Вы находитесь на странице: 1из 5

Let us illustrate The Three Interior Points Approach (TIPA).

Example 12: Let an LP problem be given as


Minimize z(x) = 2x1 x2 x3
subject to
x1 2x2 4x3
2x1 4x2 4x3
3x1 4x2 x3
x1 , x2 , x3

20
60
90
0.

Normalizing the coefficient vectors of all the constraints and the objective
function, we have
Minimize z(x) = 0.816x1 0.408x2 0.408x3
subject to
4.36 0.218x1 0.436x2 0.87x3 0
10 0.333x1 0.67x2 0.67x3 0
17.65 0.588x1 0.784x2 0.196x3 0
x1 , x2 , x3 0.

5
0

2 be the given point with objective value 14. Plug 1 x0


Let 1 x =
2
to the left hand side of the constraints to get the values 0.658, 6.313, 12.75, 5, 2,
and 2. Thus, 1 x0 is a feasible interior point. Then find two other interior points
in the plane or below the plane 2x1 x2 x3 = 14. Note
that
this plane
8
passes through 2 x0 . Consider two other points, say 2 x0 = 1 and
1

10
0
2 with objective values 18 and 23, respectively. Plug 2 x0 to
3x =
1
the left hand side of the constraints to get the values 1.31, 5.996, 11.966, 8, 1,and
1. Also, plug 3 x0 to the left hand side of the constraints to get 0.438, 4.66, 10.006,
10, 2, and 1. Thus, 2 x0 and 3 x0 are feasible interior points.Using the
descent
11.82
step of the sphere method, get three interior points 1 y 0 = 0.728 ,
0.847

12.93
13.659
0
= 1.43 and 3 y 0 = 0.5 with objective values 25.216,
2y
0.48
0.759
27.79 and 28.58, respectively. Get
of the three interior points
the average

12.803
to get another interior point y 0 = 0.886 . Apply the descent step to y 0
0.695
15.184
1

0.339 with objective value


to get another interior point 1 y =
0.339
2x1 x2 x3 = 31.046. Rename 1 y 1 as 1 x1 then update the feasible region
using 1 x1 . In the plane or below

17
2x1 x2 x3 = 31.046 , find another two points, say 2 x1 = 0.5
0.1

18
1

0.8 with objective values 34.6 and 36.81, respectively.


and 3 x =
0.01
Plug each of these to the left hand side of the constraints to get values
0,017, 3.371, 6.178, 18.76, 0.56, 0.01 and 0.0176, 11.082, 6.25, 19, 0.26, 0.1,
respectively. Thus, 2 x1 and 3 x1 are feasible interior points.
Apply

the descent
17.89
step to each of these to get the interior point 1 y 1 = 0.35 ,
0.16

18.495
18.288
1
1

0.53
0.81 with objective values 36.3, 37.57
y
=
and
y
=
2
3
0.05
0.01

18.22
and 37.399, respectively. Get the average y 1 = 0.56 . Apply descent
0.07

18.55
step again to get 1 y 2 = 0.56 with objective value
0.037
2x1 x2 x3 = 37.697. The point 1 y 2 is the best interior feasible solution. Rename 1 y 2 as 1 x2 then update the fesible region using 1 x2 . In the
2
plane or below
2x1 x2 x3 =

37.697 find another two points, say, 2 x =


18.76
19
0.56 and 3 x2 = 0.26 with objective values 38.09 and 38.36,
0.01
0.1
respectively. Plug these two points in the left hand side of the constraints to

3
get 0.349, 3.937, 7.24,17, 0.5, 0.1 and 0.0785, 3.4633, 6.437, 18, 0.8, 0.01, respectively. Thus 2 x2 and 3 x2 are interior feasible
solutions.Apply descent
step

18.79
19.36
to each of these to get the interior points 1 y 2 = 0.56 , 2 y 2 = 0.026
0.01
0.115

19.28
and 3 y 2 = 0.33 with objective values 38.155, 38.86 and 38.89, re0.0069

19.143
spectively. Get the average y 2 = 0.305 . Apply descent step again to get
0.04

19.32
3

0.028 with objective value 2x1 x2 x3 = 38.788. At this stage


1y =
0.12

19.28
we choose 3 y 2 = 0.33 with objective value 2x1 x2 x3 = 38.89 as
0.0069
the best interior feasible solution. Rename 3 y 2 as 1 x3 and update the feasible
region using 1 x2 . In the plane or below

19.8
2x1 x2 x3 = 38.89 find another two points, say 2 x3 = 0.001 and
0.001

19.9
3

0.002 with objective values 39.6 and 39.8. Plug the two points in
3x =
0.003
the left hand side of the constraints to get 0.0418, 3.405, 6.007, 19.8, 0.001, 0.001
and 0.018, 3.37, 5.947, 19.9, 0.002, 0.003, respectively. Thus 2 x3 and 3 x3 are
interior feasible solutions.
Apply

descentstep to each
of these toget the in
19.754
19.948
19.938
terior points 1 y 3 = 0.0187 ,2 y 3 = 0.001 and 3 y 3 = 0.002
0.041
0.001
0.003
with objective
values
39.56, 39.898 and 39.88, respectively.
Get theaver19.88
19.88
3
4

0.007 . Apply the descent step again to get 1 y = 0.007 with


age y =
0.015
0.015

19.948
objective value 39.782. At this stage, again we choose 2 y 3 = 0.001 .
0.001
3
4
Rename 2 y as 1 x . Due to the closeness of the last two resulting values, we

19.948
suspect that 1 x4 = 0.001 with objective value 39.89 is the near op0.001
timal. If we apply descent step again to 1 x4 , the same result we can get.
Now, 1 x3 and 1 x4 are results from the last two iterations. Let = 0.05 then
k 1 x4 1 x3 k
= 0.0035 < < 1. So, the process terminate and the answer is
3
k
1x k
20
x = 0 with objective value 40. Note that the difference between 40
0
and 39.89 is 0.11. But this is considerable because our termination point
is determined by = 0.05 and by the other positive tolerance in the descent
step which is 0.01.
Comparing TIPA and Sphere Method
0.0.1 InComparing
and Sphere
reaching the TIPA
termination
condition,Method
starting at an initial interior fea0
sible solution 1 x , the Sphere Method generates a sequence of interior feasible
solutions 1 x1 , 1 x2 , . . ., 1 xr , 1 xr+1 , where each of these points is the best
resulting point in every iteration. Thus, the Sphere Method terminates at
the (r+1)th iteration. The TIPA however, having the flexibility of choosing
arbitrarily the three initial interior feasible solutions as its starting points, can
choose points from the set Int(K) that includes 1 x0 , 1 x1 , 1 x2 , . . ., 1 xr , and
r+1
. The TIPA can choose, say 1 xr1 , as one of its three starting points.
1x
After the first iteration of the TIPA with the resulting point, say 1 y 1 , the plane
cx = c1 y 1 is used for the update of the feasible region K. All the other points
including 1 x0 , 1 x1 , 1 x2 , . . ., and 1 xr1 , are already outside from the updated
feasible region. These points will be excluded in choosing the next three initial
interior feasible solution for the second iteration. This process using TIPA can
generate less number of iterations than the Sphere Method.

BIBLIOGRAPHY

[1] Dantzig, G.B. Programming of Interdependent Activities: II Mathematical


Model, Econometrica Vol. 17, pp. 200-211, Jul. - Oct., 1949.
[2] Domingo, J.H. and Baguio, C.B., On The Sphere Method of Linear Programming Using Multiple Interior Points Approach, World Academy of
Science, Engineering and Technology, Vol:6, 2012-12-20
[3] Gass, S.I. Linear Programming, Mc Graw-Hill Book Company, 5th Edition,
New York.
[4] Murty, K.G. A New Practically Efficient Interior Point Method for LP,
Algorithmic Operations Research Vol.1, 3-19, 2006.
[5] Murty, K.G. New Sphere Methods for Linear Programs, Tutorials in Operations Research, INFORMS 2009.
[6] Murty, K.G. and Oskoorouchi, M.R. Note on Implementing the New Sphere
Method for LP using Matrix Inversions Sparingly, Optimization Letters,
Vol.3, Issue 1, pp. 137-160, January 2009.
[7] Murty, K.G. Linear and Combinatorial Programming, John Wiley and
Sons, Inc. 1976.
[8] Nash, S.G. and Sofer A. Linear and Nonlinear Programming, The Mc
Graw-Hill Book Companies Inc., New York, 1996.
[9] Schrijver, A. Theory of Linear and Integer Programming, John Wiley and
Sons, Inc, 1986.

Вам также может понравиться