Part III: Traveling salesman problems

Transcription

1 Transportation Logistics Part III: Traveling salesman problems c R.F. Hartl, S.N. Parragh 1/282

2 Motivation Motivation Why do we study the TSP? c R.F. Hartl, S.N. Parragh 2/282

3 Motivation Motivation Why do we study the TSP? it easy to formulate it is a difficult problem many significant real-world problems can be formulated as TSP c R.F. Hartl, S.N. Parragh 3/282

4 Motivation Application areas of the TSP Vehicle routing Once the assignment of customers to routes has been done. A TSP (possibly with side constraints) has to be solved for each vehicle. Cutting wallpaper Assume n sheets of wallpaper shall be cut from a single roll of paper. The amount of paper that is wasted varies depending on which sheet j is cut from the roll directly after i. We want to minimize the total wastage. c R.F. Hartl, S.N. Parragh 4/282

5 Motivation Application areas of the TSP Job sequencing n jobs shall be scheduled on a single machine. The jobs can be done in any order. Assume that we have sequence dependent setup times, i.e. depending on which job i precedes job j the time needed to adapt the machine to be able to process job j varies. All jobs shall be completed in the shortest possible time. Clustering a data array Computer wiring... c R.F. Hartl, S.N. Parragh 5/282

6 The asymmetric TSP The asymmetric TSP x ij = { 1, if arc (ij) is part of the solution, 0, otherwise. c ij = the costs to traverse arc (i,j) A...set of arcs, V...set of vertices (TSP is formulated on a complete directed graph) c R.F. Hartl, S.N. Parragh 6/282

7 The asymmetric TSP The asymmetric TSP x ij = { 1, if arc (ij) is part of the solution, 0, otherwise. c ij = the costs to traverse arc (i,j) A...set of arcs, V...set of vertices (TSP is formulated on a complete directed graph) IP Formulation c ij x ij min (1) (i,j) A x ij = 1 j V, (2) i V \{j} x ij = 1 i V, (3) j V \{i} x ij 1 S V, S 2, (4) i S j/ S x ij {0,1} (i,j) A. (5) c R.F. Hartl, S.N. Parragh 7/282

8 The asymmetric TSP The asymmetric TSP x ij = { 1, if arc (ij) is part of the solution, 0, otherwise. c ij = the costs to traverse arc (i,j) A...set of arcs, V...set of vertices (TSP is formulated on a complete directed graph) IP Formulation c ij x ij min (1) (i,j) A x ij = 1 j V, (2) i V \{j} x ij = 1 i V, (3) j V \{i} x ij 1 S V, S 2, (4) i S j/ S x ij {0,1} (i,j) A. (5) c R.F. Hartl, S.N. Parragh 8/282

9 The asymmetric TSP The asymmetric TSP - connectivity/subtour elimination Option 1 (connectivity constraints) x ij 1 S V, S 2 (6) i S j/ S Option 2 (subtour elimination constraints) x ij S 1 S V, S 2 (7) i S j S These two formulations are algebraically equivalent. They prevent the formation of subtours containing fewer than S vertices. Note that the number of subtour elimination constraints needed is 2 V V 2. c R.F. Hartl, S.N. Parragh 9/282

10 The asymmetric TSP The asymmetric TSP A solution with 2 subcycles B C A D E c R.F. Hartl, S.N. Parragh 10/282

11 The asymmetric TSP The asymmetric TSP A solution with 2 subcycles Subtour elim. for subtour A-B-C: B C A D E c R.F. Hartl, S.N. Parragh 11/282

12 The asymmetric TSP The asymmetric TSP A solution with 2 subcycles B Subtour elim. for subtour A-B-C: Option 1: x AD +x AE +x BD + x BE +x CD +x CE 1 C A D E c R.F. Hartl, S.N. Parragh 12/282

13 The asymmetric TSP The asymmetric TSP A solution with 2 subcycles B C Subtour elim. for subtour A-B-C: Option 1: x AD +x AE +x BD + x BE +x CD +x CE 1 Option 2: x AB +x AC +x BA +x BC +x CA + x CB(+x AA +x BB +x CC) 2 A D E c R.F. Hartl, S.N. Parragh 13/282

14 The asymmetric TSP The asymmetric TSP A solution with 2 subcycles B C E A D Subtour elim. for subtour A-B-C: Option 1: x AD +x AE +x BD + x BE +x CD +x CE 1 Option 2: x AB +x AC +x BA +x BC +x CA + x CB(+x AA +x BB +x CC) 2 Subtour elim. for subtour D-E: c R.F. Hartl, S.N. Parragh 14/282

15 The asymmetric TSP The asymmetric TSP A solution with 2 subcycles B C E A D Subtour elim. for subtour A-B-C: Option 1: x AD +x AE +x BD + x BE +x CD +x CE 1 Option 2: x AB +x AC +x BA +x BC +x CA + x CB(+x AA +x BB +x CC) 2 Subtour elim. for subtour D-E: Option 1: x DA +x EA +x DB + x EB +x DC +x EC 1 c R.F. Hartl, S.N. Parragh 15/282

16 The asymmetric TSP The asymmetric TSP A solution with 2 subcycles B C E A D Subtour elim. for subtour A-B-C: Option 1: x AD +x AE +x BD + x BE +x CD +x CE 1 Option 2: x AB +x AC +x BA +x BC +x CA + x CB(+x AA +x BB +x CC) 2 Subtour elim. for subtour D-E: Option 1: x DA +x EA +x DB + x EB +x DC +x EC 1 Option 2: x DE +x ED(+x DD +x EE) 1 c R.F. Hartl, S.N. Parragh 16/282

17 The asymmetric TSP A lower bound for the asymmetric TSP The ATSP is known to be NP-hard (we do not know of a polynomial time algorithm for its solution). A good lower bound on the optimal solution of the ATSP can be obtained by removing the subtour elimination constraints. The relaxed problem is the c R.F. Hartl, S.N. Parragh 17/282

18 The asymmetric TSP A lower bound for the asymmetric TSP The ATSP is known to be NP-hard (we do not know of a polynomial time algorithm for its solution). A good lower bound on the optimal solution of the ATSP can be obtained by removing the subtour elimination constraints. The relaxed problem is the Linear Assignment Problem! (c ii := such that x ii = 0) c R.F. Hartl, S.N. Parragh 18/282

19 The asymmetric TSP A lower bound for the asymmetric TSP The ATSP is known to be NP-hard (we do not know of a polynomial time algorithm for its solution). A good lower bound on the optimal solution of the ATSP can be obtained by removing the subtour elimination constraints. The relaxed problem is the Linear Assignment Problem! (c ii := such that x ii = 0) Rule: zap is a good lower bound on z ATSP if the cost matrix is strongly asymmetric. (empirical tests showed that (zatsp z AP )/z AP often < 1%). If the cost matrix is symmetric, (zatsp z AP )/z AP is often more than 30%. Source: Laporte et al. (2004) Introduction to Logistics Systems Planning and Control, p. 252ff c R.F. Hartl, S.N. Parragh 19/282

20 The asymmetric TSP ATSP: Patching heuristic Initialization Solve the AP. Let C = {C 1,...,C p } denote the set of subcycles in the optimal solution of the AP. If C = 1 STOP. Otherwise proceed with step 2. Step 1 Identify the two subcycles with the largest number of vertices. Step 2 Merge these two subcycles such that the cost increase is as small as possible and update C. If C = 1 STOP. Otherwise go back to step 2. c R.F. Hartl, S.N. Parragh 20/282

21 The asymmetric TSP ATSP: Patching heuristic - Example Initialization apply the Hungarian method. 3 B 2 D 2 A F 2 C 4 E 3 c ij A B C D E F A B C D E F c R.F. Hartl, S.N. Parragh 21/282

22 The asymmetric TSP ATSP: Patching heuristic - Example A 2 2 B C D E 2 3 F Initialization apply the Hungarian method. Several alternative AP solutions with Z = 23: Solution 1: A C F E D B A Solution 2: A C E B A, D F D Solution 3: A C B A, D F E D... c ij A B C D E F A B C D E F c R.F. Hartl, S.N. Parragh 22/282

23 The asymmetric TSP ATSP: Patching heuristic - Example A 2 2 B C D E 2 3 F Initialization apply the Hungarian method. Several alternative AP solutions with Z = 23: Solution 1: A C F E D B A Solution 2: A C E B A, D F D Solution 3: A C B A, D F E D... B D c ij A B C D E F A B C D E F A F C E c R.F. Hartl, S.N. Parragh 23/282

24 The asymmetric TSP ATSP: Patching heuristic - Example B D 2 2 A 10 4 F 2 C E 3 Steps 1 and 2 Find the best way to join the two subcycles: c ij A B C D E F A B C D E F c R.F. Hartl, S.N. Parragh 24/282

25 The asymmetric TSP ATSP: Patching heuristic - Example B D 2 2 A 10 4 F 2 C E 3 Steps 1 and 2 Find the best way to join the two subcycles: Insert 2nd tour between A and C: c ij A B C D E F A B C D E F c R.F. Hartl, S.N. Parragh 25/282

26 The asymmetric TSP ATSP: Patching heuristic - Example B D A C 10 9 E 3 F Steps 1 and 2 Find the best way to join the two subcycles: Insert 2nd tour between A and C: A D F E C = = 13 c ij A B C D E F A B C D E F c R.F. Hartl, S.N. Parragh 26/282

27 The asymmetric TSP ATSP: Patching heuristic - Example 2 B D A C c ij A B C D E F A B C D E F E 3 F Steps 1 and 2 Find the best way to join the two subcycles: Insert 2nd tour between A and C: A D F E C = = 13 A F E D C = = 13 c R.F. Hartl, S.N. Parragh 27/282

28 The asymmetric TSP ATSP: Patching heuristic - Example B D 2 2 A C c ij A B C D E F A B C D E F E F Steps 1 and 2 Find the best way to join the two subcycles: Insert 2nd tour between A and C: A D F E C = = 13 A F E D C = = 13 A E D F C = = 13 c R.F. Hartl, S.N. Parragh 28/282

29 The asymmetric TSP ATSP: Patching heuristic - Example B D 2 2 A 2 C 10 c ij A B C D E F A B C D E F E 3 F Steps 1 and 2 Find the best way to join the two subcycles: Insert 2nd tour between A and C: A D F E C = = 13 A F E D C = = 13 A E D F C = = 13 Insert 2nd tour between C and B: c R.F. Hartl, S.N. Parragh 29/282

30 The asymmetric TSP ATSP: Patching heuristic - Example A B C D c ij A B C D E F A B C D E F E 3 F Steps 1 and 2 Find the best way to join the two subcycles: Insert 2nd tour between A and C: A D F E C = = 13 A F E D C = = 13 A E D F C = = 13 Insert 2nd tour between C and B: C D F E B = = 0 c R.F. Hartl, S.N. Parragh 30/282

31 The asymmetric TSP ATSP: Patching heuristic - Example A 2 2 B C 4 D E 10 c ij A B C D E F A B C D E F F Steps 1 and 2 Find the best way to join the two subcycles: Insert 2nd tour between A and C: A D F E C = = 13 A F E D C = = 13 A E D F C = = 13 Insert 2nd tour between C and B: C D F E B = = 0 C F E D B = = 0 c R.F. Hartl, S.N. Parragh 31/282

32 The asymmetric TSP ATSP: Patching heuristic - Example A B C D c ij A B C D E F A B C D E F E F Steps 1 and 2 Find the best way to join the two subcycles: Insert 2nd tour between A and C: A D F E C = = 13 A F E D C = = 13 A E D F C = = 13 Insert 2nd tour between C and B: C D F E B = = 0 C F E D B = = 0 C E D F B = = 0 c R.F. Hartl, S.N. Parragh 32/282

33 The asymmetric TSP ATSP: Patching heuristic - Example A B C D c ij A B C D E F A B C D E F E 3 F Steps 1 and 2 Find the best way to join the two subcycles: Insert 2nd tour between A and C: A D F E C = = 13 A F E D C = = 13 A E D F C = = 13 Insert 2nd tour between C and B: C D F E B = = 0 C F E D B = = 0 C E D F B = = 0 Insert 2nd tour between B and A: c R.F. Hartl, S.N. Parragh 33/282

34 The asymmetric TSP ATSP: Patching heuristic - Example A 2 8 B C 10 c ij A B C D E F A B C D E F D E 2 3 F Steps 1 and 2 Find the best way to join the two subcycles: Insert 2nd tour between A and C: A D F E C = = 13 A F E D C = = 13 A E D F C = = 13 Insert 2nd tour between C and B: C D F E B = = 0 C F E D B = = 0 C E D F B = = 0 Insert 2nd tour between B and A: B D F E A = = 5 c R.F. Hartl, S.N. Parragh 34/282

35 The asymmetric TSP ATSP: Patching heuristic - Example A 2 B C 4 10 c ij A B C D E F A B C D E F D E 3 F Steps 1 and 2 Find the best way to join the two subcycles: Insert 2nd tour between A and C: A D F E C = = 13 A F E D C = = 13 A E D F C = = 13 Insert 2nd tour between C and B: C D F E B = = 0 C F E D B = = 0 C E D F B = = 0 Insert 2nd tour between B and A: B D F E A = = 5 B F E D A = = 5 c R.F. Hartl, S.N. Parragh 35/282

36 The asymmetric TSP ATSP: Patching heuristic - Example A 2 B C c ij A B C D E F A B C D E F D E 2 F Steps 1 and 2 Find the best way to join the two subcycles: Insert 2nd tour between A and C: A D F E C = = 13 A F E D C = = 13 A E D F C = = 13 Insert 2nd tour between C and B: C D F E B = = 0 C F E D B = = 0 C E D F B = = 0 Insert 2nd tour between B and A: B D F E A = = 5 B F E D A = = 5 B E D F A = = 13 c R.F. Hartl, S.N. Parragh 36/282

37 The asymmetric TSP ATSP: Patching heuristic - Example A 2 2 B C 4 D E 10 c ij A B C D E F A B C D E F F Steps 1 and 2 Find the best way to join the two subcycles: Insert 2nd tour between A and C: A D F E C = = 13 A F E D C = = 13 A E D F C = = 13 Insert 2nd tour between C and B: C D F E B = = 0 C F E D B = = 0 C E D F B = = 0 Insert 2nd tour between B and A: B D F E A = = 5 B F E D A = = 5 B E D F A = = 13 e.g. best solution (several): A C F E D B A Z = 23 (upper bound on z ATSP) c R.F. Hartl, S.N. Parragh 37/282

38 The symmetric TSP The symmetric TSP (STSP) Since lower and upper bounding procedures for the ATSP do not perfom well if applied to the STSP, procedures tailored to the STSP have been developed. Notation E...edge set x e = { 1, if edge e is part of the solution, 0, otherwise. c e = c ij = c ji c R.F. Hartl, S.N. Parragh 38/282

39 The symmetric TSP STSP: IP formulations j V:(j,i) E (i,j) E:i S,j/ S x ij + (i,j) E x ji + x ij c ij min (8) j V:(i,j) E (j,i) E:i S,j/ S x ij = 2 i V, (9) x ji 2 S V,2 S V /2, (10) x e {0,1} e E. (11) i < j for each edge (i,j) E Since the connectivity constraints of subset S are equivalent to those of subset V \S, we only consider S V such that S V /2. c R.F. Hartl, S.N. Parragh 39/282

40 The symmetric TSP STSP: IP formulations x e c e min e E e J(i) e E(S) (12) x e = 2 i V, (13) x e S 1 S V,2 S V /2 (14) x e {0,1} e E. (15) J(i) all edges connected to i; E(S) edges connecting vertices in subset S c R.F. Hartl, S.N. Parragh 40/282

41 The symmetric TSP STSP: a lower bound The STSP is also NP-hard. A valid lower bound on the optimal solution cost z STSP is c R.F. Hartl, S.N. Parragh 41/282

42 The symmetric TSP STSP: a lower bound The STSP is also NP-hard. A valid lower bound on the optimal solution cost zstsp is the optimal solution cost of the MST z MST c R.F. Hartl, S.N. Parragh 42/282

43 The symmetric TSP STSP: Christofides heuristic Initialization Compute a minimum spanning tree T. Step 1 Compute a least-cost perfect matching among the vertices of odd degree in T. (V D...vertices with odd degree). Add the edges of the perfect matching (M...vertices of the matching, H...graph induced by union of the edges T and M). Step 2 If there is a vertex j V of degree > 2, eliminate two edges incident in j, denote them (j,k),(j,h) with k h and add edge (k,h) (try to replace those two edges that improve the solution value the most and do not lead to subcycles). Repeat step 2 until all vertices in V have a degree of 2. c R.F. Hartl, S.N. Parragh 43/282

44 The symmetric TSP STSP: Christofides heuristic - Example O 2 A C 2 B 1 3 O A B C D E T O A B C D E T E D 1 5 T Initialization compute the minimum spanning tree (Kruskal s algorithm!) c R.F. Hartl, S.N. Parragh 44/282

45 The symmetric TSP STSP: Christofides heuristic - Example O 2 A 2 B 1 3 D 1 5 T Initialization compute the minimum spanning tree (Kruskal s algorithm!) C E Step 1 V D = {O,C,B,T} O A B C D E T O A B C D E T c R.F. Hartl, S.N. Parragh 45/282

46 The symmetric TSP STSP: Christofides heuristic - Example O 2 A 2 B 1 3 D 1 5 T Initialization compute the minimum spanning tree (Kruskal s algorithm!) C O A B C D E T O A B C D E T E Step 1 V D = {O,C,B,T} Compute a least-cost perfect matching: c R.F. Hartl, S.N. Parragh 46/282

47 The symmetric TSP STSP: Christofides heuristic - Example O 2 A C 2 B 1 3 E D 1 5 T O A B C D E T O A B C D E T Initialization compute the minimum spanning tree (Kruskal s algorithm!) Step 1 V D = {O,C,B,T} Compute a least-cost perfect matching: O B,C T: 4+10 = 14 O C,B T: 4+9 = 13 O T,B C: = 14 c R.F. Hartl, S.N. Parragh 47/282

48 The symmetric TSP STSP: Christofides heuristic - Example O 2 4 A C 2 B 1 3 E D T O A B C D E T O A B C D E T Initialization compute the minimum spanning tree (Kruskal s algorithm!) Step 1 V D = {O,C,B,T} Compute a least-cost perfect matching: O B,C T: 4+10 = 14 O C,B T: 4+9 = 13 O T,B C: = 14 Least cost perfect matching: E(M) = {(O,C),(B,T)}: 4+9 = 13 c R.F. Hartl, S.N. Parragh 48/282

49 The symmetric TSP STSP: Christofides heuristic - Example O 2 4 A C 2 B 1 3 E D T O A B C D E T O A B C D E T Initialization compute the minimum spanning tree (Kruskal s algorithm!) Step 1 V D = {O,C,B,T} Compute a least-cost perfect matching: O B,C T: 4+10 = 14 O C,B T: 4+9 = 13 O T,B C: = 14 Least cost perfect matching: E(M) = {(O,C),(B,T)}: 4+9 = 13 Step 2 j = B find shortcuts. c R.F. Hartl, S.N. Parragh 49/282

50 The symmetric TSP STSP: Christofides heuristic - Example O 2 4 A C 2 B 1 3 E D T O A B C D E T O A B C D E T Initialization compute the minimum spanning tree (Kruskal s algorithm!) Step 1 V D = {O,C,B,T} Compute a least-cost perfect matching: O B,C T: 4+10 = 14 O C,B T: 4+9 = 13 O T,B C: = 14 Least cost perfect matching: E(M) = {(O,C),(B,T)}: 4+9 = 13 Step 2 j = B find shortcuts. B is connected to A,C,E,T c R.F. Hartl, S.N. Parragh 50/282

51 The symmetric TSP STSP: Christofides heuristic - Example O 2 4 A C 2 B 1 3 E D T O A B C D E T O A B C D E T Initialization compute the minimum spanning tree (Kruskal s algorithm!) Step 1 V D = {O,C,B,T} Compute a least-cost perfect matching: O B,C T: 4+10 = 14 O C,B T: 4+9 = 13 O T,B C: = 14 Least cost perfect matching: E(M) = {(O,C),(B,T)}: 4+9 = 13 Step 2 j = B find shortcuts. B is connected to A,C,E,T possible shortcuts: A E: = 5 5 = 0 A T: = = 0 C E: = 4 4 = 0 C T: = = 0 c R.F. Hartl, S.N. Parragh 51/282

52 The symmetric TSP STSP: Christofides heuristic - Example Initialization compute the minimum spanning tree (Kruskal s algorithm!) O 2 4 A C 2 B 1 3 E D T O A B C D E T O A B C D E T Step 1 V D = {O,C,B,T} Compute a least-cost perfect matching: O B,C T: 4+10 = 14 O C,B T: 4+9 = 13 O T,B C: = 14 Least cost perfect matching: E(M) = {(O,C),(B,T)}: 4+9 = 13 Step 2 j = B find shortcuts. B is connected to A,C,E,T possible shortcuts: A E: = 5 5 = 0 A T: = = 0 C E: = 4 4 = 0 C T: = = 0 We choose C E c R.F. Hartl, S.N. Parragh 52/282

53 The symmetric TSP STSP: Christofides heuristic - Example Initialization compute the minimum spanning tree (Kruskal s algorithm!) O 2 4 A 2 B D T Step 1 V D = {O,C,B,T} Compute a least-cost perfect matching: O B,C T: 4+10 = 14 O C,B T: 4+9 = 13 O T,B C: = 14 Least cost perfect matching: E(M) = {(O,C),(B,T)}: 4+9 = 13 C 4 O A B C D E T O A B C D E T E Step 2 j = B find shortcuts. B is connected to A,C,E,T possible shortcuts: A E: = 5 5 = 0 A T: = = 0 C E: = 4 4 = 0 C T: = = 0 We choose C E z STSP = 27 z MST = 14 c R.F. Hartl, S.N. Parragh 53/282

54 Further solution methods Solution methods Heuristic methods Nearest neighbor Cheapest insertion k-opt (local search) Exact methods c R.F. Hartl, S.N. Parragh 54/282

55 Construction heuristics Nearest neighbor heuristic Initialization Set P = {r} with r being a vertex chosen arbitrarily and P denotes the (partial) TSP tour. Set h = r. Step 1 Find the vertex k V \P such that c hk = min j V\P {c hj }. Append k to the end of P. Step 2 If P = V, add r to the end of P and STOP. (P is now a Hamiltonian cycle). Otherwise, set h = k and continue with step 1. c R.F. Hartl, S.N. Parragh 55/282

56 Construction heuristics Nearest neighbor heuristic Initialization Set P = {r} with r being a vertex chosen arbitrarily and P denotes the (partial) TSP tour. Set h = r. Step 1 Find the vertex k V \P such that c hk = min j V\P {c hj }. Append k to the end of P. Step 2 If P = V, add r to the end of P and STOP. (P is now a Hamiltonian cycle). Otherwise, set h = k and continue with step 1. Hamiltonian cycle A cycle that contains all the vertices of a graph. (TSP = problem of finding the minimum cost Hamiltonian cycle in a graph) Sources: Laporte et al. (2004) Introduction to Logistics Systems Planning and Control, p. 261 E.L. Lawler J.K. Lenstra, A.H.G. Rinnooy Kan, D.B. Shmoys (Eds) (1990) The traveling salesman problem, p. 2, p. 150f. c R.F. Hartl, S.N. Parragh 56/282

57 Construction heuristics Nearest neighbor heuristic - Example A O B D T C E O A B C D E T O A B C D E T c R.F. Hartl, S.N. Parragh 57/282

58 Construction heuristics Nearest neighbor heuristic - Example A O B D T Initialization r = O, P = {O}, h = O C E O A B C D E T O A B C D E T c R.F. Hartl, S.N. Parragh 58/282

59 Construction heuristics Nearest neighbor heuristic - Example A O B D T Initialization r = O, P = {O}, h = O C E Step 1 identify the nearest neighbor to O O A B C D E T O A B C D E T c R.F. Hartl, S.N. Parragh 59/282

60 Construction heuristics Nearest neighbor heuristic - Example A O C B O A B C D E T O A B C D E T E D T Initialization r = O, P = {O}, h = O Step 1 identify the nearest neighbor to O A: k = A, P = {O,A}, Z = 2 Step 2 P < V c R.F. Hartl, S.N. Parragh 60/282

61 Construction heuristics Nearest neighbor heuristic - Example 2 A O C B O A B C D E T O A B C D E T E D T Initialization r = O, P = {O}, h = O Step 1 identify the nearest neighbor to O A: k = A, P = {O,A}, Z = 2 Step 2 P < V c R.F. Hartl, S.N. Parragh 61/282

62 Construction heuristics Nearest neighbor heuristic - Example 2 A O C B O A B C D E T O A B C D E T E D T Initialization r = O, P = {O}, h = O Step 1 identify the nearest neighbor to O A: k = A, P = {O,A}, Z = 2 Step 2 P < V Step 1 identify the nearest neighbor to A c R.F. Hartl, S.N. Parragh 62/282

63 Construction heuristics Nearest neighbor heuristic - Example 2 A O C B O A B C D E T O A B C D E T E D T Initialization r = O, P = {O}, h = O Step 1 identify the nearest neighbor to O A: k = A, P = {O,A}, Z = 2 Step 2 P < V Step 1 identify the nearest neighbor to A B: k = B, P = {O,A,B}, Z = 4 Step 2 P < V c R.F. Hartl, S.N. Parragh 63/282

64 Construction heuristics Nearest neighbor heuristic - Example O 2 A C 2 B O A B C D E T O A B C D E T E D T Initialization r = O, P = {O}, h = O Step 1 identify the nearest neighbor to O A: k = A, P = {O,A}, Z = 2 Step 2 P < V Step 1 identify the nearest neighbor to A B: k = B, P = {O,A,B}, Z = 4 Step 2 P < V c R.F. Hartl, S.N. Parragh 64/282

65 Construction heuristics Nearest neighbor heuristic - Example O 2 A C 2 B O A B C D E T O A B C D E T E D T Initialization r = O, P = {O}, h = O Step 1 identify the nearest neighbor to O A: k = A, P = {O,A}, Z = 2 Step 2 P < V Step 1 identify the nearest neighbor to A B: k = B, P = {O,A,B}, Z = 4 Step 2 P < V Step 1 identify the nearest neighbor to B c R.F. Hartl, S.N. Parragh 65/282

66 Construction heuristics Nearest neighbor heuristic - Example O 2 A C 2 B O A B C D E T O A B C D E T E D T Initialization r = O, P = {O}, h = O Step 1 identify the nearest neighbor to O A: k = A, P = {O,A}, Z = 2 Step 2 P < V Step 1 identify the nearest neighbor to A B: k = B, P = {O,A,B}, Z = 4 Step 2 P < V Step 1 identify the nearest neighbor to B C: k = C, P = {O,A,B,C}, Z = 5 Step 2 P < V c R.F. Hartl, S.N. Parragh 66/282

67 Construction heuristics Nearest neighbor heuristic - Example O 2 A C 1 2 B O A B C D E T O A B C D E T E D T Initialization r = O, P = {O}, h = O Step 1 identify the nearest neighbor to O A: k = A, P = {O,A}, Z = 2 Step 2 P < V Step 1 identify the nearest neighbor to A B: k = B, P = {O,A,B}, Z = 4 Step 2 P < V Step 1 identify the nearest neighbor to B C: k = C, P = {O,A,B,C}, Z = 5 Step 2 P < V c R.F. Hartl, S.N. Parragh 67/282

68 Construction heuristics Nearest neighbor heuristic - Example O 2 A C 1 2 B O A B C D E T O A B C D E T E D T Initialization r = O, P = {O}, h = O Step 1 identify the nearest neighbor to O A: k = A, P = {O,A}, Z = 2 Step 2 P < V Step 1 identify the nearest neighbor to A B: k = B, P = {O,A,B}, Z = 4 Step 2 P < V Step 1 identify the nearest neighbor to B C: k = C, P = {O,A,B,C}, Z = 5 Step 2 P < V... c R.F. Hartl, S.N. Parragh 68/282

69 Construction heuristics Nearest neighbor heuristic - Example O 2 A C 1 2 B O A B C D E T O A B C D E T E D T Initialization r = O, P = {O}, h = O Step 1 identify the nearest neighbor to O A: k = A, P = {O,A}, Z = 2 Step 2 P < V Step 1 identify the nearest neighbor to A B: k = B, P = {O,A,B}, Z = 4 Step 2 P < V Step 1 identify the nearest neighbor to B C: k = C, P = {O,A,B,C}, Z = 5 Step 2 P < V... P = {O,A,B,C,E,D,T}, Z = = 15 c R.F. Hartl, S.N. Parragh 69/282

70 Construction heuristics Nearest neighbor heuristic - Example O 2 A C 1 2 B 4 1 E D 5 T O A B C D E T O A B C D E T Initialization r = O, P = {O}, h = O Step 1 identify the nearest neighbor to O A: k = A, P = {O,A}, Z = 2 Step 2 P < V Step 1 identify the nearest neighbor to A B: k = B, P = {O,A,B}, Z = 4 Step 2 P < V Step 1 identify the nearest neighbor to B C: k = C, P = {O,A,B,C}, Z = 5 Step 2 P < V... P = {O,A,B,C,E,D,T}, Z = = 15 Step 2 P = V, append O to P c R.F. Hartl, S.N. Parragh 70/282

71 Construction heuristics Nearest neighbor heuristic - Example Initialization r = O, P = {O}, h = O O 2 A C B 4 1 E D 5 T O A B C D E T O A B C D E T Step 1 identify the nearest neighbor to O A: k = A, P = {O,A}, Z = 2 Step 2 P < V Step 1 identify the nearest neighbor to A B: k = B, P = {O,A,B}, Z = 4 Step 2 P < V Step 1 identify the nearest neighbor to B C: k = C, P = {O,A,B,C}, Z = 5 Step 2 P < V... P = {O,A,B,C,E,D,T}, Z = = 15 Step 2 P = V, append O to P P = {O,A,B,C,E,D,T,O} z TSP = 28 c R.F. Hartl, S.N. Parragh 71/282

72 Construction heuristics Nearest neighbor heuristic - Observations Advantages Simple and fast c R.F. Hartl, S.N. Parragh 72/282

73 Construction heuristics Nearest neighbor heuristic - Observations Advantages Simple and fast Disadvantages Solution can be of rather poor quality. (edges added in later iterations may be quite long - especially the last edge connecting back to the first vertex) c R.F. Hartl, S.N. Parragh 73/282

74 Construction heuristics Cheapest insertion heuristic Initialization arbitrarily choose the first vertex r. Initialize the TSP tour P = {r,r}, Z = 0 Iteration step If P < V +1 For each vertex i V \P find the cheapest insertion position between any j and k; j,k P and neighboring in P. Find the vertex i V \P that can be inserted the cheapest; insert it at its cheapest position (say j and k ): P = {r,...,j,i,k,...,r} and set Z = Z c j k +c j i +c i k c R.F. Hartl, S.N. Parragh 74/282

75 Construction heuristics Cheapest insertion heuristic - Example A O B D T C E O A B C D E T O A B C D E T c R.F. Hartl, S.N. Parragh 75/282

76 Construction heuristics Cheapest insertion heuristic - Example A O B D T Initialization r = O, P = {O,O} Z = 0 C E O A B C D E T O A B C D E T c R.F. Hartl, S.N. Parragh 76/282

77 Construction heuristics Cheapest insertion heuristic - Example A O C B O A B C D E T O A B C D E T E D T Initialization r = O, P = {O,O} Z = 0 Iteration 1 O A O = 2+2 = 4 O B O = 4+4 = 8 O C O = 4+4 = 8 O D O = 8+8 = 16 O E O = 7+7 = 14 O T O = = 26 c R.F. Hartl, S.N. Parragh 77/282

78 Construction heuristics Cheapest insertion heuristic - Example A O C B O A B C D E T O A B C D E T E D T Initialization r = O, P = {O,O} Z = 0 Iteration 1 O A O = 2+2 = 4 O B O = 4+4 = 8 O C O = 4+4 = 8 O D O = 8+8 = 16 O E O = 7+7 = 14 O T O = = 26 A: i = A, P = {O,A,O}, Z = 4 c R.F. Hartl, S.N. Parragh 78/282

79 Construction heuristics Cheapest insertion heuristic - Example O 2 2 A C B O A B C D E T O A B C D E T E D T Initialization r = O, P = {O,O} Z = 0 Iteration 1 O A O = 2+2 = 4 O B O = 4+4 = 8 O C O = 4+4 = 8 O D O = 8+8 = 16 O E O = 7+7 = 14 O T O = = 26 A: i = A, P = {O,A,O}, Z = 4 c R.F. Hartl, S.N. Parragh 79/282

80 Construction heuristics Cheapest insertion heuristic - Example Initialization r = O, P = {O,O} Z = 0 O 2 2 A C B O A B C D E T O A B C D E T E D T Iteration 1 O A O = 2+2 = 4 O B O = 4+4 = 8 O C O = 4+4 = 8 O D O = 8+8 = 16 O E O = 7+7 = 14 O T O = = 26 A: i = A, P = {O,A,O}, Z = 4 Iteration 2 O B A, A B O = = 4 O C A, A C O = = 5 O D A, A D O = = 12 O E A, A E O = = 10 O T A, A T O = = 24 c R.F. Hartl, S.N. Parragh 80/282

81 Construction heuristics Cheapest insertion heuristic - Example Initialization r = O, P = {O,O} Z = 0 O 2 2 A C B O A B C D E T O A B C D E T E D T Iteration 1 O A O = 2+2 = 4 O B O = 4+4 = 8 O C O = 4+4 = 8 O D O = 8+8 = 16 O E O = 7+7 = 14 O T O = = 26 A: i = A, P = {O,A,O}, Z = 4 Iteration 2 O B A, A B O = = 4 O C A, A C O = = 5 O D A, A D O = = 12 O E A, A E O = = 10 O T A, A T O = = 24 B: i = B, P = {O,B,A,O}, Z = 8 c R.F. Hartl, S.N. Parragh 81/282

82 Construction heuristics Cheapest insertion heuristic - Example Initialization r = O, P = {O,O} Z = 0 O 2 A 4 C 2 B O A B C D E T O A B C D E T E D T Iteration 1 O A O = 2+2 = 4 O B O = 4+4 = 8 O C O = 4+4 = 8 O D O = 8+8 = 16 O E O = 7+7 = 14 O T O = = 26 A: i = A, P = {O,A,O}, Z = 4 Iteration 2 O B A, A B O = = 4 O C A, A C O = = 5 O D A, A D O = = 12 O E A, A E O = = 10 O T A, A T O = = 24 B: i = B, P = {O,B,A,O}, Z = 8 c R.F. Hartl, S.N. Parragh 82/282

83 Construction heuristics Cheapest insertion heuristic - Example A 2 2 D O T 4 B C E O A B C D E T O A B C D E T c R.F. Hartl, S.N. Parragh 83/282

84 Construction heuristics Cheapest insertion heuristic - Example O 2 A 4 C 2 B O A B C D E T O A B C D E T E D T Iteration 3 O C B = = 1 B C A = = 2 A C O = = 5 O D B = = 8 B D A = = 10 A D O = = c R.F. Hartl, S.N. Parragh 84/282

85 Construction heuristics Cheapest insertion heuristic - Example O 2 A 4 C 2 B O A B C D E T O A B C D E T E D T Iteration 3 O C B = = 1 B C A = = 2 A C O = = 5 O D B = = 8 B D A = = 10 A D O = = C: i = C, P = {O,C,B,A,O}, Z = 9 c R.F. Hartl, S.N. Parragh 85/282

86 Construction heuristics Cheapest insertion heuristic - Example O 2 4 A C 2 1 B O A B C D E T O A B C D E T E D T Iteration 3 O C B = = 1 B C A = = 2 A C O = = 5 O D B = = 8 B D A = = 10 A D O = = C: i = C, P = {O,C,B,A,O}, Z = 9 c R.F. Hartl, S.N. Parragh 86/282

87 Construction heuristics Cheapest insertion heuristic - Example O 2 4 A C 2 1 B O A B C D E T O A B C D E T E D T Iteration 3 O C B = = 1 B C A = = 2 A C O = = 5 O D B = = 8 B D A = = 10 A D O = = C: i = C, P = {O,C,B,A,O}, Z = 9... c R.F. Hartl, S.N. Parragh 87/282

88 Construction heuristics Cheapest insertion heuristic - Example O 2 4 A C 2 1 B O A B C D E T O A B C D E T E D T Iteration 3 O C B = = 1 B C A = = 2 A C O = = 5 O D B = = 8 B D A = = 10 A D O = = C: i = C, P = {O,C,B,A,O}, Z = 9... P = {O,C,T,D,E,B,A,O}, z TSP = 27 c R.F. Hartl, S.N. Parragh 88/282

89 Construction heuristics Cheapest insertion heuristic - Example O 2 4 A C 2 B 10 3 E D 1 5 T O A B C D E T O A B C D E T Iteration 3 O C B = = 1 B C A = = 2 A C O = = 5 O D B = = 8 B D A = = 10 A D O = = C: i = C, P = {O,C,B,A,O}, Z = 9... P = {O,C,T,D,E,B,A,O}, z TSP = 27 c R.F. Hartl, S.N. Parragh 89/282

90 Improvement heuristics Local search Local search Local search algorithms are iterative procedures that try to improve an initial feasible solution x (0). At the kth iteration, all solutions contained in a neighborhood of the current solution x (k) are enumerated. If there exist feasible solutions that are less costly than x (k), the best solution of the neighborhood becomes the new current solution x (k+1) and the procedure is repeated. Otherwise, the procedure ends. The last current solution is the local optimum. Source: Laporte et al. (2004) Introduction to Logistics Systems Planning and Control, p. 263 c R.F. Hartl, S.N. Parragh 90/282

91 Improvement heuristics Local search Local search Local search algorithms are iterative procedures that try to improve an initial feasible solution x (0). At the kth iteration, all solutions contained in a neighborhood of the current solution x (k) are enumerated. If there exist feasible solutions that are less costly than x (k), the best solution of the neighborhood becomes the new current solution x (k+1) and the procedure is repeated. Otherwise, the procedure ends. The last current solution is the local optimum. Source: Laporte et al. (2004) Introduction to Logistics Systems Planning and Control, p. 263 Neighborhood A neighborhood is defined by all feasible solutions that can be reached by applying a given operator to a given solution (e.g.: move one vertex, exchange two vertices). c R.F. Hartl, S.N. Parragh 91/282

92 Improvement heuristics Lin-Kernighan: k-opt (local search) Initialization Let P (0) be the initial TSP solution and let z (0) TSP be the cost of P(0). Set h = 0. Step 1 Identify the best feasible solution P (h+1) that can be obtained through a k-exchange. If z (h+1) TSP z(h) TSP, STOP. Step 2 set h = h+1 and proceed with step 1. c R.F. Hartl, S.N. Parragh 92/282

93 Improvement heuristics Lin-Kernighan: k-opt (local search) Initialization Let P (0) be the initial TSP solution and let z (0) TSP be the cost of P(0). Set h = 0. Step 1 Identify the best feasible solution P (h+1) that can be obtained through a k-exchange. If z (h+1) TSP z(h) TSP, STOP. Step 2 set h = h+1 and proceed with step 1. k-exchange replace k links (edges) of the current solution with k new links, e.g. a 2-exchange: replace (i,j) and (k,l) by (i,l) and (j,k) (= inverse the sequence between j and k) c R.F. Hartl, S.N. Parragh 93/282

94 Improvement heuristics Lin-Kernighan: 2-exchange i l k j c R.F. Hartl, S.N. Parragh 94/282

95 Improvement heuristics Lin-Kernighan: 2-exchange i l i l k j k j c R.F. Hartl, S.N. Parragh 95/282

96 Improvement heuristics Lin-Kernighan: 2-opt (typical picture) r t r t c R.F. Hartl, S.N. Parragh 96/282

97 Improvement heuristics Lin-Kernighan: 2-opt - Example A O B D T C E Initial solution O C E D T B A O c R.F. Hartl, S.N. Parragh 97/282

98 Improvement heuristics Lin-Kernighan: 2-opt - Example A O B D T 2-opt C E Initial solution O C E D T B A O c R.F. Hartl, S.N. Parragh 98/282

99 Improvement heuristics Lin-Kernighan: 2-opt - Example A O B D T 2-opt O E C D T B A O C E Initial solution O C E D T B A O c R.F. Hartl, S.N. Parragh 99/282

100 Improvement heuristics Lin-Kernighan: 2-opt - Example A O B D T 2-opt O E C D T B A O O C D E T B A O C E Initial solution O C E D T B A O c R.F. Hartl, S.N. Parragh 100/282

101 Improvement heuristics Lin-Kernighan: 2-opt - Example A O B D T 2-opt O E C D T B A O O C D E T B A O O C E T D B A O C E Initial solution O C E D T B A O c R.F. Hartl, S.N. Parragh 101/282

102 Improvement heuristics Lin-Kernighan: 2-opt - Example A O C B E D T 2-opt O E C D T B A O O C D E T B A O O C E T D B A O O C E D B T A O Initial solution O C E D T B A O c R.F. Hartl, S.N. Parragh 102/282

103 Improvement heuristics Lin-Kernighan: 2-opt - Example A O C B E D T 2-opt O E C D T B A O O C D E T B A O O C E T D B A O O C E D B T A O O C E D T A B O Initial solution O C E D T B A O c R.F. Hartl, S.N. Parragh 103/282

104 Improvement heuristics Lin-Kernighan: 2-opt - Example O A C B E D T 2-opt O E C D T B A O O C D E T B A O O C E T D B A O O C E D B T A O O C E D T A B O O D E C T B A O Initial solution O C E D T B A O c R.F. Hartl, S.N. Parragh 104/282

105 Improvement heuristics Lin-Kernighan: 2-opt - Example O A C Initial solution O C E D T B A O B E D T 2-opt O E C D T B A O O C D E T B A O O C E T D B A O O C E D B T A O O C E D T A B O O D E C T B A O O C T D E B A O c R.F. Hartl, S.N. Parragh 105/282

106 Improvement heuristics Lin-Kernighan: 2-opt - Example O A C Initial solution O C E D T B A O B E D T 2-opt O E C D T B A O O C D E T B A O O C E T D B A O O C E D B T A O O C E D T A B O O D E C T B A O O C T D E B A O O C E B T D A O c R.F. Hartl, S.N. Parragh 106/282

107 Improvement heuristics Lin-Kernighan: 2-opt - Example O A C Initial solution O C E D T B A O B E D T 2-opt O E C D T B A O O C D E T B A O O C E T D B A O O C E D B T A O O C E D T A B O O D E C T B A O O C T D E B A O O C E B T D A O O C E D A B T O c R.F. Hartl, S.N. Parragh 107/282

108 Improvement heuristics Lin-Kernighan: 2-opt - Example O A C Initial solution O C E D T B A O B E D T 2-opt O E C D T B A O O C D E T B A O O C E T D B A O O C E D B T A O O C E D T A B O O D E C T B A O O C T D E B A O O C E B T D A O O C E D A B T O O T D E C B A O c R.F. Hartl, S.N. Parragh 108/282

109 Improvement heuristics Lin-Kernighan: 2-opt - Example O A C Initial solution O C E D T B A O B E D T 2-opt O E C D T B A O O C D E T B A O O C E T D B A O O C E D B T A O O C E D T A B O O D E C T B A O O C T D E B A O O C E B T D A O O C E D A B T O O T D E C B A O O C B T D E A O c R.F. Hartl, S.N. Parragh 109/282

110 Improvement heuristics Lin-Kernighan: 2-opt - Example O A C Initial solution O C E D T B A O B E D T 2-opt O E C D T B A O O C D E T B A O O C E T D B A O O C E D B T A O O C E D T A B O O D E C T B A O O C T D E B A O O C E B T D A O O C E D A B T O O T D E C B A O O C B T D E A O O C E A B T D O c R.F. Hartl, S.N. Parragh 110/282

111 Improvement heuristics Lin-Kernighan: 2-opt - Example O A C Initial solution O C E D T B A O B E D T 2-opt O E C D T B A O O C D E T B A O O C E T D B A O O C E D B T A O O C E D T A B O O D E C T B A O O C T D E B A O O C E B T D A O O C E D A B T O O T D E C B A O O C B T D E A O O C E A B T D O O B T D E C A O c R.F. Hartl, S.N. Parragh 111/282

112 Improvement heuristics Lin-Kernighan: 2-opt - Example O A C Initial solution O C E D T B A O B E D T 2-opt O E C D T B A O O C D E T B A O O C E T D B A O O C E D B T A O O C E D T A B O O D E C T B A O O C T D E B A O O C E B T D A O O C E D A B T O O T D E C B A O O C B T D E A O O C E A B T D O O B T D E C A O O C A B T D E O c R.F. Hartl, S.N. Parragh 112/282

113 Improvement heuristics Lin-Kernighan: 2-opt - Example O A C Initial solution O C E D T B A O B E D T 2-opt O E C D T B A O O C D E T B A O O C E T D B A O O C E D B T A O O C E D T A B O O D E C T B A O O C T D E B A O O C E B T D A O O C E D A B T O O T D E C B A O O C B T D E A O O C E A B T D O O B T D E C A O O C A B T D E O O A B T D E C O c R.F. Hartl, S.N. Parragh 113/282

114 Improvement heuristics Lin-Kernighan: 2-opt - Example O A C Initial solution O C E D T B A O best improvement: the best one becomes the new current solution; repeat until no further improvement. B E D T 2-opt O E C D T B A O O C D E T B A O O C E T D B A O O C E D B T A O O C E D T A B O O D E C T B A O O C T D E B A O O C E B T D A O O C E D A B T O O T D E C B A O O C B T D E A O O C E A B T D O O B T D E C A O O C A B T D E O O A B T D E C O c R.F. Hartl, S.N. Parragh 114/282

115 Improvement heuristics Lin-Kernighan: 2-opt - Example O A C Initial solution O C E D T B A O best improvement: the best one becomes the new current solution; repeat until no further improvement. B first improvement: the first improving solution becomes the new current solution; repeat until no further improvement. E D T 2-opt O E C D T B A O O C D E T B A O O C E T D B A O O C E D B T A O O C E D T A B O O D E C T B A O O C T D E B A O O C E B T D A O O C E D A B T O O T D E C B A O O C B T D E A O O C E A B T D O O B T D E C A O O C A B T D E O O A B T D E C O c R.F. Hartl, S.N. Parragh 115/282

116 Branch and bound (BB) - the concept A general method for solving (mixed) integer optimization problems (with a minimization objective). It uses two concepts: branching and bounding. Root node solve relaxation of the optimization problem (e.g., LP of IP). Set LB to the objective function value of the relaxed problem z 0 and set UB =. Put the root node into the set of unprocessed nodes Q. Repeat (until Q is empty) select a node k from the set Q, according to a certain criterion. identify on what to branch branch: partition the problem P k at k into a given number of subproblems s: P k1...p ks (one child node for each subproblem); for the solution spaces wrt the original optimization problem L(P k ) = s i=1 P k i has to hold. (e.g., 2 child nodes: in the 1st child node (1st partition) a certain characteristic is forbidden while in the 2nd child node (2nd partition), this characteristic is enforced). compute the lower bounds at the child nodes (new enforced/forbidden characteristics plus all enforced/forbidden characteristics of parent nodes); put the child nodes into Q bound: the lowest bound across all nodes Q gives the current LB; a feasible solution to the original problem at a given node gives a new UB; all open nodes associated with LBs UB can be pruned (removed from Q). c R.F. Hartl, S.N. Parragh 116/282

117 Branch and bound (BB) - the concept c R.F. Hartl, S.N. Parragh 117/282

118 Branch and bound (BB) - the concept z = 10 LB = 10 UB = c R.F. Hartl, S.N. Parragh 118/282

119 Branch and bound (BB) - the concept z = 10 z = 11 LB = 10 UB = c R.F. Hartl, S.N. Parragh 119/282

120 Branch and bound (BB) - the concept z = 10 z = 11 z = 12 LB = 11 UB = c R.F. Hartl, S.N. Parragh 120/282

121 Branch and bound (BB) - the concept z = 10 z = 11 z = 12 z = 13 LB = 11 UB = c R.F. Hartl, S.N. Parragh 121/282

122 Branch and bound (BB) - the concept z = 10 z = 11 z = 12 z = 13 z = 17 LB = 11 UB = c R.F. Hartl, S.N. Parragh 122/282

123 Branch and bound (BB) - the concept z = 10 z = 11 z = 12 z = 13 z = 17 LB = 12 UB = c R.F. Hartl, S.N. Parragh 123/282

124 Branch and bound (BB) - the concept z = 10 z = 11 z = 12 z = 13 z = 17 z = 16 LB = 12 UB = c R.F. Hartl, S.N. Parragh 124/282

125 Branch and bound (BB) - the concept z = 10 z = 11 z = 12 z = 13 z = 17 z = 16 z = 18 LB = 12 UB = 16 c R.F. Hartl, S.N. Parragh 125/282

126 Branch and bound (BB) - the concept z = 10 z = 11 z = 12 z = 13 z = 17 z = 16 z = 18 X LB = 12 UB = 16 c R.F. Hartl, S.N. Parragh 126/282

127 Branch and bound (BB) - the concept z = 10 z = 11 z = 12 z = 13 z = 17 z = 16 z = 18 X X LB = 12 UB = 16 c R.F. Hartl, S.N. Parragh 127/282

128 Branch and bound (BB) - the concept z = 10 z = 11 z = 12 z = 13 z = 17 z = 16 z = 18 X X LB = 13 UB = 16 c R.F. Hartl, S.N. Parragh 128/282

129 Branch and bound (BB) - the concept z = 10 z = 11 z = 12 z = 13 z = 17 z = 16 z = 18 X X LB = 13 UB = 14 z = 14 c R.F. Hartl, S.N. Parragh 129/282

130 Branch and bound (BB) - the concept z = 10 z = 11 z = 12 z = 13 z = 17 z = 16 z = 18 X X LB = 13 UB = 14 z = 14 z = 15 c R.F. Hartl, S.N. Parragh 130/282

131 Branch and bound (BB) - the concept z = 10 z = 11 z = 12 z = 13 z = 17 z = 16 z = 18 X X LB = 13 UB = 14 z = 14 z = 15 X c R.F. Hartl, S.N. Parragh 131/282

132 Branch and bound (BB) - the concept z = 10 z = 11 z = 12 z = 13 z = 17 z = 16 z = 18 X X LB = 13 UB = 14 z = 14 z = 15 X c R.F. Hartl, S.N. Parragh 132/282

133 Branch and bound (BB) - the concept z = 10 z = 11 z = 12 z = 13 z = 17 z = 16 z = 18 X X LB = 14 UB = 14 z = 14 z = 15 X c R.F. Hartl, S.N. Parragh 133/282

134 ATSP: Little et al. BB algorithm 1 lower bound reduction cost of the cost matrix 2 choose a node k for branching: smallest lower bound. 3 branching scheme at node k choose the arc with the largest θ value and generate two child nodes: in the first forbid all tours containing this arc (lower bound: cost value of parent node plus θ), in the second node, include this arc (exclude according row and column and set arc that would lead to a cycle to M). J.D.C. Little, K.G. Murty, D.W. Sweeney, C. Karel (1963) An Algorithm for the Traveling Salesman Problem. Operations Research 11: c R.F. Hartl, S.N. Parragh 134/282

135 ATSP: Little et al. BB algorithm θ ij is the sum of the smallest cost element in row i (excluding c ij ) and the smallest cost element in column j (excluding c ij ); only computed for arcs with c ij = 0. arc that would lead to a cycle: e.g., already committed arcs (1,2),(3,4), new committed arc (2,3); arc that would lead to a cycle: (4,1); e.g., already committed arcs (1,2),(3,4), new committed arc (5,6); arc that would lead to a cycle: (6,5). c R.F. Hartl, S.N. Parragh 135/282

136 ATSP: Little et al. BB algorithm Lower bounds let z(t) denote the costs of a tour under a matrix before row- and column-wise reduction (tour one entry in each row and column selected!); let z 1 (t) denote the cost under the matrix afterward and h the reduction constant: z(t) = h+z 1 (t) therefore, h is a valid lower bound for z(t). If c ij is set to M (forbidden), row i and column j can be reduced by their smallest entries; the sum of these two is given by θ ij. c R.F. Hartl, S.N. Parragh 136/282

137 ATSP: Little et al. BB - Example Root node k = 0 c R.F. Hartl, S.N. Parragh 137/282

138 ATSP: Little et al. BB - Example Root node k = M M M M M M c R.F. Hartl, S.N. Parragh 138/282

139 ATSP: Little et al. BB - Example Root node k = M M M M M M -2 c R.F. Hartl, S.N. Parragh 139/282