Anda di halaman 1dari 8

International Journal of Computer Information Systems,

Vol. 3, No.1,2012
A Generalized algorithm to Solve Emergency Service
Location Problems

Assem A.Tharwat
Canadian International College
Higher Institute of Engineering and Business Technology
Cairo, Egypt
Assem_tharwat@cic-cairo.com


Abstract In this article, a problem from the class of emergency
service location problems, which were studied by various authors
e.g. [2, 3], is considered. The general version of the considered
problem is NP-hard [3] but there exist special cases, that can be
solved by an efficient algorithm [9, 10]. The problem, which is
going to study here, can be interpreted also as a finite version of a
ball covering problem. We shall introduce a generalized algorithm
(in sprit of [4]) to deal with such special version of the location
problem, and finally an illustrative numerical example is given.

Keywords Emergency location; service location; Max-separable
optmization
I. INTRODUCTION
P-center problems are among the most important
problems in location theory. They can be used in models
involving the location of emergency facilities (like fire stations,
hospitals, police stations, etc.), locations of radio (TV) stations,
or in many equity models in economics [2].
The algebraic structure for the formulation and solution of
those problems involving the determination of routes on
networks arise in many different contexts. Examples for those
problems are: Networks flow problems in operations research,
Critical path analysis in project management, certain scheduling
problems, path finding problems of different kinds arise in the
design of logic networks, etc.
In this article one type from the class of emergency
service location problems is considered, which could be
considered as a generalization to that was introduced in [10].
The study motivated by the fact that "In many real life situations
the formulated problem accompanied with an empty set of
feasible solutions" to create a generalized algorithm (in spirit of
[4]), and the modified problem from the class of emergency
service location problems, which were studied in [5,6,7,8,9].
This algorithm deals with such special version of the location
problem, when the emptiness of the solution set arises.
An illustrative example in which the given algorithm is
applied, and some related points for future research are
mentioned at the end section. Description & Formulation of the
Problem
A. Description
Consider the following general n-center emergency location
problem. Suppose that there are m fixed points (places) Y
1
, Y
2
,
., Y
m
. These points have to be served (or supplied) from n
emergency service centers S
1
, S
2
, ., S
n
.












Figure I: n-center emergency location problem

Each service center S
j
has to be placed on a segment A
j
B
j
,
where A
j
& B
j
are given points and the distances d
j
= | A
j
B
j
| &
a
ij
= |Y
i
A
j
| & b
ij
= |Y
i
B
j
|, are known for all i e S ~ {1,2, ., m}
and for all j e N ~ {1,2, ., n}. Denote x
j
= |A
j
S
j
| i.e. x
j
represents the distance between A
j
and S
j
which has to be
chosen and then (as shown in the opposite figure) |S
j
B
j
| = d
j
-
x
j
and 0 s x
j
s d
j
for all j e N.
In planning emergency transport systems, it is natural to
evaluate any given proposal by the worst service it provides.
The proposal for which the worst service is as good as possible
is then accepted.
The above idea was used in the process of the
formulation of the optimization problem we are going to solve
here [5,6]; in other words, if the server is placed in a point S
j

with x
j
= |A
j
S
j
| and he must serve the point Y
i
, then he must
choose the shorter one from the given two possible routes from
S
j
to Y
i
namely the route S
j
A
j
Y
i
with the length x
j
+ a
ij
and the
other route S
j
B
j
Y
i
with the length c
ij
- x
j
, where c
ij
= d
j
+ b
ij
.
Therefor the distance (say r
ij
) which must be covered in the case
when point Y
i
is served from the center S
j
, could be represented
by the following relation:
i e S , j e N r
ij
= min {x
j
+ a
ij
, c
ij
- x
j
} ~ (x
j
+ a
ij
) . (c
ij
-
x
j
) (1)
Assume that also it is allowable for the decision-maker
to put some restrictions on the server positions, S
j
s. In other
words there are two given bounds for the decision variable x
j
,
say h
j
& H
j
i.e. the variable x
j
satisfies
0 s h
j
s x
j
s H
j
s d
j
; j e N. (2)
January Page 1 of 63 ISSN 2229 5208
International Journal of Computer Information Systems,
Vol. 3, No.1,2012
B. Formulation
Assume that, the applicant for service Y
i
is served by
the service center S
k;
k e N with the shortest distance which
must be covered from S
j
to Y
i
over all j e N. This minimum
distance is equal to, say f
i
, i.e.
f
i
(x) = min
j e N
r
ij
(x
j
) = r
ik
(x
k
) for some k e N. (3)
Assume also that the minimum distance f
i
(x) for each
applicant Y
i
must be no more than a given scalar o
i
, i.e.
f
i
(x) s o
i
i e S. (4)
Now define the function F(x) which equals the
maximum of these minimum distances over all i e S, i.e.
F (x) = max
i e S
{f
i
(x)} ~ max
i e S
{min
j e N
r
ij
(x
j
)} (5)
So we want to find that vector x for which the function
F(x) takes its minimum. Given that F(x) less than a given
parametero, which is equal to the maximum of the o
i
's ; i e S
and under the set (2) of constraints i.e.:
F (x) Min
Subject to
F(x) s o

,
h
j
s x
j
s H
j
; j e N. (P1)
The problem P1 is obviously non-convex and non-
differentiable optimization problem and it is NP-hard [3]. The
investigation of the described problem needs to assume some
excess assumptions to be no more NP-hard problem, under those
assumptions an efficient technique can be suggested to solve it.
Assumption I:
Assume that for each fixed je N, say j*, there exists a
permutation (say t
j*
), (i
1
(j*), i
2
(j*), ., i
m
(j*)) ~ (i1,i2, ., im)
of indices 1,2, ., m such that:
(a
i1 j*
,b
i1 j*
) s (a
i2 j*
,b
i2 j*
) s ...s (a
im j*
,b
im j*
) (6)
Where (a, b) s (c, d) if and only if a s c and b s d.
This assumption can be interpreted as follows: for a fixed
segment A
j
B
j
the permutation t
j*
rearrange the distances of
both A
j
, B
j
from the point Y
i
; i e S in ascending order i.e. cases
like in (Figure II) are considered and those cases like in (Figure
III) are dropped.
In Figure II: m = 4
Y
1
(4,4) Y
3
& Y
2
(2,3) Y
1

Y
3
(7,9) Y
4
& Y
4
(3,3) Y
2

In Figure III: m = 3
Y
1
(3,4) & Y
2
(7,2) & Y
3
(2,5)
We cannot rearrange them as in Figure II.











Figure II: the permutation t
j*
rearrange the distances of both A
j
, B
j
from the
point Y
i
; i e S in ascending order












Figure III: the permutation t j* doesnt rearrange the distances of both Aj, Bj
from the point Yi; i e S in ascending order.

Assumption II:
Assume that for each fixed je N, say j*, there exists a
permutation (say t
j*
), (i
1
(j*), i
2
(j*), ., i
m
(j*)) ~ (i1,i2, ., im)
of indices 1,2, ., m such that: the following sets V
i1 j*
(),V
i2 j*
(), ..,V
im j*
(), are nested , i.e.
V
i1 j*
() _ V
i2 j*
() _ ..._ V
im j*
(), (7)
Where V
ij
() = {x
j
: h
j
s x
j
s H
j
&

r
ij
(x
j
) s }, is a scalar
parameter.
For a fixed segment A
j
B
j
the permutation t
j*
rearrange
the sets V
ij
(), i e S in an ascending order. In other words, for
non-empty V
ij*
() ( i e S) the set of points for which the
distances of both A
j
, B
j
from the point Y
i
are smaller than some
fixed parameter , can be determined. Graphically speaking,
consider the cases like in Figure IV and drop the cases like in
Figure V; S = {1,2} & N = {1}.
January Page 2 of 63 ISSN 2229 5208
International Journal of Computer Information Systems,
Vol. 3, No.1,2012

Figure IV: For a fixed segment A
j
B
j
the permutation t
j*
rearrange the sets V
ij
(), i e S in an ascending order


Figure V: For a fixed segment A
j
B
j
the permutation t
j*
doesnt rearrange the
sets V
ij
(), i e S in an ascending order

Define the set M (o) = {x: h s x

s H and F (x) s o},
reformulate P1 as follows:
o Min Subject to M (o) = | (P2)
M could be interpreted as follows:
It is the set of those locations (x
1
, x
2
, ., x
n
) of the service
centers S
1
, S
2
, ., S
n
, for which the performance quality of
emergency service system, is "good enough ". Good enough in
the sense that the distance which must be covered to serve any
of the points Y
1
, Y
2
, ., Y
m
is not greater than a given scalar o.
II. THEORETICAL RESULTS & SOLUTION OF THE PROBLEM
In this section some notions, results, and a modified
algorithm to that one which given in [6] are introduced.
Define the following parameters i e S, j e N:
o
ij
= min { r
ij
(x
j
) : h
j
s x
j
s H
j
}= min { r
ij
(h
j
) , r
ij
(H
j
)} (8)

ij
= max { r
ij
(x
j
) : h
j
s x
j
s H
j
} = 1/2 ( c
ij
+ a
ij
) (9)
Let us introduce the following four results:
- The first assumption implies the second one.
- V
ij
() = | if and only if o
ij
> , see Figure VI.
- If o
ij
s s
ij
then there exist always two points x*
& x** such that: x* s x** and r
ij
(x*) = r
ij
(x**) = ,
see Figure VII.
Also x* = x** if and only if =
ij
.

- V
ij
() = [h
j
, H
j
] if and only if
ij
s

, see Figure
VIII.
The proofs of the above results were given in [6].








Figure VI: the emptiness of V
ij
()

Figure VII: the representation of V
ij
() as the union of two nonintersecting sets

Figure VIII: the set V
ij
() is the whole interval
The following three theorems give the necessary and
sufficient conditions for both the emptiness of the set V
ij
(o
i
),
and the emptiness of set M (o).

Theorem 1
Assume that
ij
> o
i
for some i e S & j e N then,
V
ij
(o
i
) = |; i e S, j e N if and only if o
i
< min
{a
ij
+ h
j
, c
ij
- H
j
}.
Proof:
Assume that V
ij
(o
i
) = | for some i e S, j e N, this happen if and
only if r
ij
(x
j
) > o
i
x
j :
h
j
s x
j
s H
j
, which equivalent to say
that
x* < h
j
& x** > H
j
.
(It is clear that: x* = o
i
- a
ij
and x** = c
ij
- o
i
), this leads to
January Page 3 of 63 ISSN 2229 5208
International Journal of Computer Information Systems,
Vol. 3, No.1,2012
V
ij
(o
i
) = | o
i
- a
ij
< h
j
& c
ij
- o
i
> H
j

o
i
< a
ij
+ h
j
& o
i
< c
ij
- H
j

o
i
< min {a
ij
+ h
j
, c
ij
- H
j
}. Q.E.D.
Corollary
Given that
ij
> o
i
, then,
min {o
i
: V
ij
(o
i
) = |} = min {a
ij
+ h
j
, c
ij
- H
j
}, i e S, j e N.

Theorem 2
Assume that the second assumption is fulfilled, then
{- seS such that V
sj
(o) = | j e N } M (o) = |.
Proof:
Assume that V
sj
(o) = | j e N for some seS, this means that
{r
sj
(x
j
) > o x
j :
h
j
s x
j
s H
j
}

j e N,
i.e. min
j e N
{r
sj
(x
j
) }

> o

; for any x
j
e [h
j
, H
j
], j e N
then F(x) = max
i e S
{ min
j e N
r
ij
(x
j
)}
> min
j e N
{r
sj
(x
j
) }

> o ; for any x
j
e [h
j
, H
j
], j e N
M (o) = |. Q.E.D.
Theorem 3
Assume that the second assumption is fulfilled, then
M (o) = | {- seS such that V
sj
(o
s
) = | j e N}.
Proof:
Assume that h
j
s x
j
s H
j
j e N. Assume the negation of the
right-hand side i.e.
i e S - j(i) e N such that V
ij(i)
(o
s
) = |, this means that:
i e S - j(i) e N such that {r
ij(i)
(x
j
) s o
s
x
j
},hence


i e S we have, min
j e N
{r
ij
(x
j
) }

s r
ij(i)
(x
j
) s o
s
x
j
,
and since
F(x) = max
i e S
{ min
j e N
r
ij
(x
j
)} = { min
j e N
r
sj
(x
j
)} for
some s e S, hence we can conclude that F(x) s o. This
completes the proof.
The following theorem gives the minimal o for which we got a
solution of P2.
Theorem 4
Let o
min
is the optimal value of P2. Then
o
min
= max
i e S
o'
i
, (10)
where o'
i
= min {: V
ij
() = | for at least one j e N}; i e S,
i.e.

o'
i
= min
j e N
{ min { a
ij
+ h
j
, c
ij
- H
j
}. (10*)

Proof:
It is obvious from the above results. Q.E.D.

We can summarize the proposed numerical procedure for
solving P1 in the following algorithm.

Algorithm I
STEP1 Input m, n, o
i
, a
ij
, c
ij
, h
j,
, for each j e N & for
each i e S. If assumption I doesn't fulfilled, go
to step STEP9.
STEP2 Calculate o = max
i e S
o
i
.
STEP3 Calculate o'
i
= min
j e N
{ min { a
ij
+ h
j
, c
ij
- H
j

} for each i e S.
Calculate o
min
= max
i e S
o'
i
.
STEP4 If o > o
min
i.e. M (o) = |, determine for each i e S an
index j(i) e N such that
V
ij(i)
(o'
i
) = |. Otherwise, go to step STEP9.
STEP5 Define the index set P
j
(o
min
),
where P
j
(o
min
) = {i : i e S; j(i) = j} for each j e N.
STEP6 Find V
j
(o
min
)

for each j e N,
where V
j
(o
min
)

= V
ij
(o'
i
) ; where i e P
j
(o
min
) = |.
STEP7 Choose the optimal decision vector x
opt
= (x
1
, x
2
, .,
x
n
) as follows:
x
j
e V
j
(o
min
)

if P
j
(o
min
) = |, and x
j
e [h
j
, H
j
]
if P
j
(o
min
) = |.
STEP8 o
min
is an optimal solution of P2, and x
opt
is
one of the optimal solutions of P1. Stop.
STEP9 There is no solution of the problem. Stop.
It is important to close this section by the following remarks.
Remark 1
To apply the above algorithm we should check whether
the first (and consequently the second) assumption is fulfilled or
not (step STEP1). If the answer is positive the emptiness of the
set M(o) (step STEP4) should be checked. If the answer of
the second question [Is M (o) is non-empty?] is "yes", the
solution of the problem will be obtained via the remaining steps
of the algorithm.
In the next section, the negative answer of the second
question will investigated, i.e. under which conditions the set M
(o) is empty.


January Page 4 of 63 ISSN 2229 5208
International Journal of Computer Information Systems,
Vol. 3, No.1,2012
Remark 2
The emptiness of the set M (o) means the emptiness of the
sets
V
sj
(o
s
) = | j e N for the s
th
row.
But if we have for some s e S: V
sj
(o
s
) = | j e N, this
means that
r
sj
(x
j
) > o
s


j e N; h
j
s x
j
s H
j

Remark 3
Theoretically, there are three possibilities to deal with the
problem of the emptiness of the set M (o).
- The first one is that, we shall assume that the maximum
distance, for serving any of the points Y
i
, belongs to the
interval [o
i
- t , o
i
+ t], where t is a non-negative given
real value. Our task will be, in this case, the obtaining in
a certain sense optimal point o
opt
from the interval [o - t ,
o + t] for which M (o
opt
) = |.
- The second possibility concerning the distances from the
end points of the line segment A
j
B
j
to any of the points
Y
i
. Assume that these distance running in the intervals
[a
ij
- t, a
ij
+ t] & [b
ij
- t, b
ij
+ t], where t is a non-negative
given real value. We have to find in a certain sense
optimal a
ij
' s & b
ij
' s (from the assumed intervals) for
which M (o) = |.
- The third direction of handling the emptiness problem is
to assume the variability of the end points of the intervals
[h
j
, H
j
], i.e. assuming that h
j
*e[h
j
- t, h
j
+ t] & H
j
*e[H
j
-
t, H
j
+ t]; j e N. Finding optimal h
j
's & H
j
's in a certain
sense (from the assumed intervals) for which M (o) = |
is required.

III. MORE THEORETICAL RESULTS & GENERALIZED ALGORITHM

Consider the following system of equations and inequalities
g
i
(x) = min
j e N
r
ij
(x
j
) = |
i
i e S. (11)
h
j
s x
j
s H
j
; j e N
Using the vector notation, the system (11) could reformulated as
follows:
G(x) = |, h s x

s H (12)
Where h = (h
1
, h
2
, ., h
n
), H = (H
1
, H
2
, ., H
n
), x = (x
1
, x
2
, .,
x
n
), and | = (|
1
, |
2
, ., |
n
).
Denote the set of all solutions of the system (11) (or
(12)) by M (|), and each component of G(x) is considered as a
function of n variables. The vector | in the system (12) can be
understood as a vector, which is attained by the left-hand side G
(x) when an approximate x e M (|) is chosen. Therefor those
|'s, for which M (|) = |, are called the attainable elements, and
the set
A = {|: M (|) = |}, (13)
is called the attainable set.
If for a given system; |* e A, then there exists a
solution of that system, with | = |* which can be obtained using
some of the method described in the literature (see e.g. [1]).
If |*e A, we have to find |
ner
e A, which has in some
sense the minimal distance from |* and accept the elements of
M (|
ner
) as an appropriate approximate solutions.
For the above purpose, consider the following
Tshebyshev distance [4]:
,, | - |*,, = max
ie S
|
i
- |
i
* (14)
and solve the following problem:
|
i
- |
i
* Min
Subject to | e A (15)
Or equivalently solve the following problem:
,, G (x) - |*,, = max
ie S
g
i
(x) - |
i
* Min
Subject to h s x

s H (P3)
There exists always at least one optimal solution x
opt
of the
above system (it is a minimization of a continuous function of x
on a compact set). Thus |
opt
= G (x
opt
) is an optimal solution of
problem (15), where |
opt
is the nearest | for which M (|) = |.
To solve (P3) let us define the set M (t) for any t eR
>0
as
follows:
M (t) = {x: h s x

s H and ,,G (x) -|*,, s t} (16)
It suitable to replace problem (P3) by the following one:
t Min
Subject to M (t) = | (P4)
so there exists always an optimal solution t
opt
to the problem
(P4) [4].
Define also the set V
ij
(t), where i e S & j e N, for any t eR
>0
as
follows:
V
ij
(t) = {x
j
: h
j
s x
j
s H
j
&

r
ij
(x
j
) s |
i
* + t},
Then it is easy to calculate the following intersection:
V
j
(t) = V
ij
(t); i e S, for each j e N.
Finally define the following parameters i e S , j e N:
q
ij
= max { min (r
ij
(h
j
) , r
ij
(H
j
)) - |
i
* , zero} (17)
q
i
= min
j e N
q
ij
& q

= max
ie S
q
i
(18)
Theorem 5
M (t) = | if and only if - seS such that V
sj
(t) = |, j e N.
Proof:
January Page 5 of 63 ISSN 2229 5208
International Journal of Computer Information Systems,
Vol. 3, No.1,2012
Assume that h
j
s x
j
s H
j
j e N, t > 0. Assume the negation of
the right-hand side i.e.
i e S - j(i) e N such that V
ij(i)
(t) = |
i e S - j(i) e N such that {r
ij(i)
(x
j
) s |
i
* + t , x
j
}
i e S min
j e N
r
ij
(x
j
)

s r
ij(i)
(x
j
) s |
i
* + t , x
j

i e S min
j e N
r
ij
(x
j
) s |
i
* + t s (max
i e S
|
i
* ) + t =
|* + t
F(x) = max
i e S
{ min
j e N
r
ij
(x
j
)} s |* + t
M (t) = | Q.E.D.
Theorem 6
For any s e S : V
sj
(t) = | if and only if t< q
sj
.
Proof:
Assume that V
sj
(t) = | for some seS, this equivalent to:
{r
sj
(x
j
) > o
i
* + t x
j :
h
j
s x
j
s H
j
}; t > 0, then,
min{r
sj
(x
j
) , h
j
s x
j
s H
j
}

> o
i
* + t ; t > 0
min{ r
ij
(h
j
) , r
ij
(H
j
)} > o
i
* + t ; t > 0
recalling equation (8)
t < min { r
ij
(h
j
) , r
ij
(H
j
)} - o
i
* ; t > 0, hence
t < max { min (r
sj
(h
j
) , r
sj
(H
j
)) - o
i
* , zero},
it means t < q
sj
. Q.E.D.

Theorem 7
For any s e S : V
sj
(t) = |, j e N if and only if t < q
s
.
Proof:
Assume that V
sj
(t) = | j e N for some seS,
t < q
sj
j e N for some seS,
t < min
j e N
q
sj
= q
s
for some seS, Q.E.D.

Theorem 8
M (t) = | if and only if t < q.
Proof:
It is clear from theorem 5, and the definition of q in (18).

Corollary
M (t) = | if and only if t > q.
Remark 4
Recalling the above algorithm (STEP1-STEP9) and the
above results, one can suggest the following generalization.
In step STEP4, let us assume that we have the case o*
< o
min
, this means that M (o*) = | for the given parameter o*,
i.e. there is no solutions of P2 and consequently there is no
solutions of P1.
Assume also that, it is available to perturb the given o*
in some interval, say [o*-t, o*+t], for some chosen t > D, where
D = max
i e S
o'
i
- min
i e S
o
i
(19)
It clear from the definition of t that o*+t > o
min
(see lemma 1).

Lemma 1
By choosing the parameter t according to relation (19),
then it is easy to prove that:
o*+t > o
min
, where o* < o
min
and o
min
is given by (10).
Proof:
From (19); t > D = max
i e S
o'
i
- min
i e S
o
i

= o
min
- min
i e S
o
i
> o
min
- max
i e S
o
i

= o
min
- o*,
which completes the prove.


Hence, according to the above results ( theorem 5 8), choosing
of t
opt
such that:
t
opt
= max
ie S
min
j e N
t
ij

where t
ij
= max{ min (r
ij
(h
j
) , r
ij
(H
j
)) - o
i
* , zero} (20)
guarantees that (see theorem 9 later):
o** > o
min
and o** e [o*- t, o*+ t] (21)
i.e. M (o**) = |, where o** = o*+ t
opt

and we can continue the remaining steps of the algorithm after
the updating of the value of the perturbed parameter o* by the
new value o**.
So the modified algorithm can be summarized follows:
STEP1 Input m, n, o
i
, a
ij
, c
ij
, h
j,
, for each j e N & for
each i e S.
If the first assumption doesnt fulfill, go to
step STEP11. Otherwise, continue.
STEP2 Calculate o = max
i e S
o
i
.
STEP3 Calculate o'
i
= min
j e N
{ min { a
ij
+ h
j
, c
ij
- H
j

} for each i e S.
Calculate o
min
= max
i e S
o'
i
.
STEP4 If o > o
min
i.e. M (o) = |, determine for each i e S an
index j(i) e N such that
V
ij(i)
(o'
i
) = |, go to step STEP7. Otherwise, continue.
STEP5 Input t. Calculate D = o
min
- min
i e S
o
i
. If t >
D, then continue. Otherwise, ask the decision
January Page 6 of 63 ISSN 2229 5208
International Journal of Computer Information Systems,
Vol. 3, No.1,2012
maker to update the value of t, and if it is not
possible go to step STEP11.
STEP6 Calculate t
opt
, t
opt
is given by (20). Let o
min
= o+ t
opt
.
Determin for each i e S an index j(i) e N such that
V
ij(i)
(o
i
+ t
opt
) = |.
STEP7 Define the index set P
j
(o
min
),
Where P
j
(o
min
) = {i : i e S; j(i) = j} for each j e N.
STEP8 Find V
j
(o
min
)

for each j e N, where V
j
(o
min
)

= V
ij

(o
i
+ t
opt
) ; and i e P
j
(o
min
) = |.
STEP9 Choose the optimal decision vector x
opt
= (x
1
, x
2
, .,
x
n
) as follows:
x
j
e V
j
(o
min
)

if P
j
(o
min
) = |, and x
j
e [h
j
, H
j
]
if P
j
(o
min
) = |.
STEP10 o
min
is an optimal solution of P2, and x
opt
is
one of the optimal solutions of P1. Stop.
STEP11 There is no solution of the problem. Stop.
Now the satisfaction of the relation (21) will give in the
following theorem.
Theorem 9
Assume that o* < o
min
, then o** = o* + t
opt
satisfies the
following two relations:
1. o** > o
min
,
2. o** e [o*-t, o*+t],
where t & o
min
are given by (19) & (10) respectively, and o* is
an arbitrary scalar smaller than o
min
.
Proof:
The first relation is equivalent to saying that:
M (o**) = | and M (o*) = | where o** = o*+t
opt
.
Now M (o*) = | means that: there exists at least one row (say
the s
th
row) for which V
sj
(o*) = | ; h
j
s x
j
s H
j
for each
column. From the definition of the sets V
ij
(o*), its emptiness
means that:
{r
sj
(x
j
) > o* x
j
: h
j
s x
j
s H
j
} je N. (22)
Since in general
t
opt
> min
j e N
max{ min (r
ij
(h
j
) , r
ij
(H
j
)) - o
i
* , zero},
then t
opt
> min
j e N
{min {r
ij
(h
j
) , r
ij
(H
j
)} - o* } ,
which could be written as follows
t
opt
> min {r
sn
(h
n
) , r
sn
(H
n
)} - o* , i.e.
o* + t
opt
> min {r
sn
(h
n
) , r
sn
(H
n
)} ; n e N.
Hence V
sn
(o*+ t
opt
) = V
sn
(o**) = |
It means that for the s
th
row we found an element j(s) =
n e N such that V
sn
(o**) = |. The same can be done for the
remaining rows (if exist) for which we have relations like (22),
and the rows for which V
ij
(o*) = | there is no problem. Then
i e S - j(i) e N such that V
ij(i)
(o**) = |. This completes the
proof of part (1).
To proof the second part, it is enough to prove that t
opt
< t for
any chosen t according to (19) and t
opt
is given by (20).
Let t > o
min
- min
i e S
o
i
, then t + min
i e S
o
i
> o
min

t + min
i e S
o
i
> max
i e S
o'
i
= max
i e S
min {: V
ij
() = | for at least one j e N}
= min{: V
sj
() = | for at least one { j e N} for some se S
= min
j e N
{ min { a
sj
+ h
j
, c
sj
- H
j
} for some se S

> min { a
sj
+ h
j
, c
sj
- H
j
} for all je N, for some se S
= min { r
sj
(h
j
) , r
sj
(H
j
) } for all jeN, for some se S
t > min { r
sj
(h
j
) , r
sj
(H
j
) }- min
i e S
o
i

> min (r
sj
(h
j
) , r
sj
(H
j
)) - o
s
* for all je N, for some se S
> min
j e N
max{ min (r
sj
(h
j
) , r
sj
(H
j
)) - o
s
* , zero} for
some se S
= max
ie S
min
j e N
max{ min (r
ij
(h
j
) , r
ij
(H
j
)) - o
i
* , zero}
t > t
opt
Q.E.D.
IV. NUMERICAL EXAMPLE
(STEP1)
Assume the following situation: m = n = 5
d = (3, 3, 3, 5, 5) & h = (1, 1, 2, 1, 3) & H = (2, 2, 4, 5, 6) &
o = (1, 2, 1, 2, 1)
, (a
1j
, b
1j
) , = , (1, 1) (5, 6) (3, 5) (2, .5) (1, 3) ,,
, (a
2j
, b
2j
) , = , (2, 2) (4, 4) (4, 6) (3, 5) (2, 4) ,,
, (a
3j
, b
3j
) , = , (3, 3) (3, 2) (5, 7) (5, 9) (4, 7) ,,
, (a
4j
, b
4j
) , = , (4, 5) (2, .5) (1, 3) (2, 1) (5, 7) ,,
, (a
5j
, b
5j
) , = , (5, 7) (1, .5) (2, 4) (4, 7) (3, 5) ,.
It is clear that the first assumption (consequently the second one)
is fulfilled since (a
ij
, b
ij
) satisfies equation (6) for each i and j.
(STEP2) & (STEP3) & (STEP4)
o = max
i e S
o
i
= 2 < o
min
= max
i e S
o'
i
= 3, where o'
i

calculated from equation (10*) for each i.

(STEP5)
If t takes the value 2, D = o
min
- min
i e S
o
i
= 2. Then t s D.
Change t to the value 2.5


January Page 7 of 63 ISSN 2229 5208
International Journal of Computer Information Systems,
Vol. 3, No.1,2012
(STEP6)
t
opt
= max
ie S
min
j e N
max{ min (r
ij
(h
j
) , r
ij
(H
j
)) - o
i
* , zero}
= max {0, 1, 2, 0, 0} = 2. Then o
min
= o+ t
opt
= 4.
V
1j
(o
1
+ t
opt
) = , [1, 2] | {4} [1, 5] {3} [4, 6] ,,
V
2j
(o
2
+ t
opt
) = , [1, 2] | | {1} [5, 6] ,,
V
3j
(o
3
+ t
opt
) = , {1, 2} [1,2] | | | ,,
V
4j
(o
4
+ t
opt
) = , | [1,2] [2, 4] [1, 5] | ,,
V
5j
(o
5
+ t
opt
) = , | [1,2] {2} [3, 4] | {6} ,,

(STEP7) & (STEP8)
P
1
= {1, 2, 3} & V
1
= {1,2}
P
2
= {3, 4, 5} & V
2
= [1,2]
P
3
= {1, 4, 5} & V
3
= {4}
P
4
= {1, 2, 4} & V
4
= {1}
P
5
= {1, 2, 5} & V
5
= {6}

(STEP9)
Choose one of the following solutions:
x
opt
= (1, x
2
, 4 , 1, 6) or x
opt
= (2, x
2
, 4 , 1, 6),
where x
2
e[1,2].
V. CONCLUSION AND FURTHER POINTS
In this article a modified problem from the class of
emergency service location problems, which were studied in
[5,6,7,8,9,10] is considered. Motivated by the fact that "In many
real life situations the formulated problem accompanied with an
empty set of feasible solutions", a generalized algorithm is
introduced (in spirit of [4]) to deal with such special version of
the location problem when the emptiness of the solution set
arises.
The case when the maximum distance, for serving any
of the points Y
i
, belongs to the interval [o
i
- t , o
i
+ t], where t is
a non-negative given real value, was studied and the optimal
point in some certain sense, from the perturbation interval [o - t
, o + t], for which M (o
opt
) = | was obtained via the generalized
algorithm. The work was concluded with an illustrative
example, in which the given algorithm is applied to explain the
importance of the studied case.
The third direction of handling the emptiness problem, i.e. the
variability of the end points of the intervals [h
j
, H
j
] for each j e
N, could be considered as a point for future research. Also
another open question is: what will happen if we change the
considered Tshebyshev distance to another distance?
VI. REFERENCES
[1] Cunningham-Green, R.A.: Mini-max Algebra, Lecture
Notes in Economics and Mathematical Systems,
Springer Verlag, 1979.
[2] Drezner, Z.: On Rectangular p-Center Problem, Naval
Research Logistics, 34(1987), pp. 229-234.
[3] Francisco J.F. Silva, Daniel Serra , Locating
Emergency Services With Priority Rules: The Priority
Queuing Covering Location Problem, JEL
Classifications: C61, L80, Working Paper Series
September 2002
[4] Hudec, O.: An Alternative p-Center Problems, 1989,
Proceedings Conference "Mathematical Optimization",
Eisenach.
[5] Tharwat, A.: Solution of Optimization Problems on
Attainable Sets of Extremal Separable Operators, 1998,
Ph.D. Thesis, Faculty of Mathematics & Physics,
Charles University, Prague, Czech Republic.
[6] Tharwat, A.A and Saleh, M.M. (2006) A
comprehensive numerical algorithm for solving service
points location problems, AMC (Applied Mathematics
and Computations), 176, pp 44-57, 2006
[7] Tharwat, A.A and Hassan, E.E. (2008), Emergency
Service Location Problem with Three Possible Routes",
the EIJ (Egyptian Informatics Journal), Vol.8, No.2, pp
287-308, December 2008
[8] Tharwat, A.A. , Saleh M.M. and Khater H.M. (2011)
An Algorithm To Solve Generalized Service Location
Problems International Journal of Computer
Information Systems, Vol. 2, No. 6, pp 21 - 21 , June,
2011
[9] Tharwat, A.A., El-Khodary, I.A.F., and Hassan, E.E.
Penalized Emergency Location Problem with Three
Possible Routes, The Egyptian Math. Society, Special
Volume, Article in Press
[10] Zimmermann, K.: Max-Separable Optimization
Problems with Uni-modal Functions, Ekonomicko-
matematicky obzor, 27(1991), N0.2, pp. 159-169.
[11] Zimmermann, K.: A Parametric Approach to Solving
One Location Problem with Additional Constraints,
1993, Proceedings of the International Conference on
Parametric Optimization and Related Topics III
Approximation Optimization, pp. 557-568.

AUTHORS PROFILE

Prof. Dr.Assem THarwat received his Ph.D degree
from Charles University, in 1998. His main field of
interest is the mathematical programming ,the
Management Science and Decision Support applications.
He got his Master degree in the field om Mathematical
Statistics from Cairo University 1989. He is a member of
the Operations Research group in both Egypt and Czech
Republic.




January Page 8 of 63 ISSN 2229 5208

Anda mungkin juga menyukai