22:52:03 +0200
changeset 686 7439dc5fe1b9
parent 643 f3792d5bb294
child 727 cab85bd7859b
child 729 be48a648d28f
child 888 5205145fabf6
permissions -rw-r--r--
Merge
kpeter@601
     1
/* -*- mode: C++; indent-tabs-mode: nil; -*-
kpeter@601
     2
 *
kpeter@601
     3
 * This file is a part of LEMON, a generic C++ optimization library.
kpeter@601
     4
 *
kpeter@601
     5
 * Copyright (C) 2003-2009
kpeter@601
     6
 * Egervary Jeno Kombinatorikus Optimalizalasi Kutatocsoport
kpeter@601
     7
 * (Egervary Research Group on Combinatorial Optimization, EGRES).
kpeter@601
     8
 *
kpeter@601
     9
 * Permission to use, modify and distribute this software is granted
kpeter@601
    10
 * provided that this copyright notice appears in all copies. For
kpeter@601
    11
 * precise terms see the accompanying LICENSE file.
kpeter@601
    12
 *
kpeter@601
    13
 * This software is provided "AS IS" with no warranty of any kind,
kpeter@601
    14
 * express or implied, and with no claim as to its suitability for any
kpeter@601
    15
 * purpose.
kpeter@601
    16
 *
kpeter@601
    17
 */
kpeter@601
    18
kpeter@601
    19
#ifndef LEMON_NETWORK_SIMPLEX_H
kpeter@601
    20
#define LEMON_NETWORK_SIMPLEX_H
kpeter@601
    21
kpeter@663
    22
/// \ingroup min_cost_flow_algs
kpeter@601
    23
///
kpeter@601
    24
/// \file
kpeter@605
    25
/// \brief Network Simplex algorithm for finding a minimum cost flow.
kpeter@601
    26
kpeter@601
    27
#include <vector>
kpeter@601
    28
#include <limits>
kpeter@601
    29
#include <algorithm>
kpeter@601
    30
kpeter@603
    31
#include <lemon/core.h>
kpeter@601
    32
#include <lemon/math.h>
kpeter@601
    33
kpeter@601
    34
namespace lemon {
kpeter@601
    35
kpeter@663
    36
  /// \addtogroup min_cost_flow_algs
kpeter@601
    37
  /// @{
kpeter@601
    38
kpeter@605
    39
  /// \brief Implementation of the primal Network Simplex algorithm
kpeter@601
    40
  /// for finding a \ref min_cost_flow "minimum cost flow".
kpeter@601
    41
  ///
kpeter@605
    42
  /// \ref NetworkSimplex implements the primal Network Simplex algorithm
kpeter@601
    43
  /// for finding a \ref min_cost_flow "minimum cost flow".
kpeter@606
    44
  /// This algorithm is a specialized version of the linear programming
kpeter@606
    45
  /// simplex method directly for the minimum cost flow problem.
kpeter@606
    46
  /// It is one of the most efficient solution methods.
kpeter@606
    47
  ///
kpeter@606
    48
  /// In general this class is the fastest implementation available
kpeter@606
    49
  /// in LEMON for the minimum cost flow problem.
kpeter@640
    50
  /// Moreover it supports both directions of the supply/demand inequality
kpeter@640
    51
  /// constraints. For more information see \ref SupplyType.
kpeter@640
    52
  ///
kpeter@640
    53
  /// Most of the parameters of the problem (except for the digraph)
kpeter@640
    54
  /// can be given using separate functions, and the algorithm can be
kpeter@640
    55
  /// executed using the \ref run() function. If some parameters are not
kpeter@640
    56
  /// specified, then default values will be used.
kpeter@601
    57
  ///
kpeter@605
    58
  /// \tparam GR The digraph type the algorithm runs on.
kpeter@641
    59
  /// \tparam V The value type used for flow amounts, capacity bounds
kpeter@607
    60
  /// and supply values in the algorithm. By default it is \c int.
kpeter@607
    61
  /// \tparam C The value type used for costs and potentials in the
kpeter@641
    62
  /// algorithm. By default it is the same as \c V.
kpeter@601
    63
  ///
kpeter@608
    64
  /// \warning Both value types must be signed and all input data must
kpeter@608
    65
  /// be integer.
kpeter@601
    66
  ///
kpeter@605
    67
  /// \note %NetworkSimplex provides five different pivot rule
kpeter@609
    68
  /// implementations, from which the most efficient one is used
kpeter@609
    69
  /// by default. For more information see \ref PivotRule.
kpeter@641
    70
  template <typename GR, typename V = int, typename C = V>
kpeter@601
    71
  class NetworkSimplex
kpeter@601
    72
  {
kpeter@605
    73
  public:
kpeter@601
    74
kpeter@642
    75
    /// The type of the flow amounts, capacity bounds and supply values
kpeter@641
    76
    typedef V Value;
kpeter@642
    77
    /// The type of the arc costs
kpeter@607
    78
    typedef C Cost;
kpeter@605
    79
kpeter@605
    80
  public:
kpeter@605
    81
kpeter@640
    82
    /// \brief Problem type constants for the \c run() function.
kpeter@605
    83
    ///
kpeter@640
    84
    /// Enum type containing the problem type constants that can be
kpeter@640
    85
    /// returned by the \ref run() function of the algorithm.
kpeter@640
    86
    enum ProblemType {
kpeter@640
    87
      /// The problem has no feasible solution (flow).
kpeter@640
    88
      INFEASIBLE,
kpeter@640
    89
      /// The problem has optimal solution (i.e. it is feasible and
kpeter@640
    90
      /// bounded), and the algorithm has found optimal flow and node
kpeter@640
    91
      /// potentials (primal and dual solutions).
kpeter@640
    92
      OPTIMAL,
kpeter@640
    93
      /// The objective function of the problem is unbounded, i.e.
kpeter@640
    94
      /// there is a directed cycle having negative total cost and
kpeter@640
    95
      /// infinite upper bound.
kpeter@640
    96
      UNBOUNDED
kpeter@640
    97
    };
kpeter@640
    98
    
kpeter@640
    99
    /// \brief Constants for selecting the type of the supply constraints.
kpeter@640
   100
    ///
kpeter@640
   101
    /// Enum type containing constants for selecting the supply type,
kpeter@640
   102
    /// i.e. the direction of the inequalities in the supply/demand
kpeter@640
   103
    /// constraints of the \ref min_cost_flow "minimum cost flow problem".
kpeter@640
   104
    ///
kpeter@663
   105
    /// The default supply type is \c GEQ, the \c LEQ type can be
kpeter@663
   106
    /// selected using \ref supplyType().
kpeter@663
   107
    /// The equality form is a special case of both supply types.
kpeter@640
   108
    enum SupplyType {
kpeter@640
   109
      /// This option means that there are <em>"greater or equal"</em>
kpeter@663
   110
      /// supply/demand constraints in the definition of the problem.
kpeter@640
   111
      GEQ,
kpeter@640
   112
      /// This option means that there are <em>"less or equal"</em>
kpeter@663
   113
      /// supply/demand constraints in the definition of the problem.
kpeter@663
   114
      LEQ
kpeter@640
   115
    };
kpeter@640
   116
    
kpeter@640
   117
    /// \brief Constants for selecting the pivot rule.
kpeter@640
   118
    ///
kpeter@640
   119
    /// Enum type containing constants for selecting the pivot rule for
kpeter@640
   120
    /// the \ref run() function.
kpeter@640
   121
    ///
kpeter@605
   122
    /// \ref NetworkSimplex provides five different pivot rule
kpeter@605
   123
    /// implementations that significantly affect the running time
kpeter@605
   124
    /// of the algorithm.
kpeter@605
   125
    /// By default \ref BLOCK_SEARCH "Block Search" is used, which
kpeter@605
   126
    /// proved to be the most efficient and the most robust on various
kpeter@605
   127
    /// test inputs according to our benchmark tests.
kpeter@605
   128
    /// However another pivot rule can be selected using the \ref run()
kpeter@605
   129
    /// function with the proper parameter.
kpeter@605
   130
    enum PivotRule {
kpeter@605
   131
kpeter@605
   132
      /// The First Eligible pivot rule.
kpeter@605
   133
      /// The next eligible arc is selected in a wraparound fashion
kpeter@605
   134
      /// in every iteration.
kpeter@605
   135
      FIRST_ELIGIBLE,
kpeter@605
   136
kpeter@605
   137
      /// The Best Eligible pivot rule.
kpeter@605
   138
      /// The best eligible arc is selected in every iteration.
kpeter@605
   139
      BEST_ELIGIBLE,
kpeter@605
   140
kpeter@605
   141
      /// The Block Search pivot rule.
kpeter@605
   142
      /// A specified number of arcs are examined in every iteration
kpeter@605
   143
      /// in a wraparound fashion and the best eligible arc is selected
kpeter@605
   144
      /// from this block.
kpeter@605
   145
      BLOCK_SEARCH,
kpeter@605
   146
kpeter@605
   147
      /// The Candidate List pivot rule.
kpeter@605
   148
      /// In a major iteration a candidate list is built from eligible arcs
kpeter@605
   149
      /// in a wraparound fashion and in the following minor iterations
kpeter@605
   150
      /// the best eligible arc is selected from this list.
kpeter@605
   151
      CANDIDATE_LIST,
kpeter@605
   152
kpeter@605
   153
      /// The Altering Candidate List pivot rule.
kpeter@605
   154
      /// It is a modified version of the Candidate List method.
kpeter@605
   155
      /// It keeps only the several best eligible arcs from the former
kpeter@605
   156
      /// candidate list and extends this list in every iteration.
kpeter@605
   157
      ALTERING_LIST
kpeter@605
   158
    };
kpeter@609
   159
    
kpeter@605
   160
  private:
kpeter@605
   161
kpeter@605
   162
    TEMPLATE_DIGRAPH_TYPEDEFS(GR);
kpeter@605
   163
kpeter@601
   164
    typedef std::vector<Arc> ArcVector;
kpeter@601
   165
    typedef std::vector<Node> NodeVector;
kpeter@601
   166
    typedef std::vector<int> IntVector;
kpeter@601
   167
    typedef std::vector<bool> BoolVector;
kpeter@642
   168
    typedef std::vector<Value> ValueVector;
kpeter@607
   169
    typedef std::vector<Cost> CostVector;
kpeter@601
   170
kpeter@601
   171
    // State constants for arcs
kpeter@601
   172
    enum ArcStateEnum {
kpeter@601
   173
      STATE_UPPER = -1,
kpeter@601
   174
      STATE_TREE  =  0,
kpeter@601
   175
      STATE_LOWER =  1
kpeter@601
   176
    };
kpeter@601
   177
kpeter@601
   178
  private:
kpeter@601
   179
kpeter@605
   180
    // Data related to the underlying digraph
kpeter@605
   181
    const GR &_graph;
kpeter@605
   182
    int _node_num;
kpeter@605
   183
    int _arc_num;
kpeter@663
   184
    int _all_arc_num;
kpeter@663
   185
    int _search_arc_num;
kpeter@605
   186
kpeter@605
   187
    // Parameters of the problem
kpeter@642
   188
    bool _have_lower;
kpeter@640
   189
    SupplyType _stype;
kpeter@641
   190
    Value _sum_supply;
kpeter@601
   191
kpeter@605
   192
    // Data structures for storing the digraph
kpeter@603
   193
    IntNodeMap _node_id;
kpeter@642
   194
    IntArcMap _arc_id;
kpeter@603
   195
    IntVector _source;
kpeter@603
   196
    IntVector _target;
kpeter@603
   197
kpeter@605
   198
    // Node and arc data
kpeter@642
   199
    ValueVector _lower;
kpeter@642
   200
    ValueVector _upper;
kpeter@642
   201
    ValueVector _cap;
kpeter@607
   202
    CostVector _cost;
kpeter@642
   203
    ValueVector _supply;
kpeter@642
   204
    ValueVector _flow;
kpeter@607
   205
    CostVector _pi;
kpeter@601
   206
kpeter@603
   207
    // Data for storing the spanning tree structure
kpeter@601
   208
    IntVector _parent;
kpeter@601
   209
    IntVector _pred;
kpeter@601
   210
    IntVector _thread;
kpeter@604
   211
    IntVector _rev_thread;
kpeter@604
   212
    IntVector _succ_num;
kpeter@604
   213
    IntVector _last_succ;
kpeter@604
   214
    IntVector _dirty_revs;
kpeter@601
   215
    BoolVector _forward;
kpeter@601
   216
    IntVector _state;
kpeter@601
   217
    int _root;
kpeter@601
   218
kpeter@601
   219
    // Temporary data used in the current pivot iteration
kpeter@603
   220
    int in_arc, join, u_in, v_in, u_out, v_out;
kpeter@603
   221
    int first, second, right, last;
kpeter@601
   222
    int stem, par_stem, new_stem;
kpeter@641
   223
    Value delta;
kpeter@601
   224
kpeter@640
   225
  public:
kpeter@640
   226
  
kpeter@640
   227
    /// \brief Constant for infinite upper bounds (capacities).
kpeter@640
   228
    ///
kpeter@640
   229
    /// Constant for infinite upper bounds (capacities).
kpeter@641
   230
    /// It is \c std::numeric_limits<Value>::infinity() if available,
kpeter@641
   231
    /// \c std::numeric_limits<Value>::max() otherwise.
kpeter@641
   232
    const Value INF;
kpeter@640
   233
kpeter@601
   234
  private:
kpeter@601
   235
kpeter@605
   236
    // Implementation of the First Eligible pivot rule
kpeter@601
   237
    class FirstEligiblePivotRule
kpeter@601
   238
    {
kpeter@601
   239
    private:
kpeter@601
   240
kpeter@601
   241
      // References to the NetworkSimplex class
kpeter@601
   242
      const IntVector  &_source;
kpeter@601
   243
      const IntVector  &_target;
kpeter@607
   244
      const CostVector &_cost;
kpeter@601
   245
      const IntVector  &_state;
kpeter@607
   246
      const CostVector &_pi;
kpeter@601
   247
      int &_in_arc;
kpeter@663
   248
      int _search_arc_num;
kpeter@601
   249
kpeter@601
   250
      // Pivot rule data
kpeter@601
   251
      int _next_arc;
kpeter@601
   252
kpeter@601
   253
    public:
kpeter@601
   254
kpeter@605
   255
      // Constructor
kpeter@601
   256
      FirstEligiblePivotRule(NetworkSimplex &ns) :
kpeter@603
   257
        _source(ns._source), _target(ns._target),
kpeter@601
   258
        _cost(ns._cost), _state(ns._state), _pi(ns._pi),
kpeter@663
   259
        _in_arc(ns.in_arc), _search_arc_num(ns._search_arc_num),
kpeter@663
   260
        _next_arc(0)
kpeter@601
   261
      {}
kpeter@601
   262
kpeter@605
   263
      // Find next entering arc
kpeter@601
   264
      bool findEnteringArc() {
kpeter@607
   265
        Cost c;
kpeter@663
   266
        for (int e = _next_arc; e < _search_arc_num; ++e) {
kpeter@601
   267
          c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@601
   268
          if (c < 0) {
kpeter@601
   269
            _in_arc = e;
kpeter@601
   270
            _next_arc = e + 1;
kpeter@601
   271
            return true;
kpeter@601
   272
          }
kpeter@601
   273
        }
kpeter@601
   274
        for (int e = 0; e < _next_arc; ++e) {
kpeter@601
   275
          c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@601
   276
          if (c < 0) {
kpeter@601
   277
            _in_arc = e;
kpeter@601
   278
            _next_arc = e + 1;
kpeter@601
   279
            return true;
kpeter@601
   280
          }
kpeter@601
   281
        }
kpeter@601
   282
        return false;
kpeter@601
   283
      }
kpeter@601
   284
kpeter@601
   285
    }; //class FirstEligiblePivotRule
kpeter@601
   286
kpeter@601
   287
kpeter@605
   288
    // Implementation of the Best Eligible pivot rule
kpeter@601
   289
    class BestEligiblePivotRule
kpeter@601
   290
    {
kpeter@601
   291
    private:
kpeter@601
   292
kpeter@601
   293
      // References to the NetworkSimplex class
kpeter@601
   294
      const IntVector  &_source;
kpeter@601
   295
      const IntVector  &_target;
kpeter@607
   296
      const CostVector &_cost;
kpeter@601
   297
      const IntVector  &_state;
kpeter@607
   298
      const CostVector &_pi;
kpeter@601
   299
      int &_in_arc;
kpeter@663
   300
      int _search_arc_num;
kpeter@601
   301
kpeter@601
   302
    public:
kpeter@601
   303
kpeter@605
   304
      // Constructor
kpeter@601
   305
      BestEligiblePivotRule(NetworkSimplex &ns) :
kpeter@603
   306
        _source(ns._source), _target(ns._target),
kpeter@601
   307
        _cost(ns._cost), _state(ns._state), _pi(ns._pi),
kpeter@663
   308
        _in_arc(ns.in_arc), _search_arc_num(ns._search_arc_num)
kpeter@601
   309
      {}
kpeter@601
   310
kpeter@605
   311
      // Find next entering arc
kpeter@601
   312
      bool findEnteringArc() {
kpeter@607
   313
        Cost c, min = 0;
kpeter@663
   314
        for (int e = 0; e < _search_arc_num; ++e) {
kpeter@601
   315
          c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@601
   316
          if (c < min) {
kpeter@601
   317
            min = c;
kpeter@601
   318
            _in_arc = e;
kpeter@601
   319
          }
kpeter@601
   320
        }
kpeter@601
   321
        return min < 0;
kpeter@601
   322
      }
kpeter@601
   323
kpeter@601
   324
    }; //class BestEligiblePivotRule
kpeter@601
   325
kpeter@601
   326
kpeter@605
   327
    // Implementation of the Block Search pivot rule
kpeter@601
   328
    class BlockSearchPivotRule
kpeter@601
   329
    {
kpeter@601
   330
    private:
kpeter@601
   331
kpeter@601
   332
      // References to the NetworkSimplex class
kpeter@601
   333
      const IntVector  &_source;
kpeter@601
   334
      const IntVector  &_target;
kpeter@607
   335
      const CostVector &_cost;
kpeter@601
   336
      const IntVector  &_state;
kpeter@607
   337
      const CostVector &_pi;
kpeter@601
   338
      int &_in_arc;
kpeter@663
   339
      int _search_arc_num;
kpeter@601
   340
kpeter@601
   341
      // Pivot rule data
kpeter@601
   342
      int _block_size;
kpeter@601
   343
      int _next_arc;
kpeter@601
   344
kpeter@601
   345
    public:
kpeter@601
   346
kpeter@605
   347
      // Constructor
kpeter@601
   348
      BlockSearchPivotRule(NetworkSimplex &ns) :
kpeter@603
   349
        _source(ns._source), _target(ns._target),
kpeter@601
   350
        _cost(ns._cost), _state(ns._state), _pi(ns._pi),
kpeter@663
   351
        _in_arc(ns.in_arc), _search_arc_num(ns._search_arc_num),
kpeter@663
   352
        _next_arc(0)
kpeter@601
   353
      {
kpeter@601
   354
        // The main parameters of the pivot rule
kpeter@663
   355
        const double BLOCK_SIZE_FACTOR = 0.5;
kpeter@601
   356
        const int MIN_BLOCK_SIZE = 10;
kpeter@601
   357
alpar@612
   358
        _block_size = std::max( int(BLOCK_SIZE_FACTOR *
kpeter@663
   359
                                    std::sqrt(double(_search_arc_num))),
kpeter@601
   360
                                MIN_BLOCK_SIZE );
kpeter@601
   361
      }
kpeter@601
   362
kpeter@605
   363
      // Find next entering arc
kpeter@601
   364
      bool findEnteringArc() {
kpeter@607
   365
        Cost c, min = 0;
kpeter@601
   366
        int cnt = _block_size;
kpeter@601
   367
        int e, min_arc = _next_arc;
kpeter@663
   368
        for (e = _next_arc; e < _search_arc_num; ++e) {
kpeter@601
   369
          c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@601
   370
          if (c < min) {
kpeter@601
   371
            min = c;
kpeter@601
   372
            min_arc = e;
kpeter@601
   373
          }
kpeter@601
   374
          if (--cnt == 0) {
kpeter@601
   375
            if (min < 0) break;
kpeter@601
   376
            cnt = _block_size;
kpeter@601
   377
          }
kpeter@601
   378
        }
kpeter@601
   379
        if (min == 0 || cnt > 0) {
kpeter@601
   380
          for (e = 0; e < _next_arc; ++e) {
kpeter@601
   381
            c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@601
   382
            if (c < min) {
kpeter@601
   383
              min = c;
kpeter@601
   384
              min_arc = e;
kpeter@601
   385
            }
kpeter@601
   386
            if (--cnt == 0) {
kpeter@601
   387
              if (min < 0) break;
kpeter@601
   388
              cnt = _block_size;
kpeter@601
   389
            }
kpeter@601
   390
          }
kpeter@601
   391
        }
kpeter@601
   392
        if (min >= 0) return false;
kpeter@601
   393
        _in_arc = min_arc;
kpeter@601
   394
        _next_arc = e;
kpeter@601
   395
        return true;
kpeter@601
   396
      }
kpeter@601
   397
kpeter@601
   398
    }; //class BlockSearchPivotRule
kpeter@601
   399
kpeter@601
   400
kpeter@605
   401
    // Implementation of the Candidate List pivot rule
kpeter@601
   402
    class CandidateListPivotRule
kpeter@601
   403
    {
kpeter@601
   404
    private:
kpeter@601
   405
kpeter@601
   406
      // References to the NetworkSimplex class
kpeter@601
   407
      const IntVector  &_source;
kpeter@601
   408
      const IntVector  &_target;
kpeter@607
   409
      const CostVector &_cost;
kpeter@601
   410
      const IntVector  &_state;
kpeter@607
   411
      const CostVector &_pi;
kpeter@601
   412
      int &_in_arc;
kpeter@663
   413
      int _search_arc_num;
kpeter@601
   414
kpeter@601
   415
      // Pivot rule data
kpeter@601
   416
      IntVector _candidates;
kpeter@601
   417
      int _list_length, _minor_limit;
kpeter@601
   418
      int _curr_length, _minor_count;
kpeter@601
   419
      int _next_arc;
kpeter@601
   420
kpeter@601
   421
    public:
kpeter@601
   422
kpeter@601
   423
      /// Constructor
kpeter@601
   424
      CandidateListPivotRule(NetworkSimplex &ns) :
kpeter@603
   425
        _source(ns._source), _target(ns._target),
kpeter@601
   426
        _cost(ns._cost), _state(ns._state), _pi(ns._pi),
kpeter@663
   427
        _in_arc(ns.in_arc), _search_arc_num(ns._search_arc_num),
kpeter@663
   428
        _next_arc(0)
kpeter@601
   429
      {
kpeter@601
   430
        // The main parameters of the pivot rule
kpeter@601
   431
        const double LIST_LENGTH_FACTOR = 1.0;
kpeter@601
   432
        const int MIN_LIST_LENGTH = 10;
kpeter@601
   433
        const double MINOR_LIMIT_FACTOR = 0.1;
kpeter@601
   434
        const int MIN_MINOR_LIMIT = 3;
kpeter@601
   435
alpar@612
   436
        _list_length = std::max( int(LIST_LENGTH_FACTOR *
kpeter@663
   437
                                     std::sqrt(double(_search_arc_num))),
kpeter@601
   438
                                 MIN_LIST_LENGTH );
kpeter@601
   439
        _minor_limit = std::max( int(MINOR_LIMIT_FACTOR * _list_length),
kpeter@601
   440
                                 MIN_MINOR_LIMIT );
kpeter@601
   441
        _curr_length = _minor_count = 0;
kpeter@601
   442
        _candidates.resize(_list_length);
kpeter@601
   443
      }
kpeter@601
   444
kpeter@601
   445
      /// Find next entering arc
kpeter@601
   446
      bool findEnteringArc() {
kpeter@607
   447
        Cost min, c;
kpeter@601
   448
        int e, min_arc = _next_arc;
kpeter@601
   449
        if (_curr_length > 0 && _minor_count < _minor_limit) {
kpeter@601
   450
          // Minor iteration: select the best eligible arc from the
kpeter@601
   451
          // current candidate list
kpeter@601
   452
          ++_minor_count;
kpeter@601
   453
          min = 0;
kpeter@601
   454
          for (int i = 0; i < _curr_length; ++i) {
kpeter@601
   455
            e = _candidates[i];
kpeter@601
   456
            c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@601
   457
            if (c < min) {
kpeter@601
   458
              min = c;
kpeter@601
   459
              min_arc = e;
kpeter@601
   460
            }
kpeter@601
   461
            if (c >= 0) {
kpeter@601
   462
              _candidates[i--] = _candidates[--_curr_length];
kpeter@601
   463
            }
kpeter@601
   464
          }
kpeter@601
   465
          if (min < 0) {
kpeter@601
   466
            _in_arc = min_arc;
kpeter@601
   467
            return true;
kpeter@601
   468
          }
kpeter@601
   469
        }
kpeter@601
   470
kpeter@601
   471
        // Major iteration: build a new candidate list
kpeter@601
   472
        min = 0;
kpeter@601
   473
        _curr_length = 0;
kpeter@663
   474
        for (e = _next_arc; e < _search_arc_num; ++e) {
kpeter@601
   475
          c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@601
   476
          if (c < 0) {
kpeter@601
   477
            _candidates[_curr_length++] = e;
kpeter@601
   478
            if (c < min) {
kpeter@601
   479
              min = c;
kpeter@601
   480
              min_arc = e;
kpeter@601
   481
            }
kpeter@601
   482
            if (_curr_length == _list_length) break;
kpeter@601
   483
          }
kpeter@601
   484
        }
kpeter@601
   485
        if (_curr_length < _list_length) {
kpeter@601
   486
          for (e = 0; e < _next_arc; ++e) {
kpeter@601
   487
            c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@601
   488
            if (c < 0) {
kpeter@601
   489
              _candidates[_curr_length++] = e;
kpeter@601
   490
              if (c < min) {
kpeter@601
   491
                min = c;
kpeter@601
   492
                min_arc = e;
kpeter@601
   493
              }
kpeter@601
   494
              if (_curr_length == _list_length) break;
kpeter@601
   495
            }
kpeter@601
   496
          }
kpeter@601
   497
        }
kpeter@601
   498
        if (_curr_length == 0) return false;
kpeter@601
   499
        _minor_count = 1;
kpeter@601
   500
        _in_arc = min_arc;
kpeter@601
   501
        _next_arc = e;
kpeter@601
   502
        return true;
kpeter@601
   503
      }
kpeter@601
   504
kpeter@601
   505
    }; //class CandidateListPivotRule
kpeter@601
   506
kpeter@601
   507
kpeter@605
   508
    // Implementation of the Altering Candidate List pivot rule
kpeter@601
   509
    class AlteringListPivotRule
kpeter@601
   510
    {
kpeter@601
   511
    private:
kpeter@601
   512
kpeter@601
   513
      // References to the NetworkSimplex class
kpeter@601
   514
      const IntVector  &_source;
kpeter@601
   515
      const IntVector  &_target;
kpeter@607
   516
      const CostVector &_cost;
kpeter@601
   517
      const IntVector  &_state;
kpeter@607
   518
      const CostVector &_pi;
kpeter@601
   519
      int &_in_arc;
kpeter@663
   520
      int _search_arc_num;
kpeter@601
   521
kpeter@601
   522
      // Pivot rule data
kpeter@601
   523
      int _block_size, _head_length, _curr_length;
kpeter@601
   524
      int _next_arc;
kpeter@601
   525
      IntVector _candidates;
kpeter@607
   526
      CostVector _cand_cost;
kpeter@601
   527
kpeter@601
   528
      // Functor class to compare arcs during sort of the candidate list
kpeter@601
   529
      class SortFunc
kpeter@601
   530
      {
kpeter@601
   531
      private:
kpeter@607
   532
        const CostVector &_map;
kpeter@601
   533
      public:
kpeter@607
   534
        SortFunc(const CostVector &map) : _map(map) {}
kpeter@601
   535
        bool operator()(int left, int right) {
kpeter@601
   536
          return _map[left] > _map[right];
kpeter@601
   537
        }
kpeter@601
   538
      };
kpeter@601
   539
kpeter@601
   540
      SortFunc _sort_func;
kpeter@601
   541
kpeter@601
   542
    public:
kpeter@601
   543
kpeter@605
   544
      // Constructor
kpeter@601
   545
      AlteringListPivotRule(NetworkSimplex &ns) :
kpeter@603
   546
        _source(ns._source), _target(ns._target),
kpeter@601
   547
        _cost(ns._cost), _state(ns._state), _pi(ns._pi),
kpeter@663
   548
        _in_arc(ns.in_arc), _search_arc_num(ns._search_arc_num),
kpeter@663
   549
        _next_arc(0), _cand_cost(ns._search_arc_num), _sort_func(_cand_cost)
kpeter@601
   550
      {
kpeter@601
   551
        // The main parameters of the pivot rule
kpeter@601
   552
        const double BLOCK_SIZE_FACTOR = 1.5;
kpeter@601
   553
        const int MIN_BLOCK_SIZE = 10;
kpeter@601
   554
        const double HEAD_LENGTH_FACTOR = 0.1;
kpeter@601
   555
        const int MIN_HEAD_LENGTH = 3;
kpeter@601
   556
alpar@612
   557
        _block_size = std::max( int(BLOCK_SIZE_FACTOR *
kpeter@663
   558
                                    std::sqrt(double(_search_arc_num))),
kpeter@601
   559
                                MIN_BLOCK_SIZE );
kpeter@601
   560
        _head_length = std::max( int(HEAD_LENGTH_FACTOR * _block_size),
kpeter@601
   561
                                 MIN_HEAD_LENGTH );
kpeter@601
   562
        _candidates.resize(_head_length + _block_size);
kpeter@601
   563
        _curr_length = 0;
kpeter@601
   564
      }
kpeter@601
   565
kpeter@605
   566
      // Find next entering arc
kpeter@601
   567
      bool findEnteringArc() {
kpeter@601
   568
        // Check the current candidate list
kpeter@601
   569
        int e;
kpeter@601
   570
        for (int i = 0; i < _curr_length; ++i) {
kpeter@601
   571
          e = _candidates[i];
kpeter@601
   572
          _cand_cost[e] = _state[e] *
kpeter@601
   573
            (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@601
   574
          if (_cand_cost[e] >= 0) {
kpeter@601
   575
            _candidates[i--] = _candidates[--_curr_length];
kpeter@601
   576
          }
kpeter@601
   577
        }
kpeter@601
   578
kpeter@601
   579
        // Extend the list
kpeter@601
   580
        int cnt = _block_size;
kpeter@603
   581
        int last_arc = 0;
kpeter@601
   582
        int limit = _head_length;
kpeter@601
   583
kpeter@663
   584
        for (int e = _next_arc; e < _search_arc_num; ++e) {
kpeter@601
   585
          _cand_cost[e] = _state[e] *
kpeter@601
   586
            (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@601
   587
          if (_cand_cost[e] < 0) {
kpeter@601
   588
            _candidates[_curr_length++] = e;
kpeter@603
   589
            last_arc = e;
kpeter@601
   590
          }
kpeter@601
   591
          if (--cnt == 0) {
kpeter@601
   592
            if (_curr_length > limit) break;
kpeter@601
   593
            limit = 0;
kpeter@601
   594
            cnt = _block_size;
kpeter@601
   595
          }
kpeter@601
   596
        }
kpeter@601
   597
        if (_curr_length <= limit) {
kpeter@601
   598
          for (int e = 0; e < _next_arc; ++e) {
kpeter@601
   599
            _cand_cost[e] = _state[e] *
kpeter@601
   600
              (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@601
   601
            if (_cand_cost[e] < 0) {
kpeter@601
   602
              _candidates[_curr_length++] = e;
kpeter@603
   603
              last_arc = e;
kpeter@601
   604
            }
kpeter@601
   605
            if (--cnt == 0) {
kpeter@601
   606
              if (_curr_length > limit) break;
kpeter@601
   607
              limit = 0;
kpeter@601
   608
              cnt = _block_size;
kpeter@601
   609
            }
kpeter@601
   610
          }
kpeter@601
   611
        }
kpeter@601
   612
        if (_curr_length == 0) return false;
kpeter@603
   613
        _next_arc = last_arc + 1;
kpeter@601
   614
kpeter@601
   615
        // Make heap of the candidate list (approximating a partial sort)
kpeter@601
   616
        make_heap( _candidates.begin(), _candidates.begin() + _curr_length,
kpeter@601
   617
                   _sort_func );
kpeter@601
   618
kpeter@601
   619
        // Pop the first element of the heap
kpeter@601
   620
        _in_arc = _candidates[0];
kpeter@601
   621
        pop_heap( _candidates.begin(), _candidates.begin() + _curr_length,
kpeter@601
   622
                  _sort_func );
kpeter@601
   623
        _curr_length = std::min(_head_length, _curr_length - 1);
kpeter@601
   624
        return true;
kpeter@601
   625
      }
kpeter@601
   626
kpeter@601
   627
    }; //class AlteringListPivotRule
kpeter@601
   628
kpeter@601
   629
  public:
kpeter@601
   630
kpeter@605
   631
    /// \brief Constructor.
kpeter@601
   632
    ///
kpeter@609
   633
    /// The constructor of the class.
kpeter@601
   634
    ///
kpeter@603
   635
    /// \param graph The digraph the algorithm runs on.
kpeter@605
   636
    NetworkSimplex(const GR& graph) :
kpeter@642
   637
      _graph(graph), _node_id(graph), _arc_id(graph),
kpeter@641
   638
      INF(std::numeric_limits<Value>::has_infinity ?
kpeter@641
   639
          std::numeric_limits<Value>::infinity() :
kpeter@641
   640
          std::numeric_limits<Value>::max())
kpeter@605
   641
    {
kpeter@640
   642
      // Check the value types
kpeter@641
   643
      LEMON_ASSERT(std::numeric_limits<Value>::is_signed,
kpeter@640
   644
        "The flow type of NetworkSimplex must be signed");
kpeter@640
   645
      LEMON_ASSERT(std::numeric_limits<Cost>::is_signed,
kpeter@640
   646
        "The cost type of NetworkSimplex must be signed");
kpeter@642
   647
        
kpeter@642
   648
      // Resize vectors
kpeter@642
   649
      _node_num = countNodes(_graph);
kpeter@642
   650
      _arc_num = countArcs(_graph);
kpeter@642
   651
      int all_node_num = _node_num + 1;
kpeter@663
   652
      int max_arc_num = _arc_num + 2 * _node_num;
kpeter@601
   653
kpeter@663
   654
      _source.resize(max_arc_num);
kpeter@663
   655
      _target.resize(max_arc_num);
kpeter@642
   656
kpeter@663
   657
      _lower.resize(_arc_num);
kpeter@663
   658
      _upper.resize(_arc_num);
kpeter@663
   659
      _cap.resize(max_arc_num);
kpeter@663
   660
      _cost.resize(max_arc_num);
kpeter@642
   661
      _supply.resize(all_node_num);
kpeter@663
   662
      _flow.resize(max_arc_num);
kpeter@642
   663
      _pi.resize(all_node_num);
kpeter@642
   664
kpeter@642
   665
      _parent.resize(all_node_num);
kpeter@642
   666
      _pred.resize(all_node_num);
kpeter@642
   667
      _forward.resize(all_node_num);
kpeter@642
   668
      _thread.resize(all_node_num);
kpeter@642
   669
      _rev_thread.resize(all_node_num);
kpeter@642
   670
      _succ_num.resize(all_node_num);
kpeter@642
   671
      _last_succ.resize(all_node_num);
kpeter@663
   672
      _state.resize(max_arc_num);
kpeter@642
   673
kpeter@642
   674
      // Copy the graph (store the arcs in a mixed order)
kpeter@642
   675
      int i = 0;
kpeter@642
   676
      for (NodeIt n(_graph); n != INVALID; ++n, ++i) {
kpeter@642
   677
        _node_id[n] = i;
kpeter@642
   678
      }
kpeter@642
   679
      int k = std::max(int(std::sqrt(double(_arc_num))), 10);
kpeter@642
   680
      i = 0;
kpeter@642
   681
      for (ArcIt a(_graph); a != INVALID; ++a) {
kpeter@642
   682
        _arc_id[a] = i;
kpeter@642
   683
        _source[i] = _node_id[_graph.source(a)];
kpeter@642
   684
        _target[i] = _node_id[_graph.target(a)];
kpeter@642
   685
        if ((i += k) >= _arc_num) i = (i % k) + 1;
kpeter@642
   686
      }
kpeter@642
   687
      
kpeter@642
   688
      // Initialize maps
kpeter@642
   689
      for (int i = 0; i != _node_num; ++i) {
kpeter@642
   690
        _supply[i] = 0;
kpeter@642
   691
      }
kpeter@642
   692
      for (int i = 0; i != _arc_num; ++i) {
kpeter@642
   693
        _lower[i] = 0;
kpeter@642
   694
        _upper[i] = INF;
kpeter@642
   695
        _cost[i] = 1;
kpeter@642
   696
      }
kpeter@642
   697
      _have_lower = false;
kpeter@642
   698
      _stype = GEQ;
kpeter@601
   699
    }
kpeter@601
   700
kpeter@609
   701
    /// \name Parameters
kpeter@609
   702
    /// The parameters of the algorithm can be specified using these
kpeter@609
   703
    /// functions.
kpeter@609
   704
kpeter@609
   705
    /// @{
kpeter@609
   706
kpeter@605
   707
    /// \brief Set the lower bounds on the arcs.
kpeter@605
   708
    ///
kpeter@605
   709
    /// This function sets the lower bounds on the arcs.
kpeter@640
   710
    /// If it is not used before calling \ref run(), the lower bounds
kpeter@640
   711
    /// will be set to zero on all arcs.
kpeter@605
   712
    ///
kpeter@605
   713
    /// \param map An arc map storing the lower bounds.
kpeter@641
   714
    /// Its \c Value type must be convertible to the \c Value type
kpeter@605
   715
    /// of the algorithm.
kpeter@605
   716
    ///
kpeter@605
   717
    /// \return <tt>(*this)</tt>
kpeter@640
   718
    template <typename LowerMap>
kpeter@640
   719
    NetworkSimplex& lowerMap(const LowerMap& map) {
kpeter@642
   720
      _have_lower = true;
kpeter@605
   721
      for (ArcIt a(_graph); a != INVALID; ++a) {
kpeter@642
   722
        _lower[_arc_id[a]] = map[a];
kpeter@605
   723
      }
kpeter@605
   724
      return *this;
kpeter@605
   725
    }
kpeter@605
   726
kpeter@605
   727
    /// \brief Set the upper bounds (capacities) on the arcs.
kpeter@605
   728
    ///
kpeter@605
   729
    /// This function sets the upper bounds (capacities) on the arcs.
kpeter@640
   730
    /// If it is not used before calling \ref run(), the upper bounds
kpeter@640
   731
    /// will be set to \ref INF on all arcs (i.e. the flow value will be
kpeter@640
   732
    /// unbounded from above on each arc).
kpeter@605
   733
    ///
kpeter@605
   734
    /// \param map An arc map storing the upper bounds.
kpeter@641
   735
    /// Its \c Value type must be convertible to the \c Value type
kpeter@605
   736
    /// of the algorithm.
kpeter@605
   737
    ///
kpeter@605
   738
    /// \return <tt>(*this)</tt>
kpeter@640
   739
    template<typename UpperMap>
kpeter@640
   740
    NetworkSimplex& upperMap(const UpperMap& map) {
kpeter@605
   741
      for (ArcIt a(_graph); a != INVALID; ++a) {
kpeter@642
   742
        _upper[_arc_id[a]] = map[a];
kpeter@605
   743
      }
kpeter@605
   744
      return *this;
kpeter@605
   745
    }
kpeter@605
   746
kpeter@605
   747
    /// \brief Set the costs of the arcs.
kpeter@605
   748
    ///
kpeter@605
   749
    /// This function sets the costs of the arcs.
kpeter@605
   750
    /// If it is not used before calling \ref run(), the costs
kpeter@605
   751
    /// will be set to \c 1 on all arcs.
kpeter@605
   752
    ///
kpeter@605
   753
    /// \param map An arc map storing the costs.
kpeter@607
   754
    /// Its \c Value type must be convertible to the \c Cost type
kpeter@605
   755
    /// of the algorithm.
kpeter@605
   756
    ///
kpeter@605
   757
    /// \return <tt>(*this)</tt>
kpeter@640
   758
    template<typename CostMap>
kpeter@640
   759
    NetworkSimplex& costMap(const CostMap& map) {
kpeter@605
   760
      for (ArcIt a(_graph); a != INVALID; ++a) {
kpeter@642
   761
        _cost[_arc_id[a]] = map[a];
kpeter@605
   762
      }
kpeter@605
   763
      return *this;
kpeter@605
   764
    }
kpeter@605
   765
kpeter@605
   766
    /// \brief Set the supply values of the nodes.
kpeter@605
   767
    ///
kpeter@605
   768
    /// This function sets the supply values of the nodes.
kpeter@605
   769
    /// If neither this function nor \ref stSupply() is used before
kpeter@605
   770
    /// calling \ref run(), the supply of each node will be set to zero.
kpeter@605
   771
    /// (It makes sense only if non-zero lower bounds are given.)
kpeter@605
   772
    ///
kpeter@605
   773
    /// \param map A node map storing the supply values.
kpeter@641
   774
    /// Its \c Value type must be convertible to the \c Value type
kpeter@605
   775
    /// of the algorithm.
kpeter@605
   776
    ///
kpeter@605
   777
    /// \return <tt>(*this)</tt>
kpeter@640
   778
    template<typename SupplyMap>
kpeter@640
   779
    NetworkSimplex& supplyMap(const SupplyMap& map) {
kpeter@605
   780
      for (NodeIt n(_graph); n != INVALID; ++n) {
kpeter@642
   781
        _supply[_node_id[n]] = map[n];
kpeter@605
   782
      }
kpeter@605
   783
      return *this;
kpeter@605
   784
    }
kpeter@605
   785
kpeter@605
   786
    /// \brief Set single source and target nodes and a supply value.
kpeter@605
   787
    ///
kpeter@605
   788
    /// This function sets a single source node and a single target node
kpeter@605
   789
    /// and the required flow value.
kpeter@605
   790
    /// If neither this function nor \ref supplyMap() is used before
kpeter@605
   791
    /// calling \ref run(), the supply of each node will be set to zero.
kpeter@605
   792
    /// (It makes sense only if non-zero lower bounds are given.)
kpeter@605
   793
    ///
kpeter@640
   794
    /// Using this function has the same effect as using \ref supplyMap()
kpeter@640
   795
    /// with such a map in which \c k is assigned to \c s, \c -k is
kpeter@640
   796
    /// assigned to \c t and all other nodes have zero supply value.
kpeter@640
   797
    ///
kpeter@605
   798
    /// \param s The source node.
kpeter@605
   799
    /// \param t The target node.
kpeter@605
   800
    /// \param k The required amount of flow from node \c s to node \c t
kpeter@605
   801
    /// (i.e. the supply of \c s and the demand of \c t).
kpeter@605
   802
    ///
kpeter@605
   803
    /// \return <tt>(*this)</tt>
kpeter@641
   804
    NetworkSimplex& stSupply(const Node& s, const Node& t, Value k) {
kpeter@642
   805
      for (int i = 0; i != _node_num; ++i) {
kpeter@642
   806
        _supply[i] = 0;
kpeter@642
   807
      }
kpeter@642
   808
      _supply[_node_id[s]] =  k;
kpeter@642
   809
      _supply[_node_id[t]] = -k;
kpeter@605
   810
      return *this;
kpeter@605
   811
    }
kpeter@609
   812
    
kpeter@640
   813
    /// \brief Set the type of the supply constraints.
kpeter@609
   814
    ///
kpeter@640
   815
    /// This function sets the type of the supply/demand constraints.
kpeter@640
   816
    /// If it is not used before calling \ref run(), the \ref GEQ supply
kpeter@609
   817
    /// type will be used.
kpeter@609
   818
    ///
kpeter@640
   819
    /// For more information see \ref SupplyType.
kpeter@609
   820
    ///
kpeter@609
   821
    /// \return <tt>(*this)</tt>
kpeter@640
   822
    NetworkSimplex& supplyType(SupplyType supply_type) {
kpeter@640
   823
      _stype = supply_type;
kpeter@609
   824
      return *this;
kpeter@609
   825
    }
kpeter@605
   826
kpeter@609
   827
    /// @}
kpeter@601
   828
kpeter@605
   829
    /// \name Execution Control
kpeter@605
   830
    /// The algorithm can be executed using \ref run().
kpeter@605
   831
kpeter@601
   832
    /// @{
kpeter@601
   833
kpeter@601
   834
    /// \brief Run the algorithm.
kpeter@601
   835
    ///
kpeter@601
   836
    /// This function runs the algorithm.
kpeter@609
   837
    /// The paramters can be specified using functions \ref lowerMap(),
kpeter@640
   838
    /// \ref upperMap(), \ref costMap(), \ref supplyMap(), \ref stSupply(), 
kpeter@642
   839
    /// \ref supplyType().
kpeter@609
   840
    /// For example,
kpeter@605
   841
    /// \code
kpeter@605
   842
    ///   NetworkSimplex<ListDigraph> ns(graph);
kpeter@640
   843
    ///   ns.lowerMap(lower).upperMap(upper).costMap(cost)
kpeter@605
   844
    ///     .supplyMap(sup).run();
kpeter@605
   845
    /// \endcode
kpeter@601
   846
    ///
kpeter@606
   847
    /// This function can be called more than once. All the parameters
kpeter@606
   848
    /// that have been given are kept for the next call, unless
kpeter@606
   849
    /// \ref reset() is called, thus only the modified parameters
kpeter@606
   850
    /// have to be set again. See \ref reset() for examples.
kpeter@642
   851
    /// However the underlying digraph must not be modified after this
kpeter@642
   852
    /// class have been constructed, since it copies and extends the graph.
kpeter@606
   853
    ///
kpeter@605
   854
    /// \param pivot_rule The pivot rule that will be used during the
kpeter@605
   855
    /// algorithm. For more information see \ref PivotRule.
kpeter@601
   856
    ///
kpeter@640
   857
    /// \return \c INFEASIBLE if no feasible flow exists,
kpeter@640
   858
    /// \n \c OPTIMAL if the problem has optimal solution
kpeter@640
   859
    /// (i.e. it is feasible and bounded), and the algorithm has found
kpeter@640
   860
    /// optimal flow and node potentials (primal and dual solutions),
kpeter@640
   861
    /// \n \c UNBOUNDED if the objective function of the problem is
kpeter@640
   862
    /// unbounded, i.e. there is a directed cycle having negative total
kpeter@640
   863
    /// cost and infinite upper bound.
kpeter@640
   864
    ///
kpeter@640
   865
    /// \see ProblemType, PivotRule
kpeter@640
   866
    ProblemType run(PivotRule pivot_rule = BLOCK_SEARCH) {
kpeter@640
   867
      if (!init()) return INFEASIBLE;
kpeter@640
   868
      return start(pivot_rule);
kpeter@601
   869
    }
kpeter@601
   870
kpeter@606
   871
    /// \brief Reset all the parameters that have been given before.
kpeter@606
   872
    ///
kpeter@606
   873
    /// This function resets all the paramaters that have been given
kpeter@609
   874
    /// before using functions \ref lowerMap(), \ref upperMap(),
kpeter@642
   875
    /// \ref costMap(), \ref supplyMap(), \ref stSupply(), \ref supplyType().
kpeter@606
   876
    ///
kpeter@606
   877
    /// It is useful for multiple run() calls. If this function is not
kpeter@606
   878
    /// used, all the parameters given before are kept for the next
kpeter@606
   879
    /// \ref run() call.
kpeter@642
   880
    /// However the underlying digraph must not be modified after this
kpeter@642
   881
    /// class have been constructed, since it copies and extends the graph.
kpeter@606
   882
    ///
kpeter@606
   883
    /// For example,
kpeter@606
   884
    /// \code
kpeter@606
   885
    ///   NetworkSimplex<ListDigraph> ns(graph);
kpeter@606
   886
    ///
kpeter@606
   887
    ///   // First run
kpeter@640
   888
    ///   ns.lowerMap(lower).upperMap(upper).costMap(cost)
kpeter@606
   889
    ///     .supplyMap(sup).run();
kpeter@606
   890
    ///
kpeter@606
   891
    ///   // Run again with modified cost map (reset() is not called,
kpeter@606
   892
    ///   // so only the cost map have to be set again)
kpeter@606
   893
    ///   cost[e] += 100;
kpeter@606
   894
    ///   ns.costMap(cost).run();
kpeter@606
   895
    ///
kpeter@606
   896
    ///   // Run again from scratch using reset()
kpeter@606
   897
    ///   // (the lower bounds will be set to zero on all arcs)
kpeter@606
   898
    ///   ns.reset();
kpeter@640
   899
    ///   ns.upperMap(capacity).costMap(cost)
kpeter@606
   900
    ///     .supplyMap(sup).run();
kpeter@606
   901
    /// \endcode
kpeter@606
   902
    ///
kpeter@606
   903
    /// \return <tt>(*this)</tt>
kpeter@606
   904
    NetworkSimplex& reset() {
kpeter@642
   905
      for (int i = 0; i != _node_num; ++i) {
kpeter@642
   906
        _supply[i] = 0;
kpeter@642
   907
      }
kpeter@642
   908
      for (int i = 0; i != _arc_num; ++i) {
kpeter@642
   909
        _lower[i] = 0;
kpeter@642
   910
        _upper[i] = INF;
kpeter@642
   911
        _cost[i] = 1;
kpeter@642
   912
      }
kpeter@642
   913
      _have_lower = false;
kpeter@640
   914
      _stype = GEQ;
kpeter@606
   915
      return *this;
kpeter@606
   916
    }
kpeter@606
   917
kpeter@601
   918
    /// @}
kpeter@601
   919
kpeter@601
   920
    /// \name Query Functions
kpeter@601
   921
    /// The results of the algorithm can be obtained using these
kpeter@601
   922
    /// functions.\n
kpeter@605
   923
    /// The \ref run() function must be called before using them.
kpeter@605
   924
kpeter@601
   925
    /// @{
kpeter@601
   926
kpeter@605
   927
    /// \brief Return the total cost of the found flow.
kpeter@605
   928
    ///
kpeter@605
   929
    /// This function returns the total cost of the found flow.
kpeter@640
   930
    /// Its complexity is O(e).
kpeter@605
   931
    ///
kpeter@605
   932
    /// \note The return type of the function can be specified as a
kpeter@605
   933
    /// template parameter. For example,
kpeter@605
   934
    /// \code
kpeter@605
   935
    ///   ns.totalCost<double>();
kpeter@605
   936
    /// \endcode
kpeter@607
   937
    /// It is useful if the total cost cannot be stored in the \c Cost
kpeter@605
   938
    /// type of the algorithm, which is the default return type of the
kpeter@605
   939
    /// function.
kpeter@605
   940
    ///
kpeter@605
   941
    /// \pre \ref run() must be called before using this function.
kpeter@642
   942
    template <typename Number>
kpeter@642
   943
    Number totalCost() const {
kpeter@642
   944
      Number c = 0;
kpeter@642
   945
      for (ArcIt a(_graph); a != INVALID; ++a) {
kpeter@642
   946
        int i = _arc_id[a];
kpeter@642
   947
        c += Number(_flow[i]) * Number(_cost[i]);
kpeter@605
   948
      }
kpeter@605
   949
      return c;
kpeter@605
   950
    }
kpeter@605
   951
kpeter@605
   952
#ifndef DOXYGEN
kpeter@607
   953
    Cost totalCost() const {
kpeter@607
   954
      return totalCost<Cost>();
kpeter@605
   955
    }
kpeter@605
   956
#endif
kpeter@605
   957
kpeter@605
   958
    /// \brief Return the flow on the given arc.
kpeter@605
   959
    ///
kpeter@605
   960
    /// This function returns the flow on the given arc.
kpeter@605
   961
    ///
kpeter@605
   962
    /// \pre \ref run() must be called before using this function.
kpeter@641
   963
    Value flow(const Arc& a) const {
kpeter@642
   964
      return _flow[_arc_id[a]];
kpeter@605
   965
    }
kpeter@605
   966
kpeter@642
   967
    /// \brief Return the flow map (the primal solution).
kpeter@601
   968
    ///
kpeter@642
   969
    /// This function copies the flow value on each arc into the given
kpeter@642
   970
    /// map. The \c Value type of the algorithm must be convertible to
kpeter@642
   971
    /// the \c Value type of the map.
kpeter@601
   972
    ///
kpeter@601
   973
    /// \pre \ref run() must be called before using this function.
kpeter@642
   974
    template <typename FlowMap>
kpeter@642
   975
    void flowMap(FlowMap &map) const {
kpeter@642
   976
      for (ArcIt a(_graph); a != INVALID; ++a) {
kpeter@642
   977
        map.set(a, _flow[_arc_id[a]]);
kpeter@642
   978
      }
kpeter@601
   979
    }
kpeter@601
   980
kpeter@605
   981
    /// \brief Return the potential (dual value) of the given node.
kpeter@605
   982
    ///
kpeter@605
   983
    /// This function returns the potential (dual value) of the
kpeter@605
   984
    /// given node.
kpeter@605
   985
    ///
kpeter@605
   986
    /// \pre \ref run() must be called before using this function.
kpeter@607
   987
    Cost potential(const Node& n) const {
kpeter@642
   988
      return _pi[_node_id[n]];
kpeter@605
   989
    }
kpeter@605
   990
kpeter@642
   991
    /// \brief Return the potential map (the dual solution).
kpeter@601
   992
    ///
kpeter@642
   993
    /// This function copies the potential (dual value) of each node
kpeter@642
   994
    /// into the given map.
kpeter@642
   995
    /// The \c Cost type of the algorithm must be convertible to the
kpeter@642
   996
    /// \c Value type of the map.
kpeter@601
   997
    ///
kpeter@601
   998
    /// \pre \ref run() must be called before using this function.
kpeter@642
   999
    template <typename PotentialMap>
kpeter@642
  1000
    void potentialMap(PotentialMap &map) const {
kpeter@642
  1001
      for (NodeIt n(_graph); n != INVALID; ++n) {
kpeter@642
  1002
        map.set(n, _pi[_node_id[n]]);
kpeter@642
  1003
      }
kpeter@601
  1004
    }
kpeter@601
  1005
kpeter@601
  1006
    /// @}
kpeter@601
  1007
kpeter@601
  1008
  private:
kpeter@601
  1009
kpeter@601
  1010
    // Initialize internal data structures
kpeter@601
  1011
    bool init() {
kpeter@605
  1012
      if (_node_num == 0) return false;
kpeter@601
  1013
kpeter@642
  1014
      // Check the sum of supply values
kpeter@642
  1015
      _sum_supply = 0;
kpeter@642
  1016
      for (int i = 0; i != _node_num; ++i) {
kpeter@642
  1017
        _sum_supply += _supply[i];
kpeter@642
  1018
      }
alpar@643
  1019
      if ( !((_stype == GEQ && _sum_supply <= 0) ||
alpar@643
  1020
             (_stype == LEQ && _sum_supply >= 0)) ) return false;
kpeter@601
  1021
kpeter@642
  1022
      // Remove non-zero lower bounds
kpeter@642
  1023
      if (_have_lower) {
kpeter@642
  1024
        for (int i = 0; i != _arc_num; ++i) {
kpeter@642
  1025
          Value c = _lower[i];
kpeter@642
  1026
          if (c >= 0) {
kpeter@642
  1027
            _cap[i] = _upper[i] < INF ? _upper[i] - c : INF;
kpeter@642
  1028
          } else {
kpeter@642
  1029
            _cap[i] = _upper[i] < INF + c ? _upper[i] - c : INF;
kpeter@642
  1030
          }
kpeter@642
  1031
          _supply[_source[i]] -= c;
kpeter@642
  1032
          _supply[_target[i]] += c;
kpeter@642
  1033
        }
kpeter@642
  1034
      } else {
kpeter@642
  1035
        for (int i = 0; i != _arc_num; ++i) {
kpeter@642
  1036
          _cap[i] = _upper[i];
kpeter@642
  1037
        }
kpeter@605
  1038
      }
kpeter@601
  1039
kpeter@609
  1040
      // Initialize artifical cost
kpeter@640
  1041
      Cost ART_COST;
kpeter@609
  1042
      if (std::numeric_limits<Cost>::is_exact) {
kpeter@663
  1043
        ART_COST = std::numeric_limits<Cost>::max() / 2 + 1;
kpeter@609
  1044
      } else {
kpeter@640
  1045
        ART_COST = std::numeric_limits<Cost>::min();
kpeter@609
  1046
        for (int i = 0; i != _arc_num; ++i) {
kpeter@640
  1047
          if (_cost[i] > ART_COST) ART_COST = _cost[i];
kpeter@609
  1048
        }
kpeter@640
  1049
        ART_COST = (ART_COST + 1) * _node_num;
kpeter@609
  1050
      }
kpeter@609
  1051
kpeter@642
  1052
      // Initialize arc maps
kpeter@642
  1053
      for (int i = 0; i != _arc_num; ++i) {
kpeter@642
  1054
        _flow[i] = 0;
kpeter@642
  1055
        _state[i] = STATE_LOWER;
kpeter@642
  1056
      }
kpeter@642
  1057
      
kpeter@601
  1058
      // Set data for the artificial root node
kpeter@601
  1059
      _root = _node_num;
kpeter@601
  1060
      _parent[_root] = -1;
kpeter@601
  1061
      _pred[_root] = -1;
kpeter@601
  1062
      _thread[_root] = 0;
kpeter@604
  1063
      _rev_thread[0] = _root;
kpeter@642
  1064
      _succ_num[_root] = _node_num + 1;
kpeter@604
  1065
      _last_succ[_root] = _root - 1;
kpeter@640
  1066
      _supply[_root] = -_sum_supply;
kpeter@663
  1067
      _pi[_root] = 0;
kpeter@601
  1068
kpeter@601
  1069
      // Add artificial arcs and initialize the spanning tree data structure
kpeter@663
  1070
      if (_sum_supply == 0) {
kpeter@663
  1071
        // EQ supply constraints
kpeter@663
  1072
        _search_arc_num = _arc_num;
kpeter@663
  1073
        _all_arc_num = _arc_num + _node_num;
kpeter@663
  1074
        for (int u = 0, e = _arc_num; u != _node_num; ++u, ++e) {
kpeter@663
  1075
          _parent[u] = _root;
kpeter@663
  1076
          _pred[u] = e;
kpeter@663
  1077
          _thread[u] = u + 1;
kpeter@663
  1078
          _rev_thread[u + 1] = u;
kpeter@663
  1079
          _succ_num[u] = 1;
kpeter@663
  1080
          _last_succ[u] = u;
kpeter@663
  1081
          _cap[e] = INF;
kpeter@663
  1082
          _state[e] = STATE_TREE;
kpeter@663
  1083
          if (_supply[u] >= 0) {
kpeter@663
  1084
            _forward[u] = true;
kpeter@663
  1085
            _pi[u] = 0;
kpeter@663
  1086
            _source[e] = u;
kpeter@663
  1087
            _target[e] = _root;
kpeter@663
  1088
            _flow[e] = _supply[u];
kpeter@663
  1089
            _cost[e] = 0;
kpeter@663
  1090
          } else {
kpeter@663
  1091
            _forward[u] = false;
kpeter@663
  1092
            _pi[u] = ART_COST;
kpeter@663
  1093
            _source[e] = _root;
kpeter@663
  1094
            _target[e] = u;
kpeter@663
  1095
            _flow[e] = -_supply[u];
kpeter@663
  1096
            _cost[e] = ART_COST;
kpeter@663
  1097
          }
kpeter@601
  1098
        }
kpeter@601
  1099
      }
kpeter@663
  1100
      else if (_sum_supply > 0) {
kpeter@663
  1101
        // LEQ supply constraints
kpeter@663
  1102
        _search_arc_num = _arc_num + _node_num;
kpeter@663
  1103
        int f = _arc_num + _node_num;
kpeter@663
  1104
        for (int u = 0, e = _arc_num; u != _node_num; ++u, ++e) {
kpeter@663
  1105
          _parent[u] = _root;
kpeter@663
  1106
          _thread[u] = u + 1;
kpeter@663
  1107
          _rev_thread[u + 1] = u;
kpeter@663
  1108
          _succ_num[u] = 1;
kpeter@663
  1109
          _last_succ[u] = u;
kpeter@663
  1110
          if (_supply[u] >= 0) {
kpeter@663
  1111
            _forward[u] = true;
kpeter@663
  1112
            _pi[u] = 0;
kpeter@663
  1113
            _pred[u] = e;
kpeter@663
  1114
            _source[e] = u;
kpeter@663
  1115
            _target[e] = _root;
kpeter@663
  1116
            _cap[e] = INF;
kpeter@663
  1117
            _flow[e] = _supply[u];
kpeter@663
  1118
            _cost[e] = 0;
kpeter@663
  1119
            _state[e] = STATE_TREE;
kpeter@663
  1120
          } else {
kpeter@663
  1121
            _forward[u] = false;
kpeter@663
  1122
            _pi[u] = ART_COST;
kpeter@663
  1123
            _pred[u] = f;
kpeter@663
  1124
            _source[f] = _root;
kpeter@663
  1125
            _target[f] = u;
kpeter@663
  1126
            _cap[f] = INF;
kpeter@663
  1127
            _flow[f] = -_supply[u];
kpeter@663
  1128
            _cost[f] = ART_COST;
kpeter@663
  1129
            _state[f] = STATE_TREE;
kpeter@663
  1130
            _source[e] = u;
kpeter@663
  1131
            _target[e] = _root;
kpeter@663
  1132
            _cap[e] = INF;
kpeter@663
  1133
            _flow[e] = 0;
kpeter@663
  1134
            _cost[e] = 0;
kpeter@663
  1135
            _state[e] = STATE_LOWER;
kpeter@663
  1136
            ++f;
kpeter@663
  1137
          }
kpeter@663
  1138
        }
kpeter@663
  1139
        _all_arc_num = f;
kpeter@663
  1140
      }
kpeter@663
  1141
      else {
kpeter@663
  1142
        // GEQ supply constraints
kpeter@663
  1143
        _search_arc_num = _arc_num + _node_num;
kpeter@663
  1144
        int f = _arc_num + _node_num;
kpeter@663
  1145
        for (int u = 0, e = _arc_num; u != _node_num; ++u, ++e) {
kpeter@663
  1146
          _parent[u] = _root;
kpeter@663
  1147
          _thread[u] = u + 1;
kpeter@663
  1148
          _rev_thread[u + 1] = u;
kpeter@663
  1149
          _succ_num[u] = 1;
kpeter@663
  1150
          _last_succ[u] = u;
kpeter@663
  1151
          if (_supply[u] <= 0) {
kpeter@663
  1152
            _forward[u] = false;
kpeter@663
  1153
            _pi[u] = 0;
kpeter@663
  1154
            _pred[u] = e;
kpeter@663
  1155
            _source[e] = _root;
kpeter@663
  1156
            _target[e] = u;
kpeter@663
  1157
            _cap[e] = INF;
kpeter@663
  1158
            _flow[e] = -_supply[u];
kpeter@663
  1159
            _cost[e] = 0;
kpeter@663
  1160
            _state[e] = STATE_TREE;
kpeter@663
  1161
          } else {
kpeter@663
  1162
            _forward[u] = true;
kpeter@663
  1163
            _pi[u] = -ART_COST;
kpeter@663
  1164
            _pred[u] = f;
kpeter@663
  1165
            _source[f] = u;
kpeter@663
  1166
            _target[f] = _root;
kpeter@663
  1167
            _cap[f] = INF;
kpeter@663
  1168
            _flow[f] = _supply[u];
kpeter@663
  1169
            _state[f] = STATE_TREE;
kpeter@663
  1170
            _cost[f] = ART_COST;
kpeter@663
  1171
            _source[e] = _root;
kpeter@663
  1172
            _target[e] = u;
kpeter@663
  1173
            _cap[e] = INF;
kpeter@663
  1174
            _flow[e] = 0;
kpeter@663
  1175
            _cost[e] = 0;
kpeter@663
  1176
            _state[e] = STATE_LOWER;
kpeter@663
  1177
            ++f;
kpeter@663
  1178
          }
kpeter@663
  1179
        }
kpeter@663
  1180
        _all_arc_num = f;
kpeter@663
  1181
      }
kpeter@601
  1182
kpeter@601
  1183
      return true;
kpeter@601
  1184
    }
kpeter@601
  1185
kpeter@601
  1186
    // Find the join node
kpeter@601
  1187
    void findJoinNode() {
kpeter@603
  1188
      int u = _source[in_arc];
kpeter@603
  1189
      int v = _target[in_arc];
kpeter@601
  1190
      while (u != v) {
kpeter@604
  1191
        if (_succ_num[u] < _succ_num[v]) {
kpeter@604
  1192
          u = _parent[u];
kpeter@604
  1193
        } else {
kpeter@604
  1194
          v = _parent[v];
kpeter@604
  1195
        }
kpeter@601
  1196
      }
kpeter@601
  1197
      join = u;
kpeter@601
  1198
    }
kpeter@601
  1199
kpeter@601
  1200
    // Find the leaving arc of the cycle and returns true if the
kpeter@601
  1201
    // leaving arc is not the same as the entering arc
kpeter@601
  1202
    bool findLeavingArc() {
kpeter@601
  1203
      // Initialize first and second nodes according to the direction
kpeter@601
  1204
      // of the cycle
kpeter@603
  1205
      if (_state[in_arc] == STATE_LOWER) {
kpeter@603
  1206
        first  = _source[in_arc];
kpeter@603
  1207
        second = _target[in_arc];
kpeter@601
  1208
      } else {
kpeter@603
  1209
        first  = _target[in_arc];
kpeter@603
  1210
        second = _source[in_arc];
kpeter@601
  1211
      }
kpeter@603
  1212
      delta = _cap[in_arc];
kpeter@601
  1213
      int result = 0;
kpeter@641
  1214
      Value d;
kpeter@601
  1215
      int e;
kpeter@601
  1216
kpeter@601
  1217
      // Search the cycle along the path form the first node to the root
kpeter@601
  1218
      for (int u = first; u != join; u = _parent[u]) {
kpeter@601
  1219
        e = _pred[u];
kpeter@640
  1220
        d = _forward[u] ?
kpeter@640
  1221
          _flow[e] : (_cap[e] == INF ? INF : _cap[e] - _flow[e]);
kpeter@601
  1222
        if (d < delta) {
kpeter@601
  1223
          delta = d;
kpeter@601
  1224
          u_out = u;
kpeter@601
  1225
          result = 1;
kpeter@601
  1226
        }
kpeter@601
  1227
      }
kpeter@601
  1228
      // Search the cycle along the path form the second node to the root
kpeter@601
  1229
      for (int u = second; u != join; u = _parent[u]) {
kpeter@601
  1230
        e = _pred[u];
kpeter@640
  1231
        d = _forward[u] ? 
kpeter@640
  1232
          (_cap[e] == INF ? INF : _cap[e] - _flow[e]) : _flow[e];
kpeter@601
  1233
        if (d <= delta) {
kpeter@601
  1234
          delta = d;
kpeter@601
  1235
          u_out = u;
kpeter@601
  1236
          result = 2;
kpeter@601
  1237
        }
kpeter@601
  1238
      }
kpeter@601
  1239
kpeter@601
  1240
      if (result == 1) {
kpeter@601
  1241
        u_in = first;
kpeter@601
  1242
        v_in = second;
kpeter@601
  1243
      } else {
kpeter@601
  1244
        u_in = second;
kpeter@601
  1245
        v_in = first;
kpeter@601
  1246
      }
kpeter@601
  1247
      return result != 0;
kpeter@601
  1248
    }
kpeter@601
  1249
kpeter@601
  1250
    // Change _flow and _state vectors
kpeter@601
  1251
    void changeFlow(bool change) {
kpeter@601
  1252
      // Augment along the cycle
kpeter@601
  1253
      if (delta > 0) {
kpeter@641
  1254
        Value val = _state[in_arc] * delta;
kpeter@603
  1255
        _flow[in_arc] += val;
kpeter@603
  1256
        for (int u = _source[in_arc]; u != join; u = _parent[u]) {
kpeter@601
  1257
          _flow[_pred[u]] += _forward[u] ? -val : val;
kpeter@601
  1258
        }
kpeter@603
  1259
        for (int u = _target[in_arc]; u != join; u = _parent[u]) {
kpeter@601
  1260
          _flow[_pred[u]] += _forward[u] ? val : -val;
kpeter@601
  1261
        }
kpeter@601
  1262
      }
kpeter@601
  1263
      // Update the state of the entering and leaving arcs
kpeter@601
  1264
      if (change) {
kpeter@603
  1265
        _state[in_arc] = STATE_TREE;
kpeter@601
  1266
        _state[_pred[u_out]] =
kpeter@601
  1267
          (_flow[_pred[u_out]] == 0) ? STATE_LOWER : STATE_UPPER;
kpeter@601
  1268
      } else {
kpeter@603
  1269
        _state[in_arc] = -_state[in_arc];
kpeter@601
  1270
      }
kpeter@601
  1271
    }
kpeter@601
  1272
kpeter@604
  1273
    // Update the tree structure
kpeter@604
  1274
    void updateTreeStructure() {
kpeter@604
  1275
      int u, w;
kpeter@604
  1276
      int old_rev_thread = _rev_thread[u_out];
kpeter@604
  1277
      int old_succ_num = _succ_num[u_out];
kpeter@604
  1278
      int old_last_succ = _last_succ[u_out];
kpeter@601
  1279
      v_out = _parent[u_out];
kpeter@601
  1280
kpeter@604
  1281
      u = _last_succ[u_in];  // the last successor of u_in
kpeter@604
  1282
      right = _thread[u];    // the node after it
kpeter@604
  1283
kpeter@604
  1284
      // Handle the case when old_rev_thread equals to v_in
kpeter@604
  1285
      // (it also means that join and v_out coincide)
kpeter@604
  1286
      if (old_rev_thread == v_in) {
kpeter@604
  1287
        last = _thread[_last_succ[u_out]];
kpeter@604
  1288
      } else {
kpeter@604
  1289
        last = _thread[v_in];
kpeter@601
  1290
      }
kpeter@601
  1291
kpeter@604
  1292
      // Update _thread and _parent along the stem nodes (i.e. the nodes
kpeter@604
  1293
      // between u_in and u_out, whose parent have to be changed)
kpeter@601
  1294
      _thread[v_in] = stem = u_in;
kpeter@604
  1295
      _dirty_revs.clear();
kpeter@604
  1296
      _dirty_revs.push_back(v_in);
kpeter@601
  1297
      par_stem = v_in;
kpeter@601
  1298
      while (stem != u_out) {
kpeter@604
  1299
        // Insert the next stem node into the thread list
kpeter@604
  1300
        new_stem = _parent[stem];
kpeter@604
  1301
        _thread[u] = new_stem;
kpeter@604
  1302
        _dirty_revs.push_back(u);
kpeter@601
  1303
kpeter@604
  1304
        // Remove the subtree of stem from the thread list
kpeter@604
  1305
        w = _rev_thread[stem];
kpeter@604
  1306
        _thread[w] = right;
kpeter@604
  1307
        _rev_thread[right] = w;
kpeter@601
  1308
kpeter@604
  1309
        // Change the parent node and shift stem nodes
kpeter@601
  1310
        _parent[stem] = par_stem;
kpeter@601
  1311
        par_stem = stem;
kpeter@601
  1312
        stem = new_stem;
kpeter@601
  1313
kpeter@604
  1314
        // Update u and right
kpeter@604
  1315
        u = _last_succ[stem] == _last_succ[par_stem] ?
kpeter@604
  1316
          _rev_thread[par_stem] : _last_succ[stem];
kpeter@601
  1317
        right = _thread[u];
kpeter@601
  1318
      }
kpeter@601
  1319
      _parent[u_out] = par_stem;
kpeter@601
  1320
      _thread[u] = last;
kpeter@604
  1321
      _rev_thread[last] = u;
kpeter@604
  1322
      _last_succ[u_out] = u;
kpeter@601
  1323
kpeter@604
  1324
      // Remove the subtree of u_out from the thread list except for
kpeter@604
  1325
      // the case when old_rev_thread equals to v_in
kpeter@604
  1326
      // (it also means that join and v_out coincide)
kpeter@604
  1327
      if (old_rev_thread != v_in) {
kpeter@604
  1328
        _thread[old_rev_thread] = right;
kpeter@604
  1329
        _rev_thread[right] = old_rev_thread;
kpeter@604
  1330
      }
kpeter@604
  1331
kpeter@604
  1332
      // Update _rev_thread using the new _thread values
kpeter@604
  1333
      for (int i = 0; i < int(_dirty_revs.size()); ++i) {
kpeter@604
  1334
        u = _dirty_revs[i];
kpeter@604
  1335
        _rev_thread[_thread[u]] = u;
kpeter@604
  1336
      }
kpeter@604
  1337
kpeter@604
  1338
      // Update _pred, _forward, _last_succ and _succ_num for the
kpeter@604
  1339
      // stem nodes from u_out to u_in
kpeter@604
  1340
      int tmp_sc = 0, tmp_ls = _last_succ[u_out];
kpeter@604
  1341
      u = u_out;
kpeter@604
  1342
      while (u != u_in) {
kpeter@604
  1343
        w = _parent[u];
kpeter@604
  1344
        _pred[u] = _pred[w];
kpeter@604
  1345
        _forward[u] = !_forward[w];
kpeter@604
  1346
        tmp_sc += _succ_num[u] - _succ_num[w];
kpeter@604
  1347
        _succ_num[u] = tmp_sc;
kpeter@604
  1348
        _last_succ[w] = tmp_ls;
kpeter@604
  1349
        u = w;
kpeter@604
  1350
      }
kpeter@604
  1351
      _pred[u_in] = in_arc;
kpeter@604
  1352
      _forward[u_in] = (u_in == _source[in_arc]);
kpeter@604
  1353
      _succ_num[u_in] = old_succ_num;
kpeter@604
  1354
kpeter@604
  1355
      // Set limits for updating _last_succ form v_in and v_out
kpeter@604
  1356
      // towards the root
kpeter@604
  1357
      int up_limit_in = -1;
kpeter@604
  1358
      int up_limit_out = -1;
kpeter@604
  1359
      if (_last_succ[join] == v_in) {
kpeter@604
  1360
        up_limit_out = join;
kpeter@601
  1361
      } else {
kpeter@604
  1362
        up_limit_in = join;
kpeter@604
  1363
      }
kpeter@604
  1364
kpeter@604
  1365
      // Update _last_succ from v_in towards the root
kpeter@604
  1366
      for (u = v_in; u != up_limit_in && _last_succ[u] == v_in;
kpeter@604
  1367
           u = _parent[u]) {
kpeter@604
  1368
        _last_succ[u] = _last_succ[u_out];
kpeter@604
  1369
      }
kpeter@604
  1370
      // Update _last_succ from v_out towards the root
kpeter@604
  1371
      if (join != old_rev_thread && v_in != old_rev_thread) {
kpeter@604
  1372
        for (u = v_out; u != up_limit_out && _last_succ[u] == old_last_succ;
kpeter@604
  1373
             u = _parent[u]) {
kpeter@604
  1374
          _last_succ[u] = old_rev_thread;
kpeter@604
  1375
        }
kpeter@604
  1376
      } else {
kpeter@604
  1377
        for (u = v_out; u != up_limit_out && _last_succ[u] == old_last_succ;
kpeter@604
  1378
             u = _parent[u]) {
kpeter@604
  1379
          _last_succ[u] = _last_succ[u_out];
kpeter@604
  1380
        }
kpeter@604
  1381
      }
kpeter@604
  1382
kpeter@604
  1383
      // Update _succ_num from v_in to join
kpeter@604
  1384
      for (u = v_in; u != join; u = _parent[u]) {
kpeter@604
  1385
        _succ_num[u] += old_succ_num;
kpeter@604
  1386
      }
kpeter@604
  1387
      // Update _succ_num from v_out to join
kpeter@604
  1388
      for (u = v_out; u != join; u = _parent[u]) {
kpeter@604
  1389
        _succ_num[u] -= old_succ_num;
kpeter@601
  1390
      }
kpeter@601
  1391
    }
kpeter@601
  1392
kpeter@604
  1393
    // Update potentials
kpeter@604
  1394
    void updatePotential() {
kpeter@607
  1395
      Cost sigma = _forward[u_in] ?
kpeter@601
  1396
        _pi[v_in] - _pi[u_in] - _cost[_pred[u_in]] :
kpeter@601
  1397
        _pi[v_in] - _pi[u_in] + _cost[_pred[u_in]];
kpeter@608
  1398
      // Update potentials in the subtree, which has been moved
kpeter@608
  1399
      int end = _thread[_last_succ[u_in]];
kpeter@608
  1400
      for (int u = u_in; u != end; u = _thread[u]) {
kpeter@608
  1401
        _pi[u] += sigma;
kpeter@601
  1402
      }
kpeter@601
  1403
    }
kpeter@601
  1404
kpeter@601
  1405
    // Execute the algorithm
kpeter@640
  1406
    ProblemType start(PivotRule pivot_rule) {
kpeter@601
  1407
      // Select the pivot rule implementation
kpeter@601
  1408
      switch (pivot_rule) {
kpeter@605
  1409
        case FIRST_ELIGIBLE:
kpeter@601
  1410
          return start<FirstEligiblePivotRule>();
kpeter@605
  1411
        case BEST_ELIGIBLE:
kpeter@601
  1412
          return start<BestEligiblePivotRule>();
kpeter@605
  1413
        case BLOCK_SEARCH:
kpeter@601
  1414
          return start<BlockSearchPivotRule>();
kpeter@605
  1415
        case CANDIDATE_LIST:
kpeter@601
  1416
          return start<CandidateListPivotRule>();
kpeter@605
  1417
        case ALTERING_LIST:
kpeter@601
  1418
          return start<AlteringListPivotRule>();
kpeter@601
  1419
      }
kpeter@640
  1420
      return INFEASIBLE; // avoid warning
kpeter@601
  1421
    }
kpeter@601
  1422
kpeter@605
  1423
    template <typename PivotRuleImpl>
kpeter@640
  1424
    ProblemType start() {
kpeter@605
  1425
      PivotRuleImpl pivot(*this);
kpeter@601
  1426
kpeter@605
  1427
      // Execute the Network Simplex algorithm
kpeter@601
  1428
      while (pivot.findEnteringArc()) {
kpeter@601
  1429
        findJoinNode();
kpeter@601
  1430
        bool change = findLeavingArc();
kpeter@640
  1431
        if (delta >= INF) return UNBOUNDED;
kpeter@601
  1432
        changeFlow(change);
kpeter@601
  1433
        if (change) {
kpeter@604
  1434
          updateTreeStructure();
kpeter@604
  1435
          updatePotential();
kpeter@601
  1436
        }
kpeter@601
  1437
      }
kpeter@640
  1438
      
kpeter@640
  1439
      // Check feasibility
kpeter@663
  1440
      for (int e = _search_arc_num; e != _all_arc_num; ++e) {
kpeter@663
  1441
        if (_flow[e] != 0) return INFEASIBLE;
kpeter@640
  1442
      }
kpeter@601
  1443
kpeter@642
  1444
      // Transform the solution and the supply map to the original form
kpeter@642
  1445
      if (_have_lower) {
kpeter@601
  1446
        for (int i = 0; i != _arc_num; ++i) {
kpeter@642
  1447
          Value c = _lower[i];
kpeter@642
  1448
          if (c != 0) {
kpeter@642
  1449
            _flow[i] += c;
kpeter@642
  1450
            _supply[_source[i]] += c;
kpeter@642
  1451
            _supply[_target[i]] -= c;
kpeter@642
  1452
          }
kpeter@601
  1453
        }
kpeter@601
  1454
      }
kpeter@663
  1455
      
kpeter@663
  1456
      // Shift potentials to meet the requirements of the GEQ/LEQ type
kpeter@663
  1457
      // optimality conditions
kpeter@663
  1458
      if (_sum_supply == 0) {
kpeter@663
  1459
        if (_stype == GEQ) {
kpeter@663
  1460
          Cost max_pot = std::numeric_limits<Cost>::min();
kpeter@663
  1461
          for (int i = 0; i != _node_num; ++i) {
kpeter@663
  1462
            if (_pi[i] > max_pot) max_pot = _pi[i];
kpeter@663
  1463
          }
kpeter@663
  1464
          if (max_pot > 0) {
kpeter@663
  1465
            for (int i = 0; i != _node_num; ++i)
kpeter@663
  1466
              _pi[i] -= max_pot;
kpeter@663
  1467
          }
kpeter@663
  1468
        } else {
kpeter@663
  1469
          Cost min_pot = std::numeric_limits<Cost>::max();
kpeter@663
  1470
          for (int i = 0; i != _node_num; ++i) {
kpeter@663
  1471
            if (_pi[i] < min_pot) min_pot = _pi[i];
kpeter@663
  1472
          }
kpeter@663
  1473
          if (min_pot < 0) {
kpeter@663
  1474
            for (int i = 0; i != _node_num; ++i)
kpeter@663
  1475
              _pi[i] -= min_pot;
kpeter@663
  1476
          }
kpeter@663
  1477
        }
kpeter@663
  1478
      }
kpeter@601
  1479
kpeter@640
  1480
      return OPTIMAL;
kpeter@601
  1481
    }
kpeter@601
  1482
kpeter@601
  1483
  }; //class NetworkSimplex
kpeter@601
  1484
kpeter@601
  1485
  ///@}
kpeter@601
  1486
kpeter@601
  1487
} //namespace lemon
kpeter@601
  1488
kpeter@601
  1489
#endif //LEMON_NETWORK_SIMPLEX_H