lemon/capacity_scaling.h
changeset 871 d3e32a777d0b
child 872 fa6f37d7a25b
     1.1 --- /dev/null	Thu Jan 01 00:00:00 1970 +0000
     1.2 +++ b/lemon/capacity_scaling.h	Thu Nov 12 23:17:34 2009 +0100
     1.3 @@ -0,0 +1,717 @@
     1.4 +/* -*- C++ -*-
     1.5 + *
     1.6 + * This file is a part of LEMON, a generic C++ optimization library
     1.7 + *
     1.8 + * Copyright (C) 2003-2008
     1.9 + * Egervary Jeno Kombinatorikus Optimalizalasi Kutatocsoport
    1.10 + * (Egervary Research Group on Combinatorial Optimization, EGRES).
    1.11 + *
    1.12 + * Permission to use, modify and distribute this software is granted
    1.13 + * provided that this copyright notice appears in all copies. For
    1.14 + * precise terms see the accompanying LICENSE file.
    1.15 + *
    1.16 + * This software is provided "AS IS" with no warranty of any kind,
    1.17 + * express or implied, and with no claim as to its suitability for any
    1.18 + * purpose.
    1.19 + *
    1.20 + */
    1.21 +
    1.22 +#ifndef LEMON_CAPACITY_SCALING_H
    1.23 +#define LEMON_CAPACITY_SCALING_H
    1.24 +
    1.25 +/// \ingroup min_cost_flow
    1.26 +///
    1.27 +/// \file
    1.28 +/// \brief Capacity scaling algorithm for finding a minimum cost flow.
    1.29 +
    1.30 +#include <vector>
    1.31 +#include <lemon/bin_heap.h>
    1.32 +
    1.33 +namespace lemon {
    1.34 +
    1.35 +  /// \addtogroup min_cost_flow
    1.36 +  /// @{
    1.37 +
    1.38 +  /// \brief Implementation of the capacity scaling algorithm for
    1.39 +  /// finding a minimum cost flow.
    1.40 +  ///
    1.41 +  /// \ref CapacityScaling implements the capacity scaling version
    1.42 +  /// of the successive shortest path algorithm for finding a minimum
    1.43 +  /// cost flow.
    1.44 +  ///
    1.45 +  /// \tparam Digraph The digraph type the algorithm runs on.
    1.46 +  /// \tparam LowerMap The type of the lower bound map.
    1.47 +  /// \tparam CapacityMap The type of the capacity (upper bound) map.
    1.48 +  /// \tparam CostMap The type of the cost (length) map.
    1.49 +  /// \tparam SupplyMap The type of the supply map.
    1.50 +  ///
    1.51 +  /// \warning
    1.52 +  /// - Arc capacities and costs should be \e non-negative \e integers.
    1.53 +  /// - Supply values should be \e signed \e integers.
    1.54 +  /// - The value types of the maps should be convertible to each other.
    1.55 +  /// - \c CostMap::Value must be signed type.
    1.56 +  ///
    1.57 +  /// \author Peter Kovacs
    1.58 +  template < typename Digraph,
    1.59 +             typename LowerMap = typename Digraph::template ArcMap<int>,
    1.60 +             typename CapacityMap = typename Digraph::template ArcMap<int>,
    1.61 +             typename CostMap = typename Digraph::template ArcMap<int>,
    1.62 +             typename SupplyMap = typename Digraph::template NodeMap<int> >
    1.63 +  class CapacityScaling
    1.64 +  {
    1.65 +    TEMPLATE_DIGRAPH_TYPEDEFS(Digraph);
    1.66 +
    1.67 +    typedef typename CapacityMap::Value Capacity;
    1.68 +    typedef typename CostMap::Value Cost;
    1.69 +    typedef typename SupplyMap::Value Supply;
    1.70 +    typedef typename Digraph::template ArcMap<Capacity> CapacityArcMap;
    1.71 +    typedef typename Digraph::template NodeMap<Supply> SupplyNodeMap;
    1.72 +    typedef typename Digraph::template NodeMap<Arc> PredMap;
    1.73 +
    1.74 +  public:
    1.75 +
    1.76 +    /// The type of the flow map.
    1.77 +    typedef typename Digraph::template ArcMap<Capacity> FlowMap;
    1.78 +    /// The type of the potential map.
    1.79 +    typedef typename Digraph::template NodeMap<Cost> PotentialMap;
    1.80 +
    1.81 +  private:
    1.82 +
    1.83 +    /// \brief Special implementation of the \ref Dijkstra algorithm
    1.84 +    /// for finding shortest paths in the residual network.
    1.85 +    ///
    1.86 +    /// \ref ResidualDijkstra is a special implementation of the
    1.87 +    /// \ref Dijkstra algorithm for finding shortest paths in the
    1.88 +    /// residual network of the digraph with respect to the reduced arc
    1.89 +    /// costs and modifying the node potentials according to the
    1.90 +    /// distance of the nodes.
    1.91 +    class ResidualDijkstra
    1.92 +    {
    1.93 +      typedef typename Digraph::template NodeMap<int> HeapCrossRef;
    1.94 +      typedef BinHeap<Cost, HeapCrossRef> Heap;
    1.95 +
    1.96 +    private:
    1.97 +
    1.98 +      // The digraph the algorithm runs on
    1.99 +      const Digraph &_graph;
   1.100 +
   1.101 +      // The main maps
   1.102 +      const FlowMap &_flow;
   1.103 +      const CapacityArcMap &_res_cap;
   1.104 +      const CostMap &_cost;
   1.105 +      const SupplyNodeMap &_excess;
   1.106 +      PotentialMap &_potential;
   1.107 +
   1.108 +      // The distance map
   1.109 +      PotentialMap _dist;
   1.110 +      // The pred arc map
   1.111 +      PredMap &_pred;
   1.112 +      // The processed (i.e. permanently labeled) nodes
   1.113 +      std::vector<Node> _proc_nodes;
   1.114 +
   1.115 +    public:
   1.116 +
   1.117 +      /// Constructor.
   1.118 +      ResidualDijkstra( const Digraph &digraph,
   1.119 +                        const FlowMap &flow,
   1.120 +                        const CapacityArcMap &res_cap,
   1.121 +                        const CostMap &cost,
   1.122 +                        const SupplyMap &excess,
   1.123 +                        PotentialMap &potential,
   1.124 +                        PredMap &pred ) :
   1.125 +        _graph(digraph), _flow(flow), _res_cap(res_cap), _cost(cost),
   1.126 +        _excess(excess), _potential(potential), _dist(digraph),
   1.127 +        _pred(pred)
   1.128 +      {}
   1.129 +
   1.130 +      /// Run the algorithm from the given source node.
   1.131 +      Node run(Node s, Capacity delta = 1) {
   1.132 +        HeapCrossRef heap_cross_ref(_graph, Heap::PRE_HEAP);
   1.133 +        Heap heap(heap_cross_ref);
   1.134 +        heap.push(s, 0);
   1.135 +        _pred[s] = INVALID;
   1.136 +        _proc_nodes.clear();
   1.137 +
   1.138 +        // Processing nodes
   1.139 +        while (!heap.empty() && _excess[heap.top()] > -delta) {
   1.140 +          Node u = heap.top(), v;
   1.141 +          Cost d = heap.prio() + _potential[u], nd;
   1.142 +          _dist[u] = heap.prio();
   1.143 +          heap.pop();
   1.144 +          _proc_nodes.push_back(u);
   1.145 +
   1.146 +          // Traversing outgoing arcs
   1.147 +          for (OutArcIt e(_graph, u); e != INVALID; ++e) {
   1.148 +            if (_res_cap[e] >= delta) {
   1.149 +              v = _graph.target(e);
   1.150 +              switch(heap.state(v)) {
   1.151 +              case Heap::PRE_HEAP:
   1.152 +                heap.push(v, d + _cost[e] - _potential[v]);
   1.153 +                _pred[v] = e;
   1.154 +                break;
   1.155 +              case Heap::IN_HEAP:
   1.156 +                nd = d + _cost[e] - _potential[v];
   1.157 +                if (nd < heap[v]) {
   1.158 +                  heap.decrease(v, nd);
   1.159 +                  _pred[v] = e;
   1.160 +                }
   1.161 +                break;
   1.162 +              case Heap::POST_HEAP:
   1.163 +                break;
   1.164 +              }
   1.165 +            }
   1.166 +          }
   1.167 +
   1.168 +          // Traversing incoming arcs
   1.169 +          for (InArcIt e(_graph, u); e != INVALID; ++e) {
   1.170 +            if (_flow[e] >= delta) {
   1.171 +              v = _graph.source(e);
   1.172 +              switch(heap.state(v)) {
   1.173 +              case Heap::PRE_HEAP:
   1.174 +                heap.push(v, d - _cost[e] - _potential[v]);
   1.175 +                _pred[v] = e;
   1.176 +                break;
   1.177 +              case Heap::IN_HEAP:
   1.178 +                nd = d - _cost[e] - _potential[v];
   1.179 +                if (nd < heap[v]) {
   1.180 +                  heap.decrease(v, nd);
   1.181 +                  _pred[v] = e;
   1.182 +                }
   1.183 +                break;
   1.184 +              case Heap::POST_HEAP:
   1.185 +                break;
   1.186 +              }
   1.187 +            }
   1.188 +          }
   1.189 +        }
   1.190 +        if (heap.empty()) return INVALID;
   1.191 +
   1.192 +        // Updating potentials of processed nodes
   1.193 +        Node t = heap.top();
   1.194 +        Cost t_dist = heap.prio();
   1.195 +        for (int i = 0; i < int(_proc_nodes.size()); ++i)
   1.196 +          _potential[_proc_nodes[i]] += _dist[_proc_nodes[i]] - t_dist;
   1.197 +
   1.198 +        return t;
   1.199 +      }
   1.200 +
   1.201 +    }; //class ResidualDijkstra
   1.202 +
   1.203 +  private:
   1.204 +
   1.205 +    // The digraph the algorithm runs on
   1.206 +    const Digraph &_graph;
   1.207 +    // The original lower bound map
   1.208 +    const LowerMap *_lower;
   1.209 +    // The modified capacity map
   1.210 +    CapacityArcMap _capacity;
   1.211 +    // The original cost map
   1.212 +    const CostMap &_cost;
   1.213 +    // The modified supply map
   1.214 +    SupplyNodeMap _supply;
   1.215 +    bool _valid_supply;
   1.216 +
   1.217 +    // Arc map of the current flow
   1.218 +    FlowMap *_flow;
   1.219 +    bool _local_flow;
   1.220 +    // Node map of the current potentials
   1.221 +    PotentialMap *_potential;
   1.222 +    bool _local_potential;
   1.223 +
   1.224 +    // The residual capacity map
   1.225 +    CapacityArcMap _res_cap;
   1.226 +    // The excess map
   1.227 +    SupplyNodeMap _excess;
   1.228 +    // The excess nodes (i.e. nodes with positive excess)
   1.229 +    std::vector<Node> _excess_nodes;
   1.230 +    // The deficit nodes (i.e. nodes with negative excess)
   1.231 +    std::vector<Node> _deficit_nodes;
   1.232 +
   1.233 +    // The delta parameter used for capacity scaling
   1.234 +    Capacity _delta;
   1.235 +    // The maximum number of phases
   1.236 +    int _phase_num;
   1.237 +
   1.238 +    // The pred arc map
   1.239 +    PredMap _pred;
   1.240 +    // Implementation of the Dijkstra algorithm for finding augmenting
   1.241 +    // shortest paths in the residual network
   1.242 +    ResidualDijkstra *_dijkstra;
   1.243 +
   1.244 +  public:
   1.245 +
   1.246 +    /// \brief General constructor (with lower bounds).
   1.247 +    ///
   1.248 +    /// General constructor (with lower bounds).
   1.249 +    ///
   1.250 +    /// \param digraph The digraph the algorithm runs on.
   1.251 +    /// \param lower The lower bounds of the arcs.
   1.252 +    /// \param capacity The capacities (upper bounds) of the arcs.
   1.253 +    /// \param cost The cost (length) values of the arcs.
   1.254 +    /// \param supply The supply values of the nodes (signed).
   1.255 +    CapacityScaling( const Digraph &digraph,
   1.256 +                     const LowerMap &lower,
   1.257 +                     const CapacityMap &capacity,
   1.258 +                     const CostMap &cost,
   1.259 +                     const SupplyMap &supply ) :
   1.260 +      _graph(digraph), _lower(&lower), _capacity(digraph), _cost(cost),
   1.261 +      _supply(digraph), _flow(NULL), _local_flow(false),
   1.262 +      _potential(NULL), _local_potential(false),
   1.263 +      _res_cap(digraph), _excess(digraph), _pred(digraph), _dijkstra(NULL)
   1.264 +    {
   1.265 +      Supply sum = 0;
   1.266 +      for (NodeIt n(_graph); n != INVALID; ++n) {
   1.267 +        _supply[n] = supply[n];
   1.268 +        _excess[n] = supply[n];
   1.269 +        sum += supply[n];
   1.270 +      }
   1.271 +      _valid_supply = sum == 0;
   1.272 +      for (ArcIt a(_graph); a != INVALID; ++a) {
   1.273 +        _capacity[a] = capacity[a];
   1.274 +        _res_cap[a] = capacity[a];
   1.275 +      }
   1.276 +
   1.277 +      // Remove non-zero lower bounds
   1.278 +      typename LowerMap::Value lcap;
   1.279 +      for (ArcIt e(_graph); e != INVALID; ++e) {
   1.280 +        if ((lcap = lower[e]) != 0) {
   1.281 +          _capacity[e] -= lcap;
   1.282 +          _res_cap[e] -= lcap;
   1.283 +          _supply[_graph.source(e)] -= lcap;
   1.284 +          _supply[_graph.target(e)] += lcap;
   1.285 +          _excess[_graph.source(e)] -= lcap;
   1.286 +          _excess[_graph.target(e)] += lcap;
   1.287 +        }
   1.288 +      }
   1.289 +    }
   1.290 +/*
   1.291 +    /// \brief General constructor (without lower bounds).
   1.292 +    ///
   1.293 +    /// General constructor (without lower bounds).
   1.294 +    ///
   1.295 +    /// \param digraph The digraph the algorithm runs on.
   1.296 +    /// \param capacity The capacities (upper bounds) of the arcs.
   1.297 +    /// \param cost The cost (length) values of the arcs.
   1.298 +    /// \param supply The supply values of the nodes (signed).
   1.299 +    CapacityScaling( const Digraph &digraph,
   1.300 +                     const CapacityMap &capacity,
   1.301 +                     const CostMap &cost,
   1.302 +                     const SupplyMap &supply ) :
   1.303 +      _graph(digraph), _lower(NULL), _capacity(capacity), _cost(cost),
   1.304 +      _supply(supply), _flow(NULL), _local_flow(false),
   1.305 +      _potential(NULL), _local_potential(false),
   1.306 +      _res_cap(capacity), _excess(supply), _pred(digraph), _dijkstra(NULL)
   1.307 +    {
   1.308 +      // Check the sum of supply values
   1.309 +      Supply sum = 0;
   1.310 +      for (NodeIt n(_graph); n != INVALID; ++n) sum += _supply[n];
   1.311 +      _valid_supply = sum == 0;
   1.312 +    }
   1.313 +
   1.314 +    /// \brief Simple constructor (with lower bounds).
   1.315 +    ///
   1.316 +    /// Simple constructor (with lower bounds).
   1.317 +    ///
   1.318 +    /// \param digraph The digraph the algorithm runs on.
   1.319 +    /// \param lower The lower bounds of the arcs.
   1.320 +    /// \param capacity The capacities (upper bounds) of the arcs.
   1.321 +    /// \param cost The cost (length) values of the arcs.
   1.322 +    /// \param s The source node.
   1.323 +    /// \param t The target node.
   1.324 +    /// \param flow_value The required amount of flow from node \c s
   1.325 +    /// to node \c t (i.e. the supply of \c s and the demand of \c t).
   1.326 +    CapacityScaling( const Digraph &digraph,
   1.327 +                     const LowerMap &lower,
   1.328 +                     const CapacityMap &capacity,
   1.329 +                     const CostMap &cost,
   1.330 +                     Node s, Node t,
   1.331 +                     Supply flow_value ) :
   1.332 +      _graph(digraph), _lower(&lower), _capacity(capacity), _cost(cost),
   1.333 +      _supply(digraph, 0), _flow(NULL), _local_flow(false),
   1.334 +      _potential(NULL), _local_potential(false),
   1.335 +      _res_cap(capacity), _excess(digraph, 0), _pred(digraph), _dijkstra(NULL)
   1.336 +    {
   1.337 +      // Remove non-zero lower bounds
   1.338 +      _supply[s] = _excess[s] =  flow_value;
   1.339 +      _supply[t] = _excess[t] = -flow_value;
   1.340 +      typename LowerMap::Value lcap;
   1.341 +      for (ArcIt e(_graph); e != INVALID; ++e) {
   1.342 +        if ((lcap = lower[e]) != 0) {
   1.343 +          _capacity[e] -= lcap;
   1.344 +          _res_cap[e] -= lcap;
   1.345 +          _supply[_graph.source(e)] -= lcap;
   1.346 +          _supply[_graph.target(e)] += lcap;
   1.347 +          _excess[_graph.source(e)] -= lcap;
   1.348 +          _excess[_graph.target(e)] += lcap;
   1.349 +        }
   1.350 +      }
   1.351 +      _valid_supply = true;
   1.352 +    }
   1.353 +
   1.354 +    /// \brief Simple constructor (without lower bounds).
   1.355 +    ///
   1.356 +    /// Simple constructor (without lower bounds).
   1.357 +    ///
   1.358 +    /// \param digraph The digraph the algorithm runs on.
   1.359 +    /// \param capacity The capacities (upper bounds) of the arcs.
   1.360 +    /// \param cost The cost (length) values of the arcs.
   1.361 +    /// \param s The source node.
   1.362 +    /// \param t The target node.
   1.363 +    /// \param flow_value The required amount of flow from node \c s
   1.364 +    /// to node \c t (i.e. the supply of \c s and the demand of \c t).
   1.365 +    CapacityScaling( const Digraph &digraph,
   1.366 +                     const CapacityMap &capacity,
   1.367 +                     const CostMap &cost,
   1.368 +                     Node s, Node t,
   1.369 +                     Supply flow_value ) :
   1.370 +      _graph(digraph), _lower(NULL), _capacity(capacity), _cost(cost),
   1.371 +      _supply(digraph, 0), _flow(NULL), _local_flow(false),
   1.372 +      _potential(NULL), _local_potential(false),
   1.373 +      _res_cap(capacity), _excess(digraph, 0), _pred(digraph), _dijkstra(NULL)
   1.374 +    {
   1.375 +      _supply[s] = _excess[s] =  flow_value;
   1.376 +      _supply[t] = _excess[t] = -flow_value;
   1.377 +      _valid_supply = true;
   1.378 +    }
   1.379 +*/
   1.380 +    /// Destructor.
   1.381 +    ~CapacityScaling() {
   1.382 +      if (_local_flow) delete _flow;
   1.383 +      if (_local_potential) delete _potential;
   1.384 +      delete _dijkstra;
   1.385 +    }
   1.386 +
   1.387 +    /// \brief Set the flow map.
   1.388 +    ///
   1.389 +    /// Set the flow map.
   1.390 +    ///
   1.391 +    /// \return \c (*this)
   1.392 +    CapacityScaling& flowMap(FlowMap &map) {
   1.393 +      if (_local_flow) {
   1.394 +        delete _flow;
   1.395 +        _local_flow = false;
   1.396 +      }
   1.397 +      _flow = &map;
   1.398 +      return *this;
   1.399 +    }
   1.400 +
   1.401 +    /// \brief Set the potential map.
   1.402 +    ///
   1.403 +    /// Set the potential map.
   1.404 +    ///
   1.405 +    /// \return \c (*this)
   1.406 +    CapacityScaling& potentialMap(PotentialMap &map) {
   1.407 +      if (_local_potential) {
   1.408 +        delete _potential;
   1.409 +        _local_potential = false;
   1.410 +      }
   1.411 +      _potential = &map;
   1.412 +      return *this;
   1.413 +    }
   1.414 +
   1.415 +    /// \name Execution control
   1.416 +
   1.417 +    /// @{
   1.418 +
   1.419 +    /// \brief Run the algorithm.
   1.420 +    ///
   1.421 +    /// This function runs the algorithm.
   1.422 +    ///
   1.423 +    /// \param scaling Enable or disable capacity scaling.
   1.424 +    /// If the maximum arc capacity and/or the amount of total supply
   1.425 +    /// is rather small, the algorithm could be slightly faster without
   1.426 +    /// scaling.
   1.427 +    ///
   1.428 +    /// \return \c true if a feasible flow can be found.
   1.429 +    bool run(bool scaling = true) {
   1.430 +      return init(scaling) && start();
   1.431 +    }
   1.432 +
   1.433 +    /// @}
   1.434 +
   1.435 +    /// \name Query Functions
   1.436 +    /// The results of the algorithm can be obtained using these
   1.437 +    /// functions.\n
   1.438 +    /// \ref lemon::CapacityScaling::run() "run()" must be called before
   1.439 +    /// using them.
   1.440 +
   1.441 +    /// @{
   1.442 +
   1.443 +    /// \brief Return a const reference to the arc map storing the
   1.444 +    /// found flow.
   1.445 +    ///
   1.446 +    /// Return a const reference to the arc map storing the found flow.
   1.447 +    ///
   1.448 +    /// \pre \ref run() must be called before using this function.
   1.449 +    const FlowMap& flowMap() const {
   1.450 +      return *_flow;
   1.451 +    }
   1.452 +
   1.453 +    /// \brief Return a const reference to the node map storing the
   1.454 +    /// found potentials (the dual solution).
   1.455 +    ///
   1.456 +    /// Return a const reference to the node map storing the found
   1.457 +    /// potentials (the dual solution).
   1.458 +    ///
   1.459 +    /// \pre \ref run() must be called before using this function.
   1.460 +    const PotentialMap& potentialMap() const {
   1.461 +      return *_potential;
   1.462 +    }
   1.463 +
   1.464 +    /// \brief Return the flow on the given arc.
   1.465 +    ///
   1.466 +    /// Return the flow on the given arc.
   1.467 +    ///
   1.468 +    /// \pre \ref run() must be called before using this function.
   1.469 +    Capacity flow(const Arc& arc) const {
   1.470 +      return (*_flow)[arc];
   1.471 +    }
   1.472 +
   1.473 +    /// \brief Return the potential of the given node.
   1.474 +    ///
   1.475 +    /// Return the potential of the given node.
   1.476 +    ///
   1.477 +    /// \pre \ref run() must be called before using this function.
   1.478 +    Cost potential(const Node& node) const {
   1.479 +      return (*_potential)[node];
   1.480 +    }
   1.481 +
   1.482 +    /// \brief Return the total cost of the found flow.
   1.483 +    ///
   1.484 +    /// Return the total cost of the found flow. The complexity of the
   1.485 +    /// function is \f$ O(e) \f$.
   1.486 +    ///
   1.487 +    /// \pre \ref run() must be called before using this function.
   1.488 +    Cost totalCost() const {
   1.489 +      Cost c = 0;
   1.490 +      for (ArcIt e(_graph); e != INVALID; ++e)
   1.491 +        c += (*_flow)[e] * _cost[e];
   1.492 +      return c;
   1.493 +    }
   1.494 +
   1.495 +    /// @}
   1.496 +
   1.497 +  private:
   1.498 +
   1.499 +    /// Initialize the algorithm.
   1.500 +    bool init(bool scaling) {
   1.501 +      if (!_valid_supply) return false;
   1.502 +
   1.503 +      // Initializing maps
   1.504 +      if (!_flow) {
   1.505 +        _flow = new FlowMap(_graph);
   1.506 +        _local_flow = true;
   1.507 +      }
   1.508 +      if (!_potential) {
   1.509 +        _potential = new PotentialMap(_graph);
   1.510 +        _local_potential = true;
   1.511 +      }
   1.512 +      for (ArcIt e(_graph); e != INVALID; ++e) (*_flow)[e] = 0;
   1.513 +      for (NodeIt n(_graph); n != INVALID; ++n) (*_potential)[n] = 0;
   1.514 +
   1.515 +      _dijkstra = new ResidualDijkstra( _graph, *_flow, _res_cap, _cost,
   1.516 +                                        _excess, *_potential, _pred );
   1.517 +
   1.518 +      // Initializing delta value
   1.519 +      if (scaling) {
   1.520 +        // With scaling
   1.521 +        Supply max_sup = 0, max_dem = 0;
   1.522 +        for (NodeIt n(_graph); n != INVALID; ++n) {
   1.523 +          if ( _supply[n] > max_sup) max_sup =  _supply[n];
   1.524 +          if (-_supply[n] > max_dem) max_dem = -_supply[n];
   1.525 +        }
   1.526 +        Capacity max_cap = 0;
   1.527 +        for (ArcIt e(_graph); e != INVALID; ++e) {
   1.528 +          if (_capacity[e] > max_cap) max_cap = _capacity[e];
   1.529 +        }
   1.530 +        max_sup = std::min(std::min(max_sup, max_dem), max_cap);
   1.531 +        _phase_num = 0;
   1.532 +        for (_delta = 1; 2 * _delta <= max_sup; _delta *= 2)
   1.533 +          ++_phase_num;
   1.534 +      } else {
   1.535 +        // Without scaling
   1.536 +        _delta = 1;
   1.537 +      }
   1.538 +
   1.539 +      return true;
   1.540 +    }
   1.541 +
   1.542 +    bool start() {
   1.543 +      if (_delta > 1)
   1.544 +        return startWithScaling();
   1.545 +      else
   1.546 +        return startWithoutScaling();
   1.547 +    }
   1.548 +
   1.549 +    /// Execute the capacity scaling algorithm.
   1.550 +    bool startWithScaling() {
   1.551 +      // Processing capacity scaling phases
   1.552 +      Node s, t;
   1.553 +      int phase_cnt = 0;
   1.554 +      int factor = 4;
   1.555 +      while (true) {
   1.556 +        // Saturating all arcs not satisfying the optimality condition
   1.557 +        for (ArcIt e(_graph); e != INVALID; ++e) {
   1.558 +          Node u = _graph.source(e), v = _graph.target(e);
   1.559 +          Cost c = _cost[e] + (*_potential)[u] - (*_potential)[v];
   1.560 +          if (c < 0 && _res_cap[e] >= _delta) {
   1.561 +            _excess[u] -= _res_cap[e];
   1.562 +            _excess[v] += _res_cap[e];
   1.563 +            (*_flow)[e] = _capacity[e];
   1.564 +            _res_cap[e] = 0;
   1.565 +          }
   1.566 +          else if (c > 0 && (*_flow)[e] >= _delta) {
   1.567 +            _excess[u] += (*_flow)[e];
   1.568 +            _excess[v] -= (*_flow)[e];
   1.569 +            (*_flow)[e] = 0;
   1.570 +            _res_cap[e] = _capacity[e];
   1.571 +          }
   1.572 +        }
   1.573 +
   1.574 +        // Finding excess nodes and deficit nodes
   1.575 +        _excess_nodes.clear();
   1.576 +        _deficit_nodes.clear();
   1.577 +        for (NodeIt n(_graph); n != INVALID; ++n) {
   1.578 +          if (_excess[n] >=  _delta) _excess_nodes.push_back(n);
   1.579 +          if (_excess[n] <= -_delta) _deficit_nodes.push_back(n);
   1.580 +        }
   1.581 +        int next_node = 0, next_def_node = 0;
   1.582 +
   1.583 +        // Finding augmenting shortest paths
   1.584 +        while (next_node < int(_excess_nodes.size())) {
   1.585 +          // Checking deficit nodes
   1.586 +          if (_delta > 1) {
   1.587 +            bool delta_deficit = false;
   1.588 +            for ( ; next_def_node < int(_deficit_nodes.size());
   1.589 +                    ++next_def_node ) {
   1.590 +              if (_excess[_deficit_nodes[next_def_node]] <= -_delta) {
   1.591 +                delta_deficit = true;
   1.592 +                break;
   1.593 +              }
   1.594 +            }
   1.595 +            if (!delta_deficit) break;
   1.596 +          }
   1.597 +
   1.598 +          // Running Dijkstra
   1.599 +          s = _excess_nodes[next_node];
   1.600 +          if ((t = _dijkstra->run(s, _delta)) == INVALID) {
   1.601 +            if (_delta > 1) {
   1.602 +              ++next_node;
   1.603 +              continue;
   1.604 +            }
   1.605 +            return false;
   1.606 +          }
   1.607 +
   1.608 +          // Augmenting along a shortest path from s to t.
   1.609 +          Capacity d = std::min(_excess[s], -_excess[t]);
   1.610 +          Node u = t;
   1.611 +          Arc e;
   1.612 +          if (d > _delta) {
   1.613 +            while ((e = _pred[u]) != INVALID) {
   1.614 +              Capacity rc;
   1.615 +              if (u == _graph.target(e)) {
   1.616 +                rc = _res_cap[e];
   1.617 +                u = _graph.source(e);
   1.618 +              } else {
   1.619 +                rc = (*_flow)[e];
   1.620 +                u = _graph.target(e);
   1.621 +              }
   1.622 +              if (rc < d) d = rc;
   1.623 +            }
   1.624 +          }
   1.625 +          u = t;
   1.626 +          while ((e = _pred[u]) != INVALID) {
   1.627 +            if (u == _graph.target(e)) {
   1.628 +              (*_flow)[e] += d;
   1.629 +              _res_cap[e] -= d;
   1.630 +              u = _graph.source(e);
   1.631 +            } else {
   1.632 +              (*_flow)[e] -= d;
   1.633 +              _res_cap[e] += d;
   1.634 +              u = _graph.target(e);
   1.635 +            }
   1.636 +          }
   1.637 +          _excess[s] -= d;
   1.638 +          _excess[t] += d;
   1.639 +
   1.640 +          if (_excess[s] < _delta) ++next_node;
   1.641 +        }
   1.642 +
   1.643 +        if (_delta == 1) break;
   1.644 +        if (++phase_cnt > _phase_num / 4) factor = 2;
   1.645 +        _delta = _delta <= factor ? 1 : _delta / factor;
   1.646 +      }
   1.647 +
   1.648 +      // Handling non-zero lower bounds
   1.649 +      if (_lower) {
   1.650 +        for (ArcIt e(_graph); e != INVALID; ++e)
   1.651 +          (*_flow)[e] += (*_lower)[e];
   1.652 +      }
   1.653 +      return true;
   1.654 +    }
   1.655 +
   1.656 +    /// Execute the successive shortest path algorithm.
   1.657 +    bool startWithoutScaling() {
   1.658 +      // Finding excess nodes
   1.659 +      for (NodeIt n(_graph); n != INVALID; ++n)
   1.660 +        if (_excess[n] > 0) _excess_nodes.push_back(n);
   1.661 +      if (_excess_nodes.size() == 0) return true;
   1.662 +      int next_node = 0;
   1.663 +
   1.664 +      // Finding shortest paths
   1.665 +      Node s, t;
   1.666 +      while ( _excess[_excess_nodes[next_node]] > 0 ||
   1.667 +              ++next_node < int(_excess_nodes.size()) )
   1.668 +      {
   1.669 +        // Running Dijkstra
   1.670 +        s = _excess_nodes[next_node];
   1.671 +        if ((t = _dijkstra->run(s)) == INVALID) return false;
   1.672 +
   1.673 +        // Augmenting along a shortest path from s to t
   1.674 +        Capacity d = std::min(_excess[s], -_excess[t]);
   1.675 +        Node u = t;
   1.676 +        Arc e;
   1.677 +        if (d > 1) {
   1.678 +          while ((e = _pred[u]) != INVALID) {
   1.679 +            Capacity rc;
   1.680 +            if (u == _graph.target(e)) {
   1.681 +              rc = _res_cap[e];
   1.682 +              u = _graph.source(e);
   1.683 +            } else {
   1.684 +              rc = (*_flow)[e];
   1.685 +              u = _graph.target(e);
   1.686 +            }
   1.687 +            if (rc < d) d = rc;
   1.688 +          }
   1.689 +        }
   1.690 +        u = t;
   1.691 +        while ((e = _pred[u]) != INVALID) {
   1.692 +          if (u == _graph.target(e)) {
   1.693 +            (*_flow)[e] += d;
   1.694 +            _res_cap[e] -= d;
   1.695 +            u = _graph.source(e);
   1.696 +          } else {
   1.697 +            (*_flow)[e] -= d;
   1.698 +            _res_cap[e] += d;
   1.699 +            u = _graph.target(e);
   1.700 +          }
   1.701 +        }
   1.702 +        _excess[s] -= d;
   1.703 +        _excess[t] += d;
   1.704 +      }
   1.705 +
   1.706 +      // Handling non-zero lower bounds
   1.707 +      if (_lower) {
   1.708 +        for (ArcIt e(_graph); e != INVALID; ++e)
   1.709 +          (*_flow)[e] += (*_lower)[e];
   1.710 +      }
   1.711 +      return true;
   1.712 +    }
   1.713 +
   1.714 +  }; //class CapacityScaling
   1.715 +
   1.716 +  ///@}
   1.717 +
   1.718 +} //namespace lemon
   1.719 +
   1.720 +#endif //LEMON_CAPACITY_SCALING_H