1.2 Guidelines for strong formulations Running time for LP usually depends on m and n ( number of...

9

Click here to load reader

description

 Hence optimal solution to a relaxation provides an upper bound on optimal value (for maximization problem). (Lower bound for minimization problem.)  Typical methods to obtain lower bound  Relaxation  Dual problem  LP relaxation widely used, but there are other types of relaxations: Lagrangian relaxation, combinatorial relaxation, semidefinite relaxation, … Purpose is to obtain lower bound  Upper bound usually obtained by finding a feasible solution. If lower bound = upper bound, it is optimal value (we may need to find the solution additionally)  We usually use divide-and- conquer. If z LP is the lower bound for a subproblem, and z’ is the current best objective value we know (upper bound) and z LP  z’, then we can discard the subproblem since the subproblem does not have a better solution. So it is important to have a good (tight) lower bound to increase the possibility of pruning the subproblem early in the divide-and-conquer (branch-and-bound). Integer Programming

Transcript of 1.2 Guidelines for strong formulations Running time for LP usually depends on m and n ( number of...

Page 1: 1.2 Guidelines for strong formulations  Running time for LP usually depends on m and n ( number of iterations O(m), O(log n)). Not critically depend on.

1.2 Guidelines for strong formulations Running time for LP usually depends on m and n ( number of iterations O(m),

O(log n)). Not critically depend on formulation (usually).For IP, the running time is very erratic on different classes of problems and also depends on the choice of formulation significantly.

Reason: most algorithms for IP are basically divide-and-conquer type. If the enumeration tree grows big, running time becomes prohibitive. recent research efforts mostly focus on preventing the growth of the enumeration tree.Also, there are strong theoretical indication that divide-and-conquer may be the best we can do to solve general IP problems (no efficient algorithms exist). However, recent advances in theory and software make it possible to solve many practically sized problems (very fast in some cases).

Integer Programming 2011 1

Page 2: 1.2 Guidelines for strong formulations  Running time for LP usually depends on m and n ( number of iterations O(m), O(log n)). Not critically depend on.

Definition 1.1: The linear relaxation of MIP is obtained by dropping the integrality requirements on integer variables (resulting problem is LP)

Def (NW, p298, in max form) A problem (RP) zR = max { zR(x) : x SR } is a relaxation of (IP) zIP = max { c’x : x S } if :

(i) S SR, and

(ii) c’x zR(x) for all x S.

(relaxation may be defined in different solution space (variables)) Prop 1.1) If RP is infeasible, so is IP. If IP is feasible, then zIP zR.

pf) From (i), first statement is true. Now suppose zIP is finite and let x0 be an optimal solution to IP. Then zIP = c’x0 zR(x0) zR.

Finally, if zIP = , (i) and (ii) imply that zR = .

Integer Programming 2011 2

Page 3: 1.2 Guidelines for strong formulations  Running time for LP usually depends on m and n ( number of iterations O(m), O(log n)). Not critically depend on.

Hence optimal solution to a relaxation provides an upper bound on optimal value (for maximization problem). (Lower bound for minimization problem.)

Typical methods to obtain lower boundRelaxationDual problem

LP relaxation widely used, but there are other types of relaxations: Lagrangian relaxation, combinatorial relaxation, semidefinite relaxation, … Purpose is to obtain lower bound

Upper bound usually obtained by finding a feasible solution.If lower bound = upper bound, it is optimal value (we may need to find the solution additionally)

We usually use divide-and- conquer. If zLP is the lower bound for a subproblem, and z’ is the current best objective value we know (upper bound) and zLP z’, then we can discard the subproblem since the subproblem does not have a better solution. So it is important to have a good (tight) lower bound to increase the possibility of pruning the subproblem early in the divide-and-conquer (branch-and-bound).

Integer Programming 2011 3

Page 4: 1.2 Guidelines for strong formulations  Running time for LP usually depends on m and n ( number of iterations O(m), O(log n)). Not critically depend on.

Suppose we have two formulations A and B for the same problem, and let PA and PB be the polyhedra defined by the LP relaxation of the formulations, respectively.Then, if PA PB , we have z* zA zB . So PA gives tighter lower bound, hence better formulation.

If an optimal solution to the relaxation is feasible to the MIP, then it is also an optimal solution to MIP.

Integer Programming 2011 4

Page 5: 1.2 Guidelines for strong formulations  Running time for LP usually depends on m and n ( number of iterations O(m), O(log n)). Not critically depend on.

Ex: Facility location problemAlternative formulation: min j N cjyj + i M j N dijxij

j N xij = 1, for i M

i M xij myj , for j N

0 xij 1 for i M, j N, yj {0, 1} for j N Let

PFL = { (x, y)’: j N xij = 1, i, xij yj, i, j

0 xij 1, 0 yj 1 }

PAFL = { (x, y)’: j N xij = 1, i, i M xij myj , j

0 xij 1, 0 yj 1 }

PFL PAFL , and the inclusion can be strict. Hence zFL zAFL .

Integer Programming 2011 5

Page 6: 1.2 Guidelines for strong formulations  Running time for LP usually depends on m and n ( number of iterations O(m), O(log n)). Not critically depend on.

Integer Programming 2011 6

PFLPAFL

conv(F)

Page 7: 1.2 Guidelines for strong formulations  Running time for LP usually depends on m and n ( number of iterations O(m), O(log n)). Not critically depend on.

Ideal Formulation (also in Chap 3, BW) Def: Given a set X Rn, the convex hull of X, denoted conv(X), is defined

as: conv(X) = { x : x = i = 1t ixi , i = 1

t i = 1, i 0 for i = 1, … , t over all finite subsets { x1, … , xt } of X }

Assume |X| finite, then Prop 1.1: conv(X) is a polyhedron. (polytope)

Prop 1.2: The extreme points of conv(X) all lie in X.

Props also hold for unbounded integer sets. (NW, p104)

Rationale: to solve IP : max { cx: x X },Solve max { cx : x conv(X) }. The problem is LP and LP has an extreme point optimal solution (simplex method can find it).

Integer Programming 2011 7

Page 8: 1.2 Guidelines for strong formulations  Running time for LP usually depends on m and n ( number of iterations O(m), O(log n)). Not critically depend on.

But conv(X) may need lots of inequalities (not a big problem) to describe and/or we may have limited knowledge about the characteristics of the inequalities ( trouble).Good approximation to conv(X) is helpful ( X conv(X) P ), we may have stronger bound.

Integer Programming 2011 8

Page 9: 1.2 Guidelines for strong formulations  Running time for LP usually depends on m and n ( number of iterations O(m), O(log n)). Not critically depend on.

Ex : The pigeon hole principlePlace n+1 pigeons into n holes in such a way that no two pigeons share the same hole. (impossible)

Formulations:(1.3) j=1

n xij = 1, i = 1, …., n+1,

xij + xkj 1, j = 1, …, n, i k, i, k = 1, …, n+1

xij {0, 1}, i = 1, …, n+1, j = 1, …, n

(1.4) j=1n xij = 1, i = 1, …., n+1,

i=1n+1 xij 1, j = 1, …., n,

xij {0, 1}, i = 1, …, n+1, j = 1, …, n

xij = 1/n for all i, j satisfies LP relaxation of (1.3), but LP relaxation of (1.4) infeasible.

Integer Programming 2011 9