lemon/network_simplex.h
author Akos Ladanyi <ladanyi@tmit.bme.hu>
Mon, 27 Apr 2009 18:05:04 +0100
changeset 637 4faa0c763906
parent 612 0c8e5c688440
parent 613 b1811c363299
child 640 6c408d864fa1
permissions -rw-r--r--
Add GLPK_ROOT_DIR variable to FindGLPK (#277)
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@601
    22
/// \ingroup min_cost_flow
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@609
    33
#include <lemon/maps.h>
kpeter@609
    34
#include <lemon/circulation.h>
kpeter@609
    35
#include <lemon/adaptors.h>
kpeter@601
    36
kpeter@601
    37
namespace lemon {
kpeter@601
    38
kpeter@601
    39
  /// \addtogroup min_cost_flow
kpeter@601
    40
  /// @{
kpeter@601
    41
kpeter@605
    42
  /// \brief Implementation of the primal Network Simplex algorithm
kpeter@601
    43
  /// for finding a \ref min_cost_flow "minimum cost flow".
kpeter@601
    44
  ///
kpeter@605
    45
  /// \ref NetworkSimplex implements the primal Network Simplex algorithm
kpeter@601
    46
  /// for finding a \ref min_cost_flow "minimum cost flow".
kpeter@606
    47
  /// This algorithm is a specialized version of the linear programming
kpeter@606
    48
  /// simplex method directly for the minimum cost flow problem.
kpeter@606
    49
  /// It is one of the most efficient solution methods.
kpeter@606
    50
  ///
kpeter@606
    51
  /// In general this class is the fastest implementation available
kpeter@606
    52
  /// in LEMON for the minimum cost flow problem.
kpeter@609
    53
  /// Moreover it supports both direction of the supply/demand inequality
kpeter@609
    54
  /// constraints. For more information see \ref ProblemType.
kpeter@601
    55
  ///
kpeter@605
    56
  /// \tparam GR The digraph type the algorithm runs on.
kpeter@607
    57
  /// \tparam F The value type used for flow amounts, capacity bounds
kpeter@607
    58
  /// and supply values in the algorithm. By default it is \c int.
kpeter@607
    59
  /// \tparam C The value type used for costs and potentials in the
kpeter@607
    60
  /// algorithm. By default it is the same as \c F.
kpeter@601
    61
  ///
kpeter@608
    62
  /// \warning Both value types must be signed and all input data must
kpeter@608
    63
  /// be integer.
kpeter@601
    64
  ///
kpeter@605
    65
  /// \note %NetworkSimplex provides five different pivot rule
kpeter@609
    66
  /// implementations, from which the most efficient one is used
kpeter@609
    67
  /// by default. For more information see \ref PivotRule.
kpeter@607
    68
  template <typename GR, typename F = int, typename C = F>
kpeter@601
    69
  class NetworkSimplex
kpeter@601
    70
  {
kpeter@605
    71
  public:
kpeter@601
    72
kpeter@607
    73
    /// The flow type of the algorithm
kpeter@607
    74
    typedef F Flow;
kpeter@607
    75
    /// The cost type of the algorithm
kpeter@607
    76
    typedef C Cost;
kpeter@609
    77
#ifdef DOXYGEN
kpeter@609
    78
    /// The type of the flow map
kpeter@609
    79
    typedef GR::ArcMap<Flow> FlowMap;
kpeter@609
    80
    /// The type of the potential map
kpeter@609
    81
    typedef GR::NodeMap<Cost> PotentialMap;
kpeter@609
    82
#else
kpeter@605
    83
    /// The type of the flow map
kpeter@607
    84
    typedef typename GR::template ArcMap<Flow> FlowMap;
kpeter@605
    85
    /// The type of the potential map
kpeter@607
    86
    typedef typename GR::template NodeMap<Cost> PotentialMap;
kpeter@609
    87
#endif
kpeter@605
    88
kpeter@605
    89
  public:
kpeter@605
    90
kpeter@605
    91
    /// \brief Enum type for selecting the pivot rule.
kpeter@605
    92
    ///
kpeter@605
    93
    /// Enum type for selecting the pivot rule for the \ref run()
kpeter@605
    94
    /// function.
kpeter@605
    95
    ///
kpeter@605
    96
    /// \ref NetworkSimplex provides five different pivot rule
kpeter@605
    97
    /// implementations that significantly affect the running time
kpeter@605
    98
    /// of the algorithm.
kpeter@605
    99
    /// By default \ref BLOCK_SEARCH "Block Search" is used, which
kpeter@605
   100
    /// proved to be the most efficient and the most robust on various
kpeter@605
   101
    /// test inputs according to our benchmark tests.
kpeter@605
   102
    /// However another pivot rule can be selected using the \ref run()
kpeter@605
   103
    /// function with the proper parameter.
kpeter@605
   104
    enum PivotRule {
kpeter@605
   105
kpeter@605
   106
      /// The First Eligible pivot rule.
kpeter@605
   107
      /// The next eligible arc is selected in a wraparound fashion
kpeter@605
   108
      /// in every iteration.
kpeter@605
   109
      FIRST_ELIGIBLE,
kpeter@605
   110
kpeter@605
   111
      /// The Best Eligible pivot rule.
kpeter@605
   112
      /// The best eligible arc is selected in every iteration.
kpeter@605
   113
      BEST_ELIGIBLE,
kpeter@605
   114
kpeter@605
   115
      /// The Block Search pivot rule.
kpeter@605
   116
      /// A specified number of arcs are examined in every iteration
kpeter@605
   117
      /// in a wraparound fashion and the best eligible arc is selected
kpeter@605
   118
      /// from this block.
kpeter@605
   119
      BLOCK_SEARCH,
kpeter@605
   120
kpeter@605
   121
      /// The Candidate List pivot rule.
kpeter@605
   122
      /// In a major iteration a candidate list is built from eligible arcs
kpeter@605
   123
      /// in a wraparound fashion and in the following minor iterations
kpeter@605
   124
      /// the best eligible arc is selected from this list.
kpeter@605
   125
      CANDIDATE_LIST,
kpeter@605
   126
kpeter@605
   127
      /// The Altering Candidate List pivot rule.
kpeter@605
   128
      /// It is a modified version of the Candidate List method.
kpeter@605
   129
      /// It keeps only the several best eligible arcs from the former
kpeter@605
   130
      /// candidate list and extends this list in every iteration.
kpeter@605
   131
      ALTERING_LIST
kpeter@605
   132
    };
kpeter@609
   133
    
kpeter@609
   134
    /// \brief Enum type for selecting the problem type.
kpeter@609
   135
    ///
kpeter@609
   136
    /// Enum type for selecting the problem type, i.e. the direction of
kpeter@609
   137
    /// the inequalities in the supply/demand constraints of the
kpeter@609
   138
    /// \ref min_cost_flow "minimum cost flow problem".
kpeter@609
   139
    ///
kpeter@609
   140
    /// The default problem type is \c GEQ, since this form is supported
kpeter@609
   141
    /// by other minimum cost flow algorithms and the \ref Circulation
kpeter@609
   142
    /// algorithm as well.
kpeter@609
   143
    /// The \c LEQ problem type can be selected using the \ref problemType()
kpeter@609
   144
    /// function.
kpeter@609
   145
    ///
kpeter@609
   146
    /// Note that the equality form is a special case of both problem type.
kpeter@609
   147
    enum ProblemType {
kpeter@609
   148
kpeter@609
   149
      /// This option means that there are "<em>greater or equal</em>"
kpeter@609
   150
      /// constraints in the defintion, i.e. the exact formulation of the
kpeter@609
   151
      /// problem is the following.
kpeter@609
   152
      /**
kpeter@609
   153
          \f[ \min\sum_{uv\in A} f(uv) \cdot cost(uv) \f]
kpeter@609
   154
          \f[ \sum_{uv\in A} f(uv) - \sum_{vu\in A} f(vu) \geq
kpeter@609
   155
              sup(u) \quad \forall u\in V \f]
kpeter@609
   156
          \f[ lower(uv) \leq f(uv) \leq upper(uv) \quad \forall uv\in A \f]
kpeter@609
   157
      */
kpeter@609
   158
      /// It means that the total demand must be greater or equal to the 
kpeter@609
   159
      /// total supply (i.e. \f$\sum_{u\in V} sup(u)\f$ must be zero or
kpeter@609
   160
      /// negative) and all the supplies have to be carried out from 
kpeter@609
   161
      /// the supply nodes, but there could be demands that are not 
kpeter@609
   162
      /// satisfied.
kpeter@609
   163
      GEQ,
kpeter@609
   164
      /// It is just an alias for the \c GEQ option.
kpeter@609
   165
      CARRY_SUPPLIES = GEQ,
kpeter@609
   166
kpeter@609
   167
      /// This option means that there are "<em>less or equal</em>"
kpeter@609
   168
      /// constraints in the defintion, i.e. the exact formulation of the
kpeter@609
   169
      /// problem is the following.
kpeter@609
   170
      /**
kpeter@609
   171
          \f[ \min\sum_{uv\in A} f(uv) \cdot cost(uv) \f]
kpeter@609
   172
          \f[ \sum_{uv\in A} f(uv) - \sum_{vu\in A} f(vu) \leq
kpeter@609
   173
              sup(u) \quad \forall u\in V \f]
kpeter@609
   174
          \f[ lower(uv) \leq f(uv) \leq upper(uv) \quad \forall uv\in A \f]
kpeter@609
   175
      */
kpeter@609
   176
      /// It means that the total demand must be less or equal to the 
kpeter@609
   177
      /// total supply (i.e. \f$\sum_{u\in V} sup(u)\f$ must be zero or
kpeter@609
   178
      /// positive) and all the demands have to be satisfied, but there
kpeter@609
   179
      /// could be supplies that are not carried out from the supply
kpeter@609
   180
      /// nodes.
kpeter@609
   181
      LEQ,
kpeter@609
   182
      /// It is just an alias for the \c LEQ option.
kpeter@609
   183
      SATISFY_DEMANDS = LEQ
kpeter@609
   184
    };
kpeter@605
   185
kpeter@605
   186
  private:
kpeter@605
   187
kpeter@605
   188
    TEMPLATE_DIGRAPH_TYPEDEFS(GR);
kpeter@605
   189
kpeter@607
   190
    typedef typename GR::template ArcMap<Flow> FlowArcMap;
kpeter@607
   191
    typedef typename GR::template ArcMap<Cost> CostArcMap;
kpeter@607
   192
    typedef typename GR::template NodeMap<Flow> FlowNodeMap;
kpeter@601
   193
kpeter@601
   194
    typedef std::vector<Arc> ArcVector;
kpeter@601
   195
    typedef std::vector<Node> NodeVector;
kpeter@601
   196
    typedef std::vector<int> IntVector;
kpeter@601
   197
    typedef std::vector<bool> BoolVector;
kpeter@607
   198
    typedef std::vector<Flow> FlowVector;
kpeter@607
   199
    typedef std::vector<Cost> CostVector;
kpeter@601
   200
kpeter@601
   201
    // State constants for arcs
kpeter@601
   202
    enum ArcStateEnum {
kpeter@601
   203
      STATE_UPPER = -1,
kpeter@601
   204
      STATE_TREE  =  0,
kpeter@601
   205
      STATE_LOWER =  1
kpeter@601
   206
    };
kpeter@601
   207
kpeter@601
   208
  private:
kpeter@601
   209
kpeter@605
   210
    // Data related to the underlying digraph
kpeter@605
   211
    const GR &_graph;
kpeter@605
   212
    int _node_num;
kpeter@605
   213
    int _arc_num;
kpeter@605
   214
kpeter@605
   215
    // Parameters of the problem
kpeter@607
   216
    FlowArcMap *_plower;
kpeter@607
   217
    FlowArcMap *_pupper;
kpeter@607
   218
    CostArcMap *_pcost;
kpeter@607
   219
    FlowNodeMap *_psupply;
kpeter@605
   220
    bool _pstsup;
kpeter@605
   221
    Node _psource, _ptarget;
kpeter@607
   222
    Flow _pstflow;
kpeter@609
   223
    ProblemType _ptype;
kpeter@601
   224
kpeter@601
   225
    // Result maps
kpeter@603
   226
    FlowMap *_flow_map;
kpeter@603
   227
    PotentialMap *_potential_map;
kpeter@601
   228
    bool _local_flow;
kpeter@601
   229
    bool _local_potential;
kpeter@601
   230
kpeter@605
   231
    // Data structures for storing the digraph
kpeter@603
   232
    IntNodeMap _node_id;
kpeter@603
   233
    ArcVector _arc_ref;
kpeter@603
   234
    IntVector _source;
kpeter@603
   235
    IntVector _target;
kpeter@603
   236
kpeter@605
   237
    // Node and arc data
kpeter@607
   238
    FlowVector _cap;
kpeter@607
   239
    CostVector _cost;
kpeter@607
   240
    FlowVector _supply;
kpeter@607
   241
    FlowVector _flow;
kpeter@607
   242
    CostVector _pi;
kpeter@601
   243
kpeter@603
   244
    // Data for storing the spanning tree structure
kpeter@601
   245
    IntVector _parent;
kpeter@601
   246
    IntVector _pred;
kpeter@601
   247
    IntVector _thread;
kpeter@604
   248
    IntVector _rev_thread;
kpeter@604
   249
    IntVector _succ_num;
kpeter@604
   250
    IntVector _last_succ;
kpeter@604
   251
    IntVector _dirty_revs;
kpeter@601
   252
    BoolVector _forward;
kpeter@601
   253
    IntVector _state;
kpeter@601
   254
    int _root;
kpeter@601
   255
kpeter@601
   256
    // Temporary data used in the current pivot iteration
kpeter@603
   257
    int in_arc, join, u_in, v_in, u_out, v_out;
kpeter@603
   258
    int first, second, right, last;
kpeter@601
   259
    int stem, par_stem, new_stem;
kpeter@607
   260
    Flow delta;
kpeter@601
   261
kpeter@601
   262
  private:
kpeter@601
   263
kpeter@605
   264
    // Implementation of the First Eligible pivot rule
kpeter@601
   265
    class FirstEligiblePivotRule
kpeter@601
   266
    {
kpeter@601
   267
    private:
kpeter@601
   268
kpeter@601
   269
      // References to the NetworkSimplex class
kpeter@601
   270
      const IntVector  &_source;
kpeter@601
   271
      const IntVector  &_target;
kpeter@607
   272
      const CostVector &_cost;
kpeter@601
   273
      const IntVector  &_state;
kpeter@607
   274
      const CostVector &_pi;
kpeter@601
   275
      int &_in_arc;
kpeter@601
   276
      int _arc_num;
kpeter@601
   277
kpeter@601
   278
      // Pivot rule data
kpeter@601
   279
      int _next_arc;
kpeter@601
   280
kpeter@601
   281
    public:
kpeter@601
   282
kpeter@605
   283
      // Constructor
kpeter@601
   284
      FirstEligiblePivotRule(NetworkSimplex &ns) :
kpeter@603
   285
        _source(ns._source), _target(ns._target),
kpeter@601
   286
        _cost(ns._cost), _state(ns._state), _pi(ns._pi),
kpeter@603
   287
        _in_arc(ns.in_arc), _arc_num(ns._arc_num), _next_arc(0)
kpeter@601
   288
      {}
kpeter@601
   289
kpeter@605
   290
      // Find next entering arc
kpeter@601
   291
      bool findEnteringArc() {
kpeter@607
   292
        Cost c;
kpeter@601
   293
        for (int e = _next_arc; e < _arc_num; ++e) {
kpeter@601
   294
          c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@601
   295
          if (c < 0) {
kpeter@601
   296
            _in_arc = e;
kpeter@601
   297
            _next_arc = e + 1;
kpeter@601
   298
            return true;
kpeter@601
   299
          }
kpeter@601
   300
        }
kpeter@601
   301
        for (int e = 0; e < _next_arc; ++e) {
kpeter@601
   302
          c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@601
   303
          if (c < 0) {
kpeter@601
   304
            _in_arc = e;
kpeter@601
   305
            _next_arc = e + 1;
kpeter@601
   306
            return true;
kpeter@601
   307
          }
kpeter@601
   308
        }
kpeter@601
   309
        return false;
kpeter@601
   310
      }
kpeter@601
   311
kpeter@601
   312
    }; //class FirstEligiblePivotRule
kpeter@601
   313
kpeter@601
   314
kpeter@605
   315
    // Implementation of the Best Eligible pivot rule
kpeter@601
   316
    class BestEligiblePivotRule
kpeter@601
   317
    {
kpeter@601
   318
    private:
kpeter@601
   319
kpeter@601
   320
      // References to the NetworkSimplex class
kpeter@601
   321
      const IntVector  &_source;
kpeter@601
   322
      const IntVector  &_target;
kpeter@607
   323
      const CostVector &_cost;
kpeter@601
   324
      const IntVector  &_state;
kpeter@607
   325
      const CostVector &_pi;
kpeter@601
   326
      int &_in_arc;
kpeter@601
   327
      int _arc_num;
kpeter@601
   328
kpeter@601
   329
    public:
kpeter@601
   330
kpeter@605
   331
      // Constructor
kpeter@601
   332
      BestEligiblePivotRule(NetworkSimplex &ns) :
kpeter@603
   333
        _source(ns._source), _target(ns._target),
kpeter@601
   334
        _cost(ns._cost), _state(ns._state), _pi(ns._pi),
kpeter@603
   335
        _in_arc(ns.in_arc), _arc_num(ns._arc_num)
kpeter@601
   336
      {}
kpeter@601
   337
kpeter@605
   338
      // Find next entering arc
kpeter@601
   339
      bool findEnteringArc() {
kpeter@607
   340
        Cost c, min = 0;
kpeter@601
   341
        for (int e = 0; e < _arc_num; ++e) {
kpeter@601
   342
          c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@601
   343
          if (c < min) {
kpeter@601
   344
            min = c;
kpeter@601
   345
            _in_arc = e;
kpeter@601
   346
          }
kpeter@601
   347
        }
kpeter@601
   348
        return min < 0;
kpeter@601
   349
      }
kpeter@601
   350
kpeter@601
   351
    }; //class BestEligiblePivotRule
kpeter@601
   352
kpeter@601
   353
kpeter@605
   354
    // Implementation of the Block Search pivot rule
kpeter@601
   355
    class BlockSearchPivotRule
kpeter@601
   356
    {
kpeter@601
   357
    private:
kpeter@601
   358
kpeter@601
   359
      // References to the NetworkSimplex class
kpeter@601
   360
      const IntVector  &_source;
kpeter@601
   361
      const IntVector  &_target;
kpeter@607
   362
      const CostVector &_cost;
kpeter@601
   363
      const IntVector  &_state;
kpeter@607
   364
      const CostVector &_pi;
kpeter@601
   365
      int &_in_arc;
kpeter@601
   366
      int _arc_num;
kpeter@601
   367
kpeter@601
   368
      // Pivot rule data
kpeter@601
   369
      int _block_size;
kpeter@601
   370
      int _next_arc;
kpeter@601
   371
kpeter@601
   372
    public:
kpeter@601
   373
kpeter@605
   374
      // Constructor
kpeter@601
   375
      BlockSearchPivotRule(NetworkSimplex &ns) :
kpeter@603
   376
        _source(ns._source), _target(ns._target),
kpeter@601
   377
        _cost(ns._cost), _state(ns._state), _pi(ns._pi),
kpeter@603
   378
        _in_arc(ns.in_arc), _arc_num(ns._arc_num), _next_arc(0)
kpeter@601
   379
      {
kpeter@601
   380
        // The main parameters of the pivot rule
kpeter@601
   381
        const double BLOCK_SIZE_FACTOR = 2.0;
kpeter@601
   382
        const int MIN_BLOCK_SIZE = 10;
kpeter@601
   383
alpar@612
   384
        _block_size = std::max( int(BLOCK_SIZE_FACTOR *
alpar@612
   385
                                    std::sqrt(double(_arc_num))),
kpeter@601
   386
                                MIN_BLOCK_SIZE );
kpeter@601
   387
      }
kpeter@601
   388
kpeter@605
   389
      // Find next entering arc
kpeter@601
   390
      bool findEnteringArc() {
kpeter@607
   391
        Cost c, min = 0;
kpeter@601
   392
        int cnt = _block_size;
kpeter@601
   393
        int e, min_arc = _next_arc;
kpeter@601
   394
        for (e = _next_arc; e < _arc_num; ++e) {
kpeter@601
   395
          c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@601
   396
          if (c < min) {
kpeter@601
   397
            min = c;
kpeter@601
   398
            min_arc = e;
kpeter@601
   399
          }
kpeter@601
   400
          if (--cnt == 0) {
kpeter@601
   401
            if (min < 0) break;
kpeter@601
   402
            cnt = _block_size;
kpeter@601
   403
          }
kpeter@601
   404
        }
kpeter@601
   405
        if (min == 0 || cnt > 0) {
kpeter@601
   406
          for (e = 0; e < _next_arc; ++e) {
kpeter@601
   407
            c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@601
   408
            if (c < min) {
kpeter@601
   409
              min = c;
kpeter@601
   410
              min_arc = e;
kpeter@601
   411
            }
kpeter@601
   412
            if (--cnt == 0) {
kpeter@601
   413
              if (min < 0) break;
kpeter@601
   414
              cnt = _block_size;
kpeter@601
   415
            }
kpeter@601
   416
          }
kpeter@601
   417
        }
kpeter@601
   418
        if (min >= 0) return false;
kpeter@601
   419
        _in_arc = min_arc;
kpeter@601
   420
        _next_arc = e;
kpeter@601
   421
        return true;
kpeter@601
   422
      }
kpeter@601
   423
kpeter@601
   424
    }; //class BlockSearchPivotRule
kpeter@601
   425
kpeter@601
   426
kpeter@605
   427
    // Implementation of the Candidate List pivot rule
kpeter@601
   428
    class CandidateListPivotRule
kpeter@601
   429
    {
kpeter@601
   430
    private:
kpeter@601
   431
kpeter@601
   432
      // References to the NetworkSimplex class
kpeter@601
   433
      const IntVector  &_source;
kpeter@601
   434
      const IntVector  &_target;
kpeter@607
   435
      const CostVector &_cost;
kpeter@601
   436
      const IntVector  &_state;
kpeter@607
   437
      const CostVector &_pi;
kpeter@601
   438
      int &_in_arc;
kpeter@601
   439
      int _arc_num;
kpeter@601
   440
kpeter@601
   441
      // Pivot rule data
kpeter@601
   442
      IntVector _candidates;
kpeter@601
   443
      int _list_length, _minor_limit;
kpeter@601
   444
      int _curr_length, _minor_count;
kpeter@601
   445
      int _next_arc;
kpeter@601
   446
kpeter@601
   447
    public:
kpeter@601
   448
kpeter@601
   449
      /// Constructor
kpeter@601
   450
      CandidateListPivotRule(NetworkSimplex &ns) :
kpeter@603
   451
        _source(ns._source), _target(ns._target),
kpeter@601
   452
        _cost(ns._cost), _state(ns._state), _pi(ns._pi),
kpeter@603
   453
        _in_arc(ns.in_arc), _arc_num(ns._arc_num), _next_arc(0)
kpeter@601
   454
      {
kpeter@601
   455
        // The main parameters of the pivot rule
kpeter@601
   456
        const double LIST_LENGTH_FACTOR = 1.0;
kpeter@601
   457
        const int MIN_LIST_LENGTH = 10;
kpeter@601
   458
        const double MINOR_LIMIT_FACTOR = 0.1;
kpeter@601
   459
        const int MIN_MINOR_LIMIT = 3;
kpeter@601
   460
alpar@612
   461
        _list_length = std::max( int(LIST_LENGTH_FACTOR *
alpar@612
   462
                                     std::sqrt(double(_arc_num))),
kpeter@601
   463
                                 MIN_LIST_LENGTH );
kpeter@601
   464
        _minor_limit = std::max( int(MINOR_LIMIT_FACTOR * _list_length),
kpeter@601
   465
                                 MIN_MINOR_LIMIT );
kpeter@601
   466
        _curr_length = _minor_count = 0;
kpeter@601
   467
        _candidates.resize(_list_length);
kpeter@601
   468
      }
kpeter@601
   469
kpeter@601
   470
      /// Find next entering arc
kpeter@601
   471
      bool findEnteringArc() {
kpeter@607
   472
        Cost min, c;
kpeter@601
   473
        int e, min_arc = _next_arc;
kpeter@601
   474
        if (_curr_length > 0 && _minor_count < _minor_limit) {
kpeter@601
   475
          // Minor iteration: select the best eligible arc from the
kpeter@601
   476
          // current candidate list
kpeter@601
   477
          ++_minor_count;
kpeter@601
   478
          min = 0;
kpeter@601
   479
          for (int i = 0; i < _curr_length; ++i) {
kpeter@601
   480
            e = _candidates[i];
kpeter@601
   481
            c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@601
   482
            if (c < min) {
kpeter@601
   483
              min = c;
kpeter@601
   484
              min_arc = e;
kpeter@601
   485
            }
kpeter@601
   486
            if (c >= 0) {
kpeter@601
   487
              _candidates[i--] = _candidates[--_curr_length];
kpeter@601
   488
            }
kpeter@601
   489
          }
kpeter@601
   490
          if (min < 0) {
kpeter@601
   491
            _in_arc = min_arc;
kpeter@601
   492
            return true;
kpeter@601
   493
          }
kpeter@601
   494
        }
kpeter@601
   495
kpeter@601
   496
        // Major iteration: build a new candidate list
kpeter@601
   497
        min = 0;
kpeter@601
   498
        _curr_length = 0;
kpeter@601
   499
        for (e = _next_arc; e < _arc_num; ++e) {
kpeter@601
   500
          c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@601
   501
          if (c < 0) {
kpeter@601
   502
            _candidates[_curr_length++] = e;
kpeter@601
   503
            if (c < min) {
kpeter@601
   504
              min = c;
kpeter@601
   505
              min_arc = e;
kpeter@601
   506
            }
kpeter@601
   507
            if (_curr_length == _list_length) break;
kpeter@601
   508
          }
kpeter@601
   509
        }
kpeter@601
   510
        if (_curr_length < _list_length) {
kpeter@601
   511
          for (e = 0; e < _next_arc; ++e) {
kpeter@601
   512
            c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@601
   513
            if (c < 0) {
kpeter@601
   514
              _candidates[_curr_length++] = e;
kpeter@601
   515
              if (c < min) {
kpeter@601
   516
                min = c;
kpeter@601
   517
                min_arc = e;
kpeter@601
   518
              }
kpeter@601
   519
              if (_curr_length == _list_length) break;
kpeter@601
   520
            }
kpeter@601
   521
          }
kpeter@601
   522
        }
kpeter@601
   523
        if (_curr_length == 0) return false;
kpeter@601
   524
        _minor_count = 1;
kpeter@601
   525
        _in_arc = min_arc;
kpeter@601
   526
        _next_arc = e;
kpeter@601
   527
        return true;
kpeter@601
   528
      }
kpeter@601
   529
kpeter@601
   530
    }; //class CandidateListPivotRule
kpeter@601
   531
kpeter@601
   532
kpeter@605
   533
    // Implementation of the Altering Candidate List pivot rule
kpeter@601
   534
    class AlteringListPivotRule
kpeter@601
   535
    {
kpeter@601
   536
    private:
kpeter@601
   537
kpeter@601
   538
      // References to the NetworkSimplex class
kpeter@601
   539
      const IntVector  &_source;
kpeter@601
   540
      const IntVector  &_target;
kpeter@607
   541
      const CostVector &_cost;
kpeter@601
   542
      const IntVector  &_state;
kpeter@607
   543
      const CostVector &_pi;
kpeter@601
   544
      int &_in_arc;
kpeter@601
   545
      int _arc_num;
kpeter@601
   546
kpeter@601
   547
      // Pivot rule data
kpeter@601
   548
      int _block_size, _head_length, _curr_length;
kpeter@601
   549
      int _next_arc;
kpeter@601
   550
      IntVector _candidates;
kpeter@607
   551
      CostVector _cand_cost;
kpeter@601
   552
kpeter@601
   553
      // Functor class to compare arcs during sort of the candidate list
kpeter@601
   554
      class SortFunc
kpeter@601
   555
      {
kpeter@601
   556
      private:
kpeter@607
   557
        const CostVector &_map;
kpeter@601
   558
      public:
kpeter@607
   559
        SortFunc(const CostVector &map) : _map(map) {}
kpeter@601
   560
        bool operator()(int left, int right) {
kpeter@601
   561
          return _map[left] > _map[right];
kpeter@601
   562
        }
kpeter@601
   563
      };
kpeter@601
   564
kpeter@601
   565
      SortFunc _sort_func;
kpeter@601
   566
kpeter@601
   567
    public:
kpeter@601
   568
kpeter@605
   569
      // Constructor
kpeter@601
   570
      AlteringListPivotRule(NetworkSimplex &ns) :
kpeter@603
   571
        _source(ns._source), _target(ns._target),
kpeter@601
   572
        _cost(ns._cost), _state(ns._state), _pi(ns._pi),
kpeter@603
   573
        _in_arc(ns.in_arc), _arc_num(ns._arc_num),
kpeter@601
   574
        _next_arc(0), _cand_cost(ns._arc_num), _sort_func(_cand_cost)
kpeter@601
   575
      {
kpeter@601
   576
        // The main parameters of the pivot rule
kpeter@601
   577
        const double BLOCK_SIZE_FACTOR = 1.5;
kpeter@601
   578
        const int MIN_BLOCK_SIZE = 10;
kpeter@601
   579
        const double HEAD_LENGTH_FACTOR = 0.1;
kpeter@601
   580
        const int MIN_HEAD_LENGTH = 3;
kpeter@601
   581
alpar@612
   582
        _block_size = std::max( int(BLOCK_SIZE_FACTOR *
alpar@612
   583
                                    std::sqrt(double(_arc_num))),
kpeter@601
   584
                                MIN_BLOCK_SIZE );
kpeter@601
   585
        _head_length = std::max( int(HEAD_LENGTH_FACTOR * _block_size),
kpeter@601
   586
                                 MIN_HEAD_LENGTH );
kpeter@601
   587
        _candidates.resize(_head_length + _block_size);
kpeter@601
   588
        _curr_length = 0;
kpeter@601
   589
      }
kpeter@601
   590
kpeter@605
   591
      // Find next entering arc
kpeter@601
   592
      bool findEnteringArc() {
kpeter@601
   593
        // Check the current candidate list
kpeter@601
   594
        int e;
kpeter@601
   595
        for (int i = 0; i < _curr_length; ++i) {
kpeter@601
   596
          e = _candidates[i];
kpeter@601
   597
          _cand_cost[e] = _state[e] *
kpeter@601
   598
            (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@601
   599
          if (_cand_cost[e] >= 0) {
kpeter@601
   600
            _candidates[i--] = _candidates[--_curr_length];
kpeter@601
   601
          }
kpeter@601
   602
        }
kpeter@601
   603
kpeter@601
   604
        // Extend the list
kpeter@601
   605
        int cnt = _block_size;
kpeter@603
   606
        int last_arc = 0;
kpeter@601
   607
        int limit = _head_length;
kpeter@601
   608
kpeter@601
   609
        for (int e = _next_arc; e < _arc_num; ++e) {
kpeter@601
   610
          _cand_cost[e] = _state[e] *
kpeter@601
   611
            (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@601
   612
          if (_cand_cost[e] < 0) {
kpeter@601
   613
            _candidates[_curr_length++] = e;
kpeter@603
   614
            last_arc = e;
kpeter@601
   615
          }
kpeter@601
   616
          if (--cnt == 0) {
kpeter@601
   617
            if (_curr_length > limit) break;
kpeter@601
   618
            limit = 0;
kpeter@601
   619
            cnt = _block_size;
kpeter@601
   620
          }
kpeter@601
   621
        }
kpeter@601
   622
        if (_curr_length <= limit) {
kpeter@601
   623
          for (int e = 0; e < _next_arc; ++e) {
kpeter@601
   624
            _cand_cost[e] = _state[e] *
kpeter@601
   625
              (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@601
   626
            if (_cand_cost[e] < 0) {
kpeter@601
   627
              _candidates[_curr_length++] = e;
kpeter@603
   628
              last_arc = e;
kpeter@601
   629
            }
kpeter@601
   630
            if (--cnt == 0) {
kpeter@601
   631
              if (_curr_length > limit) break;
kpeter@601
   632
              limit = 0;
kpeter@601
   633
              cnt = _block_size;
kpeter@601
   634
            }
kpeter@601
   635
          }
kpeter@601
   636
        }
kpeter@601
   637
        if (_curr_length == 0) return false;
kpeter@603
   638
        _next_arc = last_arc + 1;
kpeter@601
   639
kpeter@601
   640
        // Make heap of the candidate list (approximating a partial sort)
kpeter@601
   641
        make_heap( _candidates.begin(), _candidates.begin() + _curr_length,
kpeter@601
   642
                   _sort_func );
kpeter@601
   643
kpeter@601
   644
        // Pop the first element of the heap
kpeter@601
   645
        _in_arc = _candidates[0];
kpeter@601
   646
        pop_heap( _candidates.begin(), _candidates.begin() + _curr_length,
kpeter@601
   647
                  _sort_func );
kpeter@601
   648
        _curr_length = std::min(_head_length, _curr_length - 1);
kpeter@601
   649
        return true;
kpeter@601
   650
      }
kpeter@601
   651
kpeter@601
   652
    }; //class AlteringListPivotRule
kpeter@601
   653
kpeter@601
   654
  public:
kpeter@601
   655
kpeter@605
   656
    /// \brief Constructor.
kpeter@601
   657
    ///
kpeter@609
   658
    /// The constructor of the class.
kpeter@601
   659
    ///
kpeter@603
   660
    /// \param graph The digraph the algorithm runs on.
kpeter@605
   661
    NetworkSimplex(const GR& graph) :
kpeter@605
   662
      _graph(graph),
kpeter@605
   663
      _plower(NULL), _pupper(NULL), _pcost(NULL),
kpeter@609
   664
      _psupply(NULL), _pstsup(false), _ptype(GEQ),
kpeter@603
   665
      _flow_map(NULL), _potential_map(NULL),
kpeter@601
   666
      _local_flow(false), _local_potential(false),
kpeter@603
   667
      _node_id(graph)
kpeter@605
   668
    {
kpeter@607
   669
      LEMON_ASSERT(std::numeric_limits<Flow>::is_integer &&
kpeter@607
   670
                   std::numeric_limits<Flow>::is_signed,
kpeter@607
   671
        "The flow type of NetworkSimplex must be signed integer");
kpeter@607
   672
      LEMON_ASSERT(std::numeric_limits<Cost>::is_integer &&
kpeter@607
   673
                   std::numeric_limits<Cost>::is_signed,
kpeter@607
   674
        "The cost type of NetworkSimplex must be signed integer");
kpeter@605
   675
    }
kpeter@601
   676
kpeter@601
   677
    /// Destructor.
kpeter@601
   678
    ~NetworkSimplex() {
kpeter@603
   679
      if (_local_flow) delete _flow_map;
kpeter@603
   680
      if (_local_potential) delete _potential_map;
kpeter@601
   681
    }
kpeter@601
   682
kpeter@609
   683
    /// \name Parameters
kpeter@609
   684
    /// The parameters of the algorithm can be specified using these
kpeter@609
   685
    /// functions.
kpeter@609
   686
kpeter@609
   687
    /// @{
kpeter@609
   688
kpeter@605
   689
    /// \brief Set the lower bounds on the arcs.
kpeter@605
   690
    ///
kpeter@605
   691
    /// This function sets the lower bounds on the arcs.
kpeter@605
   692
    /// If neither this function nor \ref boundMaps() is used before
kpeter@605
   693
    /// calling \ref run(), the lower bounds will be set to zero
kpeter@605
   694
    /// on all arcs.
kpeter@605
   695
    ///
kpeter@605
   696
    /// \param map An arc map storing the lower bounds.
kpeter@607
   697
    /// Its \c Value type must be convertible to the \c Flow type
kpeter@605
   698
    /// of the algorithm.
kpeter@605
   699
    ///
kpeter@605
   700
    /// \return <tt>(*this)</tt>
kpeter@605
   701
    template <typename LOWER>
kpeter@605
   702
    NetworkSimplex& lowerMap(const LOWER& map) {
kpeter@605
   703
      delete _plower;
kpeter@607
   704
      _plower = new FlowArcMap(_graph);
kpeter@605
   705
      for (ArcIt a(_graph); a != INVALID; ++a) {
kpeter@605
   706
        (*_plower)[a] = map[a];
kpeter@605
   707
      }
kpeter@605
   708
      return *this;
kpeter@605
   709
    }
kpeter@605
   710
kpeter@605
   711
    /// \brief Set the upper bounds (capacities) on the arcs.
kpeter@605
   712
    ///
kpeter@605
   713
    /// This function sets the upper bounds (capacities) on the arcs.
kpeter@605
   714
    /// If none of the functions \ref upperMap(), \ref capacityMap()
kpeter@605
   715
    /// and \ref boundMaps() is used before calling \ref run(),
kpeter@605
   716
    /// the upper bounds (capacities) will be set to
kpeter@607
   717
    /// \c std::numeric_limits<Flow>::max() on all arcs.
kpeter@605
   718
    ///
kpeter@605
   719
    /// \param map An arc map storing the upper bounds.
kpeter@607
   720
    /// Its \c Value type must be convertible to the \c Flow type
kpeter@605
   721
    /// of the algorithm.
kpeter@605
   722
    ///
kpeter@605
   723
    /// \return <tt>(*this)</tt>
kpeter@605
   724
    template<typename UPPER>
kpeter@605
   725
    NetworkSimplex& upperMap(const UPPER& map) {
kpeter@605
   726
      delete _pupper;
kpeter@607
   727
      _pupper = new FlowArcMap(_graph);
kpeter@605
   728
      for (ArcIt a(_graph); a != INVALID; ++a) {
kpeter@605
   729
        (*_pupper)[a] = map[a];
kpeter@605
   730
      }
kpeter@605
   731
      return *this;
kpeter@605
   732
    }
kpeter@605
   733
kpeter@605
   734
    /// \brief Set the upper bounds (capacities) on the arcs.
kpeter@605
   735
    ///
kpeter@605
   736
    /// This function sets the upper bounds (capacities) on the arcs.
kpeter@605
   737
    /// It is just an alias for \ref upperMap().
kpeter@605
   738
    ///
kpeter@605
   739
    /// \return <tt>(*this)</tt>
kpeter@605
   740
    template<typename CAP>
kpeter@605
   741
    NetworkSimplex& capacityMap(const CAP& map) {
kpeter@605
   742
      return upperMap(map);
kpeter@605
   743
    }
kpeter@605
   744
kpeter@605
   745
    /// \brief Set the lower and upper bounds on the arcs.
kpeter@605
   746
    ///
kpeter@605
   747
    /// This function sets the lower and upper bounds on the arcs.
kpeter@605
   748
    /// If neither this function nor \ref lowerMap() is used before
kpeter@605
   749
    /// calling \ref run(), the lower bounds will be set to zero
kpeter@605
   750
    /// on all arcs.
kpeter@605
   751
    /// If none of the functions \ref upperMap(), \ref capacityMap()
kpeter@605
   752
    /// and \ref boundMaps() is used before calling \ref run(),
kpeter@605
   753
    /// the upper bounds (capacities) will be set to
kpeter@607
   754
    /// \c std::numeric_limits<Flow>::max() on all arcs.
kpeter@605
   755
    ///
kpeter@605
   756
    /// \param lower An arc map storing the lower bounds.
kpeter@605
   757
    /// \param upper An arc map storing the upper bounds.
kpeter@605
   758
    ///
kpeter@605
   759
    /// The \c Value type of the maps must be convertible to the
kpeter@607
   760
    /// \c Flow type of the algorithm.
kpeter@605
   761
    ///
kpeter@605
   762
    /// \note This function is just a shortcut of calling \ref lowerMap()
kpeter@605
   763
    /// and \ref upperMap() separately.
kpeter@605
   764
    ///
kpeter@605
   765
    /// \return <tt>(*this)</tt>
kpeter@605
   766
    template <typename LOWER, typename UPPER>
kpeter@605
   767
    NetworkSimplex& boundMaps(const LOWER& lower, const UPPER& upper) {
kpeter@605
   768
      return lowerMap(lower).upperMap(upper);
kpeter@605
   769
    }
kpeter@605
   770
kpeter@605
   771
    /// \brief Set the costs of the arcs.
kpeter@605
   772
    ///
kpeter@605
   773
    /// This function sets the costs of the arcs.
kpeter@605
   774
    /// If it is not used before calling \ref run(), the costs
kpeter@605
   775
    /// will be set to \c 1 on all arcs.
kpeter@605
   776
    ///
kpeter@605
   777
    /// \param map An arc map storing the costs.
kpeter@607
   778
    /// Its \c Value type must be convertible to the \c Cost type
kpeter@605
   779
    /// of the algorithm.
kpeter@605
   780
    ///
kpeter@605
   781
    /// \return <tt>(*this)</tt>
kpeter@605
   782
    template<typename COST>
kpeter@605
   783
    NetworkSimplex& costMap(const COST& map) {
kpeter@605
   784
      delete _pcost;
kpeter@607
   785
      _pcost = new CostArcMap(_graph);
kpeter@605
   786
      for (ArcIt a(_graph); a != INVALID; ++a) {
kpeter@605
   787
        (*_pcost)[a] = map[a];
kpeter@605
   788
      }
kpeter@605
   789
      return *this;
kpeter@605
   790
    }
kpeter@605
   791
kpeter@605
   792
    /// \brief Set the supply values of the nodes.
kpeter@605
   793
    ///
kpeter@605
   794
    /// This function sets the supply values of the nodes.
kpeter@605
   795
    /// If neither this function nor \ref stSupply() is used before
kpeter@605
   796
    /// calling \ref run(), the supply of each node will be set to zero.
kpeter@605
   797
    /// (It makes sense only if non-zero lower bounds are given.)
kpeter@605
   798
    ///
kpeter@605
   799
    /// \param map A node map storing the supply values.
kpeter@607
   800
    /// Its \c Value type must be convertible to the \c Flow type
kpeter@605
   801
    /// of the algorithm.
kpeter@605
   802
    ///
kpeter@605
   803
    /// \return <tt>(*this)</tt>
kpeter@605
   804
    template<typename SUP>
kpeter@605
   805
    NetworkSimplex& supplyMap(const SUP& map) {
kpeter@605
   806
      delete _psupply;
kpeter@605
   807
      _pstsup = false;
kpeter@607
   808
      _psupply = new FlowNodeMap(_graph);
kpeter@605
   809
      for (NodeIt n(_graph); n != INVALID; ++n) {
kpeter@605
   810
        (*_psupply)[n] = map[n];
kpeter@605
   811
      }
kpeter@605
   812
      return *this;
kpeter@605
   813
    }
kpeter@605
   814
kpeter@605
   815
    /// \brief Set single source and target nodes and a supply value.
kpeter@605
   816
    ///
kpeter@605
   817
    /// This function sets a single source node and a single target node
kpeter@605
   818
    /// and the required flow value.
kpeter@605
   819
    /// If neither this function nor \ref supplyMap() is used before
kpeter@605
   820
    /// calling \ref run(), the supply of each node will be set to zero.
kpeter@605
   821
    /// (It makes sense only if non-zero lower bounds are given.)
kpeter@605
   822
    ///
kpeter@605
   823
    /// \param s The source node.
kpeter@605
   824
    /// \param t The target node.
kpeter@605
   825
    /// \param k The required amount of flow from node \c s to node \c t
kpeter@605
   826
    /// (i.e. the supply of \c s and the demand of \c t).
kpeter@605
   827
    ///
kpeter@605
   828
    /// \return <tt>(*this)</tt>
kpeter@607
   829
    NetworkSimplex& stSupply(const Node& s, const Node& t, Flow k) {
kpeter@605
   830
      delete _psupply;
kpeter@605
   831
      _psupply = NULL;
kpeter@605
   832
      _pstsup = true;
kpeter@605
   833
      _psource = s;
kpeter@605
   834
      _ptarget = t;
kpeter@605
   835
      _pstflow = k;
kpeter@605
   836
      return *this;
kpeter@605
   837
    }
kpeter@609
   838
    
kpeter@609
   839
    /// \brief Set the problem type.
kpeter@609
   840
    ///
kpeter@609
   841
    /// This function sets the problem type for the algorithm.
kpeter@609
   842
    /// If it is not used before calling \ref run(), the \ref GEQ problem
kpeter@609
   843
    /// type will be used.
kpeter@609
   844
    ///
kpeter@609
   845
    /// For more information see \ref ProblemType.
kpeter@609
   846
    ///
kpeter@609
   847
    /// \return <tt>(*this)</tt>
kpeter@609
   848
    NetworkSimplex& problemType(ProblemType problem_type) {
kpeter@609
   849
      _ptype = problem_type;
kpeter@609
   850
      return *this;
kpeter@609
   851
    }
kpeter@605
   852
kpeter@601
   853
    /// \brief Set the flow map.
kpeter@601
   854
    ///
kpeter@601
   855
    /// This function sets the flow map.
kpeter@605
   856
    /// If it is not used before calling \ref run(), an instance will
kpeter@605
   857
    /// be allocated automatically. The destructor deallocates this
kpeter@605
   858
    /// automatically allocated map, of course.
kpeter@601
   859
    ///
kpeter@601
   860
    /// \return <tt>(*this)</tt>
kpeter@605
   861
    NetworkSimplex& flowMap(FlowMap& map) {
kpeter@601
   862
      if (_local_flow) {
kpeter@603
   863
        delete _flow_map;
kpeter@601
   864
        _local_flow = false;
kpeter@601
   865
      }
kpeter@603
   866
      _flow_map = &map;
kpeter@601
   867
      return *this;
kpeter@601
   868
    }
kpeter@601
   869
kpeter@601
   870
    /// \brief Set the potential map.
kpeter@601
   871
    ///
kpeter@605
   872
    /// This function sets the potential map, which is used for storing
kpeter@605
   873
    /// the dual solution.
kpeter@605
   874
    /// If it is not used before calling \ref run(), an instance will
kpeter@605
   875
    /// be allocated automatically. The destructor deallocates this
kpeter@605
   876
    /// automatically allocated map, of course.
kpeter@601
   877
    ///
kpeter@601
   878
    /// \return <tt>(*this)</tt>
kpeter@605
   879
    NetworkSimplex& potentialMap(PotentialMap& map) {
kpeter@601
   880
      if (_local_potential) {
kpeter@603
   881
        delete _potential_map;
kpeter@601
   882
        _local_potential = false;
kpeter@601
   883
      }
kpeter@603
   884
      _potential_map = &map;
kpeter@601
   885
      return *this;
kpeter@601
   886
    }
kpeter@609
   887
    
kpeter@609
   888
    /// @}
kpeter@601
   889
kpeter@605
   890
    /// \name Execution Control
kpeter@605
   891
    /// The algorithm can be executed using \ref run().
kpeter@605
   892
kpeter@601
   893
    /// @{
kpeter@601
   894
kpeter@601
   895
    /// \brief Run the algorithm.
kpeter@601
   896
    ///
kpeter@601
   897
    /// This function runs the algorithm.
kpeter@609
   898
    /// The paramters can be specified using functions \ref lowerMap(),
kpeter@606
   899
    /// \ref upperMap(), \ref capacityMap(), \ref boundMaps(),
kpeter@609
   900
    /// \ref costMap(), \ref supplyMap(), \ref stSupply(), 
kpeter@609
   901
    /// \ref problemType(), \ref flowMap() and \ref potentialMap().
kpeter@609
   902
    /// For example,
kpeter@605
   903
    /// \code
kpeter@605
   904
    ///   NetworkSimplex<ListDigraph> ns(graph);
kpeter@605
   905
    ///   ns.boundMaps(lower, upper).costMap(cost)
kpeter@605
   906
    ///     .supplyMap(sup).run();
kpeter@605
   907
    /// \endcode
kpeter@601
   908
    ///
kpeter@606
   909
    /// This function can be called more than once. All the parameters
kpeter@606
   910
    /// that have been given are kept for the next call, unless
kpeter@606
   911
    /// \ref reset() is called, thus only the modified parameters
kpeter@606
   912
    /// have to be set again. See \ref reset() for examples.
kpeter@606
   913
    ///
kpeter@605
   914
    /// \param pivot_rule The pivot rule that will be used during the
kpeter@605
   915
    /// algorithm. For more information see \ref PivotRule.
kpeter@601
   916
    ///
kpeter@601
   917
    /// \return \c true if a feasible flow can be found.
kpeter@605
   918
    bool run(PivotRule pivot_rule = BLOCK_SEARCH) {
kpeter@601
   919
      return init() && start(pivot_rule);
kpeter@601
   920
    }
kpeter@601
   921
kpeter@606
   922
    /// \brief Reset all the parameters that have been given before.
kpeter@606
   923
    ///
kpeter@606
   924
    /// This function resets all the paramaters that have been given
kpeter@609
   925
    /// before using functions \ref lowerMap(), \ref upperMap(),
kpeter@609
   926
    /// \ref capacityMap(), \ref boundMaps(), \ref costMap(),
kpeter@609
   927
    /// \ref supplyMap(), \ref stSupply(), \ref problemType(), 
kpeter@609
   928
    /// \ref flowMap() and \ref potentialMap().
kpeter@606
   929
    ///
kpeter@606
   930
    /// It is useful for multiple run() calls. If this function is not
kpeter@606
   931
    /// used, all the parameters given before are kept for the next
kpeter@606
   932
    /// \ref run() call.
kpeter@606
   933
    ///
kpeter@606
   934
    /// For example,
kpeter@606
   935
    /// \code
kpeter@606
   936
    ///   NetworkSimplex<ListDigraph> ns(graph);
kpeter@606
   937
    ///
kpeter@606
   938
    ///   // First run
kpeter@606
   939
    ///   ns.lowerMap(lower).capacityMap(cap).costMap(cost)
kpeter@606
   940
    ///     .supplyMap(sup).run();
kpeter@606
   941
    ///
kpeter@606
   942
    ///   // Run again with modified cost map (reset() is not called,
kpeter@606
   943
    ///   // so only the cost map have to be set again)
kpeter@606
   944
    ///   cost[e] += 100;
kpeter@606
   945
    ///   ns.costMap(cost).run();
kpeter@606
   946
    ///
kpeter@606
   947
    ///   // Run again from scratch using reset()
kpeter@606
   948
    ///   // (the lower bounds will be set to zero on all arcs)
kpeter@606
   949
    ///   ns.reset();
kpeter@606
   950
    ///   ns.capacityMap(cap).costMap(cost)
kpeter@606
   951
    ///     .supplyMap(sup).run();
kpeter@606
   952
    /// \endcode
kpeter@606
   953
    ///
kpeter@606
   954
    /// \return <tt>(*this)</tt>
kpeter@606
   955
    NetworkSimplex& reset() {
kpeter@606
   956
      delete _plower;
kpeter@606
   957
      delete _pupper;
kpeter@606
   958
      delete _pcost;
kpeter@606
   959
      delete _psupply;
kpeter@606
   960
      _plower = NULL;
kpeter@606
   961
      _pupper = NULL;
kpeter@606
   962
      _pcost = NULL;
kpeter@606
   963
      _psupply = NULL;
kpeter@606
   964
      _pstsup = false;
kpeter@609
   965
      _ptype = GEQ;
kpeter@609
   966
      if (_local_flow) delete _flow_map;
kpeter@609
   967
      if (_local_potential) delete _potential_map;
kpeter@609
   968
      _flow_map = NULL;
kpeter@609
   969
      _potential_map = NULL;
kpeter@609
   970
      _local_flow = false;
kpeter@609
   971
      _local_potential = false;
kpeter@609
   972
kpeter@606
   973
      return *this;
kpeter@606
   974
    }
kpeter@606
   975
kpeter@601
   976
    /// @}
kpeter@601
   977
kpeter@601
   978
    /// \name Query Functions
kpeter@601
   979
    /// The results of the algorithm can be obtained using these
kpeter@601
   980
    /// functions.\n
kpeter@605
   981
    /// The \ref run() function must be called before using them.
kpeter@605
   982
kpeter@601
   983
    /// @{
kpeter@601
   984
kpeter@605
   985
    /// \brief Return the total cost of the found flow.
kpeter@605
   986
    ///
kpeter@605
   987
    /// This function returns the total cost of the found flow.
kpeter@607
   988
    /// The complexity of the function is O(e).
kpeter@605
   989
    ///
kpeter@605
   990
    /// \note The return type of the function can be specified as a
kpeter@605
   991
    /// template parameter. For example,
kpeter@605
   992
    /// \code
kpeter@605
   993
    ///   ns.totalCost<double>();
kpeter@605
   994
    /// \endcode
kpeter@607
   995
    /// It is useful if the total cost cannot be stored in the \c Cost
kpeter@605
   996
    /// type of the algorithm, which is the default return type of the
kpeter@605
   997
    /// function.
kpeter@605
   998
    ///
kpeter@605
   999
    /// \pre \ref run() must be called before using this function.
kpeter@605
  1000
    template <typename Num>
kpeter@605
  1001
    Num totalCost() const {
kpeter@605
  1002
      Num c = 0;
kpeter@605
  1003
      if (_pcost) {
kpeter@605
  1004
        for (ArcIt e(_graph); e != INVALID; ++e)
kpeter@605
  1005
          c += (*_flow_map)[e] * (*_pcost)[e];
kpeter@605
  1006
      } else {
kpeter@605
  1007
        for (ArcIt e(_graph); e != INVALID; ++e)
kpeter@605
  1008
          c += (*_flow_map)[e];
kpeter@605
  1009
      }
kpeter@605
  1010
      return c;
kpeter@605
  1011
    }
kpeter@605
  1012
kpeter@605
  1013
#ifndef DOXYGEN
kpeter@607
  1014
    Cost totalCost() const {
kpeter@607
  1015
      return totalCost<Cost>();
kpeter@605
  1016
    }
kpeter@605
  1017
#endif
kpeter@605
  1018
kpeter@605
  1019
    /// \brief Return the flow on the given arc.
kpeter@605
  1020
    ///
kpeter@605
  1021
    /// This function returns the flow on the given arc.
kpeter@605
  1022
    ///
kpeter@605
  1023
    /// \pre \ref run() must be called before using this function.
kpeter@607
  1024
    Flow flow(const Arc& a) const {
kpeter@605
  1025
      return (*_flow_map)[a];
kpeter@605
  1026
    }
kpeter@605
  1027
kpeter@601
  1028
    /// \brief Return a const reference to the flow map.
kpeter@601
  1029
    ///
kpeter@601
  1030
    /// This function returns a const reference to an arc map storing
kpeter@601
  1031
    /// the found flow.
kpeter@601
  1032
    ///
kpeter@601
  1033
    /// \pre \ref run() must be called before using this function.
kpeter@601
  1034
    const FlowMap& flowMap() const {
kpeter@603
  1035
      return *_flow_map;
kpeter@601
  1036
    }
kpeter@601
  1037
kpeter@605
  1038
    /// \brief Return the potential (dual value) of the given node.
kpeter@605
  1039
    ///
kpeter@605
  1040
    /// This function returns the potential (dual value) of the
kpeter@605
  1041
    /// given node.
kpeter@605
  1042
    ///
kpeter@605
  1043
    /// \pre \ref run() must be called before using this function.
kpeter@607
  1044
    Cost potential(const Node& n) const {
kpeter@605
  1045
      return (*_potential_map)[n];
kpeter@605
  1046
    }
kpeter@605
  1047
kpeter@601
  1048
    /// \brief Return a const reference to the potential map
kpeter@601
  1049
    /// (the dual solution).
kpeter@601
  1050
    ///
kpeter@601
  1051
    /// This function returns a const reference to a node map storing
kpeter@605
  1052
    /// the found potentials, which form the dual solution of the
kpeter@605
  1053
    /// \ref min_cost_flow "minimum cost flow" problem.
kpeter@601
  1054
    ///
kpeter@601
  1055
    /// \pre \ref run() must be called before using this function.
kpeter@601
  1056
    const PotentialMap& potentialMap() const {
kpeter@603
  1057
      return *_potential_map;
kpeter@601
  1058
    }
kpeter@601
  1059
kpeter@601
  1060
    /// @}
kpeter@601
  1061
kpeter@601
  1062
  private:
kpeter@601
  1063
kpeter@601
  1064
    // Initialize internal data structures
kpeter@601
  1065
    bool init() {
kpeter@601
  1066
      // Initialize result maps
kpeter@603
  1067
      if (!_flow_map) {
kpeter@603
  1068
        _flow_map = new FlowMap(_graph);
kpeter@601
  1069
        _local_flow = true;
kpeter@601
  1070
      }
kpeter@603
  1071
      if (!_potential_map) {
kpeter@603
  1072
        _potential_map = new PotentialMap(_graph);
kpeter@601
  1073
        _local_potential = true;
kpeter@601
  1074
      }
kpeter@601
  1075
kpeter@601
  1076
      // Initialize vectors
kpeter@603
  1077
      _node_num = countNodes(_graph);
kpeter@603
  1078
      _arc_num = countArcs(_graph);
kpeter@601
  1079
      int all_node_num = _node_num + 1;
kpeter@603
  1080
      int all_arc_num = _arc_num + _node_num;
kpeter@605
  1081
      if (_node_num == 0) return false;
kpeter@601
  1082
kpeter@603
  1083
      _arc_ref.resize(_arc_num);
kpeter@603
  1084
      _source.resize(all_arc_num);
kpeter@603
  1085
      _target.resize(all_arc_num);
kpeter@601
  1086
kpeter@603
  1087
      _cap.resize(all_arc_num);
kpeter@603
  1088
      _cost.resize(all_arc_num);
kpeter@601
  1089
      _supply.resize(all_node_num);
kpeter@606
  1090
      _flow.resize(all_arc_num);
kpeter@606
  1091
      _pi.resize(all_node_num);
kpeter@601
  1092
kpeter@601
  1093
      _parent.resize(all_node_num);
kpeter@601
  1094
      _pred.resize(all_node_num);
kpeter@603
  1095
      _forward.resize(all_node_num);
kpeter@601
  1096
      _thread.resize(all_node_num);
kpeter@604
  1097
      _rev_thread.resize(all_node_num);
kpeter@604
  1098
      _succ_num.resize(all_node_num);
kpeter@604
  1099
      _last_succ.resize(all_node_num);
kpeter@606
  1100
      _state.resize(all_arc_num);
kpeter@601
  1101
kpeter@601
  1102
      // Initialize node related data
kpeter@601
  1103
      bool valid_supply = true;
kpeter@609
  1104
      Flow sum_supply = 0;
kpeter@605
  1105
      if (!_pstsup && !_psupply) {
kpeter@605
  1106
        _pstsup = true;
kpeter@605
  1107
        _psource = _ptarget = NodeIt(_graph);
kpeter@605
  1108
        _pstflow = 0;
kpeter@605
  1109
      }
kpeter@605
  1110
      if (_psupply) {
kpeter@601
  1111
        int i = 0;
kpeter@603
  1112
        for (NodeIt n(_graph); n != INVALID; ++n, ++i) {
kpeter@601
  1113
          _node_id[n] = i;
kpeter@605
  1114
          _supply[i] = (*_psupply)[n];
kpeter@609
  1115
          sum_supply += _supply[i];
kpeter@601
  1116
        }
kpeter@609
  1117
        valid_supply = (_ptype == GEQ && sum_supply <= 0) ||
kpeter@609
  1118
                       (_ptype == LEQ && sum_supply >= 0);
kpeter@601
  1119
      } else {
kpeter@601
  1120
        int i = 0;
kpeter@603
  1121
        for (NodeIt n(_graph); n != INVALID; ++n, ++i) {
kpeter@601
  1122
          _node_id[n] = i;
kpeter@601
  1123
          _supply[i] = 0;
kpeter@601
  1124
        }
kpeter@605
  1125
        _supply[_node_id[_psource]] =  _pstflow;
kpeter@609
  1126
        _supply[_node_id[_ptarget]] = -_pstflow;
kpeter@601
  1127
      }
kpeter@601
  1128
      if (!valid_supply) return false;
kpeter@601
  1129
kpeter@609
  1130
      // Infinite capacity value
kpeter@609
  1131
      Flow inf_cap =
kpeter@609
  1132
        std::numeric_limits<Flow>::has_infinity ?
kpeter@609
  1133
        std::numeric_limits<Flow>::infinity() :
kpeter@609
  1134
        std::numeric_limits<Flow>::max();
kpeter@609
  1135
kpeter@609
  1136
      // Initialize artifical cost
kpeter@609
  1137
      Cost art_cost;
kpeter@609
  1138
      if (std::numeric_limits<Cost>::is_exact) {
kpeter@609
  1139
        art_cost = std::numeric_limits<Cost>::max() / 4 + 1;
kpeter@609
  1140
      } else {
kpeter@609
  1141
        art_cost = std::numeric_limits<Cost>::min();
kpeter@609
  1142
        for (int i = 0; i != _arc_num; ++i) {
kpeter@609
  1143
          if (_cost[i] > art_cost) art_cost = _cost[i];
kpeter@609
  1144
        }
kpeter@609
  1145
        art_cost = (art_cost + 1) * _node_num;
kpeter@609
  1146
      }
kpeter@609
  1147
kpeter@609
  1148
      // Run Circulation to check if a feasible solution exists
kpeter@609
  1149
      typedef ConstMap<Arc, Flow> ConstArcMap;
kpeter@613
  1150
      ConstArcMap zero_arc_map(0), inf_arc_map(inf_cap);
kpeter@609
  1151
      FlowNodeMap *csup = NULL;
kpeter@609
  1152
      bool local_csup = false;
kpeter@609
  1153
      if (_psupply) {
kpeter@609
  1154
        csup = _psupply;
kpeter@609
  1155
      } else {
kpeter@609
  1156
        csup = new FlowNodeMap(_graph, 0);
kpeter@609
  1157
        (*csup)[_psource] =  _pstflow;
kpeter@609
  1158
        (*csup)[_ptarget] = -_pstflow;
kpeter@609
  1159
        local_csup = true;
kpeter@609
  1160
      }
kpeter@609
  1161
      bool circ_result = false;
kpeter@609
  1162
      if (_ptype == GEQ || (_ptype == LEQ && sum_supply == 0)) {
kpeter@609
  1163
        // GEQ problem type
kpeter@609
  1164
        if (_plower) {
kpeter@609
  1165
          if (_pupper) {
kpeter@609
  1166
            Circulation<GR, FlowArcMap, FlowArcMap, FlowNodeMap>
kpeter@609
  1167
              circ(_graph, *_plower, *_pupper, *csup);
kpeter@609
  1168
            circ_result = circ.run();
kpeter@609
  1169
          } else {
kpeter@609
  1170
            Circulation<GR, FlowArcMap, ConstArcMap, FlowNodeMap>
kpeter@613
  1171
              circ(_graph, *_plower, inf_arc_map, *csup);
kpeter@609
  1172
            circ_result = circ.run();
kpeter@609
  1173
          }
kpeter@609
  1174
        } else {
kpeter@609
  1175
          if (_pupper) {
kpeter@609
  1176
            Circulation<GR, ConstArcMap, FlowArcMap, FlowNodeMap>
kpeter@613
  1177
              circ(_graph, zero_arc_map, *_pupper, *csup);
kpeter@609
  1178
            circ_result = circ.run();
kpeter@609
  1179
          } else {
kpeter@609
  1180
            Circulation<GR, ConstArcMap, ConstArcMap, FlowNodeMap>
kpeter@613
  1181
              circ(_graph, zero_arc_map, inf_arc_map, *csup);
kpeter@609
  1182
            circ_result = circ.run();
kpeter@609
  1183
          }
kpeter@609
  1184
        }
kpeter@609
  1185
      } else {
kpeter@609
  1186
        // LEQ problem type
kpeter@609
  1187
        typedef ReverseDigraph<const GR> RevGraph;
kpeter@609
  1188
        typedef NegMap<FlowNodeMap> NegNodeMap;
kpeter@609
  1189
        RevGraph rgraph(_graph);
kpeter@609
  1190
        NegNodeMap neg_csup(*csup);
kpeter@609
  1191
        if (_plower) {
kpeter@609
  1192
          if (_pupper) {
kpeter@609
  1193
            Circulation<RevGraph, FlowArcMap, FlowArcMap, NegNodeMap>
kpeter@609
  1194
              circ(rgraph, *_plower, *_pupper, neg_csup);
kpeter@609
  1195
            circ_result = circ.run();
kpeter@609
  1196
          } else {
kpeter@609
  1197
            Circulation<RevGraph, FlowArcMap, ConstArcMap, NegNodeMap>
kpeter@613
  1198
              circ(rgraph, *_plower, inf_arc_map, neg_csup);
kpeter@609
  1199
            circ_result = circ.run();
kpeter@609
  1200
          }
kpeter@609
  1201
        } else {
kpeter@609
  1202
          if (_pupper) {
kpeter@609
  1203
            Circulation<RevGraph, ConstArcMap, FlowArcMap, NegNodeMap>
kpeter@613
  1204
              circ(rgraph, zero_arc_map, *_pupper, neg_csup);
kpeter@609
  1205
            circ_result = circ.run();
kpeter@609
  1206
          } else {
kpeter@609
  1207
            Circulation<RevGraph, ConstArcMap, ConstArcMap, NegNodeMap>
kpeter@613
  1208
              circ(rgraph, zero_arc_map, inf_arc_map, neg_csup);
kpeter@609
  1209
            circ_result = circ.run();
kpeter@609
  1210
          }
kpeter@609
  1211
        }
kpeter@609
  1212
      }
kpeter@609
  1213
      if (local_csup) delete csup;
kpeter@609
  1214
      if (!circ_result) return false;
kpeter@609
  1215
kpeter@601
  1216
      // Set data for the artificial root node
kpeter@601
  1217
      _root = _node_num;
kpeter@601
  1218
      _parent[_root] = -1;
kpeter@601
  1219
      _pred[_root] = -1;
kpeter@601
  1220
      _thread[_root] = 0;
kpeter@604
  1221
      _rev_thread[0] = _root;
kpeter@604
  1222
      _succ_num[_root] = all_node_num;
kpeter@604
  1223
      _last_succ[_root] = _root - 1;
kpeter@609
  1224
      _supply[_root] = -sum_supply;
kpeter@609
  1225
      if (sum_supply < 0) {
kpeter@609
  1226
        _pi[_root] = -art_cost;
kpeter@609
  1227
      } else {
kpeter@609
  1228
        _pi[_root] = art_cost;
kpeter@609
  1229
      }
kpeter@601
  1230
kpeter@601
  1231
      // Store the arcs in a mixed order
alpar@612
  1232
      int k = std::max(int(std::sqrt(double(_arc_num))), 10);
kpeter@601
  1233
      int i = 0;
kpeter@603
  1234
      for (ArcIt e(_graph); e != INVALID; ++e) {
kpeter@603
  1235
        _arc_ref[i] = e;
kpeter@601
  1236
        if ((i += k) >= _arc_num) i = (i % k) + 1;
kpeter@601
  1237
      }
kpeter@601
  1238
kpeter@601
  1239
      // Initialize arc maps
kpeter@605
  1240
      if (_pupper && _pcost) {
kpeter@605
  1241
        for (int i = 0; i != _arc_num; ++i) {
kpeter@605
  1242
          Arc e = _arc_ref[i];
kpeter@605
  1243
          _source[i] = _node_id[_graph.source(e)];
kpeter@605
  1244
          _target[i] = _node_id[_graph.target(e)];
kpeter@605
  1245
          _cap[i] = (*_pupper)[e];
kpeter@605
  1246
          _cost[i] = (*_pcost)[e];
kpeter@606
  1247
          _flow[i] = 0;
kpeter@606
  1248
          _state[i] = STATE_LOWER;
kpeter@605
  1249
        }
kpeter@605
  1250
      } else {
kpeter@605
  1251
        for (int i = 0; i != _arc_num; ++i) {
kpeter@605
  1252
          Arc e = _arc_ref[i];
kpeter@605
  1253
          _source[i] = _node_id[_graph.source(e)];
kpeter@605
  1254
          _target[i] = _node_id[_graph.target(e)];
kpeter@606
  1255
          _flow[i] = 0;
kpeter@606
  1256
          _state[i] = STATE_LOWER;
kpeter@605
  1257
        }
kpeter@605
  1258
        if (_pupper) {
kpeter@605
  1259
          for (int i = 0; i != _arc_num; ++i)
kpeter@605
  1260
            _cap[i] = (*_pupper)[_arc_ref[i]];
kpeter@605
  1261
        } else {
kpeter@605
  1262
          for (int i = 0; i != _arc_num; ++i)
kpeter@608
  1263
            _cap[i] = inf_cap;
kpeter@605
  1264
        }
kpeter@605
  1265
        if (_pcost) {
kpeter@605
  1266
          for (int i = 0; i != _arc_num; ++i)
kpeter@605
  1267
            _cost[i] = (*_pcost)[_arc_ref[i]];
kpeter@605
  1268
        } else {
kpeter@605
  1269
          for (int i = 0; i != _arc_num; ++i)
kpeter@605
  1270
            _cost[i] = 1;
kpeter@605
  1271
        }
kpeter@601
  1272
      }
kpeter@608
  1273
      
kpeter@601
  1274
      // Remove non-zero lower bounds
kpeter@605
  1275
      if (_plower) {
kpeter@601
  1276
        for (int i = 0; i != _arc_num; ++i) {
kpeter@607
  1277
          Flow c = (*_plower)[_arc_ref[i]];
kpeter@601
  1278
          if (c != 0) {
kpeter@601
  1279
            _cap[i] -= c;
kpeter@601
  1280
            _supply[_source[i]] -= c;
kpeter@601
  1281
            _supply[_target[i]] += c;
kpeter@601
  1282
          }
kpeter@601
  1283
        }
kpeter@601
  1284
      }
kpeter@601
  1285
kpeter@601
  1286
      // Add artificial arcs and initialize the spanning tree data structure
kpeter@601
  1287
      for (int u = 0, e = _arc_num; u != _node_num; ++u, ++e) {
kpeter@601
  1288
        _thread[u] = u + 1;
kpeter@604
  1289
        _rev_thread[u + 1] = u;
kpeter@604
  1290
        _succ_num[u] = 1;
kpeter@604
  1291
        _last_succ[u] = u;
kpeter@601
  1292
        _parent[u] = _root;
kpeter@601
  1293
        _pred[u] = e;
kpeter@608
  1294
        _cost[e] = art_cost;
kpeter@608
  1295
        _cap[e] = inf_cap;
kpeter@606
  1296
        _state[e] = STATE_TREE;
kpeter@609
  1297
        if (_supply[u] > 0 || (_supply[u] == 0 && sum_supply <= 0)) {
kpeter@601
  1298
          _flow[e] = _supply[u];
kpeter@601
  1299
          _forward[u] = true;
kpeter@609
  1300
          _pi[u] = -art_cost + _pi[_root];
kpeter@601
  1301
        } else {
kpeter@601
  1302
          _flow[e] = -_supply[u];
kpeter@601
  1303
          _forward[u] = false;
kpeter@609
  1304
          _pi[u] = art_cost + _pi[_root];
kpeter@601
  1305
        }
kpeter@601
  1306
      }
kpeter@601
  1307
kpeter@601
  1308
      return true;
kpeter@601
  1309
    }
kpeter@601
  1310
kpeter@601
  1311
    // Find the join node
kpeter@601
  1312
    void findJoinNode() {
kpeter@603
  1313
      int u = _source[in_arc];
kpeter@603
  1314
      int v = _target[in_arc];
kpeter@601
  1315
      while (u != v) {
kpeter@604
  1316
        if (_succ_num[u] < _succ_num[v]) {
kpeter@604
  1317
          u = _parent[u];
kpeter@604
  1318
        } else {
kpeter@604
  1319
          v = _parent[v];
kpeter@604
  1320
        }
kpeter@601
  1321
      }
kpeter@601
  1322
      join = u;
kpeter@601
  1323
    }
kpeter@601
  1324
kpeter@601
  1325
    // Find the leaving arc of the cycle and returns true if the
kpeter@601
  1326
    // leaving arc is not the same as the entering arc
kpeter@601
  1327
    bool findLeavingArc() {
kpeter@601
  1328
      // Initialize first and second nodes according to the direction
kpeter@601
  1329
      // of the cycle
kpeter@603
  1330
      if (_state[in_arc] == STATE_LOWER) {
kpeter@603
  1331
        first  = _source[in_arc];
kpeter@603
  1332
        second = _target[in_arc];
kpeter@601
  1333
      } else {
kpeter@603
  1334
        first  = _target[in_arc];
kpeter@603
  1335
        second = _source[in_arc];
kpeter@601
  1336
      }
kpeter@603
  1337
      delta = _cap[in_arc];
kpeter@601
  1338
      int result = 0;
kpeter@607
  1339
      Flow d;
kpeter@601
  1340
      int e;
kpeter@601
  1341
kpeter@601
  1342
      // Search the cycle along the path form the first node to the root
kpeter@601
  1343
      for (int u = first; u != join; u = _parent[u]) {
kpeter@601
  1344
        e = _pred[u];
kpeter@601
  1345
        d = _forward[u] ? _flow[e] : _cap[e] - _flow[e];
kpeter@601
  1346
        if (d < delta) {
kpeter@601
  1347
          delta = d;
kpeter@601
  1348
          u_out = u;
kpeter@601
  1349
          result = 1;
kpeter@601
  1350
        }
kpeter@601
  1351
      }
kpeter@601
  1352
      // Search the cycle along the path form the second node to the root
kpeter@601
  1353
      for (int u = second; u != join; u = _parent[u]) {
kpeter@601
  1354
        e = _pred[u];
kpeter@601
  1355
        d = _forward[u] ? _cap[e] - _flow[e] : _flow[e];
kpeter@601
  1356
        if (d <= delta) {
kpeter@601
  1357
          delta = d;
kpeter@601
  1358
          u_out = u;
kpeter@601
  1359
          result = 2;
kpeter@601
  1360
        }
kpeter@601
  1361
      }
kpeter@601
  1362
kpeter@601
  1363
      if (result == 1) {
kpeter@601
  1364
        u_in = first;
kpeter@601
  1365
        v_in = second;
kpeter@601
  1366
      } else {
kpeter@601
  1367
        u_in = second;
kpeter@601
  1368
        v_in = first;
kpeter@601
  1369
      }
kpeter@601
  1370
      return result != 0;
kpeter@601
  1371
    }
kpeter@601
  1372
kpeter@601
  1373
    // Change _flow and _state vectors
kpeter@601
  1374
    void changeFlow(bool change) {
kpeter@601
  1375
      // Augment along the cycle
kpeter@601
  1376
      if (delta > 0) {
kpeter@607
  1377
        Flow val = _state[in_arc] * delta;
kpeter@603
  1378
        _flow[in_arc] += val;
kpeter@603
  1379
        for (int u = _source[in_arc]; u != join; u = _parent[u]) {
kpeter@601
  1380
          _flow[_pred[u]] += _forward[u] ? -val : val;
kpeter@601
  1381
        }
kpeter@603
  1382
        for (int u = _target[in_arc]; u != join; u = _parent[u]) {
kpeter@601
  1383
          _flow[_pred[u]] += _forward[u] ? val : -val;
kpeter@601
  1384
        }
kpeter@601
  1385
      }
kpeter@601
  1386
      // Update the state of the entering and leaving arcs
kpeter@601
  1387
      if (change) {
kpeter@603
  1388
        _state[in_arc] = STATE_TREE;
kpeter@601
  1389
        _state[_pred[u_out]] =
kpeter@601
  1390
          (_flow[_pred[u_out]] == 0) ? STATE_LOWER : STATE_UPPER;
kpeter@601
  1391
      } else {
kpeter@603
  1392
        _state[in_arc] = -_state[in_arc];
kpeter@601
  1393
      }
kpeter@601
  1394
    }
kpeter@601
  1395
kpeter@604
  1396
    // Update the tree structure
kpeter@604
  1397
    void updateTreeStructure() {
kpeter@604
  1398
      int u, w;
kpeter@604
  1399
      int old_rev_thread = _rev_thread[u_out];
kpeter@604
  1400
      int old_succ_num = _succ_num[u_out];
kpeter@604
  1401
      int old_last_succ = _last_succ[u_out];
kpeter@601
  1402
      v_out = _parent[u_out];
kpeter@601
  1403
kpeter@604
  1404
      u = _last_succ[u_in];  // the last successor of u_in
kpeter@604
  1405
      right = _thread[u];    // the node after it
kpeter@604
  1406
kpeter@604
  1407
      // Handle the case when old_rev_thread equals to v_in
kpeter@604
  1408
      // (it also means that join and v_out coincide)
kpeter@604
  1409
      if (old_rev_thread == v_in) {
kpeter@604
  1410
        last = _thread[_last_succ[u_out]];
kpeter@604
  1411
      } else {
kpeter@604
  1412
        last = _thread[v_in];
kpeter@601
  1413
      }
kpeter@601
  1414
kpeter@604
  1415
      // Update _thread and _parent along the stem nodes (i.e. the nodes
kpeter@604
  1416
      // between u_in and u_out, whose parent have to be changed)
kpeter@601
  1417
      _thread[v_in] = stem = u_in;
kpeter@604
  1418
      _dirty_revs.clear();
kpeter@604
  1419
      _dirty_revs.push_back(v_in);
kpeter@601
  1420
      par_stem = v_in;
kpeter@601
  1421
      while (stem != u_out) {
kpeter@604
  1422
        // Insert the next stem node into the thread list
kpeter@604
  1423
        new_stem = _parent[stem];
kpeter@604
  1424
        _thread[u] = new_stem;
kpeter@604
  1425
        _dirty_revs.push_back(u);
kpeter@601
  1426
kpeter@604
  1427
        // Remove the subtree of stem from the thread list
kpeter@604
  1428
        w = _rev_thread[stem];
kpeter@604
  1429
        _thread[w] = right;
kpeter@604
  1430
        _rev_thread[right] = w;
kpeter@601
  1431
kpeter@604
  1432
        // Change the parent node and shift stem nodes
kpeter@601
  1433
        _parent[stem] = par_stem;
kpeter@601
  1434
        par_stem = stem;
kpeter@601
  1435
        stem = new_stem;
kpeter@601
  1436
kpeter@604
  1437
        // Update u and right
kpeter@604
  1438
        u = _last_succ[stem] == _last_succ[par_stem] ?
kpeter@604
  1439
          _rev_thread[par_stem] : _last_succ[stem];
kpeter@601
  1440
        right = _thread[u];
kpeter@601
  1441
      }
kpeter@601
  1442
      _parent[u_out] = par_stem;
kpeter@601
  1443
      _thread[u] = last;
kpeter@604
  1444
      _rev_thread[last] = u;
kpeter@604
  1445
      _last_succ[u_out] = u;
kpeter@601
  1446
kpeter@604
  1447
      // Remove the subtree of u_out from the thread list except for
kpeter@604
  1448
      // the case when old_rev_thread equals to v_in
kpeter@604
  1449
      // (it also means that join and v_out coincide)
kpeter@604
  1450
      if (old_rev_thread != v_in) {
kpeter@604
  1451
        _thread[old_rev_thread] = right;
kpeter@604
  1452
        _rev_thread[right] = old_rev_thread;
kpeter@604
  1453
      }
kpeter@604
  1454
kpeter@604
  1455
      // Update _rev_thread using the new _thread values
kpeter@604
  1456
      for (int i = 0; i < int(_dirty_revs.size()); ++i) {
kpeter@604
  1457
        u = _dirty_revs[i];
kpeter@604
  1458
        _rev_thread[_thread[u]] = u;
kpeter@604
  1459
      }
kpeter@604
  1460
kpeter@604
  1461
      // Update _pred, _forward, _last_succ and _succ_num for the
kpeter@604
  1462
      // stem nodes from u_out to u_in
kpeter@604
  1463
      int tmp_sc = 0, tmp_ls = _last_succ[u_out];
kpeter@604
  1464
      u = u_out;
kpeter@604
  1465
      while (u != u_in) {
kpeter@604
  1466
        w = _parent[u];
kpeter@604
  1467
        _pred[u] = _pred[w];
kpeter@604
  1468
        _forward[u] = !_forward[w];
kpeter@604
  1469
        tmp_sc += _succ_num[u] - _succ_num[w];
kpeter@604
  1470
        _succ_num[u] = tmp_sc;
kpeter@604
  1471
        _last_succ[w] = tmp_ls;
kpeter@604
  1472
        u = w;
kpeter@604
  1473
      }
kpeter@604
  1474
      _pred[u_in] = in_arc;
kpeter@604
  1475
      _forward[u_in] = (u_in == _source[in_arc]);
kpeter@604
  1476
      _succ_num[u_in] = old_succ_num;
kpeter@604
  1477
kpeter@604
  1478
      // Set limits for updating _last_succ form v_in and v_out
kpeter@604
  1479
      // towards the root
kpeter@604
  1480
      int up_limit_in = -1;
kpeter@604
  1481
      int up_limit_out = -1;
kpeter@604
  1482
      if (_last_succ[join] == v_in) {
kpeter@604
  1483
        up_limit_out = join;
kpeter@601
  1484
      } else {
kpeter@604
  1485
        up_limit_in = join;
kpeter@604
  1486
      }
kpeter@604
  1487
kpeter@604
  1488
      // Update _last_succ from v_in towards the root
kpeter@604
  1489
      for (u = v_in; u != up_limit_in && _last_succ[u] == v_in;
kpeter@604
  1490
           u = _parent[u]) {
kpeter@604
  1491
        _last_succ[u] = _last_succ[u_out];
kpeter@604
  1492
      }
kpeter@604
  1493
      // Update _last_succ from v_out towards the root
kpeter@604
  1494
      if (join != old_rev_thread && v_in != old_rev_thread) {
kpeter@604
  1495
        for (u = v_out; u != up_limit_out && _last_succ[u] == old_last_succ;
kpeter@604
  1496
             u = _parent[u]) {
kpeter@604
  1497
          _last_succ[u] = old_rev_thread;
kpeter@604
  1498
        }
kpeter@604
  1499
      } else {
kpeter@604
  1500
        for (u = v_out; u != up_limit_out && _last_succ[u] == old_last_succ;
kpeter@604
  1501
             u = _parent[u]) {
kpeter@604
  1502
          _last_succ[u] = _last_succ[u_out];
kpeter@604
  1503
        }
kpeter@604
  1504
      }
kpeter@604
  1505
kpeter@604
  1506
      // Update _succ_num from v_in to join
kpeter@604
  1507
      for (u = v_in; u != join; u = _parent[u]) {
kpeter@604
  1508
        _succ_num[u] += old_succ_num;
kpeter@604
  1509
      }
kpeter@604
  1510
      // Update _succ_num from v_out to join
kpeter@604
  1511
      for (u = v_out; u != join; u = _parent[u]) {
kpeter@604
  1512
        _succ_num[u] -= old_succ_num;
kpeter@601
  1513
      }
kpeter@601
  1514
    }
kpeter@601
  1515
kpeter@604
  1516
    // Update potentials
kpeter@604
  1517
    void updatePotential() {
kpeter@607
  1518
      Cost sigma = _forward[u_in] ?
kpeter@601
  1519
        _pi[v_in] - _pi[u_in] - _cost[_pred[u_in]] :
kpeter@601
  1520
        _pi[v_in] - _pi[u_in] + _cost[_pred[u_in]];
kpeter@608
  1521
      // Update potentials in the subtree, which has been moved
kpeter@608
  1522
      int end = _thread[_last_succ[u_in]];
kpeter@608
  1523
      for (int u = u_in; u != end; u = _thread[u]) {
kpeter@608
  1524
        _pi[u] += sigma;
kpeter@601
  1525
      }
kpeter@601
  1526
    }
kpeter@601
  1527
kpeter@601
  1528
    // Execute the algorithm
kpeter@605
  1529
    bool start(PivotRule pivot_rule) {
kpeter@601
  1530
      // Select the pivot rule implementation
kpeter@601
  1531
      switch (pivot_rule) {
kpeter@605
  1532
        case FIRST_ELIGIBLE:
kpeter@601
  1533
          return start<FirstEligiblePivotRule>();
kpeter@605
  1534
        case BEST_ELIGIBLE:
kpeter@601
  1535
          return start<BestEligiblePivotRule>();
kpeter@605
  1536
        case BLOCK_SEARCH:
kpeter@601
  1537
          return start<BlockSearchPivotRule>();
kpeter@605
  1538
        case CANDIDATE_LIST:
kpeter@601
  1539
          return start<CandidateListPivotRule>();
kpeter@605
  1540
        case ALTERING_LIST:
kpeter@601
  1541
          return start<AlteringListPivotRule>();
kpeter@601
  1542
      }
kpeter@601
  1543
      return false;
kpeter@601
  1544
    }
kpeter@601
  1545
kpeter@605
  1546
    template <typename PivotRuleImpl>
kpeter@601
  1547
    bool start() {
kpeter@605
  1548
      PivotRuleImpl pivot(*this);
kpeter@601
  1549
kpeter@605
  1550
      // Execute the Network Simplex algorithm
kpeter@601
  1551
      while (pivot.findEnteringArc()) {
kpeter@601
  1552
        findJoinNode();
kpeter@601
  1553
        bool change = findLeavingArc();
kpeter@601
  1554
        changeFlow(change);
kpeter@601
  1555
        if (change) {
kpeter@604
  1556
          updateTreeStructure();
kpeter@604
  1557
          updatePotential();
kpeter@601
  1558
        }
kpeter@601
  1559
      }
kpeter@601
  1560
kpeter@603
  1561
      // Copy flow values to _flow_map
kpeter@605
  1562
      if (_plower) {
kpeter@601
  1563
        for (int i = 0; i != _arc_num; ++i) {
kpeter@603
  1564
          Arc e = _arc_ref[i];
kpeter@605
  1565
          _flow_map->set(e, (*_plower)[e] + _flow[i]);
kpeter@601
  1566
        }
kpeter@601
  1567
      } else {
kpeter@601
  1568
        for (int i = 0; i != _arc_num; ++i) {
kpeter@603
  1569
          _flow_map->set(_arc_ref[i], _flow[i]);
kpeter@601
  1570
        }
kpeter@601
  1571
      }
kpeter@603
  1572
      // Copy potential values to _potential_map
kpeter@603
  1573
      for (NodeIt n(_graph); n != INVALID; ++n) {
kpeter@603
  1574
        _potential_map->set(n, _pi[_node_id[n]]);
kpeter@601
  1575
      }
kpeter@601
  1576
kpeter@601
  1577
      return true;
kpeter@601
  1578
    }
kpeter@601
  1579
kpeter@601
  1580
  }; //class NetworkSimplex
kpeter@601
  1581
kpeter@601
  1582
  ///@}
kpeter@601
  1583
kpeter@601
  1584
} //namespace lemon
kpeter@601
  1585
kpeter@601
  1586
#endif //LEMON_NETWORK_SIMPLEX_H