You are on page 1of 2

Assignment 1

1. (a) Do Exercise 2.39 from the textbook. Assume that the only optimization tool that you have
is a solver for LPs (Linear Programs), and you must be able to solve your formulation.
(b) Bring the LP formulation(s) of the previous part into standard form and give their dual(s).
(c) Bring the dual LPs of the previous part in standard form.

2. As you already know, graphs (directed or undirected) are widely used to model different settings,
and then the original problem is reduced to the solution of a graph problem. Here is a graph:
Its set of nodes is {s, v1 , v2 , . . . , vM , u1 , u2 , . . . , uM , t1 , t2 , . . . , tk } (note that every node ti , i =

v1 µ1 u1

.
. ∞ t1
∞ .
.
s ∞
∞ vj uj
µj
∞ . ∞
.
. ∞ tK

µM ∞
vM uM

Figure 1: A (directed) graph with K + 2M + 1 nodes.

1, 2, . . . , K is connected to all nodes uj , j = 1, . . . , K). A cut S of the graph G = (V, E) is just a


subset of V , and it partitions the graph into two parts S, S̄ = V \ S. If the graph is undirected
then we denote by δ(S) the edges that cross the cut S, i.e.

δ(S) = {(i, j) ∈ E : i ∈ S and j ∈ S̄}.

For example, if you assume for the moment that the edges of the graph in Figure 1 are undirected,
and S = {s} then δ(S) = {(s, vi ) : i = 1, . . . , M }. If the graph is directed, then we denote by
δ+ (S) the edges that leave the cut S, and by δ− (S) the edges that enter the cut S. For example,
in the directed graph of Figure 1, if we consider again the cut S = {s}, then δ+ (S) = {(s, vi ) : i =
1, . . . , M }, and δ− (S) = ∅. For simplicity, if w ∈ V is a node, then we denote by δ+ (w), δ− (w)
the sets δ+ ({w}), δ− ({w}).

(a) How many cuts are there in a graph with n nodes? (Note that we don’t count the sets ∅, V
as cuts.)
(b) Suppose that the numbers µj in Figure 1 are positive, and ∞ stands for a very big positive
P
number C, bigger than M j=1 µj . Let ukl be µj if k = vj , l = uj , and C if k = s, l = vj , j =
1, . . . , M or k = uj , l = ti , j = 1, . . . , M, i = 1, . . . , K (i.e., uij is the number you see on edge

1
(i, j) in Figure 1.) Consider the following LP, with one variable xe for every edge e ∈ E 1 :
P
max e∈δ+ (s) xe s.t.
P P
x
e∈δ+ (v) e − x
e∈δ− (v) e = 0, for all v ∈ V \ {s, t1 , t2 , . . . , tK }
xij ≤ uij , for all (i, j) ∈ E
xij ≥ 0, for all (i, j) ∈ E

i. How many variables, and how many constraints are there in the LP above (excluding
the last set of lower bound constraints)?
ii. Bring the LP in standard form and give its dual.
iii. Where did that LP come from? In other words, give a ”real-world” problem that can
be modeled by this LP. (hint: Assume that the original problem came from an Alberta
oil pipeline company.)
iv. This LP formulation can be applied to any directed graph G = (V, E) with a definition
of uij for every (i, j) ∈ E. With specific values for these u’s you can use any LP solver
to find the optimum. But for the specific graph of Figure 1 one can guess the optimum
value without using a solver. Can you? (Of course you can, but I want you to also
write down the value of the optimum, and why you think it is really the optimum).

1
We denote the variable corresponding to edge e = (i, j) sometimes by xe and sometimes by xij . It is the same
variable in both cases.

You might also like