Disjunctive Programming

Transcription

1 Chapter 10 Disjunctive Programming Egon Balas Introduction by Egon Balas In April 1967 I and my family arrived into the US as fresh immigrants from behind the Iron Curtain. After a fruitful semester spent with George Dantzig s group in Stanford, I started working at CMU. My debut in integer programming and entry ticket into Academia was the additive algorithm for 0-1 programming [B65], an implicit enumeration procedure based on logical tests akin to what today goes under the name of constraint propagation. As it used only additions and comparisons, it was easy to implement and was highly popular for a while. However, I was aware of its limitations and soon after I joined CMU I started investigating cutting plane procedures, trying to use for this purpose the tools of convex analysis: support functions and their level sets, maximal convex extensions, polarity, etc. During the five years starting in 1969, I proposed a number of procedures based on the central idea of intersection cuts [2] (numbered references are to those at the end of the paper, whereas mnemonicized ones are to the ones listed at the end of this introduction): Given any convex set S containing the LP optimum of a mixed integer program (MIP) but containing no feasible integer point in its interior, one can generate a valid cut by intersecting the boundary of S with the extreme rays of the cone defined by the optimal solution to the linear programming relaxation of the MIP and taking the hyperplane defined by the intersection points as the cut. The search for the most appropriate sets S in this role has led to the concept of outer polars and related constructs [3, 6]. In our days, the idea of intersection cuts has been revived in the form of cutting planes from convex sets with lattice-free interiors, and is the object of numerous investigations (e.g., [ALWW07], [BC07], [CM07], [DW07]). It was this line of research that has led to the idea of disjunctive programming, through a process outlined in section 1 of the paper below. Optimizing a function Egon Balas Carnegie Mellon University, Pittsburgh, USA eb17@andrew.cmu.edu M. Jünger et al. (eds.), 50 Years of Integer Programming , DOI / _10, Springer-Verlag Berlin Heidelberg

2 284 Egon Balas subject to a set of linear inequalities connected by conjunction or disjunction is a special type of nonconvex programming problem called disjunctive programming. Mixed 0-1 programming is its most important special case. More broadly speaking, disjunctive programming is optimization over a union of polyhedra. The basic document on disjunctive programming is the July 1974 technical report Disjunctive Programming: Properties of the convex hull of feasible points, MSRR #348, referenced as [10] in the survey below, which however has not appeared in print until 24 years later, when it was published as an invited paper with a preface by Gérard Cornuéjols and Bill Pulleyblank [B98]. The reasons for this situation are complex. The 1974 paper was not rejected, but the report I received at the end of a very long refereeing process was asking for a revision that would have involved a major rewriting effort at a time when I was engaged in other, complementary research projects. As a result, when shortly thereafter I was invited by the late Peter Hammer to prepare an introductory survey of disjunctive programming for the upcoming 1977 ARIDAM conference in Vancouver, I decided to incorporate into that survey the main results of MSSR #348 and forego its publication as a stand alone paper. This is how this survey came about. Looking back on that decision, I find that not much of substance was lost. Sections 2 6 of this survey adequately summarize the main findings of the 1974 paper, with one important exception, which I will now explain. MSRR #348 explored the basic properties of disjunctive programs from a polyhedral point of view. It gave two compact characterizations of the convex hull of a union P of polyhedra P i = {x : A i x b i }, i Q, in a higher dimensional space, a procedure we call today extended formulation. The first one describes convp as the set of points x satisfying x = (x i : i Q) for some (x i,x0 i ), i Q, such that Ai x i b i x0 i,i Q, (x0 i : i Q) = 1, xi 0 0,i Q. The second one describes the facets of convp as the vertices of the reverse polar of P, defined as P # := {y : yx 1 x P}, shown to be the set of points y satisfying y u i A i,i Q for some u i 0 such that u i b i 1, i Q. Both characterizations are linear in Q, the number of polyhedra in the union. In a certain sense the two characterizations are equivalent: projecting the higher-dimensional polyhedron of the convex hull characterization onto the x-space yields the set of all valid inequalities and conversely, taking the reverse polar of the reverse polar yields the convex hull. Thus in the survey paper I only included the second characterization, so that the first one did not appear in print until 1985 [B85]. But it is precisely this first characterization which has served as a prototype for the many extended formulations that have proved to be such prolific tools for polyhedral analysis of combinatorial problems, starting with the early studies of this type in the 1980 s [BP83], [BP89], [BLP89], and continuing with a plethora of results, the more recent ones being exemplified by [PW06], [A06], [CW08]. It is also this first characterization that was generalized to nonlinear disjunctive programming [SM99] and was extensively used in the modeling of a variety of practical situations in industry. Habent sua fata libelli, goes the Latin saying: books have their own fate. This apparently also applies to papers or theorems or discoveries. While the work on disjunctive programming, including the cutting planes that it entailed, stirred little if any enthusiasm at the time of its inception, about 15 years later when Sebas-

3 10 Disjunctive Programming 285 tian Ceria, Gérard Cornuéjols and myself recast essentially the same results in a new framework which we called lift-and-project [BCC93], the reaction was quite different. This time our work was focused on algorithmic aspects, with the cutting planes generated in rounds and embedded into an enumerative framework (branch and cut), and was accompanied by the development of an efficient computer code (MIPO, developed by Sebastian) that was able to solve many problem instances that had been impervious to solution by branch-and-bound alone. Our interest in returning to the ideas of disjunctive programming was prompted by the exciting work of Lovász and Schrijver on matrix cones [LS91]. We discovered that a streamlined version of the Lovász-Schrijver procedure was isomorphic to the disjunctive programming procedure for generating the integer hull of a 0-1 program by starting with the higher dimensional representation and projecting it onto the original space. Thus the Lovász-Schrijver Theorem according to which n applications of this procedure (n being the number of 0-1 variables) yields the integer hull, follows directly from the sequential convexification procedure for facial disjunctive programs, of which 0-1 programs are a prime example (see Section 6 below). The reader will no doubt recognize in the linear program P 1 (g,α 0) preceding Theorem 4.4 of section 4 below, the ancestor of the cut generating linear program of the lift-and-project (L&P) algorithm [BCC93], which is the specialization of P 1 (g,α 0) to the case of the disjunction x k = 0 or x k = 1. The computational success of L&P cuts triggered a strong revival of interest in cutting planes. Gérard and Sebastian soon discovered [BCCN96] that mixed integer Gomory (MIG) cuts, when used in the MIPO fashion, i.e., generated in rounds and embedded into a branch-and-bound framework, could also solve many of the problem instances unsolved at the time. Since the MIG cuts were easier to implement than the L&P ones, they were the first to find their way into the commercial codes. The combination of cutting planes with branch and bound played a central role in the revolution in the state of the art in integer programming that started in the mid-90s. The commercial implementation of lift-and-project cuts had to await the discovery of a method [BP03], [P03] for generating them directly from the LP simplex tableau, without explicit recourse to the higher dimensional cut generating linear program. Today, due to the efforts of Pierre Bonami [BB07], an open-source implementation is also publicly available [COIN-OR]. References [A06] A. Atamtürk, Strong formulations of robust mixed 0-1 programming, Mathematical Programming 108 (2006) [ALWW07] K. Andersen, Q. Louveaux, R. Weismantel and L.A. Wolsey, Inequalities from two rows of the simplex tableau, Integer Programming and Combinatorial Optimization IPCO 12 (M. Fischetti and D.P. Williamson, eds.), Springer, 2007, pp [B65] E. Balas, An additive algorithm for solving linear programs in 0-1 variables, Operations Research 13 (1965) [B85] E. Balas, Disjunctive programming and a hierarchy of relaxations for discrete optimization problems, SIAM Journal on Algebraic and Discrete Methods 6 (1985)

4 286 Egon Balas [B98] E. Balas, Disjunctive programming: Properties of the convex hull of feasible points, Invited paper with a Foreword by G. Cornuéjols and G. Pulleyblank, Discrete Applied Mathematics 89 (1998) [BB07] E. Balas and P. Bonami, New variants of lift-and-project cut generation from the LP tableau: Open source implementation and testing, Integer Programming and Combinatorial Oprtimization IPCO 12 (M. Fischetti and D.P. Williamson, eds.), Springer, 2007, pp [BC07] V. Borozan and G. Cornuéjols, Minimal inequalities for integer constraints, Technical Report, Tepper School, Carnegie Mellon University, [BCC93] E. Balas, S. Ceria and G. Cornuéjols, A lift-and-project cutting plane algorithm for mixed 0-1 programs, Mathematical Programming 58 (1993) [BCCN96] E. Balas, S. Ceria, G. Cornuéjols and N. Natraj, Gomory cuts revisited, Operations Research Letters 19 (1996) [BLP89] M. Ball, W. Liu and W.R. Pulleyblank, Two-terminal Steiner tree polyhedra, Contributions to Operations Research and Economics, MIT Press, 1989, pp [BP83] E. Balas and W.R. Pulleyblank, The perfectly matchable subgraph polytope of a bipartite graph, Networks 13 (1983) [BP89] E. Balas and W.R. Pulleyblank, The perfectly matchable subgraph polytope of an arbitrary graph, Combinatorica 9 (1989) [BP03] E. Balas and M. Perregaard, A precise correspondence between lift-and-project cuts, simple disjunctive cuts, and mixed integer Gomory cuts for 0-1 programming, Mathematical Programming 94 (2003) [CM07] G. Cornuéjols and F. Margot, On the facets of mixed integer programs with two integer variables and two constraints, Technical Report, Tepper School, Carnegie Mellon University, [COIN-OR] [CW08] M. Conforti and L.A. Wolsey, Compact formulations as a union of polyhedra, Mathematical Programming 114 (2008) [DW07] S.S. Dey and L.A. Wolsey, Lifting integer variables in minimal inequalities corresponding to lattice-free triangles, Integer Programming and Combinatorial Optimization IPCO 13 (A. Lodi, A. Panconesi, and G. Rinaldi, eds.), Springer, 2008, pp [LS91] L. Lovász and A. Schrijver, Cones of matrices and set functions and 0-1 optimization, SIAM Journal of Optimization 1 (1991) [P03] M. Perregaard, A Practical implementation of lift-and-project cuts: a computational exploration of lift-and-project with XPRESS-MP, International Symposium on Mathematical Programming, Copenhagen, August [PW06] Y. Pochet and L.A. Wolsey, Production Planning by Mixed Integer Programming, Springer, [SM99] R. Stubbs and S. Mehrotra, A branch and cut method for 0-1 mixed integer convex programming, Mathematical Programming 86 (1999)

5 10 Disjunctive Programming 287

6 288 Egon Balas The following article originally appeared as: E. Balas, Disjunctive Programming, Discrete Optimization II (P.L. Hammer, E.L. Johnson, and B.H. Korte, eds.), Annals of Discrete Mathematics 5 (1979) Copyright c 1979 North-Holland Publishing Company. Reprinted by permission from Elsevier.

7 10 Disjunctive Programming 289

8 290 Egon Balas

9 10 Disjunctive Programming 291

10 292 Egon Balas

11 10 Disjunctive Programming 293

12 294 Egon Balas

13 10 Disjunctive Programming 295

14 296 Egon Balas

15 10 Disjunctive Programming 297

16 298 Egon Balas

17 10 Disjunctive Programming 299

18 300 Egon Balas

19 10 Disjunctive Programming 301

20 302 Egon Balas

21 10 Disjunctive Programming 303

22 304 Egon Balas

23 10 Disjunctive Programming 305

24 306 Egon Balas

25 10 Disjunctive Programming 307

26 308 Egon Balas

27 10 Disjunctive Programming 309

28 310 Egon Balas

29 10 Disjunctive Programming 311

30 312 Egon Balas

31 10 Disjunctive Programming 313

32 314 Egon Balas

33 10 Disjunctive Programming 315

34 316 Egon Balas

35 10 Disjunctive Programming 317

36 318 Egon Balas

37 10 Disjunctive Programming 319

38 320 Egon Balas

39 10 Disjunctive Programming 321

40 322 Egon Balas

41 10 Disjunctive Programming 323

42 324 Egon Balas

43 10 Disjunctive Programming 325

44 326 Egon Balas

45 10 Disjunctive Programming 327

46 328 Egon Balas

47 10 Disjunctive Programming 329

48 330 Egon Balas

49 10 Disjunctive Programming 331

50 332 Egon Balas

51 10 Disjunctive Programming 333

52 334 Egon Balas

53 10 Disjunctive Programming 335

54 336 Egon Balas

55 10 Disjunctive Programming 337

56 338 Egon Balas The following article originally appeared as: E. Balas, Erratum to: Disjunctive Programming, Discrete Applied Mathematics 5 (1983) Copyright c 1983 North-Holland Publishing Company. Reprinted by permission from Elsevier.

57 10 Disjunctive Programming 339

58 340 Egon Balas