An International Journal

computers &
mathematics
with applications
PERGAMON Computers and Mathematics with Applications 38 (1999) 143-149
www.elsevier.nl/locate/camwa

Level-Continuity of Functions and Applications

H. ROMAN-FLORES*
Departamento de Matem~tica, Universidad de Tarapac~
Casilla 7D, Arica, Chile
M. ROJAS-MEDAR
IMECC-UNICAMP, CP6065, 13081-970, Campinas-SP, Brasil
(Received May 1998; revised and accepted October 1998)

A b s t r a c t - - I n this paper, we study the concepts of level-continuity and proper local maximum
points of functions defined on a topological space X and, on one hand, we establish that, under
adequate conditions, f is level-continuous if f is without proper local maximum points and, on the
other, we prove that level-convergence and variational convergence (F-convergence) of functions are
equivalents when the limit function is level-continuous. (~) 1999 Elsevier Science Ltd. All rights
reserved.

K e y w o r d s - - T o p o l o g i c a l spaces, Kuratowski limits, Variational convergence.

1. I N T R O D U C T I O N
T h e s t u d y of t h e variational convergence and its applications has been done by m a n y authors,
including De Giorgi and Franzoni [1] and A t t o u c h [2] in the setting of the calculus of variations,
Greco [3] a n d R o j a s and Rom~n-Flores [4] in convergence of fuzzy sets on locally c o m p a c t metric
spaces a n d finite-dimensional spaces, respectively.
This convergence is based on t h e Kuratowski limits and one of t h e most i m p o r t a n t properties of
t h e F-convergence is the preservation of m a x i m u m points in F-convergents sequences of functions.
More precisely, let {fn}n be a sequence of real functions on X and let Xn be a m a x i m u m point
of f~. If fn r f and xn --* x, then x is a m a x i m u m point of f and f(x) = limn-.oo fn(xn).
O n the other hand, level-continuity and level convergence has been used by the a u t h o r in
multivalued characterizations of certain class of m a x i m u m points of functions on R ~ [5] and
c o m p a c t n e s s of spaces of fuzzy sets on a metric space X , see [6].
T h e aims of this paper are, on one hand, to introduce t h e concept of level-continuity of functions
and to analyze its connections with the existence of proper local m a x i m u m points, and on the
other, t o c o m p a r e level-convergence (L-convergence) with F-convergence. This analysis is carried
out in the setting of regular topological spaces, and generalizes the results o b t a i n e d by the a u t h o r
in [5,6].

This work was partially supported by Fondecyt (Chile) through Project 1970535, and Diexa-UTA by Project
4732-97.
The authors would like to thank the referees for their valuable suggestions and comments.
*Author to whom all correspondence should be addressed.

0898-1221/99/$ - see front matter. (~) 1999 Elsevier Science Ltd. All rights reserved. Typeset by .AA/eS-'I~X
PII: S0898-1221 (99)00190-X
144 H. ROM.~N-FLORES AND M. ROJAS-MEDAR

This paper is organized as follows. In Section 2, we give the previous results that will be
used in the article. In Section 3, we introduce the concept of level-continuity of nonnegative real
functions defined on X and we study its connections with the existence of proper local maximum
points.
Finally, in Section 4, we compare L-convergence with F-convergence. Furthermore, some ex-
amples are presented.

2. PRELIMINARIES
In the sequel, all topological sp~ces will be assumed to be regular, unless specifically stated
otherwise.
DEFINITION 2.1. Let (X, T) be a topological space and let {An}heN a Sequence of subsets of X .
(i) A point x E X is a limit point of {An}n if, for every neighborhood U of x, there is an
n E N such that for ali m > n, Am M U ¢ 0.
(ii) A point x E X is a duster point of {An}n if, for every neighborhood U of x, and every
n E N, there is an m > n such that Am n U ¢ @.
(iii) liminfAn is the set of MI limit points of {An}n.
(iv) limsupAn is the set ofail duster points of{An}n.
If liminfAn -- limsupAn = A, then we say A is the limit of the sequence {An}n, or the
sequence {An}n converges to A (in the Kuratowski sense), and we write A = limAn (or A,~ K A).
PROPOSITION 2.2. H {An}n is a sequence of subsets of X then
(i) liminfAn C_ limsupAn,
(ii) liminfAn and limsupAn are dosed subsets of X ,
(iii) lim sup An = An°°__1Uken Ak,
(iv) liminfAn = N H Ukeg Ak, where H denotes an arbitrary cofinal subset of N and the
intersection is over all such H.
For more details, see [7,8].
REMARK 2.3. We recall that H is a cofinal subset of N if Vn E N, 3 m E H such that m > n.
DEFINITION 2.4. If f : X --* [0, co) is a function and a E (0, co), then we define the a-level and
the strict a-level of f by

{ f > a} = L ~ f = {x E X I f ( x ) > a}, and

{ f > a} = {x • X I f ( x ) > a},

respectively.
We observe that a </3 implies L a f D Lt~f.
DEFINITION 2.5. I f f : X --* [0, oo) is a function, then xo E X is said to be a proper local
maximum point o f f i[O < f(xo) < supxex f ( x ) and there is a neighborhood U at xo such that
f ( x ) < f(xo), for every x E U.
DEFINITION 2.6. Let f : X ---* [0, co) be and supxex f ( x ) = M (which m a y be co). We say
that f is level-continuous if ap --* a implies L~pf g Laf, Va E (0, M).
The following examples shows that continuity and level-continiuty are independent conditions.
EXAMPLE 2.7. Let X = [0, 1] be and T the usual topology generated by the usual metric on X.
Define f : X --* [0, co) by 1
l-x, ifO<x<~,

f(x) = 1 if 1
5' ~<z<l.
 L e v e l - C o n t i n u i t y of F u n c t i o n s 145

Then it is clear that f is continuous.

On the other hand, taking av = (1/2) + (l/p), p > 2, we have that

La, f = 0, 7 - , Vp.

oo
Thus, l i m s u p n a J = Np~=lUk>pLakf = Np=lUk>p[ 0 , 1 / 2 - 1/k] = [0,1/2], whereas
L1/2f = [0, 1]. Consequently, f is not level-continuous.
We observe that all points x E (1/2, 1] are proper local maximum points.
EXAMPLE 2.8. Let (X, T) be as in Example 2.7 and f : X -~ [0, co) defined by

1, if x = 1,
f(x)= 0, ifx¢l.

Then, clearly, f is not continuous.

But, for each ~ E (0, 1), we have that L a f = {1}. Therefore, f is level-continuous.
Also, we observe that all points x ~ 1 are not proper local maximum points, since f(x) = 0
for x ¢ 1.
REMARK 2.9. We observe that f : X ---* [0, oc) it is always left level-continuous, that is, if C~p/z
then L~pf K L~f. In fact, suppose that x E Np=l Uk>p L~kf" Then

x e U L.~f, Vp. (1)

k>_p

Now, if f(x) < a, then there exists P0 such that f(x) < Sk, V k > Po.
Therefore, x ~ Lakf, Vk > Po.
Because X is regular and Lapof is closed, then there exists U(x) such that U M L~pof = 0.
But L~pof 2 Uk>po L ~ f , and consequently, U n [Uk>po L~kf] = 0, that is, z ~ Uk>po L~kf, in
contradiction with (1).
So, must be f(x) > a, and consequently, l i m s u p L a J c L~f.
For the reverse inclusion, let x E L a f and H a cofinal subset of N.
Then, x E L a f implies that f(x) > (~ > ak, Vk, and therefore, x E L~kf, Vk, which implies
X E UkeH L,~f.
So, x E NH U k e g La~ f where the intersection is over all H cofinal in N.
That is, x E liminf L ~ J . Thus, we can to conclude that l i m L ~ , f = L~f.

3. L E V E L - C O N T I N U I T Y A N D
PROPER LOCAL MAXIMUM POINTS
In this section, we shall prove that under adequate assumptions, level-continuity, and nonexis-
tence of proper local maximum points are equivalent conditions.
THEOREM 3.1. Let f : X --~ [0, c~) be with supx~xf(x) = M.
If L a f is closed V a, then the following are equivalents:
(i) f is without proper local maximum points,
(ii) { f _> s} ---- {f > s}, V s E (O,M),
(iii) f is level-continuous.
PROOF.
(i) --* (ii). Let 0 < ao < M. Then because {f >_ So} is closed, it is clear that { f > s o } C_
{ f > so}. If we suppose that {f > So} ~ { f > so}, then there exists Xo E { f > a o } \ { f > so}.
Consequently, by regularity of X, there exists U(xo) such that U n { f > so} = O.
146 H. ROM.~N-FLORES AND M. ROJAS-MEDAR

But then, f(x) < so = f(xo) < M, Vx E U. Consequently, x0 is a proper local maximum
point of f, in contradiction with our hypothesis.
(ii) --* (i). Suppose that x0 is a proper local maximum point of f.
Then, 0 < f(xo) = so < M and there exists a neighborhood U(xo) of x0 such that f(x) <
f(Xo) = ol0, VX E U.
Therefore, xo ~ { f > so} and U n {f > so} = 0.
Thus, xo q {f > a o } \ { f > C~o}, and consequently, {f > so} # {f > cro}.
(iii) --. (ii). Let a e (0, M) be. We know that { f > c~} C_ {f > a}. For the reverse inclusion, let
Xo E {f > a} and choose ~p ~ ~ (strictly).
Thus, by level-continuity of f, must be L~pf ~ L~f, that is, L ~ f = Nv~=l Uk>_pL~kf.
Now, let U(xo) be an arbitrary neighborhood of xo.
If UA { f > c~} = 0, then UA {f > c~k} = 0, V k, and this implies that UM [Uk>v L~, f] = ~, Vp.
But then, Xo ¢ Uk>vL~kf, Vp.
oo
Consequently, x0 ¢ Nv=I Uk>p Lakf = L a f which is a contradiction.
Therefore, U f3 { f > a} # ~ and x0 E {f > c~}.
(ii) --* (iii). Suppose that f is not level-continuous in c~0 E (0, M).
Then there exists a sequence {ap} such that ap --* s0 and

Lapf---* Laof. (2)

Without loss of generality, due Remark 2.9, we can to suppose ~p "~ s0 (strictly).
Thus, Lakf C Laof, Vk, and because L~of is closed, we have Uk>v Lakf C L~o f for all p, that
is,
N U Lakf = l i m s u p L a J C Lao f. (3)
p----1 k>_p
On the other hand, if x E {f > so} then there is P0 such that f(x) > ak, for all k > po.
Therefore, x E L~kf, Vk > Po, and this implies that x E UkeH "L~kf for every cofinal subset H
of N.
Consequently, x E NH UkEHLakf = lim i n f L a j .
Because lim inf La~f is closed and {f > s0} C_ lim inf La, f, we can to conclude that

{f > C~o}= L~of C_l i m i n f L ~ J . (4)

Thus, by (3) and (4), we have that L~of = l i m L ~ j which contradicts (2), and the proof of
our theorem is complete. I
REMARK 3.2. Due Theorem 3.1, we can to conclude that if f is level-continuous then any local
maximum of f is a global maximum.

4. L E V E L - C O N V E R G E N C E AND 7 - C O N V E R G E N C E
Let ~'(X) = {f : X ~ [0, c¢)/L~f closed, V~}.
DEFINITION .4.1. LEVEL-CONVERGENCE. Let fn, f e Jr(X). We say that fn level-converges
to f (for short : f,~ L f) iflimL~f= = L~f, Vs.
DEFINITION 4.2. F-CONVERGENCE. Let fn, f e :7:(X). We say that fnF-converges to f (for
short : fn r f) iflimEnd(fn) = End(f), where
End(f) = {(x,~) E X × [0, c¢) I f ( x ) > ~}.

REMARK 4.3. The concept of level convergence for fuzzy sets was introduced by Kaleva and
Seikkala in [9], whereas the concept of F-convergence above is the same used by Greco et al.
in [3] and it is analogous to the hypograph convergence used by Attouch in [2].
 Level-Continuity of Functions 147

THEOREM 4.4. Let .In, f E :F(X), f level-continuous. Then, the following conditions are equiv-
alents:
(i) f. & f,
(ii) fn r f.
PROOF.
(i) --* (ii). In order to prove that fn r f it is sufficient to prove that

lim sup End(f~) C_End(f) C lim inf End(f~).

Let (x, a) E limsupEnd(f~). Then

(x,.) N U End(fk). (5)

p>l k>p

We want to prove that (x, a) e End(f), that is, f(x) > a.

If we suppose that f(x) < a, then there is e > 0 such that f(x) < a - e < a.
So, due fn L f, we obtain that x ¢ Lc,-Ef : Np>l Uk>p L~-,fk.
This implies that 3p0 such that x ~ Uk>po Lc~-Jk, and therefore, there exists U(x) such that

Un[_>UpkoL~-*fk =0. (6)

Now, we assure that [U × (a - e)] M [Uk_>poEnd(fk)] = O.

In fact,

fl>a-e, and
(y, fl) E U x (a--e, oo) N [ U End(fk)]
Lk->po J =~ 3k0 _>P0, such that (y, fl) E End(fko).

Therefore, fko (Y) >_fl > a - e.

But, due (6), y E U(x) implies that y ¢ Uk>po L a - J k .
That is, fk(Y) < a - e, Vk > P0, which is absurd. Thus, U(x) x ( a - e , ~ ) is an open in the
product topology which nonintersecting to Uk>po End(fk).
Because (x, a) E U(x) x (a - e, c~), we obtain that (x, a) ~ Uk>po End(fk).
Therefore, (z, a) 6 Np>l Uk>p End(fk), in contradiction with (5).
So, must be f(x) > a, and consequently, (x, a) E End(f).
On the other hand, let (x, a) E End(f). Then f(x) > a and, due f= L f, we obtain that

x ~ liminfLJ= = n U L,~fk. (7)

H kEH

If we suppose that (x,a) ¢ liminfEnd(fn), then there exists H0 cofinal such that (x,a)
UkeHo End(fk).
Therefore, must be to exist V(x, a) such that

V(x,a) A [ UeH End(fk)l = 0 . (8)

k o

Without loss of generality, we can suppose that V is a basic open of the product topology, that
is, V is an open of form U × (0, 7) where U is an open in X and (8, 7/) is an open interval in ~+
containing a. We note that if y E U, then V = U × (0, 7) containing the segment {y} × (8, 7).
148 H. ROM.~N-FLORES AND M. ROJAS*MEDAR

Now, we assure that the projection px(V(x,a)) is an open in X which nonintersecting

UkeHo L,~fk (we recall that Px is an open mapping).
In fact, if we suppose that px(V(x,a)) M [UkeHo Lark] ~ ~, then there exists y • px(Y(z,a))
such that fko (Y) >--a, for some /co • H0.
Therefore, y • U and there is/? _ a such that (y,/?) • V(x, a) = U x (0, ~]).
But then, (y,/?) • Y(z, a) N End(fko) C V(x, a) n [UkeHo End(fk)], in contradiction with (8).
Because

px(V(x'a)) M [=UeHoLafk]
O k and x • px(V(x,a)),

we conclude that x St Ukego Lark which, due (7), is absurd.

Summarizing, we must have (x, a) • lim inf End(fn).
Therefore, lim End(fn) = End(f), which implies that fn r f, completing the first part of our
proof.
(ii) ~ (i). Let a • [0, c~) and suppose that fn r f.
We want to prove that fn L f and, for this, it is sufficient to prove that

limsupLafn C L,~f C liminf Lafn, Vs.

Let OQ
x E limsupL,~fn = N U L.~f. (9)
n = l k>_n

If f(x) < a, then (x, a) St An°°__1 U k ~ n nnd(fk).

Therefore, 3n0 such that (x, a) St Uk>no End(fk).
Consequently, 3 V(x, a) such that

VM [ U End(fk) =0. (10)

Lk~no
Also, without loss of generality, we can to suppose that V is an open of form V = U x (8, 7).
But then, the projection U = px(V(x,a)) is a neighborhood of x which nonintersecting
Uk>no L,~fk.
In fact, i f y E UMUk>noLc~fk then 3/? _< a such that (y,/?) E V, and 3k0 _> no such that
fko(Y) >- a >/3, that is, (y,/?) • End(fko).
Thus, (y,/3) • Y M [Uk>no End(fk)] which contradicts (10).
So, U O [Uk>no Lark] = ~ but, because x • U, this implies that x St Uk>no Lark, in contradic-
tion with (9).
Hence, f(x) > a, and consequently, x • LcJ.
Therefore, lim sup L,~fn C_L,~f.
On the other hand, let x • L,~f and suppose thatf(x) > a.
Then there is e > 0 such that f(x) > a ÷ e.
So, due fn r f , we have that

(x, ~ + e) • End(f) = liminf End(fn) = N U End(fk). (11)

H kEH

Now, if we suppose that x St liminfLafn, then 3H0 cofinal such that x St Ukego Lark', and
therefore, 3 U(x) such that

UM L,~fk = ~. (12)
k o
 Level-Continuity of Functions 149

We assure that [U x (a, oo)] N Ukego E n d ( A ) = 0.

In fact, if (y, f~) 6 [U x (a, oo)] nUkeHoEnd(fk) then fko(Y) > f~ > a for some k0 e H0, and
this implies that y 6 U ~ LcJko C U N [UkeHo L,~fk], in contradiction with (12).
Thus, because (x, a + e) 6 U x (a, oo), we obtain that (x, a + e) ~ Ukego End(fk), and therefore,
( x , a + e) ¢ liminf End(f~) = E n d ( f ) which, due (11), is absurd.
So, necessarily, we must have x 6 lim i n f L a f ~ , and consequently, {f > a} is contained in
lim inf Lafn.
Finally, because liminfL~fn is closed and f is level-continuous, by Theorem 3.1, we obtain

{f > a} = {f >_a} = Laf C_liminfL,~fn.

Consequently, f~ L f and the proof is complete. |

The following example shows that, in Theorem 4.4 above, the level-continuity condition on f
can not avoided.
EXAMPLE 4.5. Let X = [0, 2] endowed with the usual topology and define (for all n _> 2)

X + l - -1,
- if 0 < x < 1,
n n
n 1
f~(x) = ~ - ~ _1 ( x - 1 ) + 1, ifl<x<2--,
n
1
2, if2-- <x<2,
n

f(x) = { 1, if0<x<l,
x, ifl<x<2.
First, we observe that f is not level-continuous. In fact, taking ap = 1 + (l/p) we have that
ap -~ 1 and L , p f = [1 + (1/p),2], Vp. Therefore, l i m / ~ p f = [1,2] whereas L l f = [0,2].
On the other hand, it is easy to see that Lafn, Laf are closed sets Vn, afn is level-continuous
for each n and fn r f, but {fn} does not converges levelwise to f . In fact, for ~ = 1 we have
Laf = [0, 2] whereas Llfn = [1, 2], Vn, consequently, l i m L l f ~ = [1, 2].
REMARK 4.6. In [10-12], the authors introduce the concept of robust set (as a generalization of
open set) and robust functions (which generalize the concept of upper semicontinuous functions),
obtaining useful results in the setting of minimization problems. We are studing the eventually
connections between these results and our work which will appear in a forthcoming paper.

REFERENCES
1. E. De Giorgi and T. Franzoni, Su un tipo di convergenza variazionale, Atti Acc. Naz. Lincei Rend. Cl. Sc.
Mat. Fis. Nat. 58 (8), 842-850, (1975).
2. H. Attouch, Variational Convergence for Function and Operators, Pitman, London, (1984).
3. G. Greco, M. Moschen and E. Quelho, On the variational convergence of fuzzy sets (to appear).
4. M. Rojas-Medar and H. Rom~n-Flores, On the equivalence of convergences of fuzzy sets, Fuzzy Sets and
Systems 80, 217-224, (1996).
5. H. Rom~n-Flores, A. Flores, A. Pach~. and L. Hern£ndez, Proc. of 7° Cong. Latino-lberoamericano de Invest.
Oper. (CLAIO'94), Santiago, Chile.
6. H. Rom~n-Flores, The compactness of E(X), Appl. Math. Lett. 11 (2), 13-17, (1998).
7. F. Hausdorff, Set Theory, Chelsea Press, New York, (1957).
8. E. Klein and A.C. Thompson, Theory of Correspondences, Wiley, New York, (1984).
9. O. Kaleva and S. Seikkala, On fuzzy metric spaces, Fuzzy Sets Syst. 12, 215-229, (1984).
10. Q. Zheng, Robust analysis and global optimization, Computers Math. Applic. 21 (6/7), 17-24, (1991).
11. Q. Zheng, Discontinuity and measurability of robuest functions in the integral global optimization, Computers
Math. Applic. 25 (10/11), 79-88, (1993).
12. E. Calperin and Q. Zheng, Global Solutions in optimal control and games, NP Research Publ., Montreal,
(1991).