Вы находитесь на странице: 1из 2

CSOR W4246 Fall, 2014

Homework 4 Theoretical part


Out: Monday, October 27, 2014
Due: 6pm, Monday, November 10, 2014
This problem set is organized as follows: the first two problems reinforce class material and the
next two problems are practical problems.
As usual, keep your answers clear and concise, and make sure that your hand-writing is legible and
that your name is clearly written on your homework. Collaboration is limited to discussion of ideas
only: you should write up your solutions entirely on your own.
For problem 3, you may use the LP solver of your choice: there are many options freely available.
Please make sure you cite which one you used.

1. (20 points) For the following network, with edge capacities as shown, find the maximum flow
from s to t along with a minimum cut.

a
6

d
1

g
12

10
1
b

s
10

20

2
c

4
5

2. (20 points) There are many variations on the maximum flow problem. For the following
two natural generalizations, show how to solve the more general problem by reducing it
to the original max-flow problem (thereby showing that these problems also admit efficient
solutions).
There are multiple sources and multiple sinks, and we wish to maximize the flow between
all sources and sinks.
Both the edges and the vertices (except for s and t) have capacities. The flow into and
out of a vertex cannot exceed the capacity of the vertex.

3. (20 points) You are given the following data points in the plane
(1, 3), (2, 5), (3, 7), (5, 11), (7, 14), (8, 15), (10, 19)
You want to find a line that approximately passes through these points (no line is a perfect
fit). Write a linear program to find the line that minimizes the maximum absolute error
max |axi + byi c|

1i7

and provide the optimal solution and its objective value.


4. (25 points) You have a data set consisting of items that a consumer has purchased on different
days and the prices at which they were purchased. You are tasked with finding bounds on
the consumers values (i.e., willingness to pay) for different items.
Specifically, let the set of items be V = {1, . . . , n}. Let t = 1, . . . , k index the days, and let
pt1 , . . . , ptn be the prices for the different items on day t. If on day t the consumer purchased,
say, item 1, this means that it was the most preferred item at the prices on that day, leading
to the inequalities:
v1 pt1 vi pti (i V ).
Let Ti be the indices of the days on which item i was purchased, as recorded in your data
set. You would like to compute upper bounds on the consumers values.
(a) (7 points) Let wij = mintTi {ptj pti }. Formulate the system of inequalities that constrains the values in terms of the variables vi and the constants wij .
(b) (5 points) Show that if v 0 = (v10 , . . . , vn0 ) and v 00 = (v100 , . . . , vn00 ) are feasible solutions to
the constraints, then so is
v 0 v 00 = (max{v10 , v100 }, . . . , max{vn0 , vn00 }).
P
From this result, you see that if you maximize the objective
iV vi subject to the
feasibility constraints, the solution will give upper bounds on the values.
(c) (13 points) Write out the linear program you have derived so far for the value bounds,
and provide its dual.
(d) Extra credit (15 points) You notice that your primal LP looks very similar to the
min-cut LP we saw in class. What problem is the dual solving in your case?
Hint: Note that if (v10 , v20 , . . . , vn0 ) is a feasible solution, then so is (v10 +c, v20 +c, . . . , vn0 +c)
for a constant c. Therefore, in the primal you can also include the constraint v1 = 0.

Вам также может понравиться