1 /* -*- mode: C++; indent-tabs-mode: nil; -*-
3 * This file is a part of LEMON, a generic C++ optimization library.
5 * Copyright (C) 2003-2010
6 * Egervary Jeno Kombinatorikus Optimalizalasi Kutatocsoport
7 * (Egervary Research Group on Combinatorial Optimization, EGRES).
9 * Permission to use, modify and distribute this software is granted
10 * provided that this copyright notice appears in all copies. For
11 * precise terms see the accompanying LICENSE file.
13 * This software is provided "AS IS" with no warranty of any kind,
14 * express or implied, and with no claim as to its suitability for any
19 #ifndef LEMON_NAGAMOCHI_IBARAKI_H
20 #define LEMON_NAGAMOCHI_IBARAKI_H
25 /// \brief Implementation of the Nagamochi-Ibaraki algorithm.
27 #include <lemon/core.h>
28 #include <lemon/bin_heap.h>
29 #include <lemon/bucket_heap.h>
30 #include <lemon/maps.h>
31 #include <lemon/radix_sort.h>
32 #include <lemon/unionfind.h>
38 /// \brief Default traits class for NagamochiIbaraki class.
40 /// Default traits class for NagamochiIbaraki class.
41 /// \param GR The undirected graph type.
42 /// \param CM Type of capacity map.
43 template <typename GR, typename CM>
44 struct NagamochiIbarakiDefaultTraits {
45 /// The type of the capacity map.
46 typedef typename CM::Value Value;
48 /// The undirected graph type the algorithm runs on.
51 /// \brief The type of the map that stores the edge capacities.
53 /// The type of the map that stores the edge capacities.
54 /// It must meet the \ref concepts::ReadMap "ReadMap" concept.
55 typedef CM CapacityMap;
57 /// \brief Instantiates a CapacityMap.
59 /// This function instantiates a \ref CapacityMap.
61 static CapacityMap *createCapacityMap(const Graph& graph)
63 static CapacityMap *createCapacityMap(const Graph&)
66 LEMON_ASSERT(false, "CapacityMap is not initialized");
67 return 0; // ignore warnings
70 /// \brief The cross reference type used by heap.
72 /// The cross reference type used by heap.
73 /// Usually \c Graph::NodeMap<int>.
74 typedef typename Graph::template NodeMap<int> HeapCrossRef;
76 /// \brief Instantiates a HeapCrossRef.
78 /// This function instantiates a \ref HeapCrossRef.
79 /// \param g is the graph, to which we would like to define the
80 /// \ref HeapCrossRef.
81 static HeapCrossRef *createHeapCrossRef(const Graph& g) {
82 return new HeapCrossRef(g);
85 /// \brief The heap type used by NagamochiIbaraki algorithm.
87 /// The heap type used by NagamochiIbaraki algorithm. It has to
88 /// maximize the priorities.
91 /// \sa NagamochiIbaraki
92 typedef BinHeap<Value, HeapCrossRef, std::greater<Value> > Heap;
94 /// \brief Instantiates a Heap.
96 /// This function instantiates a \ref Heap.
97 /// \param r is the cross reference of the heap.
98 static Heap *createHeap(HeapCrossRef& r) {
105 /// \brief Calculates the minimum cut in an undirected graph.
107 /// Calculates the minimum cut in an undirected graph with the
108 /// Nagamochi-Ibaraki algorithm. The algorithm separates the graph's
109 /// nodes into two partitions with the minimum sum of edge capacities
110 /// between the two partitions. The algorithm can be used to test
111 /// the network reliability, especially to test how many links have
112 /// to be destroyed in the network to split it to at least two
113 /// distinict subnetworks.
115 /// The complexity of the algorithm is \f$ O(nm\log(n)) \f$ but with
116 /// \ref FibHeap "Fibonacci heap" it can be decreased to
117 /// \f$ O(nm+n^2\log(n)) \f$. When the edges have unit capacities,
118 /// \c BucketHeap can be used which yields \f$ O(nm) \f$ time
121 /// \warning The value type of the capacity map should be able to
122 /// hold any cut value of the graph, otherwise the result can
124 /// \note This capacity is supposed to be integer type.
126 template <typename GR, typename CM, typename TR>
128 template <typename GR,
129 typename CM = typename GR::template EdgeMap<int>,
130 typename TR = NagamochiIbarakiDefaultTraits<GR, CM> >
132 class NagamochiIbaraki {
136 /// The type of the underlying graph.
137 typedef typename Traits::Graph Graph;
139 /// The type of the capacity map.
140 typedef typename Traits::CapacityMap CapacityMap;
141 /// The value type of the capacity map.
142 typedef typename Traits::CapacityMap::Value Value;
144 /// The heap type used by the algorithm.
145 typedef typename Traits::Heap Heap;
146 /// The cross reference type used for the heap.
147 typedef typename Traits::HeapCrossRef HeapCrossRef;
149 ///\name Named template parameters
153 struct SetUnitCapacityTraits : public Traits {
154 typedef ConstMap<typename Graph::Edge, Const<int, 1> > CapacityMap;
155 static CapacityMap *createCapacityMap(const Graph&) {
156 return new CapacityMap();
160 /// \brief \ref named-templ-param "Named parameter" for setting
161 /// the capacity map to a constMap<Edge, int, 1>() instance
163 /// \ref named-templ-param "Named parameter" for setting
164 /// the capacity map to a constMap<Edge, int, 1>() instance
165 struct SetUnitCapacity
166 : public NagamochiIbaraki<Graph, CapacityMap,
167 SetUnitCapacityTraits> {
168 typedef NagamochiIbaraki<Graph, CapacityMap,
169 SetUnitCapacityTraits> Create;
173 template <class H, class CR>
174 struct SetHeapTraits : public Traits {
175 typedef CR HeapCrossRef;
177 static HeapCrossRef *createHeapCrossRef(int num) {
178 LEMON_ASSERT(false, "HeapCrossRef is not initialized");
179 return 0; // ignore warnings
181 static Heap *createHeap(HeapCrossRef &) {
182 LEMON_ASSERT(false, "Heap is not initialized");
183 return 0; // ignore warnings
187 /// \brief \ref named-templ-param "Named parameter" for setting
188 /// heap and cross reference type
190 /// \ref named-templ-param "Named parameter" for setting heap and
191 /// cross reference type. The heap has to maximize the priorities.
192 template <class H, class CR = RangeMap<int> >
194 : public NagamochiIbaraki<Graph, CapacityMap, SetHeapTraits<H, CR> > {
195 typedef NagamochiIbaraki< Graph, CapacityMap, SetHeapTraits<H, CR> >
199 template <class H, class CR>
200 struct SetStandardHeapTraits : public Traits {
201 typedef CR HeapCrossRef;
203 static HeapCrossRef *createHeapCrossRef(int size) {
204 return new HeapCrossRef(size);
206 static Heap *createHeap(HeapCrossRef &crossref) {
207 return new Heap(crossref);
211 /// \brief \ref named-templ-param "Named parameter" for setting
212 /// heap and cross reference type with automatic allocation
214 /// \ref named-templ-param "Named parameter" for setting heap and
215 /// cross reference type with automatic allocation. They should
216 /// have standard constructor interfaces to be able to
217 /// automatically created by the algorithm (i.e. the graph should
218 /// be passed to the constructor of the cross reference and the
219 /// cross reference should be passed to the constructor of the
220 /// heap). However, external heap and cross reference objects
221 /// could also be passed to the algorithm using the \ref heap()
222 /// function before calling \ref run() or \ref init(). The heap
223 /// has to maximize the priorities.
225 template <class H, class CR = RangeMap<int> >
226 struct SetStandardHeap
227 : public NagamochiIbaraki<Graph, CapacityMap,
228 SetStandardHeapTraits<H, CR> > {
229 typedef NagamochiIbaraki<Graph, CapacityMap,
230 SetStandardHeapTraits<H, CR> > Create;
239 const CapacityMap *_capacity;
240 bool _local_capacity; // unit capacity
243 typename Graph::Node target;
253 typename Graph::Node prev, next;
255 typename Graph::Node last_rep;
259 typename Graph::template NodeMap<NodeData> *_nodes;
260 std::vector<ArcData> _arcs;
261 std::vector<EdgeData> _edges;
263 typename Graph::Node _first_node;
268 HeapCrossRef *_heap_cross_ref;
269 bool _local_heap_cross_ref;
273 typedef typename Graph::template NodeMap<typename Graph::Node> NodeList;
276 typedef typename Graph::template NodeMap<bool> MinCutMap;
279 void createStructures() {
281 _nodes = new (typename Graph::template NodeMap<NodeData>)(_graph);
284 _local_capacity = true;
285 _capacity = Traits::createCapacityMap(_graph);
287 if (!_heap_cross_ref) {
288 _local_heap_cross_ref = true;
289 _heap_cross_ref = Traits::createHeapCrossRef(_graph);
293 _heap = Traits::createHeap(*_heap_cross_ref);
296 _next_rep = new NodeList(_graph);
299 _cut_map = new MinCutMap(_graph);
304 //This is here to avoid a gcc-3.3 compilation error.
305 //It should never be called.
306 NagamochiIbaraki() {}
310 typedef NagamochiIbaraki Create;
313 /// \brief Constructor.
315 /// \param graph The graph the algorithm runs on.
316 /// \param capacity The capacity map used by the algorithm.
317 NagamochiIbaraki(const Graph& graph, const CapacityMap& capacity)
318 : _graph(graph), _capacity(&capacity), _local_capacity(false),
319 _nodes(0), _arcs(), _edges(), _min_cut(),
320 _heap_cross_ref(0), _local_heap_cross_ref(false),
321 _heap(0), _local_heap(false),
322 _next_rep(0), _cut_map(0) {}
324 /// \brief Constructor.
326 /// This constructor can be used only when the Traits class
327 /// defines how can the local capacity map be instantiated.
328 /// If the SetUnitCapacity used the algorithm automatically
329 /// constructs the capacity map.
331 ///\param graph The graph the algorithm runs on.
332 NagamochiIbaraki(const Graph& graph)
333 : _graph(graph), _capacity(0), _local_capacity(false),
334 _nodes(0), _arcs(), _edges(), _min_cut(),
335 _heap_cross_ref(0), _local_heap_cross_ref(false),
336 _heap(0), _local_heap(false),
337 _next_rep(0), _cut_map(0) {}
339 /// \brief Destructor.
342 ~NagamochiIbaraki() {
343 if (_local_capacity) delete _capacity;
344 if (_nodes) delete _nodes;
345 if (_local_heap) delete _heap;
346 if (_local_heap_cross_ref) delete _heap_cross_ref;
347 if (_next_rep) delete _next_rep;
348 if (_cut_map) delete _cut_map;
351 /// \brief Sets the heap and the cross reference used by algorithm.
353 /// Sets the heap and the cross reference used by algorithm.
354 /// If you don't use this function before calling \ref run(),
355 /// it will allocate one. The destuctor deallocates this
356 /// automatically allocated heap and cross reference, of course.
357 /// \return <tt> (*this) </tt>
358 NagamochiIbaraki &heap(Heap& hp, HeapCrossRef &cr)
360 if (_local_heap_cross_ref) {
361 delete _heap_cross_ref;
362 _local_heap_cross_ref = false;
364 _heap_cross_ref = &cr;
373 /// \name Execution control
374 /// The simplest way to execute the algorithm is to use
375 /// one of the member functions called \c run().
377 /// If you need more control on the execution,
378 /// first you must call \ref init() and then call the start()
379 /// or proper times the processNextPhase() member functions.
383 /// \brief Initializes the internal data structures.
385 /// Initializes the internal data structures.
389 int edge_num = countEdges(_graph);
390 _edges.resize(edge_num);
391 _arcs.resize(2 * edge_num);
393 typename Graph::Node prev = INVALID;
395 for (typename Graph::NodeIt n(_graph); n != INVALID; ++n) {
396 (*_cut_map)[n] = false;
397 (*_next_rep)[n] = INVALID;
398 (*_nodes)[n].last_rep = n;
399 (*_nodes)[n].first_arc = -1;
400 (*_nodes)[n].curr_arc = -1;
401 (*_nodes)[n].prev = prev;
402 if (prev != INVALID) {
403 (*_nodes)[prev].next = n;
405 (*_nodes)[n].next = INVALID;
406 (*_nodes)[n].sum = 0;
411 _first_node = typename Graph::NodeIt(_graph);
414 for (typename Graph::NodeIt n(_graph); n != INVALID; ++n) {
415 for (typename Graph::OutArcIt a(_graph, n); a != INVALID; ++a) {
416 typename Graph::Node m = _graph.target(a);
418 if (!(n < m)) continue;
420 (*_nodes)[n].sum += (*_capacity)[a];
421 (*_nodes)[m].sum += (*_capacity)[a];
423 int c = (*_nodes)[m].curr_arc;
424 if (c != -1 && _arcs[c ^ 1].target == n) {
425 _edges[c >> 1].capacity += (*_capacity)[a];
427 _edges[index].capacity = (*_capacity)[a];
429 _arcs[index << 1].prev = -1;
430 if ((*_nodes)[n].first_arc != -1) {
431 _arcs[(*_nodes)[n].first_arc].prev = (index << 1);
433 _arcs[index << 1].next = (*_nodes)[n].first_arc;
434 (*_nodes)[n].first_arc = (index << 1);
435 _arcs[index << 1].target = m;
437 (*_nodes)[m].curr_arc = (index << 1);
439 _arcs[(index << 1) | 1].prev = -1;
440 if ((*_nodes)[m].first_arc != -1) {
441 _arcs[(*_nodes)[m].first_arc].prev = ((index << 1) | 1);
443 _arcs[(index << 1) | 1].next = (*_nodes)[m].first_arc;
444 (*_nodes)[m].first_arc = ((index << 1) | 1);
445 _arcs[(index << 1) | 1].target = n;
452 typename Graph::Node cut_node = INVALID;
453 _min_cut = std::numeric_limits<Value>::max();
455 for (typename Graph::Node n = _first_node;
456 n != INVALID; n = (*_nodes)[n].next) {
457 if ((*_nodes)[n].sum < _min_cut) {
459 _min_cut = (*_nodes)[n].sum;
462 (*_cut_map)[cut_node] = true;
464 _first_node = INVALID;
470 /// \brief Processes the next phase
472 /// Processes the next phase in the algorithm. It must be called
473 /// at most one less the number of the nodes in the graph.
475 ///\return %True when the algorithm finished.
476 bool processNextPhase() {
477 if (_first_node == INVALID) return true;
480 for (typename Graph::Node n = _first_node;
481 n != INVALID; n = (*_nodes)[n].next) {
482 (*_heap_cross_ref)[n] = Heap::PRE_HEAP;
485 std::vector<typename Graph::Node> order;
486 order.reserve(_node_num);
490 Value pmc = std::numeric_limits<Value>::max();
492 _heap->push(_first_node, static_cast<Value>(0));
493 while (!_heap->empty()) {
494 typename Graph::Node n = _heap->top();
495 Value v = _heap->prio();
498 for (int a = (*_nodes)[n].first_arc; a != -1; a = _arcs[a].next) {
499 switch (_heap->state(_arcs[a].target)) {
502 Value nv = _edges[a >> 1].capacity;
503 _heap->push(_arcs[a].target, nv);
504 _edges[a >> 1].cut = nv;
508 Value nv = _edges[a >> 1].capacity + (*_heap)[_arcs[a].target];
509 _heap->decrease(_arcs[a].target, nv);
510 _edges[a >> 1].cut = nv;
512 case Heap::POST_HEAP:
517 alpha += (*_nodes)[n].sum;
521 if (!_heap->empty()) {
529 if (static_cast<int>(order.size()) < _node_num) {
530 _first_node = INVALID;
531 for (typename Graph::NodeIt n(_graph); n != INVALID; ++n) {
532 (*_cut_map)[n] = false;
534 for (int i = 0; i < static_cast<int>(order.size()); ++i) {
535 typename Graph::Node n = order[i];
536 while (n != INVALID) {
537 (*_cut_map)[n] = true;
545 if (pmc < _min_cut) {
546 for (typename Graph::NodeIt n(_graph); n != INVALID; ++n) {
547 (*_cut_map)[n] = false;
549 for (int i = 0; i < sep; ++i) {
550 typename Graph::Node n = order[i];
551 while (n != INVALID) {
552 (*_cut_map)[n] = true;
559 for (typename Graph::Node n = _first_node;
560 n != INVALID; n = (*_nodes)[n].next) {
562 for (int a = (*_nodes)[n].first_arc; a != -1; a = _arcs[a].next) {
563 if (!(_edges[a >> 1].cut < pmc)) {
565 for (int b = (*_nodes)[n].first_arc; b != -1; b = _arcs[b].next) {
566 (*_nodes)[_arcs[b].target].curr_arc = b;
570 typename Graph::Node m = _arcs[a].target;
572 for (int b = (*_nodes)[m].first_arc; b != -1; b = nb) {
574 if ((b ^ a) == 1) continue;
575 typename Graph::Node o = _arcs[b].target;
576 int c = (*_nodes)[o].curr_arc;
577 if (c != -1 && _arcs[c ^ 1].target == n) {
578 _edges[c >> 1].capacity += _edges[b >> 1].capacity;
579 (*_nodes)[n].sum += _edges[b >> 1].capacity;
580 if (_edges[b >> 1].cut < _edges[c >> 1].cut) {
581 _edges[b >> 1].cut = _edges[c >> 1].cut;
583 if (_arcs[b ^ 1].prev != -1) {
584 _arcs[_arcs[b ^ 1].prev].next = _arcs[b ^ 1].next;
586 (*_nodes)[o].first_arc = _arcs[b ^ 1].next;
588 if (_arcs[b ^ 1].next != -1) {
589 _arcs[_arcs[b ^ 1].next].prev = _arcs[b ^ 1].prev;
592 if (_arcs[a].next != -1) {
593 _arcs[_arcs[a].next].prev = b;
595 _arcs[b].next = _arcs[a].next;
598 _arcs[b ^ 1].target = n;
600 (*_nodes)[n].sum += _edges[b >> 1].capacity;
601 (*_nodes)[o].curr_arc = b;
605 if (_arcs[a].prev != -1) {
606 _arcs[_arcs[a].prev].next = _arcs[a].next;
608 (*_nodes)[n].first_arc = _arcs[a].next;
610 if (_arcs[a].next != -1) {
611 _arcs[_arcs[a].next].prev = _arcs[a].prev;
614 (*_nodes)[n].sum -= _edges[a >> 1].capacity;
615 (*_next_rep)[(*_nodes)[n].last_rep] = m;
616 (*_nodes)[n].last_rep = (*_nodes)[m].last_rep;
618 if ((*_nodes)[m].prev != INVALID) {
619 (*_nodes)[(*_nodes)[m].prev].next = (*_nodes)[m].next;
621 _first_node = (*_nodes)[m].next;
623 if ((*_nodes)[m].next != INVALID) {
624 (*_nodes)[(*_nodes)[m].next].prev = (*_nodes)[m].prev;
631 if (_node_num == 1) {
632 _first_node = INVALID;
639 /// \brief Executes the algorithm.
641 /// Executes the algorithm.
643 /// \pre init() must be called
645 while (!processNextPhase()) {}
649 /// \brief Runs %NagamochiIbaraki algorithm.
651 /// This method runs the %Min cut algorithm
653 /// \note mc.run(s) is just a shortcut of the following code.
665 /// \name Query Functions
667 /// The result of the %NagamochiIbaraki
668 /// algorithm can be obtained using these functions.\n
669 /// Before the use of these functions, either run() or start()
674 /// \brief Returns the min cut value.
676 /// Returns the min cut value if the algorithm finished.
677 /// After the first processNextPhase() it is a value of a
678 /// valid cut in the graph.
679 Value minCutValue() const {
683 /// \brief Returns a min cut in a NodeMap.
685 /// It sets the nodes of one of the two partitions to true and
686 /// the other partition to false.
687 /// \param cutMap A \ref concepts::WriteMap "writable" node map with
688 /// \c bool (or convertible) value type.
689 template <typename CutMap>
690 Value minCutMap(CutMap& cutMap) const {
691 for (typename Graph::NodeIt n(_graph); n != INVALID; ++n) {
692 cutMap.set(n, (*_cut_map)[n]);
694 return minCutValue();