A new set of hybrid heuristics for the quadratic assignment problem

Size: px
Start display at page:

Download "A new set of hybrid heuristics for the quadratic assignment problem"

Transcription

1 A new set of hybrid heuristics for the quadratic assignment problem André de São Thiago Moreira Valdir Agustinho de Melo Paulo Oswaldo Boaventura Netto Programa de Engenharia de Produção, COPPE/UFRJ CP , Rio de Janeiro, RJ, Brazil A new set of hybrid heuristics is proposed for the quadratic assignment problem, through the use of a matrix defined by Picard and Queyranne in the 70 s. This matrix has the advantage of guaranteeing the feasibility of a solution through the structural arrangement of the corresponding entries. Its use integrates the concept of pseudo-feasible solutions defined by Rangel. A modified GRASP using a taboo search, a reactive GRASP and an adaptive memory GRASP take profit of this structure. Among the result evaluation techniques utilized, a semi-qualitative comparison by pairs of results is applied. Keywords: Quadratic Assignment Problem, metaheuristics, matrix structures. 1. Introduction 1.1 A vision of the quadratic assignment problem (QAP) The classic (symmetric) QAP is the problem of allocating facilities to positions in a pairwise dependent situation, where the cost is obtained through a quadratic function depending on materials flows and distances. A graph-theoretical approach allows us to consider the QAP as the problem of allocating the vertices of a complete graph K n (f) (facilities) to those of another K n (d) (positions), the flows and distances values being associated to the edges of the corresponding graphs. Example: The permutation corresponds to (Fig. 1.1): 1 2 f 14 d 21 f 12 d 23 K n (d) 4 1 f 13 d 24 f 24 d K n (f) f 34 d f 23 d 34 Fig. 1.1: Facility allocation in the QAP A graph K n has n vertices and n(n 1)/2 edges. So the entries of each QAP data matrix can be associated to their respective edges, while the vertex positions can be permuted to obtain different solutions. The QAP objective function can be expressed in terms of permutations, (BS78) as min Z = c i N j N iπ(i)jπ(j) A QAP solution can be associated either to a vertex- (as above) or to an edge-permutation. But, as the edges cannot be independently allocated, not every edge-permutation corresponds to a feasible solution. [1747]

2 As an example, let us consider the symmetric instance (Land05), where n is the instance order; then N = n(n 1)/2 is the length of the vectors corresponding to the upper data matrix triangles: n = 5 (N = 10); D = [28, 33, 22, 20, 27, 40, 25, 30, 15, 18]; F = [10, 3, 1, 15, 6, 8, 4, 0, 10, 3], where D and F contain these entries in lexicographic order We can build a matrix Q = FD T with these vectors, for this instance, where the entry indices correspond to the lexicographic orderings of the two vectors (Table 1.1): F D Table 1.1: Cost coefficient matrix for a QAP instance This matrix contains every possible term of the QAP objective function. 1.2 The Picard-Queyranne matrix We propose a new hybrid heuristic for the quadratic assignment problem, through the use of a matrix defined by Picard and Queyranne (here referenced as PQ-matrix) in the 70 s. It represents a QAP instance through a graph H made by n copies of the complete graph K n. We can build on that structure n! different maximal cliques, each one being associated to a solution (Fig. 1.2): ( 2143) Fig. 1.2: Solution generation through the Picard-Queyranne matrix The PQ-matrix for a PQA instance corresponds to the Kronecker product of the instance data matrices: f11d V(H) = F D =. fn1d f1nd. f D nn (1.1) [1748]

3 It can also be obtained from Q by rebuilding n(n 1) n-matrices, each Q row corresponding to the content of the upper triangle of the corresponding matrix when token in lexicographic order. As we are dealing with symmetric QAPs, these matrices are then symmetrized. We can observe that two different PQ-matrices can be built (they could be called the FD- and the DF-PQmatrices), the second being obtained by interchanging F and D positions in (1.1). An important property of the PQ-matrix is that all cost coefficients of a feasible solution can be found along the same rows and columns. Example: Table 1.2 shows the lower submatrix triangle of the Land05 FD-PQ-matrix. Within it, we represent the solution (4,1,3,2,5), which corresponds to the pair set (14,34,24,54,31,21,51,23,53,52) whose elements are bold-underlined both on the matrix Q above (where they correspond to the edge-permutation (3,8,6,10,2,1,4,5,9,7)) and also in the PQ-matrix below. We call these edge-permutations relaxed solutions. Within the PQ-matrix structure, these elements appear on the same rows and columns for each row of submatrices and for each column of submatrices, characterizing this solution as being feasible Table 1.2: The PQ-matrix for Land05 So to obtain a feasible solution we have to take n 1 entries within a given submatrix column (all on the same internal column, which defines the first element) or within a given submatrix row (all on the same internal row, which defines the last element). In the example we can choose positions 1, 3, 2 and 5, all in column 4 of the first submatrix column, or we can choose the last submatrix row (positions 4, 1, 3 and 2) and within it the fifth row. Anyway, the vertexpermutation is (4,1,3,2,5). We have to choose the remaining 6 elements of our edge-permutation in the indicated positions in order to the solution to be feasible (we then obtain a feasible relaxed solution). We call the first n 1 elements the head of a permutation, the remaining ones forming its tail. We can make different tail allocations and obtain other (infeasible) solutions with the same elements in the first n - 1 positions. These solutions will be called pseudo-feasible solutions (Ra00a). [1749]

4 The pseudo-feasible solutions given by chosen line and column are the neighborhood where we will make the GRASP local search, where we look for a new feasible solution. In this process we obtain an infeasible solution by doing a 2-exchange and then we look for a new feasible solution having this 2-exchange. 2. The algorithms 2.1 GRASP (Greedy Randomized Adaptive Search Procedure) is a multistart metaheuristic proposed by Feo and Resende (FR95), composed by two steps: start solution generation, using a randomized greedy approach; local search, using convenient neighborhood and search techniques. The classical GRASP is a memoryless multistart greedy-randomized algorithm for whose details we refer to (FR95). An initial solution is obtained by the general procedure below (construction phase), where a probabilistic parameter α is applied in order to randomize a greedy element selection: Procedure GRASP( ); 1 Data Reading( ); 2 While stop criterion not satisfied do 3 BuildRandomizedGreedySolution (solution); 4 LocalSearch (solution,neighborhood(solution)); 5 AtualizeSolution (solution,bestsolutionfound); 6 EndWhile; 7 Return (bestsolutionfound); EndProcedureGRASP; The local search applied to the solution considers a neighborhood structure, which has a number of possible definitions, according to the problem structure and, even for a given problem, to the theoretical resources that are utilized. A general vision of this search is LocalSearch (solution,neighborhood(solution)); 1 While solution not locally optimal do 2 Find best solution t in the set Neighborhood (solution); 3 s = t; 4 EndWhile; EndLocalSearch; The GRASP variants studied here are: normal, reactive and adaptive memory GRASP versions (PR00, FG99); a classic taboo search as the local search heuristic (Gl89a-b); an upper value limit or filter for the initial solutions (Ra00b). A summary of these variants is presented below, where we also describe the normal GRASP here used, which is based on PQ-matrix structure (GRASP-PQ). The stopping rules were based chiefly on iteration number (here we used either 2000 or iterations) but also, for some tests, on processing time and on a target to be attained. Comparisons were made with two GRASP variants of the literature: the GRASP by Resende (RR02) the GRASP with robust tabu search by Taillard (Ta91); [1750]

5 At this stage of the research we were chiefly interested in the efficiency of the proposed search in what concerns the finding of new solutions. So the comparison with Taillard s algorithms was done by taking the same number of iterations (2000 or 20000), disregarding the time processing advantage which allows that algorithm to work with much more iterations to obtain better results. 2.2 The basic GRASP and the PQ-matrix A mixed random and minimax strategy was used to find the initial solutions for GRASP-PQ. In the algorithm, the positions indicated by letters correspond to submatrices and those indicated by numbers, to matrix elements. Procedure GRASP-PQ ( ) 1 Data reading ( ) 2 While solution not completely built do 3 Choose one among the α% greater flows f ab, not yet allocated; 4 Choose the least distance d xy, which allows for a feasible solution; 5 Allocate f ab to d xy doing P a y; P b x; < P a and P b are y and x image positions respectively > End While. The local search is done on the set of pseudo-feasible solutions corresponding to each submatrix, as follows: While the initial solution is improved do 1 For every solution tail submatrix do 2 For every row/column where the cost coefficient is lesser than that of the initial solution do 3 - generate neighbor solution; 4 - calculate cost of neighbor solution; 5 - actualize best neighbor solution; 6 End (for) 7 End (for) 8 If better neighbor solution < initial solution then Initial solution better neighbor solution; else Initial solution not improved End While 2.3 The modified GRASP variants The general idea of modifying GRASP was the classical one of giving it some type of memory (result recall) in order to allow the algorithm to use some of the experience acquired in previous trials. The reactive GRASP (PR00) examines the results obtained through a block of iterations in order to modify the assigned probabilities of a set of specified values for the random-greedy proportion parameter α of GRASP. We worked with ten values and with 100-iteration blocks. The adaptive memory GRASP (FG99) builds an elite solution set where the algorithm seeks comparison with each new starting solution in order to know if it is sufficiently different (in the sense of the existance of equal image elements) from these solutions, which allows the algorithm to enlarge its exploring of the solution set. A strategy is defined to actualize the elite solutions. We found it useful to work with 20% difference values. [1751]

6 These variants use strategies designed to give some memory to the basic memoryless GRASP, which is done through the actualization of parameter α (for reactive search) and through building of an elite solution set (for adaptive search). In the first scheme a previously defined set A of α values is used (usually we have A = m = 10) and an α value is chosen with probability p i for each iteration. The scheme begins with p i = 1/m for all I and the p i are actualized by using information given by iteration blocks usually of 100 iterations each. If M i is the solution value average corresponding to α i value and F(s*) is the best solution value until the moment, we have from where F(s*) q i = (i = 1,..., m) M i (2.1) m p = i q i / q j= 1 j. (2.2) Higher M i values improve, then, the probability of the corresponding α i be selected. The elite set S for adaptive (memory) tabu search is built by defining cost bounds best(s) and worst(s), which allow a solution to be admitted in S if its cost is better than worst(s) and it is also sufficiently different (in terms of pair composition) form every solution in S, by replacing this worst solution. Then the elite solutions are the basis for a more complete exploration of the solutions universe. For some problems the new solutions are built by using elements from elite solutions, but here we only used them to guarantee the pair difference. For reactive GRASP the commands to be included in the basic GRASP / tabu are sufficiently obvious to authorize us to avoid the repetition of the algorithm; on the other hand, the procedure for verifying differences between solutions is While (i < R) and (counter nsimil) do 1 j = 1; 2 count = 0; 3 While (j n) and (count nsimil) do 4 If (initialsolution[j] = S[i][j]) then 5 count = count + 1; 6 j = j+1; 7 EndWhile; 8 i = i+1; 9 EndWhile; 10 If (contador > nsemelhanca) then RejectSolution; 11 EndIf. 2.4 The tabu search Every variant but the basic GRASP utilized a simple tabu search strategy. Tabu search is a local search procedure which accepts the worsening of solution values in order to escape from local optima. The basic local search does not admit such strategy and as a consequence the solution path can be allowed to return to previously found solutions. This can be avoided through the use of a tabu list where a solution attribute stays during a given number of iterations (tabu time). A number of improvements have been implemented over this basic scheme, in order to deal with different situations (Glover (Gl77), (Gl86), (Gl89a), (Gl89b), Hansen (Ha86), Werra and Hertz (Wh89)). We formulated a GRASP / tabu algorithm by modifying the algorithm presented in 2.3 through the introduction of a tabu condition in Step 3 and the inclusion of present solution [1752]

7 attributes after Step 7. These attributes are the last pair whose cost coefficients were interchanged and the corresponding cost coefficients. 3. Computational results 3.1 The performance tests were run on three platforms: C compiler for Windows in a PC with AMD 2.6 XP processor (basis for reported times); C compiler for Windows in a PC with Pentium IV 3.2 processor; C compiler for Linux in a workstation with 2 Xeon 2.8 processors. The Pentium IV times were about 0.76 of equivalent AMD times and 0.85 of equivalent Xeon times. The parameter α took the values {0.1, 0.3, 0.5}, the results shown in Tables 3.1 (a) and (b) being the best of these options. The percent errors are calculated on the basis of the optimal or best known value for the instance. The tests used either 2000 or iterations, the second option being indicated by a (*) added to the instance name. We used instances from QAPLIB collection (QAPLIB): nug12 to 30, chr12a to 25a, tai12a to 50a, rou12 to 20, scr12 to 20 and els19 and also instances dre15 to 24 from Drezner (DHT04) collection (total 37 instances). For each instance we used ten seeds randomly chosen from the prime number list between 1 and 2,000,000 (Pi06). The seeds utilized were , 11399, , 2963, , , 7331, 797, and The results presented in the tables are the corresponding averages, with the exception of the Minimal Error column which gives the best result for all seeds. The column Average Improvement % gives the percent value improvement over the initial value, on the basis of this value. The column Average Gap % gives the percent gap value between the best value obtained and the best value known, on the basis of this one. When using GRASP with adaptive memory, solutions were considered for admission to the elite list through a value criterion, but the decision of using or not a solution used a pair similarity criterion with an upper limit of 20%, established by previous testing. 3.2 Discussion Tables 3.1 (a) and (b) below present the results obtained with the basic GRASP-PQ, the simple tabu version (GRASP-TB), the reactive tabu version (GRASP-RT), the adaptive memory version (GRASP-AM) and the Resende version (GRASP-RR). We can observe that the pure versions (PQ and RR) are less precise as their percent errors are greater than those of TB, RT and AM. RT seems to be more precise than AM. The least errors, both average and minimum, correspond to TB (23 and 19 cases respectively), against RT (12 and 12) and AM (13 and 10). There was a significant occurrence of zero errors for all techniques. On the other hand, RR and TB find more best solutions (13 cases each), while RT won in 7 cases. In what concerns speed PQ is much faster than RR and, among the other variants, AM is the faster as it discards initial solutions not sufficiently different from those in the elite list. The strategy seems then fairly efficient, even if we used a single selection criterion instead of a double one involving also quality of the initial solution. [1753]

8 Instance Error (%) Minimum error (%) Processing time, sec (AMD 2.6) Nsol Variant PQ TB RT AM RR PQ TB RT AM RR PQ TB RT AM RR PQ TB RT AM RR Nug ,21 6,50 6,38 5,25 2, Nug14 0 0,02 0, ,57 16,82 15,78 15,73 5, Nug ,53 24,77 22,45 21,43 7, Nug16a 0, ,59 35,08 34,32 30,81 11, Nug17 0,09 0,03 0,09 0,02 0, ,32 49,76 50,02 47,55 16, Nug18 0,28 0,12 0,1 0, ,49 65,37 58,45 62,18 20, Nug20 0, ,25 112,20 102,40 101,95 35, Nug21 0, , ,82 154,97 142,46 150,48 47, Nug25* 0,01 0 0,01 0, , , ,3 409, , Nug30* 0,29 0,13 0,17 0,40 0,08 0, , ,9 7984,0 1002,6 2994, Scr12 0, ,53 2,76 2,68 2,18 2, Scr15 6, , ,97 8,64 9,19 8,32 8, Scr20 7,54 0,003 0,03 0,01 0,01 4, ,90 33,55 28,46 30,50 36, Rou ,61 8,90 7,86 6,71 2, Rou , ,96 27,76 24,67 25,26 7, Rou20* 0,006 0,02 0,01 0,16 0, ,02 128, ,23 909,33 120,12 311, Els19 0, , ,67 70,96 62,41 65,28 33, Chr12a 1, , ,23 0,56 0,68 0,56 2, Chr15a 11,57 0 0,04 0,52 0 7, ,01 23,64 17,49 2,60 66, Chr18a* 40,68 0,19 0,52 7,05 2,83 32,67 0 0,18 0 0,18 7,22 44,16 37,29 3,98 165, Chr20a* 78,91 1,07 1,86 6,45 0,56 54, ,46 0 9,54 45,21 44,84 4,86 266, Chr20b* 47,45 3,35 4,83 8,40 4,35 36,38 0 4,18 6,44 3,83 6,69 41,11 40,66 4,75 242, Chr22a* 10,78 0,72 0,79 1,76 1,81 8,97 0 0,32 0 1,59 13,69 91,35 67,80 9,22 495, Chr22b* 17,33 1,51 1,66 2,68 1,58 15,95 0,84 0,94 1,78 1,36 13,46 78,34 83,64 8,58 450, Chr25a* 112,76 7,02 7,24 12,21 10,48 100,32 0 1,84 2,05 10,22 15,48 163,56 138,40 14,81 954, Tai12a ,72 8,97 8,60 6,40 2, Tai15a 0,13 0 0,02 0,02 0, ,55 25,32 26,37 23,75 6, Tai17a 0,44 0,35 0,28 0,49 0, ,49 50,20 47,19 49,75 13, Tai20a* 0,58 0,22 0,50 0,80 0,37 0, ,47 0,30 101, ,67 112,88 106,56 297, Tai25a* 1,08 0,64 0,79 1,43 1,67 0,55 0,15 0,37 1,02 1,55 264, , ,8 340,31 897, Tai30a* 1,32 1,29 1,21 1,74 1,60 0,93 1,02 0,56 0,85 1,53 611, ,3 7392,5 902, , Tai40a 2,44 2,28 2,42 2,31 2,35 1,80 1,76 2,08 1,99 2,09 181, , ,2 3719,5 969, Tai50a 2,80 2,72 2,73 2,71 2,81 2,27 2,56 2,41 2,42 2,52 436, ,1 8416, , Dre15 57, ,42 49, ,44 2,72 2,61 2,64 5, Dre18 83,32 3,43 1,45 1,45 4,34 68, ,65 6,69 6,06 6,36 15, Dre21* 87,19 7,92 15,79 30,17 26,81 71,91 0 7, ,79 8,57 140,19 100,57 10,93 319, Dre24* 70,81 25,10 25,76 41,92 33,43 54, ,85 33,33 20,29 260,71 186,56 20,98 602, Table 3.1 (a): Comparison among GRASP variants [1754]

9 Instance Average improvement % Average gap % Variant PQ TB RT AM RR PQ TB RT AM RR Nug12 19,81 20,56 22,67 18,13 17,19 5,53 4,66 4,89 3,95 5 Nug14 19,23 20,20 19,78 18,46 16,52 6,00 4,84 4,87 4,52 5 Nug15 21,73 22,70 22,33 20,92 19,22 5,66 4,24 4,24 3,92 5 Nug16a 18,08 20,19 20,19 18,85 17,61 5,36 4,69 4,69 4,35 5 Nug17 21,26 21,65 21,92 20,50 17,64 4,99 4,15 4,07 3,85 4 Nug18 20,31 21,07 21,18 20,55 17,77 4,91 4,25 4,22 4,09 4 Nug20 20,37 20,42 21,06 20,55 17,57 4,77 3,95 4,00 3,90 4 Nug21 24,53 25,17 25,29 24,83 22,99 5,07 4,38 4,30 4,22 4 Nug25* 21,44 22,13 21,92 21,93 19,17 4,58 3,75 3,76 3,76 4 Nug30* 19,74 21,03 20,95 20,95 19,14 4,48 3,94 3,92 3,68 4 Scr12 30,45 40,50 39,11 32,28 33,32 19,58 6,73 6,76 5,55 6 Scr15 30,51 40,28 42,15 39,67 40,67 19,38 9,65 9,41 8,84 9 Scr20 31,38 45,50 44,34 43,83 38,84 21,85 8,86 9,05 8,90 8 Rou12 17,41 17,67 17,05 14,08 13,86 5,40 5,10 5,08 4,18 5 Rou15 16,77 17,35 17,47 16,53 12,04 7,05 6,39 6,42 6,13 6 Rou20* 14,12 15,97 15,69 15,59 13,22 4,82 4,58 4,60 4,46 5 Els19 56,82 59,91 59,45 57,37 56,99 24,14 19,33 19,93 19,33 25 Chr12a 29,51 52,86 59,89 49,43 52,48 53,18 31,26 33,73 27,93 25 Chr15a 34,64 72,62 69,37 69,30 55,58 60,13 34,85 34,07 33,89 28 Chr18a* 33,90 73,49 72,10 71,88 59,23 60,03 41,34 40,52 37,58 39 Chr20a* 16,45 63,33 66,66 66,70 53,16 53,79 31,60 30,97 27,89 32 Chr20b* 13,10 62,23 66,25 66,19 48,88 54,32 26,70 25,85 23,36 27 Chr22a* 28,72 56,06 52,79 52,84 42,16 32,47 12,00 11,86 10,99 11 Chr22b* 15,76 49,25 51,94 51,86 38,67 37,12 10,73 10,79 9,95 11 Chr25a* 15,24 68,85 67,32 67,24 58,26 45,91 32,82 32,93 29,80 31 Tai12a 20,16 20,53 19,66 16,54 20,15 8,45 8,02 8,10 6,75 7 Tai15a 14,66 15,02 16,71 15,64 11,07 4,79 4,50 4,56 4,25 4 Tai17a 15,84 16,09 16,25 15,40 11,24 5,16 4,98 5,08 4,58 5 Tai20a* 13,97 14,52 15,97 15,57 12,58 5,80 5,53 5,08 4,80 5 Tai25a* 13,79 14,39 14,13 14,11 11,36 4,04 4,64 4,50 3,90 4 Tai30a* 12,49 12,60 12,65 12,62 9,62 3,65 3,57 3,69 3,14 3 Tai40a 12,55 12,40 12,42 12,42 8,95 2,56 2,62 2,49 2,59 3 Tai50a 11,70 11,95 11,79 11,79 8,05 2,13 2,12 2,13 2,15 3 Dre15 36,15 53,36 58,26 54,84 39,63 44,41 43,27 43,88 41,33 38 Dre18 27,60 57,87 61,47 60,12 47,06 46,91 42,50 43,73 42,70 42 Dre21* 25,69 62,27 60,66 60,42 44,57 43,96 44,37 40,28 32,85 34 Dre24* 35,01 62,73 61,72 61,59 45,05 43,49 35,35 34,62 26,69 31 Table 3.1 (b): Comparison among GRASP variants The average improvement was better for TB (for 19 instances) and RT (for 17 ones). On the other hand, the average gap results were much better for AM with 30 better results, while RR had 7. All these results point to the interest of further research on algorithm implementation, chiefly on the development of AM. 3.3 Comparison with GRASP by Taillard The table below shows a performance comparison between the GRASP-TB here presented and the robust tabu search by Taillard (Ta91). Whereas the speed of this last algorithm is outstanding for any such comparison, its precision was inferior with relation to that of GRASP-TB when tested in similar conditions. The tests were done on 37 instances with a defined target as stopping criterion, which was the optimal best value known for the instance. GRASP-TB presented lesser error for 17 instances, while GRASP-Taillard was better for 11 instances, a tie being observed for 9 others (NBR row). [1755]

10 Instance 4. Pair evaluation Error (%) (Taillard) Error (%) (GRASP/Tabu) Instance Error (%) (Taillard) Error (%) (GRASP/Tabu) Nug Chr18a* 0,10 0,19 Nug14 0,02 0,02 Chr20a* 3,89 1,07 Nug15 0,02 0 Chr20b* 4,42 3,35 Nug16a 0 0 Chr22a* 0,85 0,72 Nug17 0,03 0,03 Chr22b* 1,47 1,51 Nug18 0,11 0,12 Chr25a* 8,48 7,02 Nug20 0,07 0 Tai12a 0 0 Nug21 0,35 0 Tai15a 0,09 0 Nug25* 0 0 Tai17a 0,42 0,35 Nug30* 0,04 0,13 Tai20a* 0,33 0,22 Scr Tai25a* 0,54 0,64 Scr15 0,39 0 Tai30a* 0,44 1,29 Scr20 0,08 0,003 Tai40a 1,49 2,28 Rou Tai50a 1,80 2,72 Rou15 0,07 0 Dre15 14,84 0 Rou20* 0,01 0,02 Dre18 6,81 3,43 Els19 0,36 0 Dre21* 1,40 7,92 Chr12a 0 0 Dre24* 22,98 25,10 Chr15a 0,12 0 NBR A set of criteria can be used to compare the performance of a set of k algorithms by examining the different orders observed on the same parameter values. If a non-decreasing order for a parameter p is used we can say that, according to the values obtained by two algorithms, that if p(i) < p(j), then the algorithm i obtained a better result than j according to p. A sequence of C k,2 comparisons can then be made. If we have more than a parameter a given sequence can present discordances with respect to another sequence: then the distance between two sequences is the number of such discordances. The relative error between two sequences is the percent ratio distance / C k,2. We have four evaluation parameters: the average gap between final values and best value known, the complement of the average improvement from initial and final values, the execution time and the number of better solutions obtained. This last one will be presented on a suitably modified form, I(nsol, error) = 10 (error%) + nsol -1, in order to allow for cases where the optimal or best known solution is not found by the algorithm. The five algorithms under comparison were (1) GRASP-PQ, (2) GRASP-TB, (3) GRASP-RT, (4) GRASP-AM and (5) Rangel/Resende GRASP. As an example, the instance Nug15 gives the following results: A B C D E (a) avg gap 5,66 4,24 4,24 3,92 5,00 4 3,92 2 4,24 3 4,24 5 5,00 1 5,66 (b) (1-avg improv) 78,27 77,30 77,67 79,08 83, , , , , ,48 (c) I(n,sol) 0,100 0,040 0,040 0,042 0, , , , , ,100 (d) (t,sec) 3,53 24,77 22,45 21,43 7,86 1 3,53 5 7, , , ,77 The right side of the table gives the non-increasing ordering which can be found in the left side. A numerical code was used, based on the relations {(+1, >),( -1, <),(0, =)}, in order to obtain a numerical table which we can use also to compare the parameters a b c d [1756]

11 By summing pairs of rows we can evaluate discordances (corresponding to zeroes). The last column entries count their number: if they correspond, for example, to more than 75% of C 5,2 = 10, the parameter utility would be questionable. On the other hand, by considering columns we are testing the ability of pairs of algorithms to distinguish instances by difficulty, based on the current criteria Discord ab ac ad bc bd cd Discord The results for the instances below consider only pairs with 3 or less discordancies, where the indicated ordering is the same for the different parameter pairs (e.g. 23 cannot be compared through their results with Nug15 owing to different ordering scores, two +1 and two -1): Instance nug chr15a nug chr20a chr25a nug tai20a tai30a dre dre dre dre tai40a tai50a Σ We can see that GRASP-PQ is less efficient than GRASP-TB and, almost everywhere, than GRASP- RT. GRASP-AM shows itself better than GRASP-PQ but in two cases. Every algorithm but GRASP- PQ is better than algorithm (5). GRASP-RT was somewhat better than GRASP-TB and GRASP-AM was much better than (5). 5. Conclusions This paper presents some research results on a new neighborhood for local search, that of pseudofeasible QAP solutions, accessible through a structure known as the Picard-Queyranne matrix. A local search was applied to the solutions generated by a classical GRASP. A tabu search strategy was also used within the local search and two memory variants, reactive and adaptive memory GRASPs, were also tested. Comparisons were made with Taillard s robust taboo search (in what concerns the precision) and with the Resende GRASP. The results obtained show interest in the exploration of the pseudo-feasible solutions, avalable as a new neighborhood through the use of the Picard-Queyranne matrix. While some implementations need to be improved, we think this approach to develop new heuristics presents an interest for further research, both in theory and new algorithms. [1757]

12 References (BS78) Burkard, R.E. and Stratman, R.H. (1978). Numerical investigations on the quadratic assignment problem. Naval Research Logistics Quaterly 25, 1, (Ra00a) Rangel, M.C. (2000). Contribuições algébricas ao problema quadrático de alocação. D.Sc. Thesis, COPPE/UFRJ. (Ra00b) Rangel, M. C., Abreu, N.M.M., Boaventura, P.O. (2000). GRASP para o PQA: um limite de aceitação para as soluções inciais. Pesquisa Operacional 20, 1, (Gl89a) Glover, F. (1989). Tabu search Part I, ORSA Journal on Computing 1, (Gl89b) Glover, F. (1989). Tabu search Part II, ORSA Journal on Computing 2, (PR00) Prais, M. e Ribeiro, C. (2000) Reactive GRASP: an application to a matrix decomposition problem in TDMA traffic assignment. INFORMS Journal of Computing 12, 3, (FG99) Fleurent, C. e Glover, P. (1999). Improved constructive multistart strategies for the quadratic assignment problem using adaptive memory. INFORMS Journal of Computing 11, 2, (Ta91) Taillard, E. D. (1991). Robust tabu search for the quadratic assignment problem. Parallel Computing 17, pp (RR02) Resende, M.G.C. e Ribeiro, C.C.. Greedy randomized adaptive search procedures (GRASP). In Glover, F., Kochenberger, G., editores, Handbook of Metaheuristics, p Kluwer Academic Publishers, (FR95) Feo, T.A. and Resende, M.G.C. (1995). Greedy Randomized Adaptive Search Procedures. Journal of Global Optimization 6, 2, (LPR,94) Li, Y., Pardalos, P.M. e Resende, M.G.C. (1994). A greedy randomized adaptive search procedures for the quadratic assignment problem, DIMACS Series in Discrete Mathematics and Theoretical Computer Science. 16, (PR00) Prais, M. e Ribeiro, C. (2000) Reactive GRASP: an application to a matrix decomposition problem in TDMA traffic assignment. INFORMS Journal of Computing 12, 3, (Gl,77) Glover, F. (1977). Heuristics for integer programming using surrogate constraints. Decision Science 8, (Gl,86) Glover, F. (1986). Future paths for integer programming and links to artificial intelligence. Computers and Operations Research 5, (Ha,86) Hansen, P. (1986). The steepest ascent mildest descent heuristic for combinatorial programming. Congress on Numerical Methods in Combinatorial Optimization, Capri, Itália. (WH,89) Werra, D. e Hertz, A. (1989). Tabu search techniques: a tutorial and an application to neural networks. OR Spektrum 11, [1758]

AN ANT APPROACH FOR STRUCTURED QUADRATIC ASSIGNMENT PROBLEMS

AN ANT APPROACH FOR STRUCTURED QUADRATIC ASSIGNMENT PROBLEMS AN ANT APPROACH FOR STRUCTURED QUADRATIC ASSIGNMENT PROBLEMS Éric D. Taillard, Luca M. Gambardella IDSIA, Corso Elvezia 36, CH-6900 Lugano, Switzerland. Extended abstract IDSIA-22-97 ABSTRACT. The paper

More information

Metaheuristics and Local Search

Metaheuristics and Local Search Metaheuristics and Local Search 8000 Discrete optimization problems Variables x 1,..., x n. Variable domains D 1,..., D n, with D j Z. Constraints C 1,..., C m, with C i D 1 D n. Objective function f :

More information

An approach for the Class/Teacher Timetabling Problem using Graph Coloring

An approach for the Class/Teacher Timetabling Problem using Graph Coloring An approach for the Class/Teacher Timetabling Problem using Graph Coloring G. S. Bello M. C. Rangel M. C. S. Boeres Received: date / Accepted: date Keywords Timetabling Graph Coloring Metaheuristics Tabu

More information

Metaheuristics and Local Search. Discrete optimization problems. Solution approaches

Metaheuristics and Local Search. Discrete optimization problems. Solution approaches Discrete Mathematics for Bioinformatics WS 07/08, G. W. Klau, 31. Januar 2008, 11:55 1 Metaheuristics and Local Search Discrete optimization problems Variables x 1,...,x n. Variable domains D 1,...,D n,

More information

LOCAL SEARCH ALGORITHMS FOR THE MAXIMUM K-PLEX PROBLEM

LOCAL SEARCH ALGORITHMS FOR THE MAXIMUM K-PLEX PROBLEM LOCAL SEARCH ALGORITHMS FOR THE MAXIMUM K-PLEX PROBLEM By ERIKA J. SHORT A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE

More information

Variable Objective Search

Variable Objective Search Variable Objective Search Sergiy Butenko, Oleksandra Yezerska, and Balabhaskar Balasundaram Abstract This paper introduces the variable objective search framework for combinatorial optimization. The method

More information

Meta-heuristics for combinatorial optimisation

Meta-heuristics for combinatorial optimisation Meta-heuristics for combinatorial optimisation João Pedro Pedroso Centro de Investigação Operacional Faculdade de Ciências da Universidade de Lisboa and Departamento de Ciência de Computadores Faculdade

More information

A MIXED INTEGER DISJUNCTIVE MODEL FOR TRANSMISSION NETWORK EXPANSION

A MIXED INTEGER DISJUNCTIVE MODEL FOR TRANSMISSION NETWORK EXPANSION A MIXED INTEGER DISJUNCTIVE MODEL FOR TRANSMISSION NETWORK EXPANSION Laura Bahiense*, Gerson C. Oliveira (PUC/Rio), Mario Pereira*, Member, Sergio Granville*, Member Abstract: The classical non-linear

More information

HYBRIDIZATIONS OF GRASP WITH PATH-RELINKING FOR THE FAR FROM MOST STRING PROBLEM

HYBRIDIZATIONS OF GRASP WITH PATH-RELINKING FOR THE FAR FROM MOST STRING PROBLEM HYBRIDIZATIONS OF GRASP WITH PATH-RELINKING FOR THE FAR FROM MOST STRING PROBLEM DANIELE FERONE, PAOLA FESTA, AND MAURICIO G.C. RESENDE Abstract. Among the sequence selection and comparison problems, the

More information

Ruin and Recreate Principle Based Approach for the Quadratic Assignment Problem

Ruin and Recreate Principle Based Approach for the Quadratic Assignment Problem Ruin and Recreate Principle Based Approach for the Quadratic Assignment Problem Alfonsas Misevicius Kaunas University of Technology, Department of Practical Informatics, Studentu St. 50 400a, LT 3031 Kaunas,

More information

GRASP with path-relinking for the weighted maximum satisfiability problem

GRASP with path-relinking for the weighted maximum satisfiability problem GRASP with path-relinking for the weighted maximum satisfiability problem Paola Festa 1, Panos M. Pardalos 2, Leonidas S. Pitsoulis 3, and Mauricio G. C. Resende 4 1 Department of Mathematics and Applications,

More information

5. Simulated Annealing 5.1 Basic Concepts. Fall 2010 Instructor: Dr. Masoud Yaghini

5. Simulated Annealing 5.1 Basic Concepts. Fall 2010 Instructor: Dr. Masoud Yaghini 5. Simulated Annealing 5.1 Basic Concepts Fall 2010 Instructor: Dr. Masoud Yaghini Outline Introduction Real Annealing and Simulated Annealing Metropolis Algorithm Template of SA A Simple Example References

More information

GRASP FOR LINEAR INTEGER PROGRAMMING

GRASP FOR LINEAR INTEGER PROGRAMMING Chapter 1 GRASP FOR LINEAR INTEGER PROGRAMMING Teresa Neto tneto@matestvipvpt João Pedro Pedroso jpp@nccuppt Departamento de Ciência de Computadores Faculdade de Ciências da Universidade do Porto Rua do

More information

Lecture H2. Heuristic Methods: Iterated Local Search, Simulated Annealing and Tabu Search. Saeed Bastani

Lecture H2. Heuristic Methods: Iterated Local Search, Simulated Annealing and Tabu Search. Saeed Bastani Simulation Lecture H2 Heuristic Methods: Iterated Local Search, Simulated Annealing and Tabu Search Saeed Bastani saeed.bastani@eit.lth.se Spring 2017 Thanks to Prof. Arne Løkketangen at Molde University

More information

A New Heuristic for the Quadratic Assignment Problem

A New Heuristic for the Quadratic Assignment Problem JOURNAL OF APPLIED MATHEMATICS AND DECISION SCIENCES, 6(3), 143 153 Copyright c 2002, Lawrence Erlbaum Associates, Inc. A New Heuristic for the Quadratic Assignment Problem ZVI DREZNER zdrezner@fullerton.edu

More information

Single Solution-based Metaheuristics

Single Solution-based Metaheuristics Parallel Cooperative Optimization Research Group Single Solution-based Metaheuristics E-G. Talbi Laboratoire d Informatique Fondamentale de Lille Single solution-based metaheuristics Improvement of a solution.

More information

3.4 Relaxations and bounds

3.4 Relaxations and bounds 3.4 Relaxations and bounds Consider a generic Discrete Optimization problem z = min{c(x) : x X} with an optimal solution x X. In general, the algorithms generate not only a decreasing sequence of upper

More information

Multi-objective Quadratic Assignment Problem instances generator with a known optimum solution

Multi-objective Quadratic Assignment Problem instances generator with a known optimum solution Multi-objective Quadratic Assignment Problem instances generator with a known optimum solution Mădălina M. Drugan Artificial Intelligence lab, Vrije Universiteit Brussel, Pleinlaan 2, B-1050 Brussels,

More information

Hill climbing: Simulated annealing and Tabu search

Hill climbing: Simulated annealing and Tabu search Hill climbing: Simulated annealing and Tabu search Heuristic algorithms Giovanni Righini University of Milan Department of Computer Science (Crema) Hill climbing Instead of repeating local search, it is

More information

GRASP heuristics for discrete & continuous global optimization

GRASP heuristics for discrete & continuous global optimization GRASP heuristics for discrete & continuous global optimization Tutorial given at Learning and Intelligent Optimization Conference (LION 8) Gainesville, Florida February 17, 2014 Copyright @2014 AT&T Intellectual

More information

Stabilized Branch-and-cut-and-price for the Generalized Assignment Problem

Stabilized Branch-and-cut-and-price for the Generalized Assignment Problem Stabilized Branch-and-cut-and-price for the Generalized Assignment Problem Alexandre Pigatti, Marcus Poggi de Aragão Departamento de Informática, PUC do Rio de Janeiro {apigatti, poggi}@inf.puc-rio.br

More information

Ant Colony Optimization: an introduction. Daniel Chivilikhin

Ant Colony Optimization: an introduction. Daniel Chivilikhin Ant Colony Optimization: an introduction Daniel Chivilikhin 03.04.2013 Outline 1. Biological inspiration of ACO 2. Solving NP-hard combinatorial problems 3. The ACO metaheuristic 4. ACO for the Traveling

More information

XLVI Pesquisa Operacional na Gestão da Segurança Pública

XLVI Pesquisa Operacional na Gestão da Segurança Pública A linear formulation with O(n 2 ) variables for the quadratic assignment problem Serigne Gueye and Philippe Michelon Université d Avignon et des Pays du Vaucluse, Laboratoire d Informatique d Avignon (LIA),

More information

The Strongly Regular (45, 12, 3, 3) Graphs

The Strongly Regular (45, 12, 3, 3) Graphs The Strongly Regular (45, 12, 3, 3) Graphs Kris Coolsaet, Jan Degraer Department of Applied Mathematics and Computer Science Ghent University Krijgslaan 281 S9, B 9000 Gent, Belgium Kris.Coolsaet@ugent.be,

More information

Artificial Intelligence Heuristic Search Methods

Artificial Intelligence Heuristic Search Methods Artificial Intelligence Heuristic Search Methods Chung-Ang University, Jaesung Lee The original version of this content is created by School of Mathematics, University of Birmingham professor Sandor Zoltan

More information

Zebo Peng Embedded Systems Laboratory IDA, Linköping University

Zebo Peng Embedded Systems Laboratory IDA, Linköping University TDTS 01 Lecture 8 Optimization Heuristics for Synthesis Zebo Peng Embedded Systems Laboratory IDA, Linköping University Lecture 8 Optimization problems Heuristic techniques Simulated annealing Genetic

More information

Tabu Search. Biological inspiration is memory the ability to use past experiences to improve current decision making.

Tabu Search. Biological inspiration is memory the ability to use past experiences to improve current decision making. Tabu Search Developed by Fred Glover in the 1970 s. Dr Glover is a business professor at University of Colorado at Boulder. Developed specifically as a combinatorial optimization tool. Biological inspiration

More information

Tightening a Discrete Formulation of the Quadratic Assignment Problem

Tightening a Discrete Formulation of the Quadratic Assignment Problem A publication of 1309 CHEMICAL ENGINEERING TRANSACTIONS VOL. 32, 2013 Chief Editors: Sauro Pierucci, Jiří J. Klemeš Copyright 2013, AIDIC Servizi S.r.l., ISBN 978-88-95608-23-5; ISSN 1974-9791 The Italian

More information

The Core Concept for the Multidimensional Knapsack Problem

The Core Concept for the Multidimensional Knapsack Problem The Core Concept for the Multidimensional Knapsack Problem Jakob Puchinger 1, Günther R. Raidl 1, and Ulrich Pferschy 2 1 Institute of Computer Graphics and Algorithms Vienna University of Technology,

More information

Trip Distribution Modeling Milos N. Mladenovic Assistant Professor Department of Built Environment

Trip Distribution Modeling Milos N. Mladenovic Assistant Professor Department of Built Environment Trip Distribution Modeling Milos N. Mladenovic Assistant Professor Department of Built Environment 25.04.2017 Course Outline Forecasting overview and data management Trip generation modeling Trip distribution

More information

0.1 O. R. Katta G. Murty, IOE 510 Lecture slides Introductory Lecture. is any organization, large or small.

0.1 O. R. Katta G. Murty, IOE 510 Lecture slides Introductory Lecture. is any organization, large or small. 0.1 O. R. Katta G. Murty, IOE 510 Lecture slides Introductory Lecture Operations Research is the branch of science dealing with techniques for optimizing the performance of systems. System is any organization,

More information

Backbone analysis and algorithm design for the quadratic assignment problem

Backbone analysis and algorithm design for the quadratic assignment problem Science in China Series F: Information Sciences 008 SCIENCE IN CHINA PRESS Springer wwwscichinacom infoscichinacom wwwspringerlinkcom Backbone analysis and algorithm design for the quadratic assignment

More information

Effective Variable Fixing and Scoring Strategies for Binary Quadratic Programming

Effective Variable Fixing and Scoring Strategies for Binary Quadratic Programming Effective Variable Fixing and Scoring Strategies for Binary Quadratic Programming Yang Wang, Zhipeng Lü,FredGlover, and Jin-Kao Hao LERIA, Université d Angers, Boulevard Lavoisier, 4945 Angers Cedex, France

More information

Hybridizing the Cross Entropy Method: An Application to the Max-Cut Problem

Hybridizing the Cross Entropy Method: An Application to the Max-Cut Problem Hybridizing the Cross Entropy Method: An Application to the Max-Cut Problem MANUEL LAGUNA Leeds School of Business, University of Colorado at Boulder, USA laguna@colorado.edu ABRAHAM DUARTE Departamento

More information

The combinatorics of pivoting for the maximum weight clique

The combinatorics of pivoting for the maximum weight clique Operations Research Letters 32 (2004) 523 529 Operations Research Letters wwwelseviercom/locate/dsw The combinatorics of pivoting for the maximum weight clique Marco Locatelli a; ;1, Immanuel M Bomze b,

More information

A SEQUENTIAL ELIMINATION ALGORITHM FOR COMPUTING BOUNDS ON THE CLIQUE NUMBER OF A GRAPH

A SEQUENTIAL ELIMINATION ALGORITHM FOR COMPUTING BOUNDS ON THE CLIQUE NUMBER OF A GRAPH A SEQUENTIAL ELIMINATION ALGORITHM FOR COMPUTING BOUNDS ON THE CLIQUE NUMBER OF A GRAPH Bernard Gendron Département d informatique et de recherche opérationnelle and Centre de recherche sur les transports

More information

MAX-2-SAT: How Good is Tabu Search in the Worst-Case?

MAX-2-SAT: How Good is Tabu Search in the Worst-Case? MAX-2-SAT: How Good is Tabu Search in the Worst-Case? Monaldo Mastrolilli IDSIA Galleria 2, 6928 Manno, Switzerland monaldo@idsia.ch Luca Maria Gambardella IDSIA Galleria 2, 6928 Manno, Switzerland luca@idsia.ch

More information

PROBLEM SOLVING AND SEARCH IN ARTIFICIAL INTELLIGENCE

PROBLEM SOLVING AND SEARCH IN ARTIFICIAL INTELLIGENCE Artificial Intelligence, Computational Logic PROBLEM SOLVING AND SEARCH IN ARTIFICIAL INTELLIGENCE Lecture 4 Metaheuristic Algorithms Sarah Gaggl Dresden, 5th May 2017 Agenda 1 Introduction 2 Constraint

More information

A Hybrid Data Mining Metaheuristic for the p-median Problem

A Hybrid Data Mining Metaheuristic for the p-median Problem A Hybrid Data Mining Metaheuristic for the p-median Problem Alexandre Plastino Erick R. Fonseca Richard Fuchshuber Simone de L. Martins Alex A. Freitas Martino Luis Said Salhi Abstract Metaheuristics represent

More information

On a Polynomial Fractional Formulation for Independence Number of a Graph

On a Polynomial Fractional Formulation for Independence Number of a Graph On a Polynomial Fractional Formulation for Independence Number of a Graph Balabhaskar Balasundaram and Sergiy Butenko Department of Industrial Engineering, Texas A&M University, College Station, Texas

More information

Peter J. Dukes. 22 August, 2012

Peter J. Dukes. 22 August, 2012 22 August, 22 Graph decomposition Let G and H be graphs on m n vertices. A decompostion of G into copies of H is a collection {H i } of subgraphs of G such that each H i = H, and every edge of G belongs

More information

A Level-2 Reformulation Linearization Technique Bound for the Quadratic Assignment Problem

A Level-2 Reformulation Linearization Technique Bound for the Quadratic Assignment Problem University of Pennsylvania ScholarlyCommons Operations, Information and Decisions Papers Wharton Faculty Research 8-2007 A Level-2 Reformulation Linearization Technique Bound for the Quadratic Assignment

More information

Quadratic Multiple Knapsack Problem with Setups and a Solution Approach

Quadratic Multiple Knapsack Problem with Setups and a Solution Approach Proceedings of the 2012 International Conference on Industrial Engineering and Operations Management Istanbul, Turkey, July 3 6, 2012 Quadratic Multiple Knapsack Problem with Setups and a Solution Approach

More information

Heuristics for The Whitehead Minimization Problem

Heuristics for The Whitehead Minimization Problem Heuristics for The Whitehead Minimization Problem R.M. Haralick, A.D. Miasnikov and A.G. Myasnikov November 11, 2004 Abstract In this paper we discuss several heuristic strategies which allow one to solve

More information

A MIXED INTEGER QUADRATIC PROGRAMMING MODEL FOR THE LOW AUTOCORRELATION BINARY SEQUENCE PROBLEM. Jozef Kratica

A MIXED INTEGER QUADRATIC PROGRAMMING MODEL FOR THE LOW AUTOCORRELATION BINARY SEQUENCE PROBLEM. Jozef Kratica Serdica J. Computing 6 (2012), 385 400 A MIXED INTEGER QUADRATIC PROGRAMMING MODEL FOR THE LOW AUTOCORRELATION BINARY SEQUENCE PROBLEM Jozef Kratica Abstract. In this paper the low autocorrelation binary

More information

A Dynamic Programming Heuristic for the Quadratic Knapsack Problem

A Dynamic Programming Heuristic for the Quadratic Knapsack Problem A Dynamic Programming Heuristic for the Quadratic Knapsack Problem Franklin Djeumou Fomeni Adam N. Letchford Published in INFORMS Journal on Computing February 2014 Abstract It is well known that the standard

More information

Scheduling of unit-length jobs with bipartite incompatibility graphs on four uniform machines arxiv: v1 [cs.

Scheduling of unit-length jobs with bipartite incompatibility graphs on four uniform machines arxiv: v1 [cs. Scheduling of unit-length jobs with bipartite incompatibility graphs on four uniform machines arxiv:1602.01867v1 [cs.ds] 4 Feb 2016 Hanna Furmańczyk, Marek Kubale Abstract In the paper we consider the

More information

CS246 Final Exam. March 16, :30AM - 11:30AM

CS246 Final Exam. March 16, :30AM - 11:30AM CS246 Final Exam March 16, 2016 8:30AM - 11:30AM Name : SUID : I acknowledge and accept the Stanford Honor Code. I have neither given nor received unpermitted help on this examination. (signed) Directions

More information

A Comparison of Evolutionary Approaches to the Shortest Common Supersequence Problem

A Comparison of Evolutionary Approaches to the Shortest Common Supersequence Problem A Comparison of Evolutionary Approaches to the Shortest Common Supersequence Problem Carlos Cotta Dept. Lenguajes y Ciencias de la Computación, ETSI Informática, University of Málaga, Campus de Teatinos,

More information

An Optimization-Based Heuristic for the Split Delivery Vehicle Routing Problem

An Optimization-Based Heuristic for the Split Delivery Vehicle Routing Problem An Optimization-Based Heuristic for the Split Delivery Vehicle Routing Problem Claudia Archetti (1) Martin W.P. Savelsbergh (2) M. Grazia Speranza (1) (1) University of Brescia, Department of Quantitative

More information

Classical Complexity and Fixed-Parameter Tractability of Simultaneous Consecutive Ones Submatrix & Editing Problems

Classical Complexity and Fixed-Parameter Tractability of Simultaneous Consecutive Ones Submatrix & Editing Problems Classical Complexity and Fixed-Parameter Tractability of Simultaneous Consecutive Ones Submatrix & Editing Problems Rani M. R, Mohith Jagalmohanan, R. Subashini Binary matrices having simultaneous consecutive

More information

Solving the maximum edge weight clique problem via unconstrained quadratic programming

Solving the maximum edge weight clique problem via unconstrained quadratic programming European Journal of Operational Research 181 (2007) 592 597 Discrete Optimization Solving the maximum edge weight clique problem via unconstrained quadratic programming Bahram Alidaee a, Fred Glover b,

More information

A tabu search based memetic algorithm for the maximum diversity problem

A tabu search based memetic algorithm for the maximum diversity problem A tabu search based memetic algorithm for the maximum diversity problem Yang Wang a,b, Jin-Kao Hao b,, Fred Glover c, Zhipeng Lü d a School of Management, Northwestern Polytechnical University, 127 Youyi

More information

A new ILS algorithm for parallel machine scheduling problems

A new ILS algorithm for parallel machine scheduling problems J Intell Manuf (2006) 17:609 619 DOI 10.1007/s10845-006-0032-2 A new ILS algorithm for parallel machine scheduling problems Lixin Tang Jiaxiang Luo Received: April 2005 / Accepted: January 2006 Springer

More information

Totally unimodular matrices. Introduction to integer programming III: Network Flow, Interval Scheduling, and Vehicle Routing Problems

Totally unimodular matrices. Introduction to integer programming III: Network Flow, Interval Scheduling, and Vehicle Routing Problems Totally unimodular matrices Introduction to integer programming III: Network Flow, Interval Scheduling, and Vehicle Routing Problems Martin Branda Charles University in Prague Faculty of Mathematics and

More information

GRASP in Switching Input Optimal Control Synthesis

GRASP in Switching Input Optimal Control Synthesis MIC 2001-4th Metaheuristics International Conference 381 GRASP in Switching Input Optimal Control Synthesis Paola Festa Giancarlo Raiconi Dept. of Mathematics and Computer Science, University of Salerno

More information

GRASP A speedy introduction

GRASP A speedy introduction GRASP A speedy introduction thst@man..dtu.dk DTU-Management Technical University of Denmark 1 GRASP GRASP is an abbreviation for Greedy Randomized Adaptive Search Procedure. It was invented by Feo and

More information

Complexity of local search for the p-median problem

Complexity of local search for the p-median problem Complexity of local search for the p-median problem Ekaterina Alekseeva, Yuri Kochetov, Alexander Plyasunov Sobolev Institute of Mathematics, Novosibirsk {ekaterina2, jkochet, apljas}@math.nsc.ru Abstract

More information

Standard Form An LP is in standard form when: All variables are non-negativenegative All constraints are equalities Putting an LP formulation into sta

Standard Form An LP is in standard form when: All variables are non-negativenegative All constraints are equalities Putting an LP formulation into sta Chapter 4 Linear Programming: The Simplex Method An Overview of the Simplex Method Standard Form Tableau Form Setting Up the Initial Simplex Tableau Improving the Solution Calculating the Next Tableau

More information

Solving the normal equations system arising from interior po. linear programming by iterative methods

Solving the normal equations system arising from interior po. linear programming by iterative methods Solving the normal equations system arising from interior point methods for linear programming by iterative methods Aurelio Oliveira - aurelio@ime.unicamp.br IMECC - UNICAMP April - 2015 Partners Interior

More information

An artificial chemical reaction optimization algorithm for. multiple-choice; knapsack problem.

An artificial chemical reaction optimization algorithm for. multiple-choice; knapsack problem. An artificial chemical reaction optimization algorithm for multiple-choice knapsack problem Tung Khac Truong 1,2, Kenli Li 1, Yuming Xu 1, Aijia Ouyang 1, and Xiaoyong Tang 1 1 College of Information Science

More information

Lin-Kernighan Heuristic. Simulated Annealing

Lin-Kernighan Heuristic. Simulated Annealing DM63 HEURISTICS FOR COMBINATORIAL OPTIMIZATION Lecture 6 Lin-Kernighan Heuristic. Simulated Annealing Marco Chiarandini Outline 1. Competition 2. Variable Depth Search 3. Simulated Annealing DM63 Heuristics

More information

Randomized Algorithms

Randomized Algorithms Randomized Algorithms Prof. Tapio Elomaa tapio.elomaa@tut.fi Course Basics A new 4 credit unit course Part of Theoretical Computer Science courses at the Department of Mathematics There will be 4 hours

More information

Flow Shop and Job Shop Models

Flow Shop and Job Shop Models Outline DM87 SCHEDULING, TIMETABLING AND ROUTING Lecture 11 Flow Shop and Job Shop Models 1. Flow Shop 2. Job Shop Marco Chiarandini DM87 Scheduling, Timetabling and Routing 2 Outline Resume Permutation

More information

Iterated Responsive Threshold Search for the Quadratic Multiple Knapsack Problem

Iterated Responsive Threshold Search for the Quadratic Multiple Knapsack Problem Noname manuscript No. (will be inserted by the editor) Iterated Responsive Threshold Search for the Quadratic Multiple Knapsack Problem Yuning Chen Jin-Kao Hao* Accepted to Annals of Operations Research

More information

Algorithms and Complexity theory

Algorithms and Complexity theory Algorithms and Complexity theory Thibaut Barthelemy Some slides kindly provided by Fabien Tricoire University of Vienna WS 2014 Outline 1 Algorithms Overview How to write an algorithm 2 Complexity theory

More information

Introduction to integer programming III:

Introduction to integer programming III: Introduction to integer programming III: Network Flow, Interval Scheduling, and Vehicle Routing Problems Martin Branda Charles University in Prague Faculty of Mathematics and Physics Department of Probability

More information

Efficient evaluations for solving large 0 1 unconstrained quadratic optimisation problems

Efficient evaluations for solving large 0 1 unconstrained quadratic optimisation problems Int. J. Metaheuristics, Vol. 1, No. 1, 2010 3 Efficient evaluations for solving large 0 1 unconstrained quadratic optimisation problems Fred Glover* OprTek Systems, Inc., 1919 Seventh Street, Boulder,

More information

TUTORIAL ON SURROGATE CONSTRAINT APPROACHES FOR OPTIMIZATION IN GRAPHS. Fred Glover

TUTORIAL ON SURROGATE CONSTRAINT APPROACHES FOR OPTIMIZATION IN GRAPHS. Fred Glover TUTORIAL ON SURROGATE CONSTRAINT APPROACHES FOR OPTIMIZATION IN GRAPHS Fred Glover Leeds School of Business University of Colorado Boulder, CO 80309-0419 December 2002 Abstract Surrogate constraint methods

More information

Random Lifts of Graphs

Random Lifts of Graphs 27th Brazilian Math Colloquium, July 09 Plan of this talk A brief introduction to the probabilistic method. A quick review of expander graphs and their spectrum. Lifts, random lifts and their properties.

More information

MICROCANONICAL OPTIMIZATION APPLIED TO THE TRAVELING SALESMAN PROBLEM

MICROCANONICAL OPTIMIZATION APPLIED TO THE TRAVELING SALESMAN PROBLEM International Journal of Modern Physics C, Vol. 9, No. 1 (1998) 133 146 c World Scientific Publishing Company MICROCANONICAL OPTIMIZATION APPLIED TO THE TRAVELING SALESMAN PROBLEM ALEXANDRE LINHARES Computação

More information

ACO Comprehensive Exam March 17 and 18, Computability, Complexity and Algorithms

ACO Comprehensive Exam March 17 and 18, Computability, Complexity and Algorithms 1. Computability, Complexity and Algorithms (a) Let G(V, E) be an undirected unweighted graph. Let C V be a vertex cover of G. Argue that V \ C is an independent set of G. (b) Minimum cardinality vertex

More information

The coordinated scheduling of steelmaking with multi-refining and tandem transportation

The coordinated scheduling of steelmaking with multi-refining and tandem transportation roceedings of the 17th World Congress The International Federation of Automatic Control The coordinated scheduling of steelmaking with multi-refining and tandem transportation Jing Guan*, Lixin Tang*,

More information

f-flip strategies for unconstrained binary quadratic programming

f-flip strategies for unconstrained binary quadratic programming Ann Oper Res (2016 238:651 657 DOI 10.1007/s10479-015-2076-1 NOTE f-flip strategies for unconstrained binary quadratic programming Fred Glover 1 Jin-Kao Hao 2,3 Published online: 11 December 2015 Springer

More information

Integer Programming Formulations for the Minimum Weighted Maximal Matching Problem

Integer Programming Formulations for the Minimum Weighted Maximal Matching Problem Optimization Letters manuscript No. (will be inserted by the editor) Integer Programming Formulations for the Minimum Weighted Maximal Matching Problem Z. Caner Taşkın Tınaz Ekim Received: date / Accepted:

More information

The Quadratic Assignment Problem

The Quadratic Assignment Problem The Quadratic Assignment Problem Joel L. O. ; Lalla V. ; Mbongo J. ; Ali M. M. ; Tuyishimire E.; Sawyerr B. A. Supervisor Suriyakat W. (Not in the picture) (MISG 2013) QAP 1 / 30 Introduction Introduction

More information

Numerical optimization

Numerical optimization THE UNIVERSITY OF WESTERN ONTARIO LONDON ONTARIO Paul Klein Office: SSC 408 Phone: 661-111 ext. 857 Email: paul.klein@uwo.ca URL: www.ssc.uwo.ca/economics/faculty/klein/ Numerical optimization In these

More information

A quadratic penalty algorithm for linear programming and its application to linearizations of quadratic assignment problems

A quadratic penalty algorithm for linear programming and its application to linearizations of quadratic assignment problems A quadratic penalty algorithm for linear programming and its application to linearizations of quadratic assignment problems I. L. Galabova J. A. J. Hall University of Edinburgh School of Mathematics and

More information

arxiv: v2 [cs.ds] 22 Feb 2012

arxiv: v2 [cs.ds] 22 Feb 2012 Simple heuristics for the assembly line worker assignment and balancing problem arxiv:1003.3676v2 [cs.ds] 22 Feb 2012 Mayron César O. Moreira, Alysson M. Costa Instituto de Ciências Matemáticas e de Computação,

More information

Overview of course. Introduction to Optimization, DIKU Monday 12 November David Pisinger

Overview of course. Introduction to Optimization, DIKU Monday 12 November David Pisinger Introduction to Optimization, DIKU 007-08 Monday November David Pisinger Lecture What is OR, linear models, standard form, slack form, simplex repetition, graphical interpretation, extreme points, basic

More information

Introduction into Vehicle Routing Problems and other basic mixed-integer problems

Introduction into Vehicle Routing Problems and other basic mixed-integer problems Introduction into Vehicle Routing Problems and other basic mixed-integer problems Martin Branda Charles University in Prague Faculty of Mathematics and Physics Department of Probability and Mathematical

More information

Lecture notes for Analysis of Algorithms : Markov decision processes

Lecture notes for Analysis of Algorithms : Markov decision processes Lecture notes for Analysis of Algorithms : Markov decision processes Lecturer: Thomas Dueholm Hansen June 6, 013 Abstract We give an introduction to infinite-horizon Markov decision processes (MDPs) with

More information

Beta Damping Quantum Behaved Particle Swarm Optimization

Beta Damping Quantum Behaved Particle Swarm Optimization Beta Damping Quantum Behaved Particle Swarm Optimization Tarek M. Elbarbary, Hesham A. Hefny, Atef abel Moneim Institute of Statistical Studies and Research, Cairo University, Giza, Egypt tareqbarbary@yahoo.com,

More information

A hybrid heuristic for minimizing weighted carry-over effects in round robin tournaments

A hybrid heuristic for minimizing weighted carry-over effects in round robin tournaments A hybrid heuristic for minimizing weighted carry-over effects Allison C. B. Guedes Celso C. Ribeiro Summary Optimization problems in sports Preliminary definitions The carry-over effects minimization problem

More information

Variable Neighborhood Decomposition Search

Variable Neighborhood Decomposition Search Journal of Heuristics, 7: 335 350, 2001 c 2001 Kluwer Academic Publishers Variable Neighborhood Decomposition Search PIERRE HANSEN AND NENAD MLADENOVIĆ Gerad, Ecole des Hautes Etudes Commerciales, Montreal,

More information

Large-Scale 3D En-Route Conflict Resolution

Large-Scale 3D En-Route Conflict Resolution Large-Scale 3D En-Route Conflict Resolution Cyril Allignol, Nicolas Barnier, Nicolas Durand, Alexandre Gondran and Ruixin Wang allignol,barnier,durand,gondran,wangrx @recherche.enac.fr ATM 2017 Seattle

More information

Aditya Bhaskara CS 5968/6968, Lecture 1: Introduction and Review 12 January 2016

Aditya Bhaskara CS 5968/6968, Lecture 1: Introduction and Review 12 January 2016 Lecture 1: Introduction and Review We begin with a short introduction to the course, and logistics. We then survey some basics about approximation algorithms and probability. We also introduce some of

More information

Available online at ScienceDirect. Procedia Technology 25 (2016 )

Available online at   ScienceDirect. Procedia Technology 25 (2016 ) Available online at www.sciencedirect.com ScienceDirect Procedia Technology 25 (2016 ) 998 1005 Global Colloquium in Recent Advancement and Effectual Researches in Engineering, Science and Technology (RAEREST

More information

Packing triangles in regular tournaments

Packing triangles in regular tournaments Packing triangles in regular tournaments Raphael Yuster Abstract We prove that a regular tournament with n vertices has more than n2 11.5 (1 o(1)) pairwise arc-disjoint directed triangles. On the other

More information

Worst-case design of structures using stopping rules in k-adaptive random sampling approach

Worst-case design of structures using stopping rules in k-adaptive random sampling approach 10 th World Congress on Structural and Multidisciplinary Optimization May 19-4, 013, Orlando, Florida, USA Worst-case design of structures using stopping rules in k-adaptive random sampling approach Makoto

More information

Worst case analysis for a general class of on-line lot-sizing heuristics

Worst case analysis for a general class of on-line lot-sizing heuristics Worst case analysis for a general class of on-line lot-sizing heuristics Wilco van den Heuvel a, Albert P.M. Wagelmans a a Econometric Institute and Erasmus Research Institute of Management, Erasmus University

More information

Research Article An Auxiliary Function Method for Global Minimization in Integer Programming

Research Article An Auxiliary Function Method for Global Minimization in Integer Programming Mathematical Problems in Engineering Volume 2011, Article ID 402437, 13 pages doi:10.1155/2011/402437 Research Article An Auxiliary Function Method for Global Minimization in Integer Programming Hongwei

More information

Linear-Time Algorithms for Finding Tucker Submatrices and Lekkerkerker-Boland Subgraphs

Linear-Time Algorithms for Finding Tucker Submatrices and Lekkerkerker-Boland Subgraphs Linear-Time Algorithms for Finding Tucker Submatrices and Lekkerkerker-Boland Subgraphs Nathan Lindzey, Ross M. McConnell Colorado State University, Fort Collins CO 80521, USA Abstract. Tucker characterized

More information

COMPARISON OF ITERATIVE SEARCHES FOR THE QUADRATIC ASSIGNMENT PROBLEM

COMPARISON OF ITERATIVE SEARCHES FOR THE QUADRATIC ASSIGNMENT PROBLEM To appear in Location Science 3. CRT-989 COMPARISON OF ITERATIVE SEARCHES FOR THE QUADRATIC ASSIGNMENT PROBLEM Centre de recherche sur les transports Université de Montréal C. P. 6128, succursale Centre-Ville

More information

Combining Memory and Landmarks with Predictive State Representations

Combining Memory and Landmarks with Predictive State Representations Combining Memory and Landmarks with Predictive State Representations Michael R. James and Britton Wolfe and Satinder Singh Computer Science and Engineering University of Michigan {mrjames, bdwolfe, baveja}@umich.edu

More information

NILS: a Neutrality-based Iterated Local Search and its application to Flowshop Scheduling

NILS: a Neutrality-based Iterated Local Search and its application to Flowshop Scheduling NILS: a Neutrality-based Iterated Local Search and its application to Flowshop Scheduling Marie-Eleonore Marmion, Clarisse Dhaenens, Laetitia Jourdan, Arnaud Liefooghe, Sébastien Verel To cite this version:

More information

Mathematics for Decision Making: An Introduction. Lecture 8

Mathematics for Decision Making: An Introduction. Lecture 8 Mathematics for Decision Making: An Introduction Lecture 8 Matthias Köppe UC Davis, Mathematics January 29, 2009 8 1 Shortest Paths and Feasible Potentials Feasible Potentials Suppose for all v V, there

More information

The Tactical Berth Allocation Problem with Quay-Crane Assignment and Transshipment Quadratic Costs

The Tactical Berth Allocation Problem with Quay-Crane Assignment and Transshipment Quadratic Costs The Tactical Berth Allocation Problem with Quay-Crane Assignment and Transshipment Quadratic Costs Models and Heuristics Ilaria Vacca Transport and Mobility Laboratory, EPFL joint work with Matteo Salani,

More information

National Center for Geographic Information and Analysis. A Comparison of Strategies for Data Storage Reduction in Location-Allocation Problems

National Center for Geographic Information and Analysis. A Comparison of Strategies for Data Storage Reduction in Location-Allocation Problems National Center for Geographic Information and Analysis A Comparison of Strategies for Data Storage Reduction in Location-Allocation Problems by Paul A. Sorensen Richard L. Church University of California,

More information

An Exact Algorithm for the Steiner Tree Problem with Delays

An Exact Algorithm for the Steiner Tree Problem with Delays Electronic Notes in Discrete Mathematics 36 (2010) 223 230 www.elsevier.com/locate/endm An Exact Algorithm for the Steiner Tree Problem with Delays Valeria Leggieri 1 Dipartimento di Matematica, Università

More information