Beruflich Dokumente
Kultur Dokumente
r,(x,)=p,(1-e-~'"),
This paper discusses the classical resource allocation problem. By very elementary arguments on Lagrangian duality it is
shown that this problem can be reduced to a single one-dimensional maximization of a differentiable concave function.
Moreover, a simple graphical method is developed and applied
to a family of well-known problems from the literature.
p,,k,>O
2. An example
1. Introduction
(4)
t--]
s u b j e c t to
F ( X ) = max ~ r,(x,),
(1)
subject to
x.
O<~x,<~X
(5)
for i = 1,2 ..... N,
(6)
Ex,=x.
(21
t=l
a,~x,<<,b,,
i = 1 ..... N,
(3)
North-Holland
European Journal of Operational Research 17 (1984) 31 - 34
32
subject
to
~. xt(u* ) = X,
(7)
i-I
0 = logo ( u ) ,
x,*=A,+ B,a,
s,=
or
log, (u)
"k,
k,
log, ( piki )
k,
'
(12)
(13)
Ex*(O)=X
(14)
t--]
log, (p,k,)
(11)
where
A, =
O~x,<~X.
(10)
(9) becomes
(8)
subject to
x" =
(9)
and the function (9(X) obtained from (14) is piecewise linear. The graph of (9(X) is found by identifying the points at which (9(X) changes its slope.
Once we have found 0(X), given X we can find
0", that replaced in the graph of x,*((9) will give
the optimal values of x,*. This straightforward
approach is illustrated in the next section by solving a numerical example.
X ~
Fig. I. T h e f u n c t i o n x,*(u).
R. V.V. Vidal
33
Table 1
A numerical example; X = l0 b. N = 4
Activity
p,
k,
p,k,
3-106
3-10 -~
9
4.106
2.10 -6
8
2.106
10 -6
2
106
10 -6
1
.-.* 1 -
Table 2
A and B parameters
:</I0;:
Activity
A,
- B,
0,73.106
0,33.106
1,04.106
0,5.106
0,69.106
106
0
100
-1
x~ = 5.76- 105,
-1
X,/lO ~
1
2
3
4
5
6
7
8
2.21
2.08
0.69
0.08
0
- 0.31
-0.818
-1
0
0.0436
1.1973
2.3136
2.42
3.142
3.818
4
x~' = x~ = 0.
Notice that we are able to generate easily the
optimal solution for all permissible values of X,
and the numerical values can directly be read from
the graphs or easily calculated, if more precision is
needed, using a pocket calculator. Moreover, our
approach is not an algorithm but a straightforward
procedure based on some geometrical properties of
the problem that are easy to understand and therefore easy to implement.
Table 3
The function 0 ( X )
xl2:i
4. A family of problems
The approach described in Section 3 to solve
problem (4)-(5) can be applied to other problems
of the type (1)-(3) as far as we are able to find
similar suitable transformations as (10) and (11).
Table 4 gives other examples from the literature
where such transformations can be found.
34
Table 4
Other examples
r, ( x, )
A,
B,
Reference
p, (1 - k , / x , )
__1
v/p~,
Srikantan [9]
p ,.k , > 0
~[ff
- p,x, t
p, > O,k > 2
u I/*-1
( - p , xf )x l / k
p , / ( a , - x, )
p ,,a , > 0
__1
Ilrff
a,
-V~ ,
Jensen [31
p,
Luss and
Gupta [5]
p, logc(1 + m , x i)
p,.m , > 0
p , x , / ( a , + x,)
p,.a, > 0
m~
__1
- m,
Luenberger 161
p2
Sivazlian and
Stanfel [81
2p,~,
p, > 0
Klein {4I
- I
U2
5. Conclusions
[3]
[41
[5]
[6]
[7]
[8]
[9]
[10]
References
[11]
[1] G.R. Bitran and A.C. Hax, On the solution of convex
knapsack problems with bounded variables, Proc. 9th International
Mathematical
Programming
Symposium,
[12]