A Binarisation Approach to Non-Convex Quadratically Constrained Quadratic Programs
|
|
- Denis Underwood
- 5 years ago
- Views:
Transcription
1 A Binarisation Approach to Non-Convex Quadratically Constrained Quadratic Programs Laura Galli Adam N. Letchford February 2015 Abstract The global optimisation of non-convex quadratically constrained quadratic programs is a notoriously difficult problem, being not only N P-hard in the strong sense, but also very difficult in practice. We present a new heuristic approach to this problem, which enables one to obtain solutions of good quality in reasonable computing times. The heuristic consists of four phases: binarisation, convexification, branchand-bound and local optimisation. Computational results, on boxconstrained and point packing instances, are encouraging. Keywords: quadratically constrained quadratic programming, mixedinteger nonlinear programming, heuristics. 1 Introduction A quadratically constrained quadratic program (QCQP) is an optimisation problem that can be written in the following form: inf x T Q 0 x + c 0 x (1) s.t. x T Q k x + c k x h k (k = 1,..., m) (2) x R n, (3) where the Q k are symmetric matrices of order n, the c k are n-vectors and the h k are scalars. (We write inf rather than min because it is possible that the infimum is not attainable.) If all of the Q k are positive semidefinite (psd) matrices, then QCQPs are convex optimization problems, and therefore can be solved efficiently (see, e.g., [10]). Non-convex QCQPs, on the other hand, are N P-hard. In Dipartimento di Informatica, Università di Pisa, Largo B. Pontecorvo 3, Pisa, Italy. laura.galli@unipi.it Department of Management Science, Lancaster University, Lancaster LA1 4YX, United Kingdom. A.N.Letchford@lancaster.ac.uk 1
2 fact, just checking unboundedness of a quadratic form over the non-negative orthant is N P-complete in the strong sense [28]. From now on, when we say QCQPs, we mean non-convex ones, unless otherwise stated. Solving QCQPs to proven optimality is very hard in practice as well in theory (e.g., [4, 5, 6, 23, 27]). One obvious source of this difficulty is the possibility of an exponentially large number of local minima. Another, less obvious, source of difficulty is that variables can take irrational values at the optimum, which means that one must either be content with solutions that are optimal only up to some given accuracy, or somehow represent solutions implicitly (for example, via the KKT conditions, as in [11]). On the other hand, QCQPs are a powerful modelling tool. A folklore result (see, e.g., [22, 29]) is that mixed 0-1 linear and quadratic programs can be modelled as QCQPs, since the condition x i {0, 1} is equivalent to the quadratic constraint x i = x 2 i. Also, Shor [36] pointed out that any optimisation problem in which the objective and constraint functions are polynomials can be transformed to a QCQP, by adding suitable variables and constraints. (For example, one can replace a cubic term x i x j x k with the quadratic term x i x jk, where x jk is a new variable, if one also adds the quadratic equation x jk = x j x j.) Several other applications of QCQPs can be found in, e.g., [2, 14, 24, 30]. In this paper, we present a new heuristic for QCQPs, which is aimed mainly at loosely-constrained QCQPs, where it is fairly easy to find a feasible solution. The heuristic exploits the fact that there now exist very good software packages for solving convex QCQPs with binary variables. It consists of four phases: binarisation, convexification, branch-and-bound and local optimisation. In our computational experiments, on standard test instances, the heuristic typically found solutions of very good quality in reasonable computing times. The structure of the paper is as follows. The literature is reviewed in Section 2, the heuristic is described in Section 3, the computational results are given in Section 4, and some concluding remarks are made in Section 5. Remark: The idea of a box will appear several times in the paper. For us, it means a region of the form {x R n : l x u}, where l {R { }} n, u {R { }} n and l < u. 2 Literature Review We now review some of the literature on QCQPs. Since the literature is vast, we only review works of direct relevance. We refer the reader to Section 5 of [13] for a more detailed survey. 2
3 2.1 LP-based approaches Like any other global optimisation problem, QCQPs can be solved (to any fixed precision) using the spatial branch-and-bound method of McCormick [26]. Suppose that each variable is bounded, i.e., l i x i u i for all i. Then, we can construct a linear programming (LP) relaxation in the following way. For each pair i, j for which the quadratic term x i x j appears in the problem, substitute the term with a new variable X ij, and add the constraints X ij l i x j + l j x i l i l j, X ij u i x j + u j x i u i u j X ij l i x j + u j x i l i u j, X ij l j x i + u i x j l j u i. This LP relaxation can then be embedded in a branch-and-bound scheme, in which branching is performed by partitioning the domain of an x variable into two intervals. To our knowledge, the only pure spatial branch-andbound algorithm for QCQPs was that of Al-Khayyal et al. [2]. An enhanced version of spatial branch-and-bound, called branch-andreduce, uses feasibility and optimality arguments to reduce variable domains (Tawarmalani & Sahinidis [38]). Raber [30] presented a branch-and-reduce algorithm for QCQPs. He used a non-standard branching rule, in which the solution space is partitioned into simplices rather than boxes. Linderoth [23] presented another variant, in which the solution space is partitioned into Cartesian products of triangles and rectangles. He also used second orderconic relaxations in addition to linear ones. Bao et al. [6] presented a third version, in which linear over- and under-estimators are derived for certain quadratic functions with multiple terms, instead of treating each quadratic term independently. Sherali & Tuncbilek [35] presented a spatial branch-and-bound algorithm for general polynomial optimisation problems, in which LP relaxations are constructed via the Reformulation-Linearisation Technique (RLT). For nonconvex QPs, the constraints generated by the RLT are obtained by multiplying pairs of linear constraints together, and then linearising by substituting quadratic terms with the X ij variables mentioned above. These constraints include the ones of McCormick [26] as a special case. Audet et al. [5] specialised the RLT approach to QCQP instances in which there are a substantial number of linear constraints, and also showed how to tighten the relaxations using cutting planes. Saxena et al. [33, 34] showed how to strengthen the RLT relaxation further, using specialised disjunctions derived from Eigenvalue considerations. Their algorithm works also for mixed-integer QCQPs, provided the integer-constrained variables are bounded. Misener & Floudas [27] showed how to construct strong mixed-integer linear programming (MILP) relaxations of mixed-integer QCQPs, rather than just linear ones. Although one must use branch-and-bound even to 3
4 solve these relaxations, the advantage is that one can use the power of existing MILP algorithms, which are now highly sophisticated. 2.2 SDP-based approaches A completely different body of work deals with semidefinite programming (SDP) relaxations of QCQPs and related problems. These papers group the X ij variables in a matrix X = xx T, and introduce the augmented matrix variable Y = ( 1 x )( ) 1 T ( 1 x T = x x X Then, in addition to the linear constraints of McCormick [26], one can add the constraint that Y must be psd. The SDP relaxation originated in Shor [36], but was developed by Ramana [31] and Fujie & Kojima [16]. Burer & Vandenbussche [11] presented a finite exact algorithm for nonconvex QP (rather than QCQP) that uses SDP relaxations together with a specialised branching rule based on complementarity constraints. Luo et al. [24] explored SDP relaxations of QCQP instances arising in signal processing, and propose a rounding heuristic. Bao et al. [7] discussed the doubly-non-negative (DNN) relaxation, which strengthens the SDP relaxation by imposing non-negativity on the off-diagonal elements of the matrix X. Burer & Chen [12] used the DNN relaxation to improve the exact algorithm presented in [11]. Anstreicher [3] considered relaxations for QCQPs that include box constraints, and shows that a combination of SDP and RLT is far superior than either method alone. (Note that the combined SDP / RLT relaxation is stronger than the DNN relaxation.) Still stronger SDP relaxations were presented in Zheng et al. [41]. Another method, called difference-of-convex (DC) programming, expresses each quadratic objective and constraint function as the difference between a convex quadratic function and a concave one (e.g., Bomze & Locatelli [9], Zheng et al. [40]). This enables one to compute bounds by using existing bounding techniques for convex and concave problems. Recently, however, Anstreicher [4] showed that the combined SDP / RLT relaxation dominates all of the standard DC relaxations. 2.3 The binary case We now recall a few key facts about QCQPs in which all variables are required to be binary (0-1 QCQPs). As mentioned in the introduction, they can be viewed as a special kind of QCQP. Clearly, 0-1 QCQPs are N P-hard in the strong sense, since the same is true even for the less general 0-1 LP. On the hand, they are easier than general QCQPs in two senses. First, the issue of irrational numbers and ). 4
5 limited precision does not arise, which makes finite termination easy to achieve. Second, non-convex instances can be easily convexified, without adding any new variables, using the identities x 2 i x i = 0 for all i. Hammer & Rubin [19] suggested to simply make each matrix Q k psd, by subtracting the minimum Eigenvalue from all diagonal entries, and adjusting the linear term accordingly. A more sophisticated convexification approach, for 0-1 QPs only, was presented in [8]. We extended that approach to 0-1 QCQPs in [17]. The lower bound obtained by solving the continuous relaxation of the resulting convexified 0-1 QCQP is as strong as the standard SDP bound, which is described in, e.g., [22, 29, 32]. We will also need a classical result of Fortet [15], which states that 0-1 QCQPs can be easily converted into 0-1 LPs, at the cost of adding the X ij variables mentioned above. Indeed, in this case, the McCormick inequalities take the much simpler form X ij 0, X ij x i, X ij x j and X ij x i +x j 1. Then, when x i and x j are forced to take binary values, X ij equals x i x j. Some alternative strategies for linearising quadratic problems with 0-1 variables are surveyed in Section 5 of [13]. We do not go into details here, for the sake of brevity. 3 Our Approach We now describe our approach. As mentioned in the introduction, it consists of four phases. These phases are covered in the following four subsections. 3.1 Binarisation By binarisation, we mean the construction of a 0-1 QCQP that approximates the original QCQP, in the following sense: any feasible solution to the 0-1 QCQP can be easily mapped to a feasible solution of the original QCQP with the same cost. Suppose initially that the QCQP contains explicit bounds of the form l x u, where the components of l and u are finite. That is, x is constrained to lie within a finite box. Using the simple scaling x i (x i l i )/(u i l i ) for all i, we can assume that x [0, 1] n. (If u i l i = 0 for some i, then x i is a constant and can be eliminated from the problem.) The most obvious binarisation strategy is then to select a positive integer p, and use only p bits of precision when assigning values to the x variables. The 0-1 QCQP can then be constructed as follows. For i = 1,..., n and for k = 1... p, introduce a binary variable z ik, equal to the kth bit of the binary representation of x i. Then eliminate the x variables from (1) (3) using the identities: p x i = 2 k z ik (i = 1,..., n). k=1 5
6 The resulting 0-1 QCQP has np variables. Note however that using the standard bit-representation gives a lopsided approximation to the original problem, in the sense that the x variables can take the value 0 but not the value 1. We prefer to use a representation that is symmetrical over the variable domains. The easiest way to do this is to impose the restriction that each x variable must take on a value that is a multiple of 1/(2 p 1). One can then use the modified identities: x i = 1 2 p 1 p 2 p k z ik (i = 1,..., n). (4) k=1 We call this binarisation strategy 1. Now, suppose that x is an optimal solution to the original QCQP. If we use binarisation strategy 1, then there exists a feasible solution x in our restricted set for which x i x i 1 2(2 p 1) value 1 2(2 p 1) for all i. Informally speaking, the can be thought of as the maximum error, even if, in general, there is no guarantee that the cost of x will be close to that of x. In order to reduce the maximum error, but with the same number of bits, one can impose instead the restriction that each x variable must take on a value that is equal to 1/2 p+1 plus a multiple of 1/2 p. To do this, it suffices to use the following modified identities: x i = 1 p 2 p k z ik (i = 1,..., n). (5) k=1 We call this binarisation strategy 2. With this strategy, the maximum error drops to 2 (p+1). It is of course not obvious how to select the right binarisation strategy, or the appropriate value of p, for a given 0-1 QCQP instance. For this reason, we consider several combinations in our computational experiments (see Section 4). The 0-1 QCQP obtained via either binarisation strategy can be written in the form: min z T Q0 z + c 0 z s.t. z T Qj z + c j z h j (j = 1,..., m) z {0, 1} np, where Q j and c j indicate that the corresponding coefficients must be computed according to (4) or (5). Now consider the case in which the QCQP does not have explicit finite lower and upper bounds. In most cases of practical interest, one can easily derive finite bounds using one s knowledge of the problem under consideration. If those bounds are rather loose, then one can compute tighter bounds 6
7 by solving a series of LPs or SDPs (two per variable), or by using any of the domain reduction techniques that are used in modern branch-and-reduce algorithms (see Subsection 2.1). 3.2 Convexification Now that we have a 0-1 QCQP that approximates the original QCQP, the next step is to make it convex. As mentioned in Subsection 2.3, there are two main options available. The first is to perturb the diagonals of the Q j in order to make them psd, as done in [17]. The second is convert the 0-1 QCQP into a 0-1 LP, using the Fortet construction. Observe that the first option leads to a convex 0-1 QCQP with np variables and m constraints, whereas the second leads to a 0-1 LP with O(n 2 p 2 ) variables and m+o(n 2 p 2 ) constraints. It is actually not obvious how to select between these two convexification options a priori. There is no dominance relation in general between the lower bounds obtained by solving the continuous relaxations of the two resulting problems. The first option yields a model with far fewer variables and constraints, but the existing software packages for solving 0-1 LPs are far more sophisticated than those for solving 0-1 QCQPs. Moreover, the Fortet constraints, though numerous, are extremely sparse. On the other hand, they can cause the LP relaxations to suffer from degeneracy. For these reasons, we consider both options in our computational experiments (again, see Section 4). 3.3 Branch-and-bound At this point, we have a convex 0-1 QCQP or 0-1 LP that approximates the original QCQP. The next step is to solve it. For the 0-1 LP, things are straightforward: we can just solve it with the classical branch-and-bound approach, using the simplex method to solve the LP relaxations. Rather than coding a branch-and-bound solver of our own, we decided to use the mixed-integer LP solver that is available in the IBM software package CPLEX (version 12.6). We remark that CPLEX has its own internal routines for, e.g., fixing variables, generating cutting planes, producing heuristic solutions, and selecting the variable on which to branch. We decided to use default settings, rather than experimenting with the (many) parameters available in CPLEX. For the convex 0-1 QCQP formulation, things are more complicated. For one thing, a variety of generic exact algorithms are available for convex mixed-integer nonlinear programs, such as outer approximation, generalised Benders decomposition, the extended cutting-plane method, and LP/NLPbased branch-and-bound (see the survey [18]). Moreover, we also have the option of converting the convex 0-1 QCQP into a 0-1 second order cone 7
8 program (SOCP). The motivation for doing this is that the SOCP relaxations can be solved quite quickly via interior-point methods (see, e.g., [1]). For ease of implementation, we decided to use the mixed-integer SOCP solver that is available in CPLEX. That solver provides two algorithmic options. Although the CPLEX manual does not give full details, our understanding is that the first option is plain branch-and-bound, in which the SOCP relaxations are solved via the Barrier algorithm; whereas the second option is a branch-and-cut scheme, in which the SOCP relaxations are solved only approximately via Kelley s cutting-plane algorithm [21]. Again, we just use the default parameters in both cases. 3.4 Local optimisation If the original QCQP is tightly constrained, or we use too few bits, there is a chance that the binarised and convexified problem is infeasible. In that case, our heuristic fails. Otherwise, the optimal solution z {0, 1} np to the latter problem can be converted, using the bit representation described in Subsection 3.1, into a feasible solution x R n to the original QCQP. Observe that x is not guaranteed even to be a local minimum for the QCQP. Our fourth and final phase is therefore to move from x to a nearby local minimum. To do this, we use the open-source non-linear programming solver IPOPT [20]. As we understand it, IPOPT is a primal-dual interiorpoint solver with a logarithmic barrier function, in which a line-search is conducted in each Newton iteration. As in the case of CPLEX, we use default parameter settings. 4 Computational Results In order to ascertain the potential of our algorithm, we ran experiments on two different problems: non-convex quadratic programs with box-constraints and point packing problems. These problems and experiments are described in the following two subsections. In both experiments, we used the callable library of CPLEX 12.6 to solve all 0-1 LPs and 0-1 QCQPs, and IPOPT to perform the local optimisation step. The experiments were performed on a 64-bit 2.3 Ghz AMD Opteron 6376 processor with 16Gb RAM, under the Ubuntu 12.4 operating system. All programs were implemented in C++ and compiled with gcc Non-convex QP with box constraints A non-convex quadratic program with box-constraints (QPB) takes the form: max { x T Qx + c x : x [0, 1] n}, 8
9 Strategy 1 Strategy 2 n p = 1 p = 2 p = 3 p = 1 p = 2 p = 3 IPOPT Table 1: Average percentage gaps for box-constrained QP instances: no local optimisation where Q Q n n and c Q n. This problem, which is strongly N P-hard, has received much attention (e.g., [3, 4, 11, 16, 39]). We used the instances described in [11, 39], which were created as follows: given a density %, each entry in Q and c is set to zero with probability (100 )%, and to a random integer between 50 and 50 with probability %. For eighteen different combinations of n and %, there are three random instances. So there are 54 instances in total. The instances and their optimal solution values were kindly given to us by Kurt Anstreicher. First, we ran our heuristic without the local optimisation step. Table 1 shows the quality of the bounds obtained with this setting. Each row corresponds to a particular combination of n and. For each of the two binarisation strategies, and for numbers of bits ranging from 1 to 3, we report the average, over the three instances of the given type, of the percentage integrality gap (i.e., the percentage gap between the profit of our heuristic solution and the optimal profit). For interest, we also include the average percentage integrality gap of solutions obtained using IPOPT alone, using five different random feasible initial solutions as starting points. We were surprised to see that, for these instances, binarisation strategy 1 performs very well, whereas strategy 2 performs extremely poorly. Not only that, but just one bit is enough to get very good solutions when strategy 1 9
10 Strategy 1 Strategy 2 n p = 1 p = 2 p = 3 p = 1 p = 2 p = 3 IPOPT Table 2: Average percentage gaps for box-constrained QP instances: with local optimisation is used. Note that having only 1 bit in this case corresponds to restricting all variables to be binary. This implies that, for these instances, there exist near-optimal solutions such that no variables take fractional values. As for IPOPT, it obtains solutions that are reasonably good, but almost always worse than those obtained with strategy 1. Next, we wished to apply the local optimisation step to the solutions obtained with our heuristic. Here, we immediately faced a technicality: IPOPT needs to start from solutions that are in the strict interior of the feasible region, but many of the solutions obtained with binarisation strategy 1 lay on the boundary. To deal with this, we simply moved the initial heuristic solutions a little towards the centre of the unit hypercube, before passing them to IPOPT. (More precisely, we changed x i to 0.9x i for i = 1,..., n.) Table 2 shows the quality of the bounds obtained with the local optimisation step. It can be seen that the solutions obtained with binarisation strategy 1 plus local optimisation are of exceptionally good quality. The addition of local optimisation also greatly improves the quality of the solutions obtained with strategy 2, but not enough to obtain results as good as those obtained with strategy 1 (or even with IPOPT alone). Table 3 shows the time taken to solve the convex 0-1 QCQPs and the time 10
11 Solving 0-1 QCQP Str. Bits Simplex IPM Kelley IPOPT Table 3: Average times (in seconds) for QPB instances for the local optimisation step, in seconds, averaged over all 54 instances, for various parameter settings. The column headings simplex, IPM and Kelley refer to the manner in which the continuous relaxations of the convex 0-1 QCQPs are solved: via the Fortet linearisation and the simplex method, via SOC reformulation and interior-point methods, or via SOC reformulation and the Kelley cutting plane method, respectively. We see that the IPM option is fastest, and that the time taken by the additional local optimisation step is negligible. In Table 4, we present more detailed running times for our preferred option, which uses binarisation strategy 1, with the IPM option to solve the 0-1 QCQPs, followed by local optimisation. We see that the running time depends heavily on the number of bits used. We remark that our heuristic compares very favourably in terms of running time with the current fastest exact algorithm for QPB, due to Burer and Vandenbussche [11]. Indeed, their exact algorithm took around 45,000 seconds for the instances with n = 50, on a machine that is only a little slower than ours. If we use only two bits, we obtain solutions that are frequently optimal, and always within 0.04% of optimal, while taking about one-hundredth of the time taken by their algorithm. 4.2 Point packing Next, we tested our algorithm on the point packing (PP) problem, which is a much-studied problem in global optimisation (see, e.g., [3, 4, 25, 37]). Given an integer n 2, the problem is to place n points in the unit square, in such a way that the minimum distance between any pair of points is maximised. Letting x i and y i denote the horizontal and vertical coordinates of point i, respectively, and letting θ denote the square of the objective function, PP 11
12 Number of bits n IPOPT Table 4: More detailed average times for QPB instances can be formulated as follows: max θ (6) s.t. θ (x i x j ) 2 + (y i y j ) 2 (1 i < j n) (7) x, y [0, 1] n (8) θ R +. (9) Since this formulation has a linear objective and (non-convex) quadratic constraints, it can be tackled with our method. It turns out that PP has four peculiar properties that make it quite different from QPB: 1. For any feasible solution obtained using binarisation strategy 2, one can obtain a better feasible solution using strategy 1. Accordingly, we used only strategy 1 in our experiments. 2. For a fixed number of bits p, there are 2 2p possible point locations. Therefore, in order to obtain a packing in which no two points coincide, the number of bits must be at least 1 p = 2 log 2 n. In our experiments, we set p to 3, which is greater than p for n
13 3. The binarisation of θ can cause the 0-1 QCQP solver to miss good packings. To see why, suppose that we use strategy 1 with 3 bits, together with the trivial bounds 0 θ 2. The binarisation then effectively forces θ to be a multiple of (2 0)/(2 3 1) = 2/7. Then, if an optimal packing has an objective value of 0.3 and a sub-optimal one has an objective value of 0.4, they will both be treated as if they had a value of 2/ To address this issue, it helps to use a tighter upper bound on θ. We used the following bound, which is given in Szabó et al. [37]: 2 (n 1) (n 1) n 1 (n 1) (n 1) We remark that the upper bounds obtained by Anstreicher [3, 4], using various combinations of RLT and SDP constraints, are all weaker than this one for n PP exhibits a high degree of symmetry. Indeed, given any feasible solution, we can rotate or reflect the unit square, or permute the points, and obtain another solution of the same profit. This symmetry makes the 0-1 QCQP hard to solve. To address this, we add the following constraints to the formulation (6) (9): n i=1 x i n/2 n i=1 y i n/2 i=1 y i i=1 x i 0 x 1 x 2... x n 1. Similar (though not identical) symmetry constraints were used, e.g., by Anstreicher [3, 4]. The best way to solve the resulting 0-1 QCQPs turned out to be to apply the Fortet linearisation and then solve the resulting 0-1 LP by the simplex method. The results that we obtained with this option are displayed in Table 5. The first column displays the number of points, n. To explain the second two columns, consider the triple (x, y, θ ) corresponding to the optimal 0-1 QCQP solution. The column headed %gap1 gives the percentage integrality gap for θ itself, whereas the column headed %gap2 gives the gap for the true (non-discretised) value of θ corresponding to the packing represented by (x, y ). The next column gives the time, in seconds, taken to solve the 0-1 QCQP. The next two columns give the gap obtained after the local optimisation, and the time taken to perform the local optimisation. The last two columns give the gaps and times obtained using IPOPT alone, averaged over 100 random starting points. 13
14 Initial Loc. Opt IPOPT n %gap1 %gap2 time %gap time %gap time Table 5: Results for PP instances, with strategy 1 and three bits. The results here are quite different to those obtained for QPB. We see that the initial solutions obtained by our approach are of rather poor quality, and almost always worse than those obtained with IPOPT. On the other hand, when local optimisation is applied, we obtain solutions of extremely good quality in every case, something that IPOPT alone fails to do. As for running times, the bottleneck for our approach is again the solution of the 0-1 QCQP. 5 Conclusion Non-convex QCQPs are very hard to solve to proven optimality, in both theory and practice. We have presented a new heuristic for them, which is based on the solution of a convex 0-1 QCQP that approximates the original instance. Although convex 0-1 QCQPs are also N P-hard, they are much easier to solve in practice than non-convex QCQPs. The computational results show that our heuristic find extremely good solutions, with reasonable computing times in the case of QPB, but rather excessive running times in the case of PP. There are five potential ways to improve our heuristic that we believe would be worth considering: In order to speed up the solution of the 0-1 QCQPs, it might help to reformulate them, using the procedure described in our paper [17], mentioned in Subsection 2.3. (Unfortunately, we were unable to test this, since CPLEX appears to do some kind of internal reformulation of its own, that we were unable to deactivate.) When solving the 0-1 QCQP, it might help to use a specialised rule for selecting the z variable upon which to branch. Intuitively, for any 14
15 fixed i, it would make sense to branch on z ik before branching on z ik for some k > k, since this corresponds to making a bigger change in the domain of x i. One could use different numbers of bits for different x variables. To do this effectively, one would need heuristic rules for deciding which variables merit more precision than others. One could derive conditions under which one can binarise only a subset of the x variables, yet still convexify the resulting (mixed) 0-1 QCQP. For example, one could binarise only the variables that have non-zero entries in at least one matrix Q k that is not psd. (Note that, in the case of PP, this would enable us to treat θ as continuous.) One could develop methods for handling infeasible 0-1 QCQPs. For example, one could relax the constraints initially, solve the modified 0-1 QCQP to get a nearly feasible QCQP solution, and repair that solution to make it feasible. Finally, on the theoretical side, one could attempt to prove that, if the QCQP satisfies certain conditions, then running our heuristic with a certain number of bits is guaranteed to yield a solution of a certain quality. (Note that such a result could not be proved for general QCQPs, since even finding a feasible solution of a general QCQP is N P-hard in the strong sense.) Acknowledgement We thank Professor Kurt Anstreicher for providing us with the QPB instances and their optimal solution values. References [1] F. Alizadeh & D. Goldfarb (2003) Second-order cone programming. Math. Program., 95, [2] F.A. Al-Khayyal, C. Larsen & T. Van Voorhis (1995) A relaxation method for non-convex quadratically constrained quadratic programs. J. Glob. Optim., 6, [3] K.M. Anstreicher (2009) Semidefinite programming versus the reformulation-linearization technique for non-convex quadratically constrained quadratic programming. J. Glob. Optim., 43, [4] K.M. Anstreicher (2012) On convex relaxations for quadratically constrained quadratic programming. Math. Program., 136,
16 [5] C. Audet, P. Hansen, B. Jaumard & G. Savard (2000) A branch-and-cut algorithm for nonconvex quadratically constrained quadratic programming. Math. Program., 87, [6] X. Bao, N.V. Sahinidis & M. Tawarmalani (2009) Multiterm polyhedral relaxations for nonconvex, quadratically-constrained quadratic programs. Optim. Methods & Softw., 24, [7] X. Bao, N.V. Sahinidis & M. Tawarmalani (2011) Semidefinite relaxations for quadratically constrained quadratic programming: a review and comparisons. Math. Program., 129, [8] A. Billionnet, S. Elloumi & M.-C. Plateau (2009) Improving the performance of standard solvers for quadratic 0-1 programs by a tight convex reformulation: the QCR method. Discr. Appl. Math., 157, [9] I.M. Bomze & M. Locatelli (2004) Undominated d.c. decompositions of quadratic functions and applications to branch-and-bound approaches. Comput. Optim. & Appl., 28, [10] S. Boyd & L. Vandenberghe (2004) Convex Optimization. Cambridge: Cambridge University Press. [11] S.A. Burer & D. Vandenbussche (2008) A finite branch-and-bound algorithm for nonconvex quadratic programming via semidefinite relaxations. Math. Program., 113, [12] S.A. Burer & J. Chen (2012) Globally solving nonconvex quadratic programming problems via completely positive programming. Math. Program. Comput., 4, [13] S.A. Burer & A.N. Letchford (2012) Non-convex mixed-integer nonlinear programming: a survey. Surveys in Oper. Res. & Mgmt. Sci., 17, [14] E. Feron (1999) Nonconvex quadratic programming, semidefinite relaxations and randomization algorithms in information and decision systems. In T. E. Djaferis & I. Schick (eds.) System Theory: Modeling Analysis and Control. Dortrecht: Kluwer. [15] R. Fortet (1959) L Algèbre de Boole et ses applications en recherche opérationnelle. Cahiers Centre Etudes Rech. Oper., 4, [16] T. Fujie & M. Kojima (1997) Semidefinite programming relaxation for nonconvex quadratic programs. J. Glob. Optim., 10,
17 [17] L. Galli & A.N. Letchford (2014) A compact variant of the QCR method for quadratically constrained quadratic 0-1 programs. Optim. Lett., 8, [18] I.E. Grossmann (2002) Review of nonlinear mixed-integer and disjunctive techniques. Optim. & Eng., 3, [19] P.L. Hammer & A.A. Rubin (1970) Some remarks on quadratic programming with 01 variables. RAIRO, 3, [20] IPOPT (Interior-Point Optimizer). Managed by A. Wächter & S. Wigerske. Available at [21] J.E. Kelley (1960) The cutting-plane method for solving convex programs. SIAM Journal, 8, [22] C. Lemarechal & F. Oustry (2001) SDP relaxations in combinatorial optimization from a Lagrangian viewpoint. In N. Hadjisawas & P.M. Pardalos (eds.) Advances in Convex Analysis and Global Optimization. Dortrecht: Kluwer. [23] J. Linderoth (2005) A simplicial branch-and-bound algorithm for solving quadratically constrained quadratic programs. Math. Program., 103, [24] Z.-Q. Luo, W.-K. Ma, A.M.-C. So, Y. Ye & S. Zhang (2010) Semidefinite relaxation of quadratic optimization problems. IEEE Signal Processing Magazine, 27, [25] M.C. Markót & T. Csendes (2005) A new verified optimization technique for the packing circles in a unit square problem. SIAM J. Optim., 16, [26] G.P. McCormick (1976) Computability of global solutions to factorable nonconvex programs: Part I Convex underestimating problems. Math. Program., 10, [27] R. Misener & C.A. Floudas (2013) GloMIQO: Global mixed-integer quadratic optimizer. J. Glob. Optim, 57, [28] K.G. Murty & S.N. Kabadi (1987) Some N P-complete problems in quadratic and nonlinear programming. Math. Program., 39, [29] S. Poljak, F. Rendl & H. Wolkowicz (1995) A recipe for semidefinite relaxation for (0,1)-quadratic programming. J. Glob. Optim., 7, [30] U. Raber (1998) A simplicial branch-and-bound method for solving nonconvex all-quadratic programs. J. Glob. Optim., 13,
18 [31] M. Ramana (1993) An Algorithmic Analysis of Multiquadratic and Semidefinite Programming Problems. PhD thesis, Johns Hopkins University, Baltimore, MD. [32] F. Roupin (2004) From linear to semidefinite programming: an algorithm to obtain semidefinite relaxations for bivalent quadratic problems. J. Comb. Optim., 8, [33] A. Saxena, P. Bonami & J. Lee (2010) Convex relaxations of nonconvex mixed integer quadratically constrained programs: extended formulations. Math. Program., 124, [34] A. Saxena, P. Bonami & J. Lee (2011) Convex relaxations of nonconvex mixed integer quadratically constrained programs: projected formulations. Math. Program., 130, [35] H.D. Sherali & C.H. Tuncbilek (1992) A global optimization algorithm for polynomial programming problems using a reformulation linearization technique. J. Glob. Optim., 2, [36] N.Z. Shor (1987) Quadratic optimization problems. Soviet J. Comput. Syst. Sci., 25, 111. [37] P.G. Szabó, M.C. Markót & T. Csendes (2005) Global optimization in geometry circle packing into the square. In C. Audet, P. Hansen & G. Savard (eds.) Essays and Surveys in Global Optimization, pp Dordrecht: Kluwer. [38] M. Tawarmalani & N.V. Sahinidis (2003) Convexification and Global Optimization in Continuous and Mixed-Integer Nonlinear Programming. Dortrecht: Kluwer. [39] D. Vandenbussche & G.L. Nemhauser (2005) A branch-and-cut algorithm for nonconvex quadatic programs with box constraints. Math. Program., 102, [40] X.J. Zheng, X.L. Sun, D. Li. (2011) Nonconvex quadratically constrained quadratic programming: best D.C. decompositions and their SDP representations. J. Glob. Optim., 50, [41] X.J. Zheng, X.L. Sun, D. Li. (2011) Convex relaxations for nonconvex quadratically constrained quadratic programming: matrix cone decomposition and polyhedral approximation. Math. Program., 129,
A Note on Representations of Linear Inequalities in Non-Convex Mixed-Integer Quadratic Programs
A Note on Representations of Linear Inequalities in Non-Convex Mixed-Integer Quadratic Programs Adam N. Letchford Daniel J. Grainger To appear in Operations Research Letters Abstract In the literature
More informationConvexification of Mixed-Integer Quadratically Constrained Quadratic Programs
Convexification of Mixed-Integer Quadratically Constrained Quadratic Programs Laura Galli 1 Adam N. Letchford 2 Lancaster, April 2011 1 DEIS, University of Bologna, Italy 2 Department of Management Science,
More informationOnline generation via offline selection - Low dimensional linear cuts from QP SDP relaxation -
Online generation via offline selection - Low dimensional linear cuts from QP SDP relaxation - Radu Baltean-Lugojan Ruth Misener Computational Optimisation Group Department of Computing Pierre Bonami Andrea
More informationCutting Planes for RLT Relaxations of Mixed 0-1 Polynomial Programs
Cutting Planes for RLT Relaxations of Mixed 0-1 Polynomial Programs Franklin Djeumou Fomeni Konstantinos Kaparis Adam N. Letchford To Appear in Mathematical Programming Abstract The Reformulation-Linearization
More informationCutting Planes for RLT Relaxations of Mixed 0-1 Polynomial Programs
Cutting Planes for RLT Relaxations of Mixed 0-1 Polynomial Programs Franklin Djeumou Fomeni Konstantinos Kaparis Adam N. Letchford December 2013 Abstract The Reformulation-Linearization Technique (RLT),
More informationNonconvex Quadratic Programming: Return of the Boolean Quadric Polytope
Nonconvex Quadratic Programming: Return of the Boolean Quadric Polytope Kurt M. Anstreicher Dept. of Management Sciences University of Iowa Seminar, Chinese University of Hong Kong, October 2009 We consider
More informationOn Non-Convex Quadratic Programming with Box Constraints
On Non-Convex Quadratic Programming with Box Constraints Samuel Burer Adam N. Letchford July 2008 Abstract Non-Convex Quadratic Programming with Box Constraints is a fundamental N P-hard global optimisation
More informationSemidefinite Relaxations for Non-Convex Quadratic Mixed-Integer Programming
Semidefinite Relaxations for Non-Convex Quadratic Mixed-Integer Programming Christoph Buchheim 1 and Angelika Wiegele 2 1 Fakultät für Mathematik, Technische Universität Dortmund christoph.buchheim@tu-dortmund.de
More informationConvex Quadratic Relaxations of Nonconvex Quadratically Constrained Quadratic Progams
Convex Quadratic Relaxations of Nonconvex Quadratically Constrained Quadratic Progams John E. Mitchell, Jong-Shi Pang, and Bin Yu Original: June 10, 2011 Abstract Nonconvex quadratic constraints can be
More informationUsing quadratic convex reformulation to tighten the convex relaxation of a quadratic program with complementarity constraints
Noname manuscript No. (will be inserted by the editor) Using quadratic conve reformulation to tighten the conve relaation of a quadratic program with complementarity constraints Lijie Bai John E. Mitchell
More informationCutting Planes for First Level RLT Relaxations of Mixed 0-1 Programs
Cutting Planes for First Level RLT Relaxations of Mixed 0-1 Programs 1 Cambridge, July 2013 1 Joint work with Franklin Djeumou Fomeni and Adam N. Letchford Outline 1. Introduction 2. Literature Review
More informationComparing Convex Relaxations for Quadratically Constrained Quadratic Programming
Comparing Convex Relaxations for Quadratically Constrained Quadratic Programming Kurt M. Anstreicher Dept. of Management Sciences University of Iowa European Workshop on MINLP, Marseille, April 2010 The
More informationThe Trust Region Subproblem with Non-Intersecting Linear Constraints
The Trust Region Subproblem with Non-Intersecting Linear Constraints Samuel Burer Boshi Yang February 21, 2013 Abstract This paper studies an extended trust region subproblem (etrs in which the trust region
More informationSolving Box-Constrained Nonconvex Quadratic Programs
Noname manuscript No. (will be inserted by the editor) Solving Box-Constrained Nonconvex Quadratic Programs Pierre Bonami Oktay Günlük Jeff Linderoth June 13, 2016 Abstract We present effective computational
More informationOn Valid Inequalities for Quadratic Programming with Continuous Variables and Binary Indicators
On Valid Inequalities for Quadratic Programming with Continuous Variables and Binary Indicators Hongbo Dong and Jeff Linderoth Wisconsin Institutes for Discovery University of Wisconsin-Madison, USA, hdong6,linderoth}@wisc.edu
More informationOn the Lovász Theta Function and Some Variants
On the Lovász Theta Function and Some Variants Laura Galli Adam N. Letchford March 2017. To appear in Discrete Optimization. Abstract The Lovász theta function of a graph is a well-known upper bound on
More informationMixed-Integer Nonlinear Programming
Mixed-Integer Nonlinear Programming Claudia D Ambrosio CNRS researcher LIX, École Polytechnique, France pictures taken from slides by Leo Liberti MPRO PMA 2016-2017 Motivating Applications Nonlinear Knapsack
More informationSemidefinite Programming Basics and Applications
Semidefinite Programming Basics and Applications Ray Pörn, principal lecturer Åbo Akademi University Novia University of Applied Sciences Content What is semidefinite programming (SDP)? How to represent
More informationRelaxations and Randomized Methods for Nonconvex QCQPs
Relaxations and Randomized Methods for Nonconvex QCQPs Alexandre d Aspremont, Stephen Boyd EE392o, Stanford University Autumn, 2003 Introduction While some special classes of nonconvex problems can be
More information12. Interior-point methods
12. Interior-point methods Convex Optimization Boyd & Vandenberghe inequality constrained minimization logarithmic barrier function and central path barrier method feasibility and phase I methods complexity
More informationMonomial-wise Optimal Separable Underestimators for Mixed-Integer Polynomial Optimization
Monomial-wise Optimal Separable Underestimators for Mixed-Integer Polynomial Optimization Christoph Buchheim Claudia D Ambrosio Received: date / Accepted: date Abstract In this paper we introduce a new
More informationThe Ongoing Development of CSDP
The Ongoing Development of CSDP Brian Borchers Department of Mathematics New Mexico Tech Socorro, NM 87801 borchers@nmt.edu Joseph Young Department of Mathematics New Mexico Tech (Now at Rice University)
More informationA finite branch-and-bound algorithm for nonconvex quadratic programming via semidefinite relaxations
Math. Program., Ser. A (2008) 3:259 282 DOI 0.007/s007-006-0080-6 FULL LENGTH PAPER A finite branch-and-bound algorithm for nonconvex quadratic programming via semidefinite relaxations Samuel Burer Dieter
More informationCSC Linear Programming and Combinatorial Optimization Lecture 12: The Lift and Project Method
CSC2411 - Linear Programming and Combinatorial Optimization Lecture 12: The Lift and Project Method Notes taken by Stefan Mathe April 28, 2007 Summary: Throughout the course, we have seen the importance
More informationQuadratic reformulation techniques for 0-1 quadratic programs
OSE SEMINAR 2014 Quadratic reformulation techniques for 0-1 quadratic programs Ray Pörn CENTER OF EXCELLENCE IN OPTIMIZATION AND SYSTEMS ENGINEERING ÅBO AKADEMI UNIVERSITY ÅBO NOVEMBER 14th 2014 2 Structure
More informationExtended Linear Formulation for Binary Quadratic Problems
Noname manuscript No. (will be inserted by the editor) Extended Linear Formulation for Binary Quadratic Problems Fabio Furini Emiliano Traversi Received: date / Accepted: date Abstract We propose and test
More informationarxiv: v1 [math.oc] 5 Jun 2013
A note on QUBO instances defined on Chimera graphs Sanjeeb Dash IBM T. J. Watson Research Center arxiv:1306.1202v1 [math.oc] 5 Jun 2013 May 9, 2017 Abstract McGeogh and Wang (2013) recently obtained optimal
More informationValid Inequalities and Convex Hulls for Multilinear Functions
Electronic Notes in Discrete Mathematics 36 (2010) 805 812 www.elsevier.com/locate/endm Valid Inequalities and Convex Hulls for Multilinear Functions Pietro Belotti Department of Industrial and Systems
More informationA NEW SECOND-ORDER CONE PROGRAMMING RELAXATION FOR MAX-CUT PROBLEMS
Journal of the Operations Research Society of Japan 2003, Vol. 46, No. 2, 164-177 2003 The Operations Research Society of Japan A NEW SECOND-ORDER CONE PROGRAMMING RELAXATION FOR MAX-CUT PROBLEMS Masakazu
More informationResearch Note. A New Infeasible Interior-Point Algorithm with Full Nesterov-Todd Step for Semi-Definite Optimization
Iranian Journal of Operations Research Vol. 4, No. 1, 2013, pp. 88-107 Research Note A New Infeasible Interior-Point Algorithm with Full Nesterov-Todd Step for Semi-Definite Optimization B. Kheirfam We
More informationMIT Algebraic techniques and semidefinite optimization February 14, Lecture 3
MI 6.97 Algebraic techniques and semidefinite optimization February 4, 6 Lecture 3 Lecturer: Pablo A. Parrilo Scribe: Pablo A. Parrilo In this lecture, we will discuss one of the most important applications
More informationA new family of facet defining inequalities for the maximum edge-weighted clique problem
A new family of facet defining inequalities for the maximum edge-weighted clique problem Franklin Djeumou Fomeni June 2016 Abstract This paper considers a family of cutting planes, recently developed for
More informationSolving Mixed-Integer Nonlinear Programs
Solving Mixed-Integer Nonlinear Programs (with SCIP) Ambros M. Gleixner Zuse Institute Berlin MATHEON Berlin Mathematical School 5th Porto Meeting on Mathematics for Industry, April 10 11, 2014, Porto
More informationConic optimization under combinatorial sparsity constraints
Conic optimization under combinatorial sparsity constraints Christoph Buchheim and Emiliano Traversi Abstract We present a heuristic approach for conic optimization problems containing sparsity constraints.
More informationA Dynamic Programming Heuristic for the Quadratic Knapsack Problem
A Dynamic Programming Heuristic for the Quadratic Knapsack Problem Franklin Djeumou Fomeni Adam N. Letchford March 2012 Abstract It is well known that the standard (linear) knapsack problem can be solved
More informationA General Framework for Convex Relaxation of Polynomial Optimization Problems over Cones
Research Reports on Mathematical and Computing Sciences Series B : Operations Research Department of Mathematical and Computing Sciences Tokyo Institute of Technology 2-12-1 Oh-Okayama, Meguro-ku, Tokyo
More informationSDP relaxation for a strategic pricing bilevel problem in electricity markets
SDP relaxation for a strategic pricing bilevel problem in electricity markets Marcia Fampa Instituto de Matemática-DCC and COPPE Universidade Federal do Rio de Janeiro C.P. 68530, 21945-970 Rio de Janeiro,
More informationComputer Sciences Department
Computer Sciences Department Valid Inequalities for the Pooling Problem with Binary Variables Claudia D Ambrosio Jeff Linderoth James Luedtke Technical Report #682 November 200 Valid Inequalities for the
More information6-1 The Positivstellensatz P. Parrilo and S. Lall, ECC
6-1 The Positivstellensatz P. Parrilo and S. Lall, ECC 2003 2003.09.02.10 6. The Positivstellensatz Basic semialgebraic sets Semialgebraic sets Tarski-Seidenberg and quantifier elimination Feasibility
More informationCSC Linear Programming and Combinatorial Optimization Lecture 10: Semidefinite Programming
CSC2411 - Linear Programming and Combinatorial Optimization Lecture 10: Semidefinite Programming Notes taken by Mike Jamieson March 28, 2005 Summary: In this lecture, we introduce semidefinite programming
More informationRLT-POS: Reformulation-Linearization Technique (RLT)-based Optimization Software for Polynomial Programming Problems
RLT-POS: Reformulation-Linearization Technique (RLT)-based Optimization Software for Polynomial Programming Problems E. Dalkiran 1 H.D. Sherali 2 1 The Department of Industrial and Systems Engineering,
More informationSecond Order Cone Programming Relaxation of Nonconvex Quadratic Optimization Problems
Second Order Cone Programming Relaxation of Nonconvex Quadratic Optimization Problems Sunyoung Kim Department of Mathematics, Ewha Women s University 11-1 Dahyun-dong, Sudaemoon-gu, Seoul 120-750 Korea
More informationRelations between Semidefinite, Copositive, Semi-infinite and Integer Programming
Relations between Semidefinite, Copositive, Semi-infinite and Integer Programming Author: Faizan Ahmed Supervisor: Dr. Georg Still Master Thesis University of Twente the Netherlands May 2010 Relations
More informationIndicator Constraints in Mixed-Integer Programming
Indicator Constraints in Mixed-Integer Programming Andrea Lodi University of Bologna, Italy - andrea.lodi@unibo.it Amaya Nogales-Gómez, Universidad de Sevilla, Spain Pietro Belotti, FICO, UK Matteo Fischetti,
More informationCuts for Conic Mixed-Integer Programming
Cuts for Conic Mixed-Integer Programming Alper Atamtürk and Vishnu Narayanan Department of Industrial Engineering and Operations Research, University of California, Berkeley, CA 94720-1777 USA atamturk@berkeley.edu,
More informationObtaining Tighter Relaxations of Mathematical Programs with Complementarity Constraints
Obtaining Tighter Relaxations of Mathematical Programs with Complementarity Constraints John E. Mitchell, Jong-Shi Pang, and Bin Yu Original: February 19, 2011. Revised: October 11, 2011 Abstract The class
More information12. Interior-point methods
12. Interior-point methods Convex Optimization Boyd & Vandenberghe inequality constrained minimization logarithmic barrier function and central path barrier method feasibility and phase I methods complexity
More informationCopositive Programming and Combinatorial Optimization
Copositive Programming and Combinatorial Optimization Franz Rendl http://www.math.uni-klu.ac.at Alpen-Adria-Universität Klagenfurt Austria joint work with I.M. Bomze (Wien) and F. Jarre (Düsseldorf) IMA
More informationConvex Relaxations of Non-Convex Mixed Integer Quadratically Constrained Programs: Projected Formulations
Anureet Saxena Pierre Bonami Jon Lee Convex Relaxations of Non-Convex Mixed Integer Quadratically Constrained Programs: Projected Formulations November 19, 008 Abstract A common way to produce a convex
More informationLifting for conic mixed-integer programming
Math. Program., Ser. A DOI 1.17/s117-9-282-9 FULL LENGTH PAPER Lifting for conic mixed-integer programming Alper Atamtürk Vishnu Narayanan Received: 13 March 28 / Accepted: 28 January 29 The Author(s)
More informationDecomposition-based Methods for Large-scale Discrete Optimization p.1
Decomposition-based Methods for Large-scale Discrete Optimization Matthew V Galati Ted K Ralphs Department of Industrial and Systems Engineering Lehigh University, Bethlehem, PA, USA Départment de Mathématiques
More informationLagrange Duality. Daniel P. Palomar. Hong Kong University of Science and Technology (HKUST)
Lagrange Duality Daniel P. Palomar Hong Kong University of Science and Technology (HKUST) ELEC5470 - Convex Optimization Fall 2017-18, HKUST, Hong Kong Outline of Lecture Lagrangian Dual function Dual
More informationMustapha Ç. Pinar 1. Communicated by Jean Abadie
RAIRO Operations Research RAIRO Oper. Res. 37 (2003) 17-27 DOI: 10.1051/ro:2003012 A DERIVATION OF LOVÁSZ THETA VIA AUGMENTED LAGRANGE DUALITY Mustapha Ç. Pinar 1 Communicated by Jean Abadie Abstract.
More informationOptimality, Duality, Complementarity for Constrained Optimization
Optimality, Duality, Complementarity for Constrained Optimization Stephen Wright University of Wisconsin-Madison May 2014 Wright (UW-Madison) Optimality, Duality, Complementarity May 2014 1 / 41 Linear
More informationA 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 informationarxiv: v1 [math.co] 7 Jun 2012
LINEAR PROGRAMMING RELAXATIONS OF QUADRATICALLY CONSTRAINED QUADRATIC PROGRAMS ANDREA QUALIZZA 1, PIETRO BELOTTI 2, AND FRANÇOIS MARGOT 1,3 arxiv:1206.1633v1 [math.co] 7 Jun 2012 Abstract. We investigate
More informationarxiv: v1 [math.oc] 23 Nov 2012
arxiv:1211.5406v1 [math.oc] 23 Nov 2012 The equivalence between doubly nonnegative relaxation and semidefinite relaxation for binary quadratic programming problems Abstract Chuan-Hao Guo a,, Yan-Qin Bai
More informationA Continuation Approach Using NCP Function for Solving Max-Cut Problem
A Continuation Approach Using NCP Function for Solving Max-Cut Problem Xu Fengmin Xu Chengxian Ren Jiuquan Abstract A continuous approach using NCP function for approximating the solution of the max-cut
More informationLinear Programming Versus Convex Quadratic Programming for the Module Allocation Problem
Linear Programming Versus Convex Quadratic Programming for the Module Allocation Problem Sourour Elloumi 1 April 13, 2005 Abstract We consider the Module Allocation Problem with Non-Uniform communication
More informationSparse Optimization Lecture: Basic Sparse Optimization Models
Sparse Optimization Lecture: Basic Sparse Optimization Models Instructor: Wotao Yin July 2013 online discussions on piazza.com Those who complete this lecture will know basic l 1, l 2,1, and nuclear-norm
More informationFast ADMM for Sum of Squares Programs Using Partial Orthogonality
Fast ADMM for Sum of Squares Programs Using Partial Orthogonality Antonis Papachristodoulou Department of Engineering Science University of Oxford www.eng.ox.ac.uk/control/sysos antonis@eng.ox.ac.uk with
More informationConstraint Qualification Failure in Action
Constraint Qualification Failure in Action Hassan Hijazi a,, Leo Liberti b a The Australian National University, Data61-CSIRO, Canberra ACT 2601, Australia b CNRS, LIX, Ecole Polytechnique, 91128, Palaiseau,
More informationAn exact reformulation algorithm for large nonconvex NLPs involving bilinear terms
An exact reformulation algorithm for large nonconvex NLPs involving bilinear terms Leo Liberti CNRS LIX, École Polytechnique, F-91128 Palaiseau, France. E-mail: liberti@lix.polytechnique.fr. Constantinos
More information15-780: LinearProgramming
15-780: LinearProgramming J. Zico Kolter February 1-3, 2016 1 Outline Introduction Some linear algebra review Linear programming Simplex algorithm Duality and dual simplex 2 Outline Introduction Some linear
More informationBi-Perspective Functions for Mixed-Integer Fractional Programs with Indicator Variables
Bi-Perspective Functions for Mixed-Integer Fractional Programs with Indicator Variables Adam N. Letchford Qiang Ni Zhaoyu Zhong September 2017 Abstract Perspective functions have long been used to convert
More informationBCOL RESEARCH REPORT 07.04
BCOL RESEARCH REPORT 07.04 Industrial Engineering & Operations Research University of California, Berkeley, CA 94720-1777 LIFTING FOR CONIC MIXED-INTEGER PROGRAMMING ALPER ATAMTÜRK AND VISHNU NARAYANAN
More informationSolving large Semidefinite Programs - Part 1 and 2
Solving large Semidefinite Programs - Part 1 and 2 Franz Rendl http://www.math.uni-klu.ac.at Alpen-Adria-Universität Klagenfurt Austria F. Rendl, Singapore workshop 2006 p.1/34 Overview Limits of Interior
More informationAn 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 informationInterior-Point Methods for Linear Optimization
Interior-Point Methods for Linear Optimization Robert M. Freund and Jorge Vera March, 204 c 204 Robert M. Freund and Jorge Vera. All rights reserved. Linear Optimization with a Logarithmic Barrier Function
More informationHeuristics and Upper Bounds for a Pooling Problem with Cubic Constraints
Heuristics and Upper Bounds for a Pooling Problem with Cubic Constraints Matthew J. Real, Shabbir Ahmed, Helder Inàcio and Kevin Norwood School of Chemical & Biomolecular Engineering 311 Ferst Drive, N.W.
More informationA PREDICTOR-CORRECTOR PATH-FOLLOWING ALGORITHM FOR SYMMETRIC OPTIMIZATION BASED ON DARVAY'S TECHNIQUE
Yugoslav Journal of Operations Research 24 (2014) Number 1, 35-51 DOI: 10.2298/YJOR120904016K A PREDICTOR-CORRECTOR PATH-FOLLOWING ALGORITHM FOR SYMMETRIC OPTIMIZATION BASED ON DARVAY'S TECHNIQUE BEHROUZ
More informationDSOS/SDOS Programming: New Tools for Optimization over Nonnegative Polynomials
DSOS/SDOS Programming: New Tools for Optimization over Nonnegative Polynomials Amir Ali Ahmadi Princeton University Dept. of Operations Research and Financial Engineering (ORFE) Joint work with: Anirudha
More informationHandout 6: Some Applications of Conic Linear Programming
ENGG 550: Foundations of Optimization 08 9 First Term Handout 6: Some Applications of Conic Linear Programming Instructor: Anthony Man Cho So November, 08 Introduction Conic linear programming CLP, and
More informationarxiv: v3 [math.oc] 13 Jul 2018
Globally solving Non-Convex Quadratic Programs via Linear Integer Programming techniques Wei Xia, Juan Vera 2 and Luis F. Zuluaga arxiv:5.02423v3 [math.oc] 3 Jul 208 Department of Industrial and Systems
More informationA note on QUBO instances defined on Chimera graphs
A note on QUBO instances defined on Chimera graphs Sanjeeb Dash IBM T. J. Watson Research Center July 1, 2013 Abstract McGeoch and Wang (2013) recently obtained optimal or near-optimal solutions to some
More information5. Duality. Lagrangian
5. Duality Convex Optimization Boyd & Vandenberghe Lagrange dual problem weak and strong duality geometric interpretation optimality conditions perturbation and sensitivity analysis examples generalized
More informationMixed Integer Non Linear Programming
Mixed Integer Non Linear Programming Claudia D Ambrosio CNRS Research Scientist CNRS & LIX, École Polytechnique MPRO PMA 2016-2017 Outline What is a MINLP? Dealing with nonconvexities Global Optimization
More informationInderjit Dhillon The University of Texas at Austin
Inderjit Dhillon The University of Texas at Austin ( Universidad Carlos III de Madrid; 15 th June, 2012) (Based on joint work with J. Brickell, S. Sra, J. Tropp) Introduction 2 / 29 Notion of distance
More informationSDP Relaxations for MAXCUT
SDP Relaxations for MAXCUT from Random Hyperplanes to Sum-of-Squares Certificates CATS @ UMD March 3, 2017 Ahmed Abdelkader MAXCUT SDP SOS March 3, 2017 1 / 27 Overview 1 MAXCUT, Hardness and UGC 2 LP
More informationA CONIC DANTZIG-WOLFE DECOMPOSITION APPROACH FOR LARGE SCALE SEMIDEFINITE PROGRAMMING
A CONIC DANTZIG-WOLFE DECOMPOSITION APPROACH FOR LARGE SCALE SEMIDEFINITE PROGRAMMING Kartik Krishnan Advanced Optimization Laboratory McMaster University Joint work with Gema Plaza Martinez and Tamás
More informationInteger Programming ISE 418. Lecture 8. Dr. Ted Ralphs
Integer Programming ISE 418 Lecture 8 Dr. Ted Ralphs ISE 418 Lecture 8 1 Reading for This Lecture Wolsey Chapter 2 Nemhauser and Wolsey Sections II.3.1, II.3.6, II.4.1, II.4.2, II.5.4 Duality for Mixed-Integer
More informationIntroduction to Semidefinite Programming I: Basic properties a
Introduction to Semidefinite Programming I: Basic properties and variations on the Goemans-Williamson approximation algorithm for max-cut MFO seminar on Semidefinite Programming May 30, 2010 Semidefinite
More information4y Springer NONLINEAR INTEGER PROGRAMMING
NONLINEAR INTEGER PROGRAMMING DUAN LI Department of Systems Engineering and Engineering Management The Chinese University of Hong Kong Shatin, N. T. Hong Kong XIAOLING SUN Department of Mathematics Shanghai
More informationl p -Norm Constrained Quadratic Programming: Conic Approximation Methods
OUTLINE l p -Norm Constrained Quadratic Programming: Conic Approximation Methods Wenxun Xing Department of Mathematical Sciences Tsinghua University, Beijing Email: wxing@math.tsinghua.edu.cn OUTLINE OUTLINE
More informationConvex Optimization. (EE227A: UC Berkeley) Lecture 6. Suvrit Sra. (Conic optimization) 07 Feb, 2013
Convex Optimization (EE227A: UC Berkeley) Lecture 6 (Conic optimization) 07 Feb, 2013 Suvrit Sra Organizational Info Quiz coming up on 19th Feb. Project teams by 19th Feb Good if you can mix your research
More information1. Introduction. Consider the following quadratic binary optimization problem,
ON DUALITY GAP IN BINARY QUADRATIC PROGRAMMING XIAOLING SUN, CHUNLI LIU, DUAN LI, AND JIANJUN GAO Abstract. We present in this paper new results on the duality gap between the binary quadratic optimization
More informationA semidefinite relaxation scheme for quadratically constrained quadratic problems with an additional linear constraint
Iranian Journal of Operations Research Vol. 2, No. 2, 20, pp. 29-34 A semidefinite relaxation scheme for quadratically constrained quadratic problems with an additional linear constraint M. Salahi Semidefinite
More informationc 2000 Society for Industrial and Applied Mathematics
SIAM J. OPIM. Vol. 10, No. 3, pp. 750 778 c 2000 Society for Industrial and Applied Mathematics CONES OF MARICES AND SUCCESSIVE CONVEX RELAXAIONS OF NONCONVEX SES MASAKAZU KOJIMA AND LEVEN UNÇEL Abstract.
More informationConvex Quadratic Approximation
Convex Quadratic Approximation J. Ben Rosen 1 and Roummel F. Marcia 2 Abstract. For some applications it is desired to approximate a set of m data points in IR n with a convex quadratic function. Furthermore,
More informationAdvances in Convex Optimization: Theory, Algorithms, and Applications
Advances in Convex Optimization: Theory, Algorithms, and Applications Stephen Boyd Electrical Engineering Department Stanford University (joint work with Lieven Vandenberghe, UCLA) ISIT 02 ISIT 02 Lausanne
More informationA Penalized Quadratic Convex Reformulation Method for Random Quadratic Unconstrained Binary Optimization
A Penalized Quadratic Convex Reformulation Method for Random Quadratic Unconstrained Binary Optimization Karthik Natarajan Dongjian Shi Kim-Chuan Toh June 15, 2013 Abstract The Quadratic Convex Reformulation
More informationNew Global Algorithms for Quadratic Programming with A Few Negative Eigenvalues Based on Alternative Direction Method and Convex Relaxation
New Global Algorithms for Quadratic Programming with A Few Negative Eigenvalues Based on Alternative Direction Method and Convex Relaxation Hezhi Luo Xiaodi Bai Gino Lim Jiming Peng December 3, 07 Abstract
More informationSoftware for Integer and Nonlinear Optimization
Software for Integer and Nonlinear Optimization Sven Leyffer, leyffer@mcs.anl.gov Mathematics & Computer Science Division Argonne National Laboratory Roger Fletcher & Jeff Linderoth Advanced Methods and
More informationAn Exact Algorithm for a Resource Allocation Problem in Mobile Wireless Communications
An Exact Algorithm for a Resource Allocation Problem in Mobile Wireless Communications Adam N. Letchford Qiang Ni Zhaoyu Zhong To appear in Computational Optimization & Applications Abstract We consider
More informationMixed-Integer Nonlinear Decomposition Toolbox for Pyomo (MindtPy)
Mario R. Eden, Marianthi Ierapetritou and Gavin P. Towler (Editors) Proceedings of the 13 th International Symposium on Process Systems Engineering PSE 2018 July 1-5, 2018, San Diego, California, USA 2018
More informationOptimization over Structured Subsets of Positive Semidefinite Matrices via Column Generation
Optimization over Structured Subsets of Positive Semidefinite Matrices via Column Generation Amir Ali Ahmadi Princeton, ORFE a a a@princeton.edu Sanjeeb Dash IBM Research sanjeebd@us.ibm.com March 8, 2016
More informationLecture 6: Conic Optimization September 8
IE 598: Big Data Optimization Fall 2016 Lecture 6: Conic Optimization September 8 Lecturer: Niao He Scriber: Juan Xu Overview In this lecture, we finish up our previous discussion on optimality conditions
More informationGlobal Quadratic Minimization over Bivalent Constraints: Necessary and Sufficient Global Optimality Condition
Global Quadratic Minimization over Bivalent Constraints: Necessary and Sufficient Global Optimality Condition Guoyin Li Communicated by X.Q. Yang Abstract In this paper, we establish global optimality
More informationDeterministic Methods for Detecting Redundant Linear. Constraints in Semidefinite Programming
Deterministic Methods for Detecting Redundant Linear Constraints in Semidefinite Programming Daniel Stover Department of Mathematics and Statistics Northen Arizona University,Flagstaff, AZ 86001. July
More informationELE539A: Optimization of Communication Systems Lecture 16: Pareto Optimization and Nonconvex Optimization
ELE539A: Optimization of Communication Systems Lecture 16: Pareto Optimization and Nonconvex Optimization Professor M. Chiang Electrical Engineering Department, Princeton University March 16, 2007 Lecture
More information