lemon/network_simplex.h
author Alpar Juttner <alpar@cs.elte.hu>
Tue, 20 Dec 2011 19:43:58 +0100
changeset 971 22201ee8e437
parent 921 140c953ad5d1
parent 919 e0cef67fe565
child 984 fcb6ad1e67d0
child 1003 16f55008c863
permissions -rw-r--r--
Merge #432 and #433
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
 *
alpar@877
     5
 * Copyright (C) 2003-2010
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@755
    43
  /// for finding a \ref min_cost_flow "minimum cost flow"
kpeter@755
    44
  /// \ref amo93networkflows, \ref dantzig63linearprog,
kpeter@755
    45
  /// \ref kellyoneill91netsimplex.
kpeter@812
    46
  /// This algorithm is a highly efficient specialized version of the
kpeter@812
    47
  /// linear programming simplex method directly for the minimum cost
kpeter@812
    48
  /// flow problem.
kpeter@606
    49
  ///
kpeter@919
    50
  /// In general, \ref NetworkSimplex and \ref CostScaling are the fastest
kpeter@919
    51
  /// implementations available in LEMON for this problem.
kpeter@919
    52
  /// Furthermore, this class supports both directions of the supply/demand
kpeter@919
    53
  /// inequality constraints. For more information, see \ref SupplyType.
kpeter@640
    54
  ///
kpeter@640
    55
  /// Most of the parameters of the problem (except for the digraph)
kpeter@640
    56
  /// can be given using separate functions, and the algorithm can be
kpeter@640
    57
  /// executed using the \ref run() function. If some parameters are not
kpeter@640
    58
  /// specified, then default values will be used.
kpeter@601
    59
  ///
kpeter@605
    60
  /// \tparam GR The digraph type the algorithm runs on.
kpeter@812
    61
  /// \tparam V The number type used for flow amounts, capacity bounds
kpeter@786
    62
  /// and supply values in the algorithm. By default, it is \c int.
kpeter@812
    63
  /// \tparam C The number type used for costs and potentials in the
kpeter@786
    64
  /// algorithm. By default, it is the same as \c V.
kpeter@601
    65
  ///
kpeter@921
    66
  /// \warning Both \c V and \c C must be signed number types.
kpeter@921
    67
  /// \warning All input data (capacities, supply values, and costs) must
kpeter@608
    68
  /// be integer.
kpeter@601
    69
  ///
kpeter@605
    70
  /// \note %NetworkSimplex provides five different pivot rule
kpeter@609
    71
  /// implementations, from which the most efficient one is used
kpeter@786
    72
  /// by default. For more information, see \ref PivotRule.
kpeter@641
    73
  template <typename GR, typename V = int, typename C = V>
kpeter@601
    74
  class NetworkSimplex
kpeter@601
    75
  {
kpeter@605
    76
  public:
kpeter@601
    77
kpeter@642
    78
    /// The type of the flow amounts, capacity bounds and supply values
kpeter@641
    79
    typedef V Value;
kpeter@642
    80
    /// The type of the arc costs
kpeter@607
    81
    typedef C Cost;
kpeter@605
    82
kpeter@605
    83
  public:
kpeter@605
    84
kpeter@640
    85
    /// \brief Problem type constants for the \c run() function.
kpeter@605
    86
    ///
kpeter@640
    87
    /// Enum type containing the problem type constants that can be
kpeter@640
    88
    /// returned by the \ref run() function of the algorithm.
kpeter@640
    89
    enum ProblemType {
kpeter@640
    90
      /// The problem has no feasible solution (flow).
kpeter@640
    91
      INFEASIBLE,
kpeter@640
    92
      /// The problem has optimal solution (i.e. it is feasible and
kpeter@640
    93
      /// bounded), and the algorithm has found optimal flow and node
kpeter@640
    94
      /// potentials (primal and dual solutions).
kpeter@640
    95
      OPTIMAL,
kpeter@640
    96
      /// The objective function of the problem is unbounded, i.e.
kpeter@640
    97
      /// there is a directed cycle having negative total cost and
kpeter@640
    98
      /// infinite upper bound.
kpeter@640
    99
      UNBOUNDED
kpeter@640
   100
    };
alpar@877
   101
kpeter@640
   102
    /// \brief Constants for selecting the type of the supply constraints.
kpeter@640
   103
    ///
kpeter@640
   104
    /// Enum type containing constants for selecting the supply type,
kpeter@640
   105
    /// i.e. the direction of the inequalities in the supply/demand
kpeter@640
   106
    /// constraints of the \ref min_cost_flow "minimum cost flow problem".
kpeter@640
   107
    ///
kpeter@663
   108
    /// The default supply type is \c GEQ, the \c LEQ type can be
kpeter@663
   109
    /// selected using \ref supplyType().
kpeter@663
   110
    /// The equality form is a special case of both supply types.
kpeter@640
   111
    enum SupplyType {
kpeter@640
   112
      /// This option means that there are <em>"greater or equal"</em>
kpeter@663
   113
      /// supply/demand constraints in the definition of the problem.
kpeter@640
   114
      GEQ,
kpeter@640
   115
      /// This option means that there are <em>"less or equal"</em>
kpeter@663
   116
      /// supply/demand constraints in the definition of the problem.
kpeter@663
   117
      LEQ
kpeter@640
   118
    };
alpar@877
   119
kpeter@640
   120
    /// \brief Constants for selecting the pivot rule.
kpeter@640
   121
    ///
kpeter@640
   122
    /// Enum type containing constants for selecting the pivot rule for
kpeter@640
   123
    /// the \ref run() function.
kpeter@640
   124
    ///
kpeter@605
   125
    /// \ref NetworkSimplex provides five different pivot rule
kpeter@605
   126
    /// implementations that significantly affect the running time
kpeter@605
   127
    /// of the algorithm.
kpeter@786
   128
    /// By default, \ref BLOCK_SEARCH "Block Search" is used, which
kpeter@919
   129
    /// turend out to be the most efficient and the most robust on various
kpeter@812
   130
    /// test inputs.
kpeter@786
   131
    /// However, another pivot rule can be selected using the \ref run()
kpeter@605
   132
    /// function with the proper parameter.
kpeter@605
   133
    enum PivotRule {
kpeter@605
   134
kpeter@786
   135
      /// The \e First \e Eligible pivot rule.
kpeter@605
   136
      /// The next eligible arc is selected in a wraparound fashion
kpeter@605
   137
      /// in every iteration.
kpeter@605
   138
      FIRST_ELIGIBLE,
kpeter@605
   139
kpeter@786
   140
      /// The \e Best \e Eligible pivot rule.
kpeter@605
   141
      /// The best eligible arc is selected in every iteration.
kpeter@605
   142
      BEST_ELIGIBLE,
kpeter@605
   143
kpeter@786
   144
      /// The \e Block \e Search pivot rule.
kpeter@605
   145
      /// A specified number of arcs are examined in every iteration
kpeter@605
   146
      /// in a wraparound fashion and the best eligible arc is selected
kpeter@605
   147
      /// from this block.
kpeter@605
   148
      BLOCK_SEARCH,
kpeter@605
   149
kpeter@786
   150
      /// The \e Candidate \e List pivot rule.
kpeter@605
   151
      /// In a major iteration a candidate list is built from eligible arcs
kpeter@605
   152
      /// in a wraparound fashion and in the following minor iterations
kpeter@605
   153
      /// the best eligible arc is selected from this list.
kpeter@605
   154
      CANDIDATE_LIST,
kpeter@605
   155
kpeter@786
   156
      /// The \e Altering \e Candidate \e List pivot rule.
kpeter@605
   157
      /// It is a modified version of the Candidate List method.
kpeter@605
   158
      /// It keeps only the several best eligible arcs from the former
kpeter@605
   159
      /// candidate list and extends this list in every iteration.
kpeter@605
   160
      ALTERING_LIST
kpeter@605
   161
    };
alpar@877
   162
kpeter@605
   163
  private:
kpeter@605
   164
kpeter@605
   165
    TEMPLATE_DIGRAPH_TYPEDEFS(GR);
kpeter@605
   166
kpeter@601
   167
    typedef std::vector<int> IntVector;
kpeter@642
   168
    typedef std::vector<Value> ValueVector;
kpeter@607
   169
    typedef std::vector<Cost> CostVector;
kpeter@895
   170
    typedef std::vector<signed char> CharVector;
kpeter@919
   171
    // Note: vector<signed char> is used instead of vector<ArcState> and
kpeter@895
   172
    // vector<ArcDirection> for efficiency reasons
kpeter@601
   173
kpeter@601
   174
    // State constants for arcs
kpeter@862
   175
    enum ArcState {
kpeter@601
   176
      STATE_UPPER = -1,
kpeter@601
   177
      STATE_TREE  =  0,
kpeter@601
   178
      STATE_LOWER =  1
kpeter@601
   179
    };
kpeter@601
   180
kpeter@895
   181
    // Direction constants for tree arcs
kpeter@895
   182
    enum ArcDirection {
kpeter@895
   183
      DIR_DOWN = -1,
kpeter@895
   184
      DIR_UP   =  1
kpeter@895
   185
    };
kpeter@862
   186
kpeter@601
   187
  private:
kpeter@601
   188
kpeter@605
   189
    // Data related to the underlying digraph
kpeter@605
   190
    const GR &_graph;
kpeter@605
   191
    int _node_num;
kpeter@605
   192
    int _arc_num;
kpeter@663
   193
    int _all_arc_num;
kpeter@663
   194
    int _search_arc_num;
kpeter@605
   195
kpeter@605
   196
    // Parameters of the problem
kpeter@642
   197
    bool _have_lower;
kpeter@640
   198
    SupplyType _stype;
kpeter@641
   199
    Value _sum_supply;
kpeter@601
   200
kpeter@605
   201
    // Data structures for storing the digraph
kpeter@603
   202
    IntNodeMap _node_id;
kpeter@642
   203
    IntArcMap _arc_id;
kpeter@603
   204
    IntVector _source;
kpeter@603
   205
    IntVector _target;
kpeter@830
   206
    bool _arc_mixing;
kpeter@603
   207
kpeter@605
   208
    // Node and arc data
kpeter@642
   209
    ValueVector _lower;
kpeter@642
   210
    ValueVector _upper;
kpeter@642
   211
    ValueVector _cap;
kpeter@607
   212
    CostVector _cost;
kpeter@642
   213
    ValueVector _supply;
kpeter@642
   214
    ValueVector _flow;
kpeter@607
   215
    CostVector _pi;
kpeter@601
   216
kpeter@603
   217
    // Data for storing the spanning tree structure
kpeter@601
   218
    IntVector _parent;
kpeter@601
   219
    IntVector _pred;
kpeter@601
   220
    IntVector _thread;
kpeter@604
   221
    IntVector _rev_thread;
kpeter@604
   222
    IntVector _succ_num;
kpeter@604
   223
    IntVector _last_succ;
kpeter@895
   224
    CharVector _pred_dir;
kpeter@895
   225
    CharVector _state;
kpeter@604
   226
    IntVector _dirty_revs;
kpeter@601
   227
    int _root;
kpeter@601
   228
kpeter@601
   229
    // Temporary data used in the current pivot iteration
kpeter@603
   230
    int in_arc, join, u_in, v_in, u_out, v_out;
kpeter@641
   231
    Value delta;
kpeter@601
   232
kpeter@811
   233
    const Value MAX;
kpeter@663
   234
kpeter@640
   235
  public:
alpar@877
   236
kpeter@640
   237
    /// \brief Constant for infinite upper bounds (capacities).
kpeter@640
   238
    ///
kpeter@640
   239
    /// Constant for infinite upper bounds (capacities).
kpeter@641
   240
    /// It is \c std::numeric_limits<Value>::infinity() if available,
kpeter@641
   241
    /// \c std::numeric_limits<Value>::max() otherwise.
kpeter@641
   242
    const Value INF;
kpeter@640
   243
kpeter@601
   244
  private:
kpeter@601
   245
kpeter@605
   246
    // Implementation of the First Eligible pivot rule
kpeter@601
   247
    class FirstEligiblePivotRule
kpeter@601
   248
    {
kpeter@601
   249
    private:
kpeter@601
   250
kpeter@601
   251
      // References to the NetworkSimplex class
kpeter@601
   252
      const IntVector  &_source;
kpeter@601
   253
      const IntVector  &_target;
kpeter@607
   254
      const CostVector &_cost;
kpeter@895
   255
      const CharVector &_state;
kpeter@607
   256
      const CostVector &_pi;
kpeter@601
   257
      int &_in_arc;
kpeter@663
   258
      int _search_arc_num;
kpeter@601
   259
kpeter@601
   260
      // Pivot rule data
kpeter@601
   261
      int _next_arc;
kpeter@601
   262
kpeter@601
   263
    public:
kpeter@601
   264
kpeter@605
   265
      // Constructor
kpeter@601
   266
      FirstEligiblePivotRule(NetworkSimplex &ns) :
kpeter@603
   267
        _source(ns._source), _target(ns._target),
kpeter@601
   268
        _cost(ns._cost), _state(ns._state), _pi(ns._pi),
kpeter@663
   269
        _in_arc(ns.in_arc), _search_arc_num(ns._search_arc_num),
kpeter@663
   270
        _next_arc(0)
kpeter@601
   271
      {}
kpeter@601
   272
kpeter@605
   273
      // Find next entering arc
kpeter@601
   274
      bool findEnteringArc() {
kpeter@607
   275
        Cost c;
kpeter@839
   276
        for (int e = _next_arc; e != _search_arc_num; ++e) {
kpeter@601
   277
          c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@601
   278
          if (c < 0) {
kpeter@601
   279
            _in_arc = e;
kpeter@601
   280
            _next_arc = e + 1;
kpeter@601
   281
            return true;
kpeter@601
   282
          }
kpeter@601
   283
        }
kpeter@839
   284
        for (int e = 0; e != _next_arc; ++e) {
kpeter@601
   285
          c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@601
   286
          if (c < 0) {
kpeter@601
   287
            _in_arc = e;
kpeter@601
   288
            _next_arc = e + 1;
kpeter@601
   289
            return true;
kpeter@601
   290
          }
kpeter@601
   291
        }
kpeter@601
   292
        return false;
kpeter@601
   293
      }
kpeter@601
   294
kpeter@601
   295
    }; //class FirstEligiblePivotRule
kpeter@601
   296
kpeter@601
   297
kpeter@605
   298
    // Implementation of the Best Eligible pivot rule
kpeter@601
   299
    class BestEligiblePivotRule
kpeter@601
   300
    {
kpeter@601
   301
    private:
kpeter@601
   302
kpeter@601
   303
      // References to the NetworkSimplex class
kpeter@601
   304
      const IntVector  &_source;
kpeter@601
   305
      const IntVector  &_target;
kpeter@607
   306
      const CostVector &_cost;
kpeter@895
   307
      const CharVector &_state;
kpeter@607
   308
      const CostVector &_pi;
kpeter@601
   309
      int &_in_arc;
kpeter@663
   310
      int _search_arc_num;
kpeter@601
   311
kpeter@601
   312
    public:
kpeter@601
   313
kpeter@605
   314
      // Constructor
kpeter@601
   315
      BestEligiblePivotRule(NetworkSimplex &ns) :
kpeter@603
   316
        _source(ns._source), _target(ns._target),
kpeter@601
   317
        _cost(ns._cost), _state(ns._state), _pi(ns._pi),
kpeter@663
   318
        _in_arc(ns.in_arc), _search_arc_num(ns._search_arc_num)
kpeter@601
   319
      {}
kpeter@601
   320
kpeter@605
   321
      // Find next entering arc
kpeter@601
   322
      bool findEnteringArc() {
kpeter@607
   323
        Cost c, min = 0;
kpeter@839
   324
        for (int e = 0; e != _search_arc_num; ++e) {
kpeter@601
   325
          c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@601
   326
          if (c < min) {
kpeter@601
   327
            min = c;
kpeter@601
   328
            _in_arc = e;
kpeter@601
   329
          }
kpeter@601
   330
        }
kpeter@601
   331
        return min < 0;
kpeter@601
   332
      }
kpeter@601
   333
kpeter@601
   334
    }; //class BestEligiblePivotRule
kpeter@601
   335
kpeter@601
   336
kpeter@605
   337
    // Implementation of the Block Search pivot rule
kpeter@601
   338
    class BlockSearchPivotRule
kpeter@601
   339
    {
kpeter@601
   340
    private:
kpeter@601
   341
kpeter@601
   342
      // References to the NetworkSimplex class
kpeter@601
   343
      const IntVector  &_source;
kpeter@601
   344
      const IntVector  &_target;
kpeter@607
   345
      const CostVector &_cost;
kpeter@895
   346
      const CharVector &_state;
kpeter@607
   347
      const CostVector &_pi;
kpeter@601
   348
      int &_in_arc;
kpeter@663
   349
      int _search_arc_num;
kpeter@601
   350
kpeter@601
   351
      // Pivot rule data
kpeter@601
   352
      int _block_size;
kpeter@601
   353
      int _next_arc;
kpeter@601
   354
kpeter@601
   355
    public:
kpeter@601
   356
kpeter@605
   357
      // Constructor
kpeter@601
   358
      BlockSearchPivotRule(NetworkSimplex &ns) :
kpeter@603
   359
        _source(ns._source), _target(ns._target),
kpeter@601
   360
        _cost(ns._cost), _state(ns._state), _pi(ns._pi),
kpeter@663
   361
        _in_arc(ns.in_arc), _search_arc_num(ns._search_arc_num),
kpeter@663
   362
        _next_arc(0)
kpeter@601
   363
      {
kpeter@601
   364
        // The main parameters of the pivot rule
kpeter@839
   365
        const double BLOCK_SIZE_FACTOR = 1.0;
kpeter@601
   366
        const int MIN_BLOCK_SIZE = 10;
kpeter@601
   367
alpar@612
   368
        _block_size = std::max( int(BLOCK_SIZE_FACTOR *
kpeter@663
   369
                                    std::sqrt(double(_search_arc_num))),
kpeter@601
   370
                                MIN_BLOCK_SIZE );
kpeter@601
   371
      }
kpeter@601
   372
kpeter@605
   373
      // Find next entering arc
kpeter@601
   374
      bool findEnteringArc() {
kpeter@607
   375
        Cost c, min = 0;
kpeter@601
   376
        int cnt = _block_size;
kpeter@727
   377
        int e;
kpeter@839
   378
        for (e = _next_arc; e != _search_arc_num; ++e) {
kpeter@601
   379
          c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@601
   380
          if (c < min) {
kpeter@601
   381
            min = c;
kpeter@727
   382
            _in_arc = e;
kpeter@601
   383
          }
kpeter@601
   384
          if (--cnt == 0) {
kpeter@727
   385
            if (min < 0) goto search_end;
kpeter@601
   386
            cnt = _block_size;
kpeter@601
   387
          }
kpeter@601
   388
        }
kpeter@839
   389
        for (e = 0; e != _next_arc; ++e) {
kpeter@727
   390
          c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@727
   391
          if (c < min) {
kpeter@727
   392
            min = c;
kpeter@727
   393
            _in_arc = e;
kpeter@727
   394
          }
kpeter@727
   395
          if (--cnt == 0) {
kpeter@727
   396
            if (min < 0) goto search_end;
kpeter@727
   397
            cnt = _block_size;
kpeter@601
   398
          }
kpeter@601
   399
        }
kpeter@601
   400
        if (min >= 0) return false;
kpeter@727
   401
kpeter@727
   402
      search_end:
kpeter@601
   403
        _next_arc = e;
kpeter@601
   404
        return true;
kpeter@601
   405
      }
kpeter@601
   406
kpeter@601
   407
    }; //class BlockSearchPivotRule
kpeter@601
   408
kpeter@601
   409
kpeter@605
   410
    // Implementation of the Candidate List pivot rule
kpeter@601
   411
    class CandidateListPivotRule
kpeter@601
   412
    {
kpeter@601
   413
    private:
kpeter@601
   414
kpeter@601
   415
      // References to the NetworkSimplex class
kpeter@601
   416
      const IntVector  &_source;
kpeter@601
   417
      const IntVector  &_target;
kpeter@607
   418
      const CostVector &_cost;
kpeter@895
   419
      const CharVector &_state;
kpeter@607
   420
      const CostVector &_pi;
kpeter@601
   421
      int &_in_arc;
kpeter@663
   422
      int _search_arc_num;
kpeter@601
   423
kpeter@601
   424
      // Pivot rule data
kpeter@601
   425
      IntVector _candidates;
kpeter@601
   426
      int _list_length, _minor_limit;
kpeter@601
   427
      int _curr_length, _minor_count;
kpeter@601
   428
      int _next_arc;
kpeter@601
   429
kpeter@601
   430
    public:
kpeter@601
   431
kpeter@601
   432
      /// Constructor
kpeter@601
   433
      CandidateListPivotRule(NetworkSimplex &ns) :
kpeter@603
   434
        _source(ns._source), _target(ns._target),
kpeter@601
   435
        _cost(ns._cost), _state(ns._state), _pi(ns._pi),
kpeter@663
   436
        _in_arc(ns.in_arc), _search_arc_num(ns._search_arc_num),
kpeter@663
   437
        _next_arc(0)
kpeter@601
   438
      {
kpeter@601
   439
        // The main parameters of the pivot rule
kpeter@727
   440
        const double LIST_LENGTH_FACTOR = 0.25;
kpeter@601
   441
        const int MIN_LIST_LENGTH = 10;
kpeter@601
   442
        const double MINOR_LIMIT_FACTOR = 0.1;
kpeter@601
   443
        const int MIN_MINOR_LIMIT = 3;
kpeter@601
   444
alpar@612
   445
        _list_length = std::max( int(LIST_LENGTH_FACTOR *
kpeter@663
   446
                                     std::sqrt(double(_search_arc_num))),
kpeter@601
   447
                                 MIN_LIST_LENGTH );
kpeter@601
   448
        _minor_limit = std::max( int(MINOR_LIMIT_FACTOR * _list_length),
kpeter@601
   449
                                 MIN_MINOR_LIMIT );
kpeter@601
   450
        _curr_length = _minor_count = 0;
kpeter@601
   451
        _candidates.resize(_list_length);
kpeter@601
   452
      }
kpeter@601
   453
kpeter@601
   454
      /// Find next entering arc
kpeter@601
   455
      bool findEnteringArc() {
kpeter@607
   456
        Cost min, c;
kpeter@727
   457
        int e;
kpeter@601
   458
        if (_curr_length > 0 && _minor_count < _minor_limit) {
kpeter@601
   459
          // Minor iteration: select the best eligible arc from the
kpeter@601
   460
          // current candidate list
kpeter@601
   461
          ++_minor_count;
kpeter@601
   462
          min = 0;
kpeter@601
   463
          for (int i = 0; i < _curr_length; ++i) {
kpeter@601
   464
            e = _candidates[i];
kpeter@601
   465
            c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@601
   466
            if (c < min) {
kpeter@601
   467
              min = c;
kpeter@727
   468
              _in_arc = e;
kpeter@601
   469
            }
kpeter@727
   470
            else if (c >= 0) {
kpeter@601
   471
              _candidates[i--] = _candidates[--_curr_length];
kpeter@601
   472
            }
kpeter@601
   473
          }
kpeter@727
   474
          if (min < 0) return true;
kpeter@601
   475
        }
kpeter@601
   476
kpeter@601
   477
        // Major iteration: build a new candidate list
kpeter@601
   478
        min = 0;
kpeter@601
   479
        _curr_length = 0;
kpeter@839
   480
        for (e = _next_arc; e != _search_arc_num; ++e) {
kpeter@601
   481
          c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@601
   482
          if (c < 0) {
kpeter@601
   483
            _candidates[_curr_length++] = e;
kpeter@601
   484
            if (c < min) {
kpeter@601
   485
              min = c;
kpeter@727
   486
              _in_arc = e;
kpeter@601
   487
            }
kpeter@727
   488
            if (_curr_length == _list_length) goto search_end;
kpeter@601
   489
          }
kpeter@601
   490
        }
kpeter@839
   491
        for (e = 0; e != _next_arc; ++e) {
kpeter@727
   492
          c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@727
   493
          if (c < 0) {
kpeter@727
   494
            _candidates[_curr_length++] = e;
kpeter@727
   495
            if (c < min) {
kpeter@727
   496
              min = c;
kpeter@727
   497
              _in_arc = e;
kpeter@601
   498
            }
kpeter@727
   499
            if (_curr_length == _list_length) goto search_end;
kpeter@601
   500
          }
kpeter@601
   501
        }
kpeter@601
   502
        if (_curr_length == 0) return false;
alpar@877
   503
alpar@877
   504
      search_end:
kpeter@601
   505
        _minor_count = 1;
kpeter@601
   506
        _next_arc = e;
kpeter@601
   507
        return true;
kpeter@601
   508
      }
kpeter@601
   509
kpeter@601
   510
    }; //class CandidateListPivotRule
kpeter@601
   511
kpeter@601
   512
kpeter@605
   513
    // Implementation of the Altering Candidate List pivot rule
kpeter@601
   514
    class AlteringListPivotRule
kpeter@601
   515
    {
kpeter@601
   516
    private:
kpeter@601
   517
kpeter@601
   518
      // References to the NetworkSimplex class
kpeter@601
   519
      const IntVector  &_source;
kpeter@601
   520
      const IntVector  &_target;
kpeter@607
   521
      const CostVector &_cost;
kpeter@895
   522
      const CharVector &_state;
kpeter@607
   523
      const CostVector &_pi;
kpeter@601
   524
      int &_in_arc;
kpeter@663
   525
      int _search_arc_num;
kpeter@601
   526
kpeter@601
   527
      // Pivot rule data
kpeter@601
   528
      int _block_size, _head_length, _curr_length;
kpeter@601
   529
      int _next_arc;
kpeter@601
   530
      IntVector _candidates;
kpeter@607
   531
      CostVector _cand_cost;
kpeter@601
   532
kpeter@601
   533
      // Functor class to compare arcs during sort of the candidate list
kpeter@601
   534
      class SortFunc
kpeter@601
   535
      {
kpeter@601
   536
      private:
kpeter@607
   537
        const CostVector &_map;
kpeter@601
   538
      public:
kpeter@607
   539
        SortFunc(const CostVector &map) : _map(map) {}
kpeter@601
   540
        bool operator()(int left, int right) {
kpeter@601
   541
          return _map[left] > _map[right];
kpeter@601
   542
        }
kpeter@601
   543
      };
kpeter@601
   544
kpeter@601
   545
      SortFunc _sort_func;
kpeter@601
   546
kpeter@601
   547
    public:
kpeter@601
   548
kpeter@605
   549
      // Constructor
kpeter@601
   550
      AlteringListPivotRule(NetworkSimplex &ns) :
kpeter@603
   551
        _source(ns._source), _target(ns._target),
kpeter@601
   552
        _cost(ns._cost), _state(ns._state), _pi(ns._pi),
kpeter@663
   553
        _in_arc(ns.in_arc), _search_arc_num(ns._search_arc_num),
kpeter@663
   554
        _next_arc(0), _cand_cost(ns._search_arc_num), _sort_func(_cand_cost)
kpeter@601
   555
      {
kpeter@601
   556
        // The main parameters of the pivot rule
kpeter@727
   557
        const double BLOCK_SIZE_FACTOR = 1.0;
kpeter@601
   558
        const int MIN_BLOCK_SIZE = 10;
kpeter@601
   559
        const double HEAD_LENGTH_FACTOR = 0.1;
kpeter@601
   560
        const int MIN_HEAD_LENGTH = 3;
kpeter@601
   561
alpar@612
   562
        _block_size = std::max( int(BLOCK_SIZE_FACTOR *
kpeter@663
   563
                                    std::sqrt(double(_search_arc_num))),
kpeter@601
   564
                                MIN_BLOCK_SIZE );
kpeter@601
   565
        _head_length = std::max( int(HEAD_LENGTH_FACTOR * _block_size),
kpeter@601
   566
                                 MIN_HEAD_LENGTH );
kpeter@601
   567
        _candidates.resize(_head_length + _block_size);
kpeter@601
   568
        _curr_length = 0;
kpeter@601
   569
      }
kpeter@601
   570
kpeter@605
   571
      // Find next entering arc
kpeter@601
   572
      bool findEnteringArc() {
kpeter@601
   573
        // Check the current candidate list
kpeter@601
   574
        int e;
kpeter@895
   575
        Cost c;
kpeter@839
   576
        for (int i = 0; i != _curr_length; ++i) {
kpeter@601
   577
          e = _candidates[i];
kpeter@895
   578
          c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@895
   579
          if (c < 0) {
kpeter@895
   580
            _cand_cost[e] = c;
kpeter@895
   581
          } else {
kpeter@601
   582
            _candidates[i--] = _candidates[--_curr_length];
kpeter@601
   583
          }
kpeter@601
   584
        }
kpeter@601
   585
kpeter@601
   586
        // Extend the list
kpeter@601
   587
        int cnt = _block_size;
kpeter@601
   588
        int limit = _head_length;
kpeter@601
   589
kpeter@839
   590
        for (e = _next_arc; e != _search_arc_num; ++e) {
kpeter@895
   591
          c = _state[e] * (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@895
   592
          if (c < 0) {
kpeter@895
   593
            _cand_cost[e] = c;
kpeter@601
   594
            _candidates[_curr_length++] = e;
kpeter@601
   595
          }
kpeter@601
   596
          if (--cnt == 0) {
kpeter@727
   597
            if (_curr_length > limit) goto search_end;
kpeter@601
   598
            limit = 0;
kpeter@601
   599
            cnt = _block_size;
kpeter@601
   600
          }
kpeter@601
   601
        }
kpeter@839
   602
        for (e = 0; e != _next_arc; ++e) {
kpeter@727
   603
          _cand_cost[e] = _state[e] *
kpeter@727
   604
            (_cost[e] + _pi[_source[e]] - _pi[_target[e]]);
kpeter@727
   605
          if (_cand_cost[e] < 0) {
kpeter@727
   606
            _candidates[_curr_length++] = e;
kpeter@727
   607
          }
kpeter@727
   608
          if (--cnt == 0) {
kpeter@727
   609
            if (_curr_length > limit) goto search_end;
kpeter@727
   610
            limit = 0;
kpeter@727
   611
            cnt = _block_size;
kpeter@601
   612
          }
kpeter@601
   613
        }
kpeter@601
   614
        if (_curr_length == 0) return false;
alpar@877
   615
kpeter@727
   616
      search_end:
kpeter@601
   617
kpeter@601
   618
        // Make heap of the candidate list (approximating a partial sort)
kpeter@601
   619
        make_heap( _candidates.begin(), _candidates.begin() + _curr_length,
kpeter@601
   620
                   _sort_func );
kpeter@601
   621
kpeter@601
   622
        // Pop the first element of the heap
kpeter@601
   623
        _in_arc = _candidates[0];
kpeter@727
   624
        _next_arc = e;
kpeter@601
   625
        pop_heap( _candidates.begin(), _candidates.begin() + _curr_length,
kpeter@601
   626
                  _sort_func );
kpeter@601
   627
        _curr_length = std::min(_head_length, _curr_length - 1);
kpeter@601
   628
        return true;
kpeter@601
   629
      }
kpeter@601
   630
kpeter@601
   631
    }; //class AlteringListPivotRule
kpeter@601
   632
kpeter@601
   633
  public:
kpeter@601
   634
kpeter@605
   635
    /// \brief Constructor.
kpeter@601
   636
    ///
kpeter@609
   637
    /// The constructor of the class.
kpeter@601
   638
    ///
kpeter@603
   639
    /// \param graph The digraph the algorithm runs on.
kpeter@896
   640
    /// \param arc_mixing Indicate if the arcs will be stored in a
alpar@877
   641
    /// mixed order in the internal data structure.
kpeter@896
   642
    /// In general, it leads to similar performance as using the original
kpeter@896
   643
    /// arc order, but it makes the algorithm more robust and in special
kpeter@896
   644
    /// cases, even significantly faster. Therefore, it is enabled by default.
kpeter@896
   645
    NetworkSimplex(const GR& graph, bool arc_mixing = true) :
kpeter@642
   646
      _graph(graph), _node_id(graph), _arc_id(graph),
kpeter@830
   647
      _arc_mixing(arc_mixing),
kpeter@811
   648
      MAX(std::numeric_limits<Value>::max()),
kpeter@641
   649
      INF(std::numeric_limits<Value>::has_infinity ?
kpeter@811
   650
          std::numeric_limits<Value>::infinity() : MAX)
kpeter@605
   651
    {
kpeter@812
   652
      // Check the number types
kpeter@641
   653
      LEMON_ASSERT(std::numeric_limits<Value>::is_signed,
kpeter@640
   654
        "The flow type of NetworkSimplex must be signed");
kpeter@640
   655
      LEMON_ASSERT(std::numeric_limits<Cost>::is_signed,
kpeter@640
   656
        "The cost type of NetworkSimplex must be signed");
kpeter@601
   657
kpeter@830
   658
      // Reset data structures
kpeter@729
   659
      reset();
kpeter@601
   660
    }
kpeter@601
   661
kpeter@609
   662
    /// \name Parameters
kpeter@609
   663
    /// The parameters of the algorithm can be specified using these
kpeter@609
   664
    /// functions.
kpeter@609
   665
kpeter@609
   666
    /// @{
kpeter@609
   667
kpeter@605
   668
    /// \brief Set the lower bounds on the arcs.
kpeter@605
   669
    ///
kpeter@605
   670
    /// This function sets the lower bounds on the arcs.
kpeter@640
   671
    /// If it is not used before calling \ref run(), the lower bounds
kpeter@640
   672
    /// will be set to zero on all arcs.
kpeter@605
   673
    ///
kpeter@605
   674
    /// \param map An arc map storing the lower bounds.
kpeter@641
   675
    /// Its \c Value type must be convertible to the \c Value type
kpeter@605
   676
    /// of the algorithm.
kpeter@605
   677
    ///
kpeter@605
   678
    /// \return <tt>(*this)</tt>
kpeter@640
   679
    template <typename LowerMap>
kpeter@640
   680
    NetworkSimplex& lowerMap(const LowerMap& map) {
kpeter@642
   681
      _have_lower = true;
kpeter@605
   682
      for (ArcIt a(_graph); a != INVALID; ++a) {
kpeter@642
   683
        _lower[_arc_id[a]] = map[a];
kpeter@605
   684
      }
kpeter@605
   685
      return *this;
kpeter@605
   686
    }
kpeter@605
   687
kpeter@605
   688
    /// \brief Set the upper bounds (capacities) on the arcs.
kpeter@605
   689
    ///
kpeter@605
   690
    /// This function sets the upper bounds (capacities) on the arcs.
kpeter@640
   691
    /// If it is not used before calling \ref run(), the upper bounds
kpeter@640
   692
    /// will be set to \ref INF on all arcs (i.e. the flow value will be
kpeter@812
   693
    /// unbounded from above).
kpeter@605
   694
    ///
kpeter@605
   695
    /// \param map An arc map storing the upper bounds.
kpeter@641
   696
    /// Its \c Value type must be convertible to the \c Value type
kpeter@605
   697
    /// of the algorithm.
kpeter@605
   698
    ///
kpeter@605
   699
    /// \return <tt>(*this)</tt>
kpeter@640
   700
    template<typename UpperMap>
kpeter@640
   701
    NetworkSimplex& upperMap(const UpperMap& map) {
kpeter@605
   702
      for (ArcIt a(_graph); a != INVALID; ++a) {
kpeter@642
   703
        _upper[_arc_id[a]] = map[a];
kpeter@605
   704
      }
kpeter@605
   705
      return *this;
kpeter@605
   706
    }
kpeter@605
   707
kpeter@605
   708
    /// \brief Set the costs of the arcs.
kpeter@605
   709
    ///
kpeter@605
   710
    /// This function sets the costs of the arcs.
kpeter@605
   711
    /// If it is not used before calling \ref run(), the costs
kpeter@605
   712
    /// will be set to \c 1 on all arcs.
kpeter@605
   713
    ///
kpeter@605
   714
    /// \param map An arc map storing the costs.
kpeter@607
   715
    /// Its \c Value type must be convertible to the \c Cost type
kpeter@605
   716
    /// of the algorithm.
kpeter@605
   717
    ///
kpeter@605
   718
    /// \return <tt>(*this)</tt>
kpeter@640
   719
    template<typename CostMap>
kpeter@640
   720
    NetworkSimplex& costMap(const CostMap& map) {
kpeter@605
   721
      for (ArcIt a(_graph); a != INVALID; ++a) {
kpeter@642
   722
        _cost[_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 supply values of the nodes.
kpeter@605
   728
    ///
kpeter@605
   729
    /// This function sets the supply values of the nodes.
kpeter@605
   730
    /// If neither this function nor \ref stSupply() is used before
kpeter@605
   731
    /// calling \ref run(), the supply of each node will be set to zero.
kpeter@605
   732
    ///
kpeter@605
   733
    /// \param map A node map storing the supply values.
kpeter@641
   734
    /// Its \c Value type must be convertible to the \c Value type
kpeter@605
   735
    /// of the algorithm.
kpeter@605
   736
    ///
kpeter@605
   737
    /// \return <tt>(*this)</tt>
kpeter@919
   738
    ///
kpeter@919
   739
    /// \sa supplyType()
kpeter@640
   740
    template<typename SupplyMap>
kpeter@640
   741
    NetworkSimplex& supplyMap(const SupplyMap& map) {
kpeter@605
   742
      for (NodeIt n(_graph); n != INVALID; ++n) {
kpeter@642
   743
        _supply[_node_id[n]] = map[n];
kpeter@605
   744
      }
kpeter@605
   745
      return *this;
kpeter@605
   746
    }
kpeter@605
   747
kpeter@605
   748
    /// \brief Set single source and target nodes and a supply value.
kpeter@605
   749
    ///
kpeter@605
   750
    /// This function sets a single source node and a single target node
kpeter@605
   751
    /// and the required flow value.
kpeter@605
   752
    /// If neither this function nor \ref supplyMap() is used before
kpeter@605
   753
    /// calling \ref run(), the supply of each node will be set to zero.
kpeter@605
   754
    ///
kpeter@640
   755
    /// Using this function has the same effect as using \ref supplyMap()
kpeter@919
   756
    /// with a map in which \c k is assigned to \c s, \c -k is
kpeter@640
   757
    /// assigned to \c t and all other nodes have zero supply value.
kpeter@640
   758
    ///
kpeter@605
   759
    /// \param s The source node.
kpeter@605
   760
    /// \param t The target node.
kpeter@605
   761
    /// \param k The required amount of flow from node \c s to node \c t
kpeter@605
   762
    /// (i.e. the supply of \c s and the demand of \c t).
kpeter@605
   763
    ///
kpeter@605
   764
    /// \return <tt>(*this)</tt>
kpeter@641
   765
    NetworkSimplex& stSupply(const Node& s, const Node& t, Value k) {
kpeter@642
   766
      for (int i = 0; i != _node_num; ++i) {
kpeter@642
   767
        _supply[i] = 0;
kpeter@642
   768
      }
kpeter@642
   769
      _supply[_node_id[s]] =  k;
kpeter@642
   770
      _supply[_node_id[t]] = -k;
kpeter@605
   771
      return *this;
kpeter@605
   772
    }
alpar@877
   773
kpeter@640
   774
    /// \brief Set the type of the supply constraints.
kpeter@609
   775
    ///
kpeter@640
   776
    /// This function sets the type of the supply/demand constraints.
kpeter@640
   777
    /// If it is not used before calling \ref run(), the \ref GEQ supply
kpeter@609
   778
    /// type will be used.
kpeter@609
   779
    ///
kpeter@786
   780
    /// For more information, see \ref SupplyType.
kpeter@609
   781
    ///
kpeter@609
   782
    /// \return <tt>(*this)</tt>
kpeter@640
   783
    NetworkSimplex& supplyType(SupplyType supply_type) {
kpeter@640
   784
      _stype = supply_type;
kpeter@609
   785
      return *this;
kpeter@609
   786
    }
kpeter@605
   787
kpeter@609
   788
    /// @}
kpeter@601
   789
kpeter@605
   790
    /// \name Execution Control
kpeter@605
   791
    /// The algorithm can be executed using \ref run().
kpeter@605
   792
kpeter@601
   793
    /// @{
kpeter@601
   794
kpeter@601
   795
    /// \brief Run the algorithm.
kpeter@601
   796
    ///
kpeter@601
   797
    /// This function runs the algorithm.
kpeter@609
   798
    /// The paramters can be specified using functions \ref lowerMap(),
alpar@877
   799
    /// \ref upperMap(), \ref costMap(), \ref supplyMap(), \ref stSupply(),
kpeter@642
   800
    /// \ref supplyType().
kpeter@609
   801
    /// For example,
kpeter@605
   802
    /// \code
kpeter@605
   803
    ///   NetworkSimplex<ListDigraph> ns(graph);
kpeter@640
   804
    ///   ns.lowerMap(lower).upperMap(upper).costMap(cost)
kpeter@605
   805
    ///     .supplyMap(sup).run();
kpeter@605
   806
    /// \endcode
kpeter@601
   807
    ///
kpeter@830
   808
    /// This function can be called more than once. All the given parameters
kpeter@830
   809
    /// are kept for the next call, unless \ref resetParams() or \ref reset()
kpeter@830
   810
    /// is used, thus only the modified parameters have to be set again.
kpeter@830
   811
    /// If the underlying digraph was also modified after the construction
kpeter@830
   812
    /// of the class (or the last \ref reset() call), then the \ref reset()
kpeter@830
   813
    /// function must be called.
kpeter@606
   814
    ///
kpeter@605
   815
    /// \param pivot_rule The pivot rule that will be used during the
kpeter@786
   816
    /// algorithm. For more information, see \ref PivotRule.
kpeter@601
   817
    ///
kpeter@640
   818
    /// \return \c INFEASIBLE if no feasible flow exists,
kpeter@640
   819
    /// \n \c OPTIMAL if the problem has optimal solution
kpeter@640
   820
    /// (i.e. it is feasible and bounded), and the algorithm has found
kpeter@640
   821
    /// optimal flow and node potentials (primal and dual solutions),
kpeter@640
   822
    /// \n \c UNBOUNDED if the objective function of the problem is
kpeter@640
   823
    /// unbounded, i.e. there is a directed cycle having negative total
kpeter@640
   824
    /// cost and infinite upper bound.
kpeter@640
   825
    ///
kpeter@640
   826
    /// \see ProblemType, PivotRule
kpeter@830
   827
    /// \see resetParams(), reset()
kpeter@640
   828
    ProblemType run(PivotRule pivot_rule = BLOCK_SEARCH) {
kpeter@640
   829
      if (!init()) return INFEASIBLE;
kpeter@640
   830
      return start(pivot_rule);
kpeter@601
   831
    }
kpeter@601
   832
kpeter@606
   833
    /// \brief Reset all the parameters that have been given before.
kpeter@606
   834
    ///
kpeter@606
   835
    /// This function resets all the paramaters that have been given
kpeter@609
   836
    /// before using functions \ref lowerMap(), \ref upperMap(),
kpeter@642
   837
    /// \ref costMap(), \ref supplyMap(), \ref stSupply(), \ref supplyType().
kpeter@606
   838
    ///
kpeter@830
   839
    /// It is useful for multiple \ref run() calls. Basically, all the given
kpeter@830
   840
    /// parameters are kept for the next \ref run() call, unless
kpeter@830
   841
    /// \ref resetParams() or \ref reset() is used.
kpeter@830
   842
    /// If the underlying digraph was also modified after the construction
kpeter@830
   843
    /// of the class or the last \ref reset() call, then the \ref reset()
kpeter@830
   844
    /// function must be used, otherwise \ref resetParams() is sufficient.
kpeter@606
   845
    ///
kpeter@606
   846
    /// For example,
kpeter@606
   847
    /// \code
kpeter@606
   848
    ///   NetworkSimplex<ListDigraph> ns(graph);
kpeter@606
   849
    ///
kpeter@606
   850
    ///   // First run
kpeter@640
   851
    ///   ns.lowerMap(lower).upperMap(upper).costMap(cost)
kpeter@606
   852
    ///     .supplyMap(sup).run();
kpeter@606
   853
    ///
kpeter@830
   854
    ///   // Run again with modified cost map (resetParams() is not called,
kpeter@606
   855
    ///   // so only the cost map have to be set again)
kpeter@606
   856
    ///   cost[e] += 100;
kpeter@606
   857
    ///   ns.costMap(cost).run();
kpeter@606
   858
    ///
kpeter@830
   859
    ///   // Run again from scratch using resetParams()
kpeter@606
   860
    ///   // (the lower bounds will be set to zero on all arcs)
kpeter@830
   861
    ///   ns.resetParams();
kpeter@640
   862
    ///   ns.upperMap(capacity).costMap(cost)
kpeter@606
   863
    ///     .supplyMap(sup).run();
kpeter@606
   864
    /// \endcode
kpeter@606
   865
    ///
kpeter@606
   866
    /// \return <tt>(*this)</tt>
kpeter@830
   867
    ///
kpeter@830
   868
    /// \see reset(), run()
kpeter@830
   869
    NetworkSimplex& resetParams() {
kpeter@642
   870
      for (int i = 0; i != _node_num; ++i) {
kpeter@642
   871
        _supply[i] = 0;
kpeter@642
   872
      }
kpeter@642
   873
      for (int i = 0; i != _arc_num; ++i) {
kpeter@642
   874
        _lower[i] = 0;
kpeter@642
   875
        _upper[i] = INF;
kpeter@642
   876
        _cost[i] = 1;
kpeter@642
   877
      }
kpeter@642
   878
      _have_lower = false;
kpeter@640
   879
      _stype = GEQ;
kpeter@606
   880
      return *this;
kpeter@606
   881
    }
kpeter@606
   882
kpeter@830
   883
    /// \brief Reset the internal data structures and all the parameters
kpeter@830
   884
    /// that have been given before.
kpeter@830
   885
    ///
kpeter@830
   886
    /// This function resets the internal data structures and all the
kpeter@830
   887
    /// paramaters that have been given before using functions \ref lowerMap(),
kpeter@830
   888
    /// \ref upperMap(), \ref costMap(), \ref supplyMap(), \ref stSupply(),
kpeter@830
   889
    /// \ref supplyType().
kpeter@830
   890
    ///
kpeter@830
   891
    /// It is useful for multiple \ref run() calls. Basically, all the given
kpeter@830
   892
    /// parameters are kept for the next \ref run() call, unless
kpeter@830
   893
    /// \ref resetParams() or \ref reset() is used.
kpeter@830
   894
    /// If the underlying digraph was also modified after the construction
kpeter@830
   895
    /// of the class or the last \ref reset() call, then the \ref reset()
kpeter@830
   896
    /// function must be used, otherwise \ref resetParams() is sufficient.
kpeter@830
   897
    ///
kpeter@830
   898
    /// See \ref resetParams() for examples.
kpeter@830
   899
    ///
kpeter@830
   900
    /// \return <tt>(*this)</tt>
kpeter@830
   901
    ///
kpeter@830
   902
    /// \see resetParams(), run()
kpeter@830
   903
    NetworkSimplex& reset() {
kpeter@830
   904
      // Resize vectors
kpeter@830
   905
      _node_num = countNodes(_graph);
kpeter@830
   906
      _arc_num = countArcs(_graph);
kpeter@830
   907
      int all_node_num = _node_num + 1;
kpeter@830
   908
      int max_arc_num = _arc_num + 2 * _node_num;
kpeter@830
   909
kpeter@830
   910
      _source.resize(max_arc_num);
kpeter@830
   911
      _target.resize(max_arc_num);
kpeter@830
   912
kpeter@830
   913
      _lower.resize(_arc_num);
kpeter@830
   914
      _upper.resize(_arc_num);
kpeter@830
   915
      _cap.resize(max_arc_num);
kpeter@830
   916
      _cost.resize(max_arc_num);
kpeter@830
   917
      _supply.resize(all_node_num);
kpeter@830
   918
      _flow.resize(max_arc_num);
kpeter@830
   919
      _pi.resize(all_node_num);
kpeter@830
   920
kpeter@830
   921
      _parent.resize(all_node_num);
kpeter@830
   922
      _pred.resize(all_node_num);
kpeter@895
   923
      _pred_dir.resize(all_node_num);
kpeter@830
   924
      _thread.resize(all_node_num);
kpeter@830
   925
      _rev_thread.resize(all_node_num);
kpeter@830
   926
      _succ_num.resize(all_node_num);
kpeter@830
   927
      _last_succ.resize(all_node_num);
kpeter@830
   928
      _state.resize(max_arc_num);
kpeter@830
   929
kpeter@830
   930
      // Copy the graph
kpeter@830
   931
      int i = 0;
kpeter@830
   932
      for (NodeIt n(_graph); n != INVALID; ++n, ++i) {
kpeter@830
   933
        _node_id[n] = i;
kpeter@830
   934
      }
kpeter@830
   935
      if (_arc_mixing) {
kpeter@830
   936
        // Store the arcs in a mixed order
kpeter@896
   937
        const int skip = std::max(_arc_num / _node_num, 3);
kpeter@830
   938
        int i = 0, j = 0;
kpeter@830
   939
        for (ArcIt a(_graph); a != INVALID; ++a) {
kpeter@830
   940
          _arc_id[a] = i;
kpeter@830
   941
          _source[i] = _node_id[_graph.source(a)];
kpeter@830
   942
          _target[i] = _node_id[_graph.target(a)];
kpeter@896
   943
          if ((i += skip) >= _arc_num) i = ++j;
kpeter@830
   944
        }
kpeter@830
   945
      } else {
kpeter@830
   946
        // Store the arcs in the original order
kpeter@830
   947
        int i = 0;
kpeter@830
   948
        for (ArcIt a(_graph); a != INVALID; ++a, ++i) {
kpeter@830
   949
          _arc_id[a] = i;
kpeter@830
   950
          _source[i] = _node_id[_graph.source(a)];
kpeter@830
   951
          _target[i] = _node_id[_graph.target(a)];
kpeter@830
   952
        }
kpeter@830
   953
      }
alpar@877
   954
kpeter@830
   955
      // Reset parameters
kpeter@830
   956
      resetParams();
kpeter@830
   957
      return *this;
kpeter@830
   958
    }
alpar@877
   959
kpeter@601
   960
    /// @}
kpeter@601
   961
kpeter@601
   962
    /// \name Query Functions
kpeter@601
   963
    /// The results of the algorithm can be obtained using these
kpeter@601
   964
    /// functions.\n
kpeter@605
   965
    /// The \ref run() function must be called before using them.
kpeter@605
   966
kpeter@601
   967
    /// @{
kpeter@601
   968
kpeter@605
   969
    /// \brief Return the total cost of the found flow.
kpeter@605
   970
    ///
kpeter@605
   971
    /// This function returns the total cost of the found flow.
kpeter@640
   972
    /// Its complexity is O(e).
kpeter@605
   973
    ///
kpeter@605
   974
    /// \note The return type of the function can be specified as a
kpeter@605
   975
    /// template parameter. For example,
kpeter@605
   976
    /// \code
kpeter@605
   977
    ///   ns.totalCost<double>();
kpeter@605
   978
    /// \endcode
kpeter@607
   979
    /// It is useful if the total cost cannot be stored in the \c Cost
kpeter@605
   980
    /// type of the algorithm, which is the default return type of the
kpeter@605
   981
    /// function.
kpeter@605
   982
    ///
kpeter@605
   983
    /// \pre \ref run() must be called before using this function.
kpeter@642
   984
    template <typename Number>
kpeter@642
   985
    Number totalCost() const {
kpeter@642
   986
      Number c = 0;
kpeter@642
   987
      for (ArcIt a(_graph); a != INVALID; ++a) {
kpeter@642
   988
        int i = _arc_id[a];
kpeter@642
   989
        c += Number(_flow[i]) * Number(_cost[i]);
kpeter@605
   990
      }
kpeter@605
   991
      return c;
kpeter@605
   992
    }
kpeter@605
   993
kpeter@605
   994
#ifndef DOXYGEN
kpeter@607
   995
    Cost totalCost() const {
kpeter@607
   996
      return totalCost<Cost>();
kpeter@605
   997
    }
kpeter@605
   998
#endif
kpeter@605
   999
kpeter@605
  1000
    /// \brief Return the flow on the given arc.
kpeter@605
  1001
    ///
kpeter@605
  1002
    /// This function returns the flow on the given arc.
kpeter@605
  1003
    ///
kpeter@605
  1004
    /// \pre \ref run() must be called before using this function.
kpeter@641
  1005
    Value flow(const Arc& a) const {
kpeter@642
  1006
      return _flow[_arc_id[a]];
kpeter@605
  1007
    }
kpeter@605
  1008
kpeter@642
  1009
    /// \brief Return the flow map (the primal solution).
kpeter@601
  1010
    ///
kpeter@642
  1011
    /// This function copies the flow value on each arc into the given
kpeter@642
  1012
    /// map. The \c Value type of the algorithm must be convertible to
kpeter@642
  1013
    /// the \c Value type of the map.
kpeter@601
  1014
    ///
kpeter@601
  1015
    /// \pre \ref run() must be called before using this function.
kpeter@642
  1016
    template <typename FlowMap>
kpeter@642
  1017
    void flowMap(FlowMap &map) const {
kpeter@642
  1018
      for (ArcIt a(_graph); a != INVALID; ++a) {
kpeter@642
  1019
        map.set(a, _flow[_arc_id[a]]);
kpeter@642
  1020
      }
kpeter@601
  1021
    }
kpeter@601
  1022
kpeter@605
  1023
    /// \brief Return the potential (dual value) of the given node.
kpeter@605
  1024
    ///
kpeter@605
  1025
    /// This function returns the potential (dual value) of the
kpeter@605
  1026
    /// given node.
kpeter@605
  1027
    ///
kpeter@605
  1028
    /// \pre \ref run() must be called before using this function.
kpeter@607
  1029
    Cost potential(const Node& n) const {
kpeter@642
  1030
      return _pi[_node_id[n]];
kpeter@605
  1031
    }
kpeter@605
  1032
kpeter@642
  1033
    /// \brief Return the potential map (the dual solution).
kpeter@601
  1034
    ///
kpeter@642
  1035
    /// This function copies the potential (dual value) of each node
kpeter@642
  1036
    /// into the given map.
kpeter@642
  1037
    /// The \c Cost type of the algorithm must be convertible to the
kpeter@642
  1038
    /// \c Value type of the map.
kpeter@601
  1039
    ///
kpeter@601
  1040
    /// \pre \ref run() must be called before using this function.
kpeter@642
  1041
    template <typename PotentialMap>
kpeter@642
  1042
    void potentialMap(PotentialMap &map) const {
kpeter@642
  1043
      for (NodeIt n(_graph); n != INVALID; ++n) {
kpeter@642
  1044
        map.set(n, _pi[_node_id[n]]);
kpeter@642
  1045
      }
kpeter@601
  1046
    }
kpeter@601
  1047
kpeter@601
  1048
    /// @}
kpeter@601
  1049
kpeter@601
  1050
  private:
kpeter@601
  1051
kpeter@601
  1052
    // Initialize internal data structures
kpeter@601
  1053
    bool init() {
kpeter@605
  1054
      if (_node_num == 0) return false;
kpeter@601
  1055
kpeter@642
  1056
      // Check the sum of supply values
kpeter@642
  1057
      _sum_supply = 0;
kpeter@642
  1058
      for (int i = 0; i != _node_num; ++i) {
kpeter@642
  1059
        _sum_supply += _supply[i];
kpeter@642
  1060
      }
alpar@643
  1061
      if ( !((_stype == GEQ && _sum_supply <= 0) ||
alpar@643
  1062
             (_stype == LEQ && _sum_supply >= 0)) ) return false;
kpeter@601
  1063
kpeter@642
  1064
      // Remove non-zero lower bounds
kpeter@642
  1065
      if (_have_lower) {
kpeter@642
  1066
        for (int i = 0; i != _arc_num; ++i) {
kpeter@642
  1067
          Value c = _lower[i];
kpeter@642
  1068
          if (c >= 0) {
kpeter@811
  1069
            _cap[i] = _upper[i] < MAX ? _upper[i] - c : INF;
kpeter@642
  1070
          } else {
kpeter@811
  1071
            _cap[i] = _upper[i] < MAX + c ? _upper[i] - c : INF;
kpeter@642
  1072
          }
kpeter@642
  1073
          _supply[_source[i]] -= c;
kpeter@642
  1074
          _supply[_target[i]] += c;
kpeter@642
  1075
        }
kpeter@642
  1076
      } else {
kpeter@642
  1077
        for (int i = 0; i != _arc_num; ++i) {
kpeter@642
  1078
          _cap[i] = _upper[i];
kpeter@642
  1079
        }
kpeter@605
  1080
      }
kpeter@601
  1081
kpeter@609
  1082
      // Initialize artifical cost
kpeter@640
  1083
      Cost ART_COST;
kpeter@609
  1084
      if (std::numeric_limits<Cost>::is_exact) {
kpeter@663
  1085
        ART_COST = std::numeric_limits<Cost>::max() / 2 + 1;
kpeter@609
  1086
      } else {
kpeter@888
  1087
        ART_COST = 0;
kpeter@609
  1088
        for (int i = 0; i != _arc_num; ++i) {
kpeter@640
  1089
          if (_cost[i] > ART_COST) ART_COST = _cost[i];
kpeter@609
  1090
        }
kpeter@640
  1091
        ART_COST = (ART_COST + 1) * _node_num;
kpeter@609
  1092
      }
kpeter@609
  1093
kpeter@642
  1094
      // Initialize arc maps
kpeter@642
  1095
      for (int i = 0; i != _arc_num; ++i) {
kpeter@642
  1096
        _flow[i] = 0;
kpeter@642
  1097
        _state[i] = STATE_LOWER;
kpeter@642
  1098
      }
alpar@877
  1099
kpeter@601
  1100
      // Set data for the artificial root node
kpeter@601
  1101
      _root = _node_num;
kpeter@601
  1102
      _parent[_root] = -1;
kpeter@601
  1103
      _pred[_root] = -1;
kpeter@601
  1104
      _thread[_root] = 0;
kpeter@604
  1105
      _rev_thread[0] = _root;
kpeter@642
  1106
      _succ_num[_root] = _node_num + 1;
kpeter@604
  1107
      _last_succ[_root] = _root - 1;
kpeter@640
  1108
      _supply[_root] = -_sum_supply;
kpeter@663
  1109
      _pi[_root] = 0;
kpeter@601
  1110
kpeter@601
  1111
      // Add artificial arcs and initialize the spanning tree data structure
kpeter@663
  1112
      if (_sum_supply == 0) {
kpeter@663
  1113
        // EQ supply constraints
kpeter@663
  1114
        _search_arc_num = _arc_num;
kpeter@663
  1115
        _all_arc_num = _arc_num + _node_num;
kpeter@663
  1116
        for (int u = 0, e = _arc_num; u != _node_num; ++u, ++e) {
kpeter@663
  1117
          _parent[u] = _root;
kpeter@663
  1118
          _pred[u] = e;
kpeter@663
  1119
          _thread[u] = u + 1;
kpeter@663
  1120
          _rev_thread[u + 1] = u;
kpeter@663
  1121
          _succ_num[u] = 1;
kpeter@663
  1122
          _last_succ[u] = u;
kpeter@663
  1123
          _cap[e] = INF;
kpeter@663
  1124
          _state[e] = STATE_TREE;
kpeter@663
  1125
          if (_supply[u] >= 0) {
kpeter@895
  1126
            _pred_dir[u] = DIR_UP;
kpeter@663
  1127
            _pi[u] = 0;
kpeter@663
  1128
            _source[e] = u;
kpeter@663
  1129
            _target[e] = _root;
kpeter@663
  1130
            _flow[e] = _supply[u];
kpeter@663
  1131
            _cost[e] = 0;
kpeter@663
  1132
          } else {
kpeter@895
  1133
            _pred_dir[u] = DIR_DOWN;
kpeter@663
  1134
            _pi[u] = ART_COST;
kpeter@663
  1135
            _source[e] = _root;
kpeter@663
  1136
            _target[e] = u;
kpeter@663
  1137
            _flow[e] = -_supply[u];
kpeter@663
  1138
            _cost[e] = ART_COST;
kpeter@663
  1139
          }
kpeter@601
  1140
        }
kpeter@601
  1141
      }
kpeter@663
  1142
      else if (_sum_supply > 0) {
kpeter@663
  1143
        // LEQ supply constraints
kpeter@663
  1144
        _search_arc_num = _arc_num + _node_num;
kpeter@663
  1145
        int f = _arc_num + _node_num;
kpeter@663
  1146
        for (int u = 0, e = _arc_num; u != _node_num; ++u, ++e) {
kpeter@663
  1147
          _parent[u] = _root;
kpeter@663
  1148
          _thread[u] = u + 1;
kpeter@663
  1149
          _rev_thread[u + 1] = u;
kpeter@663
  1150
          _succ_num[u] = 1;
kpeter@663
  1151
          _last_succ[u] = u;
kpeter@663
  1152
          if (_supply[u] >= 0) {
kpeter@895
  1153
            _pred_dir[u] = DIR_UP;
kpeter@663
  1154
            _pi[u] = 0;
kpeter@663
  1155
            _pred[u] = e;
kpeter@663
  1156
            _source[e] = u;
kpeter@663
  1157
            _target[e] = _root;
kpeter@663
  1158
            _cap[e] = INF;
kpeter@663
  1159
            _flow[e] = _supply[u];
kpeter@663
  1160
            _cost[e] = 0;
kpeter@663
  1161
            _state[e] = STATE_TREE;
kpeter@663
  1162
          } else {
kpeter@895
  1163
            _pred_dir[u] = DIR_DOWN;
kpeter@663
  1164
            _pi[u] = ART_COST;
kpeter@663
  1165
            _pred[u] = f;
kpeter@663
  1166
            _source[f] = _root;
kpeter@663
  1167
            _target[f] = u;
kpeter@663
  1168
            _cap[f] = INF;
kpeter@663
  1169
            _flow[f] = -_supply[u];
kpeter@663
  1170
            _cost[f] = ART_COST;
kpeter@663
  1171
            _state[f] = STATE_TREE;
kpeter@663
  1172
            _source[e] = u;
kpeter@663
  1173
            _target[e] = _root;
kpeter@663
  1174
            _cap[e] = INF;
kpeter@663
  1175
            _flow[e] = 0;
kpeter@663
  1176
            _cost[e] = 0;
kpeter@663
  1177
            _state[e] = STATE_LOWER;
kpeter@663
  1178
            ++f;
kpeter@663
  1179
          }
kpeter@663
  1180
        }
kpeter@663
  1181
        _all_arc_num = f;
kpeter@663
  1182
      }
kpeter@663
  1183
      else {
kpeter@663
  1184
        // GEQ supply constraints
kpeter@663
  1185
        _search_arc_num = _arc_num + _node_num;
kpeter@663
  1186
        int f = _arc_num + _node_num;
kpeter@663
  1187
        for (int u = 0, e = _arc_num; u != _node_num; ++u, ++e) {
kpeter@663
  1188
          _parent[u] = _root;
kpeter@663
  1189
          _thread[u] = u + 1;
kpeter@663
  1190
          _rev_thread[u + 1] = u;
kpeter@663
  1191
          _succ_num[u] = 1;
kpeter@663
  1192
          _last_succ[u] = u;
kpeter@663
  1193
          if (_supply[u] <= 0) {
kpeter@895
  1194
            _pred_dir[u] = DIR_DOWN;
kpeter@663
  1195
            _pi[u] = 0;
kpeter@663
  1196
            _pred[u] = e;
kpeter@663
  1197
            _source[e] = _root;
kpeter@663
  1198
            _target[e] = u;
kpeter@663
  1199
            _cap[e] = INF;
kpeter@663
  1200
            _flow[e] = -_supply[u];
kpeter@663
  1201
            _cost[e] = 0;
kpeter@663
  1202
            _state[e] = STATE_TREE;
kpeter@663
  1203
          } else {
kpeter@895
  1204
            _pred_dir[u] = DIR_UP;
kpeter@663
  1205
            _pi[u] = -ART_COST;
kpeter@663
  1206
            _pred[u] = f;
kpeter@663
  1207
            _source[f] = u;
kpeter@663
  1208
            _target[f] = _root;
kpeter@663
  1209
            _cap[f] = INF;
kpeter@663
  1210
            _flow[f] = _supply[u];
kpeter@663
  1211
            _state[f] = STATE_TREE;
kpeter@663
  1212
            _cost[f] = ART_COST;
kpeter@663
  1213
            _source[e] = _root;
kpeter@663
  1214
            _target[e] = u;
kpeter@663
  1215
            _cap[e] = INF;
kpeter@663
  1216
            _flow[e] = 0;
kpeter@663
  1217
            _cost[e] = 0;
kpeter@663
  1218
            _state[e] = STATE_LOWER;
kpeter@663
  1219
            ++f;
kpeter@663
  1220
          }
kpeter@663
  1221
        }
kpeter@663
  1222
        _all_arc_num = f;
kpeter@663
  1223
      }
kpeter@601
  1224
kpeter@601
  1225
      return true;
kpeter@601
  1226
    }
kpeter@601
  1227
kpeter@601
  1228
    // Find the join node
kpeter@601
  1229
    void findJoinNode() {
kpeter@603
  1230
      int u = _source[in_arc];
kpeter@603
  1231
      int v = _target[in_arc];
kpeter@601
  1232
      while (u != v) {
kpeter@604
  1233
        if (_succ_num[u] < _succ_num[v]) {
kpeter@604
  1234
          u = _parent[u];
kpeter@604
  1235
        } else {
kpeter@604
  1236
          v = _parent[v];
kpeter@604
  1237
        }
kpeter@601
  1238
      }
kpeter@601
  1239
      join = u;
kpeter@601
  1240
    }
kpeter@601
  1241
kpeter@601
  1242
    // Find the leaving arc of the cycle and returns true if the
kpeter@601
  1243
    // leaving arc is not the same as the entering arc
kpeter@601
  1244
    bool findLeavingArc() {
kpeter@601
  1245
      // Initialize first and second nodes according to the direction
kpeter@601
  1246
      // of the cycle
kpeter@895
  1247
      int first, second;
kpeter@603
  1248
      if (_state[in_arc] == STATE_LOWER) {
kpeter@603
  1249
        first  = _source[in_arc];
kpeter@603
  1250
        second = _target[in_arc];
kpeter@601
  1251
      } else {
kpeter@603
  1252
        first  = _target[in_arc];
kpeter@603
  1253
        second = _source[in_arc];
kpeter@601
  1254
      }
kpeter@603
  1255
      delta = _cap[in_arc];
kpeter@601
  1256
      int result = 0;
kpeter@895
  1257
      Value c, d;
kpeter@601
  1258
      int e;
kpeter@601
  1259
kpeter@895
  1260
      // Search the cycle form the first node to the join node
kpeter@601
  1261
      for (int u = first; u != join; u = _parent[u]) {
kpeter@601
  1262
        e = _pred[u];
kpeter@895
  1263
        d = _flow[e];
kpeter@895
  1264
        if (_pred_dir[u] == DIR_DOWN) {
kpeter@895
  1265
          c = _cap[e];
kpeter@895
  1266
          d = c >= MAX ? INF : c - d;
kpeter@895
  1267
        }
kpeter@601
  1268
        if (d < delta) {
kpeter@601
  1269
          delta = d;
kpeter@601
  1270
          u_out = u;
kpeter@601
  1271
          result = 1;
kpeter@601
  1272
        }
kpeter@601
  1273
      }
kpeter@895
  1274
kpeter@895
  1275
      // Search the cycle form the second node to the join node
kpeter@601
  1276
      for (int u = second; u != join; u = _parent[u]) {
kpeter@601
  1277
        e = _pred[u];
kpeter@895
  1278
        d = _flow[e];
kpeter@895
  1279
        if (_pred_dir[u] == DIR_UP) {
kpeter@895
  1280
          c = _cap[e];
kpeter@895
  1281
          d = c >= MAX ? INF : c - d;
kpeter@895
  1282
        }
kpeter@601
  1283
        if (d <= delta) {
kpeter@601
  1284
          delta = d;
kpeter@601
  1285
          u_out = u;
kpeter@601
  1286
          result = 2;
kpeter@601
  1287
        }
kpeter@601
  1288
      }
kpeter@601
  1289
kpeter@601
  1290
      if (result == 1) {
kpeter@601
  1291
        u_in = first;
kpeter@601
  1292
        v_in = second;
kpeter@601
  1293
      } else {
kpeter@601
  1294
        u_in = second;
kpeter@601
  1295
        v_in = first;
kpeter@601
  1296
      }
kpeter@601
  1297
      return result != 0;
kpeter@601
  1298
    }
kpeter@601
  1299
kpeter@601
  1300
    // Change _flow and _state vectors
kpeter@601
  1301
    void changeFlow(bool change) {
kpeter@601
  1302
      // Augment along the cycle
kpeter@601
  1303
      if (delta > 0) {
kpeter@641
  1304
        Value val = _state[in_arc] * delta;
kpeter@603
  1305
        _flow[in_arc] += val;
kpeter@603
  1306
        for (int u = _source[in_arc]; u != join; u = _parent[u]) {
kpeter@895
  1307
          _flow[_pred[u]] -= _pred_dir[u] * val;
kpeter@601
  1308
        }
kpeter@603
  1309
        for (int u = _target[in_arc]; u != join; u = _parent[u]) {
kpeter@895
  1310
          _flow[_pred[u]] += _pred_dir[u] * val;
kpeter@601
  1311
        }
kpeter@601
  1312
      }
kpeter@601
  1313
      // Update the state of the entering and leaving arcs
kpeter@601
  1314
      if (change) {
kpeter@603
  1315
        _state[in_arc] = STATE_TREE;
kpeter@601
  1316
        _state[_pred[u_out]] =
kpeter@601
  1317
          (_flow[_pred[u_out]] == 0) ? STATE_LOWER : STATE_UPPER;
kpeter@601
  1318
      } else {
kpeter@603
  1319
        _state[in_arc] = -_state[in_arc];
kpeter@601
  1320
      }
kpeter@601
  1321
    }
kpeter@601
  1322
kpeter@604
  1323
    // Update the tree structure
kpeter@604
  1324
    void updateTreeStructure() {
kpeter@604
  1325
      int old_rev_thread = _rev_thread[u_out];
kpeter@604
  1326
      int old_succ_num = _succ_num[u_out];
kpeter@604
  1327
      int old_last_succ = _last_succ[u_out];
kpeter@601
  1328
      v_out = _parent[u_out];
kpeter@601
  1329
kpeter@895
  1330
      // Check if u_in and u_out coincide
kpeter@895
  1331
      if (u_in == u_out) {
kpeter@895
  1332
        // Update _parent, _pred, _pred_dir
kpeter@895
  1333
        _parent[u_in] = v_in;
kpeter@895
  1334
        _pred[u_in] = in_arc;
kpeter@895
  1335
        _pred_dir[u_in] = u_in == _source[in_arc] ? DIR_UP : DIR_DOWN;
kpeter@604
  1336
kpeter@895
  1337
        // Update _thread and _rev_thread
kpeter@895
  1338
        if (_thread[v_in] != u_out) {
kpeter@895
  1339
          int after = _thread[old_last_succ];
kpeter@895
  1340
          _thread[old_rev_thread] = after;
kpeter@895
  1341
          _rev_thread[after] = old_rev_thread;
kpeter@895
  1342
          after = _thread[v_in];
kpeter@895
  1343
          _thread[v_in] = u_out;
kpeter@895
  1344
          _rev_thread[u_out] = v_in;
kpeter@895
  1345
          _thread[old_last_succ] = after;
kpeter@895
  1346
          _rev_thread[after] = old_last_succ;
kpeter@895
  1347
        }
kpeter@604
  1348
      } else {
kpeter@895
  1349
        // Handle the case when old_rev_thread equals to v_in
kpeter@895
  1350
        // (it also means that join and v_out coincide)
kpeter@895
  1351
        int thread_continue = old_rev_thread == v_in ?
kpeter@895
  1352
          _thread[old_last_succ] : _thread[v_in];
kpeter@601
  1353
kpeter@895
  1354
        // Update _thread and _parent along the stem nodes (i.e. the nodes
kpeter@895
  1355
        // between u_in and u_out, whose parent have to be changed)
kpeter@895
  1356
        int stem = u_in;              // the current stem node
kpeter@895
  1357
        int par_stem = v_in;          // the new parent of stem
kpeter@895
  1358
        int next_stem;                // the next stem node
kpeter@895
  1359
        int last = _last_succ[u_in];  // the last successor of stem
kpeter@895
  1360
        int before, after = _thread[last];
kpeter@895
  1361
        _thread[v_in] = u_in;
kpeter@895
  1362
        _dirty_revs.clear();
kpeter@895
  1363
        _dirty_revs.push_back(v_in);
kpeter@895
  1364
        while (stem != u_out) {
kpeter@895
  1365
          // Insert the next stem node into the thread list
kpeter@895
  1366
          next_stem = _parent[stem];
kpeter@895
  1367
          _thread[last] = next_stem;
kpeter@895
  1368
          _dirty_revs.push_back(last);
kpeter@601
  1369
kpeter@895
  1370
          // Remove the subtree of stem from the thread list
kpeter@895
  1371
          before = _rev_thread[stem];
kpeter@895
  1372
          _thread[before] = after;
kpeter@895
  1373
          _rev_thread[after] = before;
kpeter@601
  1374
kpeter@895
  1375
          // Change the parent node and shift stem nodes
kpeter@895
  1376
          _parent[stem] = par_stem;
kpeter@895
  1377
          par_stem = stem;
kpeter@895
  1378
          stem = next_stem;
kpeter@601
  1379
kpeter@895
  1380
          // Update last and after
kpeter@895
  1381
          last = _last_succ[stem] == _last_succ[par_stem] ?
kpeter@895
  1382
            _rev_thread[par_stem] : _last_succ[stem];
kpeter@895
  1383
          after = _thread[last];
kpeter@895
  1384
        }
kpeter@895
  1385
        _parent[u_out] = par_stem;
kpeter@895
  1386
        _thread[last] = thread_continue;
kpeter@895
  1387
        _rev_thread[thread_continue] = last;
kpeter@895
  1388
        _last_succ[u_out] = last;
kpeter@601
  1389
kpeter@895
  1390
        // Remove the subtree of u_out from the thread list except for
kpeter@895
  1391
        // the case when old_rev_thread equals to v_in
kpeter@895
  1392
        if (old_rev_thread != v_in) {
kpeter@895
  1393
          _thread[old_rev_thread] = after;
kpeter@895
  1394
          _rev_thread[after] = old_rev_thread;
kpeter@895
  1395
        }
kpeter@604
  1396
kpeter@895
  1397
        // Update _rev_thread using the new _thread values
kpeter@895
  1398
        for (int i = 0; i != int(_dirty_revs.size()); ++i) {
kpeter@895
  1399
          int u = _dirty_revs[i];
kpeter@895
  1400
          _rev_thread[_thread[u]] = u;
kpeter@895
  1401
        }
kpeter@604
  1402
kpeter@895
  1403
        // Update _pred, _pred_dir, _last_succ and _succ_num for the
kpeter@895
  1404
        // stem nodes from u_out to u_in
kpeter@895
  1405
        int tmp_sc = 0, tmp_ls = _last_succ[u_out];
kpeter@895
  1406
        for (int u = u_out, p = _parent[u]; u != u_in; u = p, p = _parent[u]) {
kpeter@895
  1407
          _pred[u] = _pred[p];
kpeter@895
  1408
          _pred_dir[u] = -_pred_dir[p];
kpeter@895
  1409
          tmp_sc += _succ_num[u] - _succ_num[p];
kpeter@895
  1410
          _succ_num[u] = tmp_sc;
kpeter@895
  1411
          _last_succ[p] = tmp_ls;
kpeter@895
  1412
        }
kpeter@895
  1413
        _pred[u_in] = in_arc;
kpeter@895
  1414
        _pred_dir[u_in] = u_in == _source[in_arc] ? DIR_UP : DIR_DOWN;
kpeter@895
  1415
        _succ_num[u_in] = old_succ_num;
kpeter@604
  1416
      }
kpeter@604
  1417
kpeter@604
  1418
      // Update _last_succ from v_in towards the root
kpeter@895
  1419
      int up_limit_out = _last_succ[join] == v_in ? join : -1;
kpeter@895
  1420
      int last_succ_out = _last_succ[u_out];
kpeter@895
  1421
      for (int u = v_in; u != -1 && _last_succ[u] == v_in; u = _parent[u]) {
kpeter@895
  1422
        _last_succ[u] = last_succ_out;
kpeter@604
  1423
      }
kpeter@895
  1424
kpeter@604
  1425
      // Update _last_succ from v_out towards the root
kpeter@604
  1426
      if (join != old_rev_thread && v_in != old_rev_thread) {
kpeter@895
  1427
        for (int u = v_out; u != up_limit_out && _last_succ[u] == old_last_succ;
kpeter@604
  1428
             u = _parent[u]) {
kpeter@604
  1429
          _last_succ[u] = old_rev_thread;
kpeter@604
  1430
        }
kpeter@895
  1431
      }
kpeter@895
  1432
      else if (last_succ_out != old_last_succ) {
kpeter@895
  1433
        for (int u = v_out; u != up_limit_out && _last_succ[u] == old_last_succ;
kpeter@604
  1434
             u = _parent[u]) {
kpeter@895
  1435
          _last_succ[u] = last_succ_out;
kpeter@604
  1436
        }
kpeter@604
  1437
      }
kpeter@604
  1438
kpeter@604
  1439
      // Update _succ_num from v_in to join
kpeter@895
  1440
      for (int u = v_in; u != join; u = _parent[u]) {
kpeter@604
  1441
        _succ_num[u] += old_succ_num;
kpeter@604
  1442
      }
kpeter@604
  1443
      // Update _succ_num from v_out to join
kpeter@895
  1444
      for (int u = v_out; u != join; u = _parent[u]) {
kpeter@604
  1445
        _succ_num[u] -= old_succ_num;
kpeter@601
  1446
      }
kpeter@601
  1447
    }
kpeter@601
  1448
kpeter@895
  1449
    // Update potentials in the subtree that has been moved
kpeter@604
  1450
    void updatePotential() {
kpeter@895
  1451
      Cost sigma = _pi[v_in] - _pi[u_in] -
kpeter@895
  1452
                   _pred_dir[u_in] * _cost[in_arc];
kpeter@608
  1453
      int end = _thread[_last_succ[u_in]];
kpeter@608
  1454
      for (int u = u_in; u != end; u = _thread[u]) {
kpeter@608
  1455
        _pi[u] += sigma;
kpeter@601
  1456
      }
kpeter@601
  1457
    }
kpeter@601
  1458
kpeter@839
  1459
    // Heuristic initial pivots
kpeter@839
  1460
    bool initialPivots() {
kpeter@839
  1461
      Value curr, total = 0;
kpeter@839
  1462
      std::vector<Node> supply_nodes, demand_nodes;
kpeter@839
  1463
      for (NodeIt u(_graph); u != INVALID; ++u) {
kpeter@839
  1464
        curr = _supply[_node_id[u]];
kpeter@839
  1465
        if (curr > 0) {
kpeter@839
  1466
          total += curr;
kpeter@839
  1467
          supply_nodes.push_back(u);
kpeter@839
  1468
        }
kpeter@839
  1469
        else if (curr < 0) {
kpeter@839
  1470
          demand_nodes.push_back(u);
kpeter@839
  1471
        }
kpeter@839
  1472
      }
kpeter@839
  1473
      if (_sum_supply > 0) total -= _sum_supply;
kpeter@839
  1474
      if (total <= 0) return true;
kpeter@839
  1475
kpeter@839
  1476
      IntVector arc_vector;
kpeter@839
  1477
      if (_sum_supply >= 0) {
kpeter@839
  1478
        if (supply_nodes.size() == 1 && demand_nodes.size() == 1) {
kpeter@839
  1479
          // Perform a reverse graph search from the sink to the source
kpeter@839
  1480
          typename GR::template NodeMap<bool> reached(_graph, false);
kpeter@839
  1481
          Node s = supply_nodes[0], t = demand_nodes[0];
kpeter@839
  1482
          std::vector<Node> stack;
kpeter@839
  1483
          reached[t] = true;
kpeter@839
  1484
          stack.push_back(t);
kpeter@839
  1485
          while (!stack.empty()) {
kpeter@839
  1486
            Node u, v = stack.back();
kpeter@839
  1487
            stack.pop_back();
kpeter@839
  1488
            if (v == s) break;
kpeter@839
  1489
            for (InArcIt a(_graph, v); a != INVALID; ++a) {
kpeter@839
  1490
              if (reached[u = _graph.source(a)]) continue;
kpeter@839
  1491
              int j = _arc_id[a];
kpeter@839
  1492
              if (_cap[j] >= total) {
kpeter@839
  1493
                arc_vector.push_back(j);
kpeter@839
  1494
                reached[u] = true;
kpeter@839
  1495
                stack.push_back(u);
kpeter@839
  1496
              }
kpeter@839
  1497
            }
kpeter@839
  1498
          }
kpeter@839
  1499
        } else {
kpeter@839
  1500
          // Find the min. cost incomming arc for each demand node
kpeter@839
  1501
          for (int i = 0; i != int(demand_nodes.size()); ++i) {
kpeter@839
  1502
            Node v = demand_nodes[i];
kpeter@839
  1503
            Cost c, min_cost = std::numeric_limits<Cost>::max();
kpeter@839
  1504
            Arc min_arc = INVALID;
kpeter@839
  1505
            for (InArcIt a(_graph, v); a != INVALID; ++a) {
kpeter@839
  1506
              c = _cost[_arc_id[a]];
kpeter@839
  1507
              if (c < min_cost) {
kpeter@839
  1508
                min_cost = c;
kpeter@839
  1509
                min_arc = a;
kpeter@839
  1510
              }
kpeter@839
  1511
            }
kpeter@839
  1512
            if (min_arc != INVALID) {
kpeter@839
  1513
              arc_vector.push_back(_arc_id[min_arc]);
kpeter@839
  1514
            }
kpeter@839
  1515
          }
kpeter@839
  1516
        }
kpeter@839
  1517
      } else {
kpeter@839
  1518
        // Find the min. cost outgoing arc for each supply node
kpeter@839
  1519
        for (int i = 0; i != int(supply_nodes.size()); ++i) {
kpeter@839
  1520
          Node u = supply_nodes[i];
kpeter@839
  1521
          Cost c, min_cost = std::numeric_limits<Cost>::max();
kpeter@839
  1522
          Arc min_arc = INVALID;
kpeter@839
  1523
          for (OutArcIt a(_graph, u); a != INVALID; ++a) {
kpeter@839
  1524
            c = _cost[_arc_id[a]];
kpeter@839
  1525
            if (c < min_cost) {
kpeter@839
  1526
              min_cost = c;
kpeter@839
  1527
              min_arc = a;
kpeter@839
  1528
            }
kpeter@839
  1529
          }
kpeter@839
  1530
          if (min_arc != INVALID) {
kpeter@839
  1531
            arc_vector.push_back(_arc_id[min_arc]);
kpeter@839
  1532
          }
kpeter@839
  1533
        }
kpeter@839
  1534
      }
kpeter@839
  1535
kpeter@839
  1536
      // Perform heuristic initial pivots
kpeter@839
  1537
      for (int i = 0; i != int(arc_vector.size()); ++i) {
kpeter@839
  1538
        in_arc = arc_vector[i];
kpeter@839
  1539
        if (_state[in_arc] * (_cost[in_arc] + _pi[_source[in_arc]] -
kpeter@839
  1540
            _pi[_target[in_arc]]) >= 0) continue;
kpeter@839
  1541
        findJoinNode();
kpeter@839
  1542
        bool change = findLeavingArc();
kpeter@839
  1543
        if (delta >= MAX) return false;
kpeter@839
  1544
        changeFlow(change);
kpeter@839
  1545
        if (change) {
kpeter@839
  1546
          updateTreeStructure();
kpeter@839
  1547
          updatePotential();
kpeter@839
  1548
        }
kpeter@839
  1549
      }
kpeter@839
  1550
      return true;
kpeter@839
  1551
    }
kpeter@839
  1552
kpeter@601
  1553
    // Execute the algorithm
kpeter@640
  1554
    ProblemType start(PivotRule pivot_rule) {
kpeter@601
  1555
      // Select the pivot rule implementation
kpeter@601
  1556
      switch (pivot_rule) {
kpeter@605
  1557
        case FIRST_ELIGIBLE:
kpeter@601
  1558
          return start<FirstEligiblePivotRule>();
kpeter@605
  1559
        case BEST_ELIGIBLE:
kpeter@601
  1560
          return start<BestEligiblePivotRule>();
kpeter@605
  1561
        case BLOCK_SEARCH:
kpeter@601
  1562
          return start<BlockSearchPivotRule>();
kpeter@605
  1563
        case CANDIDATE_LIST:
kpeter@601
  1564
          return start<CandidateListPivotRule>();
kpeter@605
  1565
        case ALTERING_LIST:
kpeter@601
  1566
          return start<AlteringListPivotRule>();
kpeter@601
  1567
      }
kpeter@640
  1568
      return INFEASIBLE; // avoid warning
kpeter@601
  1569
    }
kpeter@601
  1570
kpeter@605
  1571
    template <typename PivotRuleImpl>
kpeter@640
  1572
    ProblemType start() {
kpeter@605
  1573
      PivotRuleImpl pivot(*this);
kpeter@601
  1574
kpeter@839
  1575
      // Perform heuristic initial pivots
kpeter@839
  1576
      if (!initialPivots()) return UNBOUNDED;
kpeter@839
  1577
kpeter@605
  1578
      // Execute the Network Simplex algorithm
kpeter@601
  1579
      while (pivot.findEnteringArc()) {
kpeter@601
  1580
        findJoinNode();
kpeter@601
  1581
        bool change = findLeavingArc();
kpeter@811
  1582
        if (delta >= MAX) return UNBOUNDED;
kpeter@601
  1583
        changeFlow(change);
kpeter@601
  1584
        if (change) {
kpeter@604
  1585
          updateTreeStructure();
kpeter@604
  1586
          updatePotential();
kpeter@601
  1587
        }
kpeter@601
  1588
      }
alpar@877
  1589
kpeter@640
  1590
      // Check feasibility
kpeter@663
  1591
      for (int e = _search_arc_num; e != _all_arc_num; ++e) {
kpeter@663
  1592
        if (_flow[e] != 0) return INFEASIBLE;
kpeter@640
  1593
      }
kpeter@601
  1594
kpeter@642
  1595
      // Transform the solution and the supply map to the original form
kpeter@642
  1596
      if (_have_lower) {
kpeter@601
  1597
        for (int i = 0; i != _arc_num; ++i) {
kpeter@642
  1598
          Value c = _lower[i];
kpeter@642
  1599
          if (c != 0) {
kpeter@642
  1600
            _flow[i] += c;
kpeter@642
  1601
            _supply[_source[i]] += c;
kpeter@642
  1602
            _supply[_target[i]] -= c;
kpeter@642
  1603
          }
kpeter@601
  1604
        }
kpeter@601
  1605
      }
alpar@877
  1606
kpeter@663
  1607
      // Shift potentials to meet the requirements of the GEQ/LEQ type
kpeter@663
  1608
      // optimality conditions
kpeter@663
  1609
      if (_sum_supply == 0) {
kpeter@663
  1610
        if (_stype == GEQ) {
kpeter@888
  1611
          Cost max_pot = -std::numeric_limits<Cost>::max();
kpeter@663
  1612
          for (int i = 0; i != _node_num; ++i) {
kpeter@663
  1613
            if (_pi[i] > max_pot) max_pot = _pi[i];
kpeter@663
  1614
          }
kpeter@663
  1615
          if (max_pot > 0) {
kpeter@663
  1616
            for (int i = 0; i != _node_num; ++i)
kpeter@663
  1617
              _pi[i] -= max_pot;
kpeter@663
  1618
          }
kpeter@663
  1619
        } else {
kpeter@663
  1620
          Cost min_pot = std::numeric_limits<Cost>::max();
kpeter@663
  1621
          for (int i = 0; i != _node_num; ++i) {
kpeter@663
  1622
            if (_pi[i] < min_pot) min_pot = _pi[i];
kpeter@663
  1623
          }
kpeter@663
  1624
          if (min_pot < 0) {
kpeter@663
  1625
            for (int i = 0; i != _node_num; ++i)
kpeter@663
  1626
              _pi[i] -= min_pot;
kpeter@663
  1627
          }
kpeter@663
  1628
        }
kpeter@663
  1629
      }
kpeter@601
  1630
kpeter@640
  1631
      return OPTIMAL;
kpeter@601
  1632
    }
kpeter@601
  1633
kpeter@601
  1634
  }; //class NetworkSimplex
kpeter@601
  1635
kpeter@601
  1636
  ///@}
kpeter@601
  1637
kpeter@601
  1638
} //namespace lemon
kpeter@601
  1639
kpeter@601
  1640
#endif //LEMON_NETWORK_SIMPLEX_H