Abstract
We introduce the conditional p-dispersion problem (c-pDP), an incremental variant of the p-dispersion problem (pDP). In the c-pDP, one is given a set N of n points, a symmetric dissimilarity matrix D of dimensions \(n\times n\), an integer \(p\ge 1\) and a set \(Q\subseteq N\) of cardinality \(q\ge 1\). The objective is to select a set \(P\subset N\setminus Q\) of cardinality p that maximizes the minimal dissimilarity between every pair of selected vertices, i.e., \(z(P\cup Q) {:}{=}\min \{D(i, j), i, j\in P\cup Q\}\). The set Q may model a predefined subset of preferences or hard location constraints in incremental network design. We adapt the state-of-the-art algorithm for the pDP to the c-pDP and include an ad-hoc acceleration mechanism designed to leverage the information provided by the set Q to further reduce the size of the problem instance. We perform exhaustive computational experiments and show that the proposed acceleration mechanism helps reduce the total computational time by a factor of five on average. We also assess the scalability of the algorithm and derive sensitivity analyses.
Similar content being viewed by others
References
Aloise, D., Contardo, C.: A sampling-based exact algorithm for the solution of the minimax diameter clustering problem. J. Glob. Optim. 71, 613–630 (2018)
Belotti, P., Labbé, M., Maffioli, F., Ndiaye, M.M.: A branch-and-cut method for the obnoxious p-median problem. 4OR 5(4), 299–314 (2006)
Berman, O., Drezner, Z.: A new formulation for the conditional p-median and p-center problems. Oper. Res. Lett. 36(4), 481–483 (2008)
Berman, O., Simchi-Levi, D.: Conditional location problems on networks. Transp. Sci. 24(1), 77–78 (1990)
Boland, N., Dethridge, J., Dumitrescu, I.: Accelerated label setting algorithms for the elementary resource constrained shortest path problem. Oper. Res. Lett. 34(1), 58–68 (2006)
Calik, H., Labbé, M., Yaman, H.: p-center problems. In: Location Science, pp. 79–92. Springer, Berlin (2015)
Callaghan, B., Salhi, S., Brimberg, J.: Optimal solutions for the continuous p-centre problem and related-neighbour and conditional problems: a relaxation-based algorithm. J. Oper. Res. Soc. 70(2), 192–211 (2019)
Chen, D., Chen, R.: New relaxation-based algorithms for the optimal solution of the continuous and discrete p-center problems. Comput. Oper. Res. 36(5), 1646–1655 (2009)
Chen, D., Chen, R.: A relaxation-based algorithm for solving the conditional p-center problem. Oper. Res. Lett. 38(3), 215–217 (2010)
Chen, R., Handler, G.Y.: Relaxation method for the solution of the minimax location-allocation problem in euclidean space. Nav. Res. Logist. (NRL) 34(6), 775–788 (1987)
Chen, R., Handler, Y.: The conditional p-center problem in the plane. Nav. Res. Logist. (NRL) 40(1), 117–127 (1993)
Chvatal, V.: Linear Programming. Series of Books in the Mathematical Sciences. W. H. Freeman, New York (1983)
Contardo, C.: Decremental clustering for the solution of p-dispersion problems to proven optimality. techreport G-2019-22, Cahiers du GERAD (2019)
Contardo, C., Sefair, J.A.: A progressive approximation approach for the exact solution of sparse large-scale binary interdiction games. techreport G-2019-49, Cahiers du GERAD (2019)
Contardo, C., Iori, M., Kramer, R.: A scalable exact algorithm for the vertex p-center problem. Comput. Oper. Res. 103, 211–220 (2019)
Daskin, M.S., Maass, K.L.: The p-median problem. In: Location Science, pp. 21–45. Springer, Berlin (2015)
Delattre, M., Hansen, P.: Bicriterion cluster analysis. IEEE Trans. Pattern Anal. Mach. Intell. 2(4), 277–291 (1980)
Downey, R.G., Fellows, M.R.: Parameterized Complexity. Springer, Berlin (2012)
Drezner, T., Drezner, Z.: Sequential location of two facilities: comparing random to optimal location of the first facility. Ann. Oper. Res. 246(1–2), 5–18 (2016)
Drezner, Z.: Conditional p-center problems. Transp. Sci. 23(1), 51–53 (1989)
Drezner, Z.: On the conditional p-median problem. Comput. Oper. Res. 22(5), 525–530 (1995)
Drezner, Z., Erkut, E.: Solving the continuous p-dispersion problem using non-linear programming. J. Oper. Res. Soc. 46(4), 516–520 (1995)
Drezner, Z., Kalczynski, P., Salhi, S.: The planar multiple obnoxious facilities location problem: a voronoi based heuristic. Omega 87, 105–116 (2019)
Erkut, E.: The discrete p-dispersion problem. Eur. J. Oper. Res. 46(1), 48–60 (1990)
Gauthier, J.B., Desrosiers, J., Lübbecke, M.E.: About the minimum mean cycle-canceling algorithm. Discret. Appl. Math. 196, 115–134 (2015)
Hooker, J.N., Garfinkel, R.S., Chen, C.: Finite dominating sets for network location problems. Oper. Res. 39(1), 100–118 (1991)
Irawan, C.A., Salhi, S., Scaparra, M.P.: An adaptive multiphase approach for large unconditional and conditional p-median problems. Eur. J. Oper. Res. 237(2), 590–605 (2014)
Irnich, S., Desaulniers, G., Desrosiers, J., Hadjar, A.: Path-reduced costs for eliminating arcs in routing and scheduling. INFORMS J. Comput. 22(2), 297–313 (2010)
Kuby, M.J.: Programming models for facility dispersion: The p-dispersion and maxisum dispersion problems. Geogr. Anal. 19(4), 315–329 (1987)
Laporte, G., Nickel, S., da Gama, F.S.: Location Science. Springer, Berlin (2015)
Laporte, G., Nickel, S., da Gama, F.S.: Location science. Springer, Berlin (2019)
Martinelli, R., Pecin, D., Poggi, M.: Efficient elementary and restricted non-elementary route pricing. Eur. J. Oper. Res. 239(1), 102–111 (2014)
Pisinger, D.: Upper bounds and exact algorithms for p-dispersion problems. Comput. Oper. Res. 33(5), 1380–1398 (2006)
Reinelt, G.: TSPLIB-A traveling salesman problem library. INFORMS J. Comput. 3(4), 376–384 (1991). https://doi.org/10.1287/ijoc.3.4.376
Righini, G., Salani, M.: New dynamic programming algorithms for the resource constrained elementary shortest path problem. Netw. Int. J. 51(3), 155–170 (2008)
Saboonchi, B., Hansen, P., Perron, S.: Maxminmin p-dispersion problem: a variable neighborhood search approach. Comput. Oper. Res. 52, 251–259 (2014)
Sayah, D., Irnich, S.: A new compact formulation for the discrete p-dispersion problem. Eur. J. Oper. Res. 256(1), 62–67 (2017)
Statman, M.: How many stocks make a diversified portfolio? J. Financ. Quant. Anal. 22(3), 353–363 (1987). https://doi.org/10.2307/2330969
Acknowledgements
The authors thank the Canadian Natural Sciences and Engineering Research Council (NSERC) under Discovery Grants 2017-06106 and 2020-06311 for its financial support. We also thank the GERAD for providing access to its computing infrastructure. Finally, we thank the referees for their valuable comments.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
A Impact of the choice of construction strategy for the set Q
In this section, we present three construction strategies (greedy, optimal and random) for the initial set Q. The greedy strategy consists of selecting the set Q in an iterative manner such that, at every iteration, one vertex is added to the solution according to its impact on the objective function. The optimal strategy consists of solving a \(\texttt {pDP}{}\) by setting \(p = q\) using the algorithm proposed by Contardo [13]. Note that we use one of the optimal solutions even though there might be alternative optimal solutions if there is a lot of symmetry. The third strategy consists of selecting the set Q randomly.
Table 8 presents summarized results for the construction strategies of the set Q. For each construction strategy (greedy, optimal and random), we report the average deviation in percentage from the best known solution computed as (\(z^* - z^*_{best}\))/\(z^*_{best}\), where \(z^*\) is the optimal solution found according to a given initial set Q and \(z^*_{best}\) is the best solution for a given instance independently on how Q is generated (Average \(\Delta \) \(z^*\) (%)), the worst deviation in percentage from the best known solution (Worst (%)), the number of best known solutions (# Best), the average computational time in seconds (Average CPU time (s)), the number of instances solved to optimality within the prescribed time limit (# Solved), and the average percentage of eliminated vertices with our graph reduction technique (Average elim. (%)). In addition, Figs. 14, 15 and 16 present the average time, the average deviation with respect to the best known solution value, and the average percentage of eliminated vertices according to the construction strategy as well as the number of instances, respectively.
By analyzing the results, we can note that randomly constructing the set Q yields the worst solutions (\(-\) 74.4% on average compared to less than \(-\) 2% on average for the two other strategies). In addition, the total computational time is the slowest which is due to the symmetry between the solutions because the value of the initial lower bound is the worst one. By comparing the greedy and the optimal construction strategies, one can realize that the greedy construction strategy yields better results for our set of instances. In fact, it yields the best average deviation from the best known optimal solution (\(-\) 1.4% on average compared to \(-\) 1.9% on average with the optimal construction) and its worst solution is better than the worst solutions with the two other construction strategies. In addition, this strategy has the lowest computational time and the highest number of instances solved to optimality within the prescribed time limit. Finally, our graph reduction technique performs best with this strategy. With our set of instances, greedily constructing the set Q outperforms both the optimal construction and the random construction as it yields the best solutions and has the lowest computational time. In practice, obtaining the set Q with a greedy construction algorithm is much faster than obtaining the set Q by solving to optimality the pDP. Therefore, it seems more practical the generate the set Q greedily.
B Detailed results
In this section, we presented detailed results. Tables 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23 and 24 present the detailed results for the values of \(p = \{5, 10, 15, 20\}\) and \(q = \{5, 20, 15, 20\}\). In each table, the first column contains the name of the instance (Instance). Then, for each construction strategy for the set Q, i.e., greedy construction, optimal construction, and random construction, we report the total computational time in seconds (Sec), the optimal solution value (\(z^*\)), and the number of eliminated vertices with our graph reduction technique (Elim). If no solution was found within the prescribed time limit, the cells are left blank.
C Detailed computational time with the naive algorithm
In this section, we present the total computational time for the naive and the ad hoc algorithms. Tables 25, 26, 27 and 28 present the detailed results for the values of \(p = \{5, 10, 15, 20\}\), respectively. In each table, the first column contains the name of the instance (Instance). Then, for each value of q, i.e., \(q = \{5, 10, 15, 20\}\), we report the total computational time in seconds used with the proposed ad hoc algorithm (ad hoc) and with the naive algorithm (naive). If no optimal solution was found within the prescribed time limit, the cells are left blank.
D Detailed computational time with increasing values of q and p
In this section, we present the total computational time for the results obtained with increasing values of q and p, that is \(q, p \in \{10, 20, 40, 80\}\), and for the instances with \(|N|\le 10{,}000\). Tables 29, 30, 31 and 32 present the detailed results for the values of \(p = \{10, 20, 40, 80\}\), respectively. In each table, the first column contains the name of the instance (Instance). Then, for each value of q, i.e., \(q = \{10, 20, 40, 80\}\), we report the total computational time in seconds. If no optimal solution was found within the prescribed time limit, the cells are left blank.
Rights and permissions
About this article
Cite this article
Cherkesly, M., Contardo, C. The conditional p-dispersion problem. J Glob Optim 81, 23–83 (2021). https://doi.org/10.1007/s10898-020-00962-4
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10898-020-00962-4