IOE 518: Introduction to IP, Winter 2012 BIP formulations Page 16 c Marina A. Epelman
IOE 518: Introduction to IP, Winter 2012 BIP formulations Page 17 c Marina A. Epelman
Set covering problem: formal statement
Set cover:
a collection T {1, . . . , n} such that [j2T Mj = M.
Set packing:
a collection T {1, . . . , n} such that Mj \ Mk = ; for all
j, k 2 T , j 6= k.
IOE 518: Introduction to IP, Winter 2012 BIP formulations Page 19 c Marina A. Epelman
Related problem: set partition
Set partition:
a collection T {1, . . . , n} which is both a cover and a packing.
IOE 518: Introduction to IP, Winter 2012 BIP formulations Page 20 c Marina A. Epelman
Constraints (1), (2), (3) (the cut-set formulation) or (1), (2), (4)
(the subtour elimination formulation), together with binary
restrictions on the variables, are valid formulations of the TSP.
How many variables and constraints do these formulations have?
IOE 518: Introduction to IP, Winter 2012 TSP Page 22 c Marina A. Epelman
Initial inventory is 0.
Variables:
I xt amount producedin period t
I st total inventory at the end of period t
I yt = 1 if production takes place in period t
IOE 518: Introduction to IP, Winter 2012 MIP formulations Page 25 c Marina A. Epelman
Uncapacitated lot sizing (ULS) continued
Formulation
Pn Pn Pn
min t=1 pt xt + t=1 ht st + t=1 ft yt
s.t. st 1 + xt = dt + st , t = 1, . . . , n (Inventory balance)
xt Myt , t = 1, . . . , n (Forcing constraints)
st , xt 0, yt 2 {0, 1}, t = 1, . . . , n
s0 = 0
Same variables:
yj = 1 if depot j is built, yj = 0 otherwise;
xij fraction of demand of client i supplied by depot j
Alternative formulation:
XX X
min cij xij + f j yj
i2M
X j2N j2N
s.t. xij = 1 8i 2 M (Demand satisfied)
j2N
X
xij myj 8j 2 N (Combined forcing constraints)
i2M
xij 0 8i 2 M, j 2 N
yj 2 {0, 1} 8j 2 N
IOE 518: Introduction to IP, Winter 2012 MIP formulations Page 28 c Marina A. Epelman
Polyhedron
A subset of Rn (that can be) described by a finite set of linear
constraints P = {x 2 Rn : Ax b} is a polyhedron.
IOE 518: Introduction to IP, Winter 2012 Strength of formulations Page 29 c Marina A. Epelman
Formulations for a 2D Set
X = {(1, 1), (2, 1), (3, 1), (1, 2), (2, 2), (3, 2), (2, 3)}.
IOE 518: Introduction to IP, Winter 2012 Strength of formulations Page 30 c Marina A. Epelman
IOE 518: Introduction to IP, Winter 2012 Strength of formulations Page 32 c Marina A. Epelman
which is an LP.
I Note: X P, so z IP z LP
I Suppose P1 P2 are both formulations for X . Then
z IP z LP1 z LP2
IOE 518: Introduction to IP, Winter 2012 Strength of formulations Page 33 c Marina A. Epelman
Comparing formulations for UFL
X
P1 = {(x, y ) 2 Rmn+n : xij = 1 8i, xij yj 8i, j
j2N
P P
I
Let (x, y ) 2 P1 . Then i2M xij i2M yj = myj . Hence,
P1 P2 .
I Suppose, for simplicity, that m = kn, with k 2 and integer.
Let each depot serve k clients, assigning xij = 1 appropriately,
and let yj = k/m 8j. Then (x, y ) 2 P2 \ P1
Conclusion: formulation P1 is better than P2 .
IOE 518: Introduction to IP, Winter 2012 Strength of formulations Page 34 c Marina A. Epelman
Convex hull
Given a set S Rn , the convexPhull of S, denoted Pt by conv(S), is
t i
defined as conv(S) = {x : x = i=1 i x , i=1 i = 1, i
0 i = 1, . . . , t over all finite subsets {x , . . . , x t } S}.
1
Proposition 1.1
conv(S) is a polyhedron if S is finite, or if it is the set of
feasible solutions of some MIP (not true for an arbitrary set S!).
which is an LP.
I Solution x LP will be found at an extreme point of P, and if
x LP 2 X , then we found the optimal solution of (IP) just by
solving (RP).
Proposition 1.2
All extreme points of conv(X ) lie in X .