[1] ONOVSKIJ M. JA., BOLTJANSKIJ V. G. & SARYMSAKOV T. E. Topological Sem@ei& Izd. SamGU, Tashkent (1960). (In Russian.)
[2] BRB~STED A., Fiied points and partia1 orders, Proc. Am. math. Sot. 60, 365-366 (1976).
[3] DAY M. M., Nomed Linear Spuces, Springer, Berlin (1973).
[4] D~ORD N. & SCHWARTZ J. T., Linear Operators I, Interscience, New York (1958).
[5] E~SE~JFELD J.& LAKSHMIKAMHAM V., Comparison principle and nonlinear contractions in abstract spaces, 1. math. Analysis Appk. 49, 504-511 (1975).
[6] EISE~LD J. & LAKSHMIKA~M V., Fixed point theorems through abstract cones, J. math. Analysis Appfic. 52, 25-35 (1975).
[7] EISENFELD J. & LAKSHMIKAN?HAM V., Fixed point theorems on closed sets through abstract cones, Technical Report NO. 39, University of Texas at Arlington (March 1976).
[8] EKELAM) I., Nonconvex minimization problems, Bull. Am. math. Sot. (NJ.) 1, 44-74 (1979).
[9] ISBELL J. R., Uniform Spaces, Am. Math. Sot. Surveys No. 12, Providence, RI (1964).
[10] KIRK W. A. & CARISTI J., Mapping theorems in metric and Banach spaces, BUN. Acad. PO/on. Sci. Ser. Math. 23, 891-894 (1975).
[11] KRASNOSEL’SKIJ M. A., Positive Solutions of Operator Equations, Nordhoff, Groningen (1964).
[12] MCAR~UR C. W., In what spaces is every closed normal cone regular ?. Proc. Edinb. Moth. Sot. 17 (Series II),121-12s (1970).
[13] NAMES A. B., Nonconvdx minimization principle in ordered regular Banach spaces, Mathematics (Cfuj) 23 (46), 43-48 (1981).
[14] NEMETH A. B., Summation criteria for regular cones with applications, “Babes-Bolyai” Univ. Faculty of Math. Research Semin., preprint No. 4, pp. 99-124 (1981).
[15] N~METH A. B., Normal cone valued metrics and nonconvex vector minimization principle, “Babe3-Bolyai” Univ. Faculty of Math. Research Semin., preprint No. 1 (1983).
[16] N~METH A. B., Nonconvex vector minimization principles with and without the axiom of choice, “Babe%-Bolyai” Univ. Faculty of Math. Research Semin., preprint No. 1 (1983).
[17] PERESSI~JI A. L., Ordered Topological Vector Spaces, Harper & Row, New York (1967).
Paper (preprint) in HTML form
1986-Nemeth-A NONCONVEX VECTOR MINIMIZATION PROBLEM
A NONCONVEX VECTOR MINIMIZATION PROBLEM
A. B. NémethInstitutul de Matematicá, str. Republicii Nr. 37,C.P. Nr. 68, 3400 CLUJ-NAPOCA Romania
(Received 1 August 1984: received for publication 29 July 1985)
Key words and phrases: Ordered regular vector spaces, nonconvex vector minimization. axiom of choice.
INTRODUCTION
In ORDER to produce a vector minimization principle which contains Ekeland's variational theorem [8] as well as the results of [13, 14], we have to introduce cone valued metrics. Working with these metrics provides noteworthy technical facilities in some application oriented investigations, the most relevant one from our point of view being the fixed point theory. From the results of Eisenfeld and Lakshmikantham [5-7] comes the idea of application of regular cone valued metrics, which will play an important role in this paper.
The normal cone valued metrics induce uniformizable topologies and every uniformizable Hausdorff topology can be induced by a normal and regular cone valued metric according to a result due to Antonovskij, Boltjanskij and Sarymsakov [1]. The above considerations, largely exposed in our preprint [15] will only be summarized in Section 5. Problems concerning a special sort of relativized regularity considered first in [14] will be given in Sections 3 and 4, after showing by some examples in Section 2 the consistency of the notion introduced. The principal result of this paper is theorem 6.1, which constitutes a general nonconvex vector minimization principle containing Ekeland's variational principle [8] and the results of [13, 14]. The principle comprised by theorem 6.1 is in fact a criterion for the above-mentioned relativized regularity of a cone. This is asserted by theorem 6.2. This paper ends with the deduction from theorem 6.1 generalizations of some results due to Eisenfeld and Lakshmikantham [7], among others an ordered vector space variant of the Kirk-Caristi fixed point theorem [10].
The proofs use the axiom of choice. In [16] we have shown that a denumerable variant of this axiom suffices when EE is a Fréchet space. This is the case also when E=RE=R with the natural ordering. The relations of this axiom with the fixed point theorem cited above and with other ordering principles of the analysis were considered by Brønsted [2] (see also the assertion in [16. Section 1]).
1. DEFINITIONS
Let EE be a vector space over the reals and let KK be an acute convex cone in it, i.e. a set having the properties: (i) K+K sub KK+K \subset K; (ii) tK sub Kt K \subset K for each nonnegative real number tt; (iii) K nn(-K)={0}K \cap(-K)=\{0\}. We shall refer to KK with these properties simply as to a cone in EE, and shall suppose throughout that K!={0}K \neq\{0\}. Putting x <= yx \leqslant y whenever y-xy-x is in KK, we obtain a reflexive, transitive and antisymmetric order relation on EE, which is translation invariant and invariant with respect to the multiplication with nonnegative reals. It is called the order induced by KK or simply the KK-order in EE. The vector space EE endowed with an order relation as defined
above is called ordered vector space and the cone KK inducing the order in it is called its positive cone. KK bounded, KK monotone etc. will mean bounded, monotone etc, with respect to the order induced by KK.
The set AA in the ordered vector space EE with the positive cone KK will be called full if A=(A+K)nn(A-K)A= (A+K) \cap(A-K).
Suppose that EE is an ordered vector space endowed with a locally convex vector space topology which is Hausdorff. The positive cone KK in EE is said to be normal if the zero element of EE has a neighbourhood basis B(0)B(0) consisting of full sets.
The cone K_(0)K_{0} in KK is called KK bound regular (sequentially KK bound regular) if each K_(0)K_{0} increasing and KK order bounded net (sequence) in K_(0)K_{0} converges to an element of K_(0)K_{0}. If KK is itself KK bound regular (sequentially KK bound regular) then it is called regular (sequentially regular).
If K_(0)K_{0} is a KK bound regular subcone of KK then it is obviously a regular cone.
The cone KK is called fully regular if each of its KK increasing and topologically bounded nets is convergent to one of its elements.
A stronger notion of relativized regularity was considered in [11, 1.8.4]. Various concepts of bound regularity were introduced and used in [14].
2. EXAMPLES OF REGULAR CONES
2.1. Let KK be a normal cone of the locally convex Hausdorff space EE. Suppose that K_(0)K_{0} is subcone of KK which is complete and has the property that its linear span is complete, metrizable and does not contain any subspace isomorphic with the Banach space c_(0)c_{0} of the real sequences converging to 0 in its usual norm. According to theorem 1 of McArthur in [12], K_(0)K_{0} is fully regular. Since KK is normal, every K_(0)K_{0} increasing KK order bounded net in K_(0)K_{0} is topologically bounded too (see for example [17, II, proposition 1.4]) and hence convergent by the full regularity of K_(0)K_{0}. The limit of such a net is in K_(0)K_{0}, the latter being complete by hypothesis. We have thus shown that K_(0)K_{0} is a KK bound regular subcone of KK.
2.2. Suppose that BB is a complete, convex and bounded subset of the locally convex Hausdorff space EE. Suppose BB does not contain 0 . Let K_(0)K_{0} be the cone generated by BB. Then K_(0)K_{0} is a KK bound regular subcone of every normal cone KK containing it. Indeed, if we consider the subspace sp K_(0)K_{0} of EE, then this space can be endowed with a norm considering the unit ball in it the convex circled hull of BB. This topology on sp K_(0)K_{0} is finer than that induced from EE and sp K_(0)K_{0} is a Banach space with respect to the norm. In this space K_(0)K_{0} admits a plastering and hence it is fully regular according to theorem 1.12 of Krasnosel'skij in [11]. Now, since KK is a normal cone containing K_(0)K_{0} the reasoning in 2.1 applies to conclude K_(0)K_{0} is KK bound regular.
2.3. Let SS be a nonempty set and let R^(S)R^{S} be the vector space of all the real functions defined on SS, endowed with the topology of the pointwise convergence. This topology is in fact the direct product topology of R^(S)R^{S} and converts this space in a locally convex Hausdorff space. The cone R_(+)^(S)R_{+}^{S} of all the nonnegative functions in R^(S)R^{S} is normal since the generating family of seminorms p_(s)p_{s} on R^(S)R^{S} defined by p_(s)(x)=|x(s)|p_{s}(x)=|x(s)| is monotone on it ([17, II, proposition 1.5]). It is also regular, since the convergence in R^(S)R^{S} is the pointwise one and hence the regularity of R_(+)^(S)R_{+}^{S} is a direct consequence of the regularity of R_(+)R_{+}, the nonnegative half line.
2.4. Let R^(S)R^{S} be the vector space considered at 2.3. Denote by c_(0)(S)c_{0}(S) the subspace consisting of the functions xx having the property that for every positive real epsi\varepsilon the set {s in S:∣x(s) > epsi}\{s \in S: \mid x(s)>\varepsilon\} is finite. Let us define a norm on c_(0)(S)c_{0}(S) by putting ||x||=max_(s)|x(s)|\|x\|=\max _{s}|x(s)|. Equipped with this norm, c_(0)(S)c_{0}(S) becomes a Banach space ([3, II.2]). The cone c_(0)^(+)(S)c_{0}^{+}(S) of the nonnegative functions in c_(0)(S)c_{0}(S) is normal and regular.
2.5. Consider the Banach space cc of all the convergent sequences of real numbers endowed with the sup norm. Then c^(+)c^{+}, the cone of sequences with nonnegative terms in normal. Let c_(0)c_{0} be the subspace in cc of the sequences converging to 0 . Then c_(0)^(+)=c^(+)nnc_(0)c_{0}^{+}=c^{+} \cap c_{0} is a regular cone which is neither c^(+)c^{+}bound regular, nor fully regular.
2.6. Let KK be a generating closed normal cone in the barelled space EE. The set L^(+)L^{+}of all the linear and continuous operators AA with the property that A(K)sub KA(K) \subset K forms a regular cone in the space L(E)L(E) of all the linear and continuous operators acting in EE and equipped with the topology of simple convergence (see [15, 9.1]).
2.7. Let HH be a real Hilbert space. The linear and continuous operator AA acting in HH is called positive if (Ax,x) >= 0(A x, x) \geqslant 0 for every xx in HH. Let the vector space L(H)L(H) of all the linear and continuous operators acting in HH be endowed with the topology of simple convergence. Then the set L^(+)L^{+}of all the positive operators forms a regular cone in L(H)([15,9.2])L(H)([15,9.2]).
3. A SUMMATION CRITERION FOR REGULARITY
The reduction of definitions of regularity to criterions which use a denumerable set of terms is very desirable from a technical point of view. In this direction we observe first of all that:
{:(3.1){:[" a complete cone "K_(0)" in "K" is "K" bound regular if and only if "],[" it is sequentially "K" bound regular. "]}:}\left.\begin{array}{l}
\text { a complete cone } K_{0} \text { in } K \text { is } K \text { bound regular if and only if } \tag{3.1}\\
\text { it is sequentially } K \text { bound regular. }
\end{array}\right\}
Proof of (3.1). The "only if" part is immediate. For the converse implication let us assume that K_(0)K_{0} is sequentially KK bound regular but is not KK bound regular. Then there exists the K_(0)K_{0} increasing KK bounded net (x_(i))_(i in I)\left(x_{i}\right)_{i \in I} in K_(0)K_{0} which is not Cauchy. That is, there exists a neighbourhood UU of 0 such that for any ii in II there exists x_(j)x_{j} and x_(k)x_{k} with j >= ij \geqslant i and k >= ik \geqslant i such that x_(j)-x_(k)!in Ux_{j}-x_{k} \notin U. We can suppose x_(k)-x_(j)inK_(0)x_{k}-x_{j} \in K_{0} since ( x_(i)x_{i} ) is K_(0)K_{0} increasing. Fix ii and consider x_(j)x_{j} and x_(k)x_{k} as above. Put x_(1)=x_(j)x_{1}=x_{j} and x_(2)=x_(k)x_{2}=x_{k}. Starting with x_(k)x_{k} instead of x_(i)x_{i}, we can determine x_(3)x_{3} and x_(4)x_{4} in ( x_(i)x_{i} ) so as to have x_(3)-x_(2)inK_(0),x_(4)-x_(5)inK_(0)x_{3}-x_{2} \in K_{0}, x_{4}-x_{5} \in K_{0} and x_(4)-x_(3)!in Ux_{4}-x_{3} \notin U. Continuing this procedure we obtain a sequence which is K_(0)K_{0} increasing and KK bounded, but is not Cauchy, contradicting the hypothesis.
The above assertion is used in establishing the following summation criterion:
{:(3.2){:[" The complete subcone "K_(0)" of the cone "K" is "K" bound regular if and "],[" only if for every sequence "(x_(n))" of "K_(0)" the condition "x_(n)!in U],[" for any "n" and for some neighbourhood "U" of "0" implies that the "],[" set "{s_(m):m in N}" where "s_(m)=sum_(n=1)^(m)x_(n)","" cannot be "K" order bounded. "]}:}\left.\begin{array}{l}
\text { The complete subcone } K_{0} \text { of the cone } K \text { is } K \text { bound regular if and } \tag{3.2}\\
\text { only if for every sequence }\left(x_{n}\right) \text { of } K_{0} \text { the condition } x_{n} \notin U \\
\text { for any } n \text { and for some neighbourhood } U \text { of } 0 \text { implies that the } \\
\text { set }\left\{s_{m}: m \in N\right\} \text { where } s_{m}=\sum_{n=1}^{m} x_{n}, \text { cannot be } K \text { order bounded. }
\end{array}\right\}
Proof of (3.2). Suppose that there is a sequence (x_(n))\left(x_{n}\right) in K_(0)K_{0} with x_(n)!in Ux_{n} \notin U for every nn and for some neighbourhood UU of 0 such that {s_(m)=sum_(n=1)^(m)x_(n):m in N}\left\{s_{m}=\sum_{n=1}^{m} x_{n}: m \in N\right\} is KK order bounded. Then ( s_(m)s_{m} ) forms a K_(0)K_{0} increasing KK bounded sequence which is not convergent, that is, K_(0)K_{0} cannot be sequentially KK bound regular.
Assume now that K_(0)K_{0} does not be KK bound regular, and hence neither sequentially KK bound regular by (3.1). Hence K_(0)K_{0} contains a K_(0)K_{0} increasing KK order bounded sequence (z_(n))\left(z_{n}\right) with the property that there exists a neighbourhood UU of 0 such that z_(n+1)-z_(n)!in Uz_{n+1}-z_{n} \notin U for every nn. We can assume z_(1)=0z_{1}=0. Then the elements x_(n)=z_(n+1)-z_(n)x_{n}=z_{n+1}-z_{n} form a sequence in K_(0)K_{0} such that
{s_(m)=sum_(n=1)^(m)x_(n)=z_(m+1):m in N}\left\{s_{m}=\sum_{n=1}^{m} x_{n}=z_{m+1}: m \in N\right\}
is KK order bounded. This contradicts the condition of the criterion and completes the proof.
Summation criteria for various types of regularity were considered by McArthur [12] and by the author in [13, 14]. The summation techniques in some problems concerning the sequentially regular cones had already been used by Krasnosel'skij in [11, theorems 1.6 and 1.7].
4. NEAR TO MINIMUM POINT CRITERION FOR REGULARITY
Let EE be a vector space ordered by the cone KK. Suppose that MM is a set in EE and HH is a subset of KK. The point xx in MM will be said an HH near to minimum point of MM if
(x-H-K)nn M=O/.(x-H-K) \cap M=\varnothing .
Another regularity criterion for a cone can be stated in terms of near to minimality. It is:
the complete subcone K_(0)K_{0} of the cone KK in the locally convex space EE is KK bound regular if and only if for each nonempty set HH in K_(0)K_{0} with the property that E\\HE \backslash H is a neighbourhood of 0 , every KK lower bounded subset of EE has HH near to minimum points.
Proof of (4.1). Suppose that MM is a set of EE which is KK lower bounded by bb. Assume that HH is as in (4.1) and there are no HH near to minimum points in MM. Then
(x-H-K)nn M!=O/(x-H-K) \cap M \neq \varnothing
for each xx in MM. Let us consider x_(1)in Mx_{1} \in M arbitrarily and choose
x_(n+1)in(x_(n)-H-K)nn M.x_{n+1} \in\left(x_{n}-H-K\right) \cap M .
n=1,2,dotsn=1,2, \ldots. Then we have
x_(n+1) <= x_(n)-h_(n),n in Nx_{n+1} \leqslant x_{n}-h_{n}, n \in N
for some h_(n)h_{n} in HH, where <=\leqslant denotes the KK order in EE. Summing these relations from n=1n=1 to n=mn=m, we get
By the hypothesis on HH, there exists a neighbourhood UU of 0 in EE so as to have h_(n)!in Uh_{n} \notin U for every nn. This, together with the above relation show, via the criterion 3.2, that K_(0)K_{0} cannot be KK bound regular.
To prove the "if" part of the assertion, assume that every subset of EE which is KK bounded from below has HH near to minimum points for each HH as in 4.1, but K_(0)K_{0} is not sequentially KK bound regular. Using the criterion 3.2 again, it follows that there exist the elements x_(n)x_{n} in K_(0)K_{0} without some neighbourhood UU of 0 such that the sums
{sum_(n=1)^(m)x_(n),m in N}\left\{\sum_{n=1}^{m} x_{n}, m \in N\right\}
have a KK upper bound bb. This means that the set
M={-sum_(n=1)^(m)x_(n):m in N}M=\left\{-\sum_{n=1}^{m} x_{n}: m \in N\right\}
is KK order bounded from below by -b-b. Let xx be an arbitrarily chosen element of MM. Then x=-sum_(n=1)^(m)x_(n)x=-\sum_{n=1}^{m} x_{n} for some mm. We have
x-x_(m+1)=-sum_(n=1)^(m+1)x_(n)in Mx-x_{m+1}=-\sum_{n=1}^{m+1} x_{n} \in M
and
x-x_(m+1)in x-H sub x-H-Kx-x_{m+1} \in x-H \subset x-H-K
where we have denoted by HH the set {x_(n):n in N}\left\{x_{n}: n \in N\right\}. That is. we have got
(x-H-K)nn M!=O/(x-H-K) \cap M \neq \varnothing
for every xx in MM. Since HH satisfies the hypothesis in the proposition, this is a contradiction.
We shall use criterion 4.1 in the following slightly modified form:
{:(4.2){:[" The complete subcone "K_(0)" of the cone "K" in the locally convex "],[" space "E" is "K" bound regular if and only if for each "H" in "K_(0)(H!=O/)],[" with the property that "ELH" is a neighbourhood of "0", every set "M" in "],[E" which has "K" order bounded "K" lower sections, i.e. which contains "],[" a point "z" such that the set "]}:}\left.\begin{array}{l}
\text { The complete subcone } K_{0} \text { of the cone } K \text { in the locally convex } \tag{4.2}\\
\text { space } E \text { is } K \text { bound regular if and only if for each } H \text { in } K_{0}(H \neq \varnothing) \\
\text { with the property that } E L H \text { is a neighbourhood of } 0 \text {, every set } M \text { in } \\
E \text { which has } K \text { order bounded } K \text { lower sections, i.e. which contains } \\
\text { a point } z \text { such that the set }
\end{array}\right\}
(z-K)nn M(z-K) \cap M
is KK order bounded, has HH near to minimum points.
5. CONE VALUED METRICS
Let EE be an ordered vector space with the positive cone KK. A KK metric rr on a nonempty set VV is a mapping rr of V xx VV \times V into KK which satisfies for arbitrary elements u,vu, v and zz in VV the following conditions: (i) r(u,u)=0r(u, u)=0; (ii) r(u,v)=0r(u, v)=0 implies u=vu=v; (iii) r(u,v)=r(v,u)r(u, v)=r(v, u); (iv) r(u,z) <= r(u,v)+r(v,z)r(u, z) \leqslant r(u, v)+r(v, z), where <=\leqslant stands for the KK ordering.
If KK is a normal cone in the locally convex Hausdorff space EE, then the sets
V(U,a)={u in V:r(a,u)in U},U in B(0)V(U, a)=\{u \in V: r(a, u) \in U\}, U \in B(0)
with B(0)B(0) a neighbourhood basis in EE consisting of full sets, and a runs over VV, form a
neighbourhood basis for a Hausdorff topology on VV. The resulting topology is a uniformizable one; it is called the topology induced by rr on VV. Cauchy nets and complexity can be defined in a natural way (for details see [15]). Antonovskij, Boltjanskij and Sarymsakov have shown in [ 1,11.41,11.4 and 11.5] that a completely regular Hausdorff topological space is KK metrizable with KK the positive cone R_(+)^(S)R_{+}^{S} of the space R^(S)R^{S} considered in 2.3. We have observed that R_(+)^(S)R_{+}^{S} is a normal and regular cone. That is, every completely regular Hausdorff topological space is KK metrizable with KK a normal and regular cone. Hence by [9, I.15], the Hausdorff topological spaces which are KK metrizable by a KK metric of this kind are quite those which are uniformizable.
Regular cone valued metrics were considered by Eisenfeld and Lakshmikantham in [5-7] for the case when KK is a regular cone of a Banach space. Since criteria using a denumerable set of terms also work for regular cones in nonmetrizable locally convex Hausdorff spaces, we have as an immediate consequence of one of them, the criterion 3.2, the following assertion:
Let VV be a set endowed with a K_(0)K_{0} metric rr, where K_(0)K_{0} is a KK bound regular 7 subcone of the normal cone KK in the locally convex Hausdorff space EE. Suppose that (v_(n))\left(v_{n}\right) is a sequence in VV such that the set
{:(5.1){sum_(n=1)^(m)r(v_(n+1),v_(n)):m in N}:}\begin{equation*}
\left\{\sum_{n=1}^{m} r\left(v_{n+1}, v_{n}\right): m \in N\right\} \tag{5.1}
\end{equation*}
is KK order bounded. Then (v_(n))\left(v_{n}\right) is Cauchy in the topology on VV induced by the K_(0)K_{0} metric rr.
6. NONCONVEX MINIMIZATION
Let VV be a topological space and let FF be an operator from VV to the ordered topological vector space EE. We shall say that FF is submonotone if from the conditions:
(i) limv_(nu)=v\lim v_{\nu}=v, where (v_(nu))_(nu in I)\left(v_{\nu}\right)_{\nu \in I} is a net in VV indexed by the totally ordered set II;
(ii) F(v_(nu)) <= F(v_(mu))F\left(v_{\nu}\right) \leqslant F\left(v_{\mu}\right) whenever nu >= mu\nu \geqslant \mu, it follows that
F(v) <= F(v_(nu))" for every "nu" in "I.F(v) \leqslant F\left(v_{\nu}\right) \text { for every } \nu \text { in } I .
Observe that submonotonicity is a feeble sort of lower semicontinuity of operators with values in ordered vector spaces. Various related but stronger notions were considered in [57] and in [13-15].
The main result of our note is the following.
THEOREM 6.1. Let EE be a locally convex Hausdorff space and let KK be a closed normal cone in EE. Suppose that K_(0)K_{0} is a KK bound regular complete subcone of KK.
Let (V,r)(V, r) be a complete K_(0)K_{0} metric space, and let F:V rarr EF: V \rightarrow E be a submonotone operator with respect to the KK ordering.
Suppose that FF has KK order bounded KK lower sections, i.e. that there exists at least an element zz in VV such that the set
(i) (F(z)-K)nn F(V)(F(z)-K) \cap F(V) has a KK lower bound.
Then for every zz with the property (i) and for every positive real epsi\varepsilon there is a vv so as to have
(ii) F(z)-F(v)-epsi r(z,v)in KF(z)-F(v)-\varepsilon r(z, v) \in K
and
(iii) F(v)-F(w)-epsi r(v,w)!in KF(v)-F(w)-\varepsilon r(v, w) \notin K whenever w in V\\{v}w \in V \backslash\{v\}.
Let UU be a neighbourhood of 0 in EE. If H=K\\U!=O/H=K \backslash U \neq \varnothing, then for a zz with the property (i) there exists a uu in VV such that
(iv) F(z)-F(u)in KF(z)-F(u) \in K
and
(v) (F(u)-epsi H-K)nn F(V)=O/(F(u)-\varepsilon H-K) \cap F(V)=\varnothing.
For every uu with this property there is an element vv in VV satisfying (iii) and the condition (ii) with uu instead of zz, and such that
(vi) r(u,v)in Ur(u, v) \in U.
Proof. Define the relation -<\prec on F(V)F(V) by putting F(p)-<F(q)F(p) \prec F(q) if
F(q)-F(p)-epsi r(p,q)in K.F(q)-F(p)-\varepsilon r(p, q) \in K .
It is straightforward that -<\prec is reflexive, transitive and antisymmetric, hence an order relation on F(V)F(V). Apply Hausdorff's theorem (see [4, I.2.6]) to determine a subset ZZ in F(V)F(V) which is totally ordered with respect to the relation -<\prec, has F(z)F(z) as supremum, and is maximal with respect to the set theoretic inclusion. We shall show that ZZ contains its infimum with respect to -<\prec.
Let us introduce a relation <=\leqslant in F^(-1)(Z)=V_(0)F^{-1}(Z)=V_{0} by putting p <= qp \leqslant q if F(p) < F(q)F(p)<F(q). Then V_(0)V_{0} will be totally ordered with respect to <=\leqslant and the filter of its lower sections is Cauchy. To verify this, let us assume the contrary: there exists a neighbourhood U^(')U^{\prime} of 0 such that for each ss in V_(0)V_{0} there are pp and qq in V_(0),p <= sV_{0}, p \leqslant s and q <= sq \leqslant s, such that r(p,q)!inU^(')r(p, q) \notin U^{\prime}. Fix ss and let pp and qq be as above. We can suppose p <= qp \leqslant q. Put v_(1)=q,v_(2)=pv_{1}=q, v_{2}=p. Then r(v_(2),v_(1))!inU^(')r\left(v_{2}, v_{1}\right) \notin U^{\prime} and F(v_(1))-F(v_(2))-epsi r(v_(2),v_(1))in KF\left(v_{1}\right)-F\left(v_{2}\right)-\varepsilon r\left(v_{2}, v_{1}\right) \in K. Starting with pp instead of ss we can continue this procedure. Accordingly we can determine the decreasing sequence ( v_(n)v_{n} ) in V_(0)V_{0} such that
{:(6.1)r(v_(2k),v_(2k-1))!inU^(')quad" for every "k.:}\begin{equation*}
r\left(v_{2 k}, v_{2 k-1}\right) \notin U^{\prime} \quad \text { for every } k . \tag{6.1}
\end{equation*}
From the definition of the relation <=\leqslant on V_(0)V_{0} we have also
F(v_(n))-F(v_(n+1))-epsi r(v_(n+1),v_(n))in K" for each "n". "F\left(v_{n}\right)-F\left(v_{n+1}\right)-\varepsilon r\left(v_{n+1}, v_{n}\right) \in K \text { for each } n \text {. }
By summing this relation from n=1n=1 to n=mn=m, we get
F(v_(1))-F(v_(n+1))-epsisum_(n=1)^(m)r(v_(n+1),v_(n))in K.F\left(v_{1}\right)-F\left(v_{n+1}\right)-\varepsilon \sum_{n=1}^{m} r\left(v_{n+1}, v_{n}\right) \in K .
Since the elements F(v_(n))F\left(v_{n}\right) are all in the set (i), they have a KK lower bound, say y_(0)y_{0}. Adding with F(v_(n+1))-y_(0)in KF\left(v_{n+1}\right)-y_{0} \in K, the above relation yields
F(v_(1))-y_(0)-epsisum_(n=1)^(m)r(v_(n+1),v_(n))in K.F\left(v_{1}\right)-y_{0}-\varepsilon \sum_{n=1}^{m} r\left(v_{n+1}, v_{n}\right) \in K .
The obtained relation shows that the sums
sum_(n=1)^(m)r(v_(n+1),v_(n)),m in N\sum_{n=1}^{m} r\left(v_{n+1}, v_{n}\right), m \in N
are KK order bounded, wherefrom we get, via the assertion (5.1), a contradiction with (6.1).
The obtained contradiction shows that the lower sections of V_(0)V_{0} form a Cauchy filter, which converges by the completeness of VV to vv.
Since FF is submonotone with respect to the KK order, we have
{:(6.2)F(p)-F(v)in K:}\begin{equation*}
F(p)-F(v) \in K \tag{6.2}
\end{equation*}
for every pp in V_(0)V_{0}. Let qq be arbitrary in V_(0)V_{0}. For every p <= qp \leqslant q we have
F(q)-F(p)-epsi r(p,q)in KF(q)-F(p)-\varepsilon r(p, q) \in K
which by adding to (6.2) yields
F(q)-F(v)-epsi r(p,q)in K.F(q)-F(v)-\varepsilon r(p, q) \in K .
Letting p rarr vp \rightarrow v in this relation. taking into account KK is closed, it follows
F(q)-F(v)-epsi r(v,q)in K.F(q)-F(v)-\varepsilon r(v, q) \in K .
that is. F(v)-<F(q)F(v) \prec F(q) for each F(q)F(q) in ZZ. Now, since ZZ is maximal, F(v)F(v) must be in ZZ and it is the infimum of ZZ with respect to -<\prec.
The last assertion implies also that there does not exist any ww in V\\{v}V \backslash\{v\} so as to have F(w) < F(v)F(w)<F(v). Thus we have proved the relations (ii) and (iii).
If HH is the set defined in the theorem, then by 4.2. with F(V)F(V) for MM and with epsi H\varepsilon H for HH, we conclude the existence in VV of uu with the properties (iv) and (v).
If we proceed as above taking uu in place of zz, we can get a vv in VV so as to have (iii) and (ii) with uu instead of zz, that is. to have the relation
this relation together with (6.3) contradict (v).
We shall show that the principal result of theorem 6.1 consisting in the existence of a vv so as to have relation (iii), which may be considered a nonconvex vector minimization principle, is the best possible with respect to the order relation in EE, or, in other words, it characterizes the KK bound regular subcones. More precisely, we have the following.
Theorem 6.2. Let KK be a closed normal cone of the locally convex Hausdorff space EE and let K_(0)K_{0} be a complete subcone of KK. Then the minimization principle comprised in the existence of a vv satisfying (iii) of theorem 6.1, holds for every K_(0)K_{0} metric space ( V,rV, r ) and every KK submonotone mapping from VV to EE which has KK bounded KK lower sections, if and only if K_(0)K_{0} is KK bound regular.
Proof. The 'if' part is contained in theorem 6.1. For the converse implication let us suppose-
that K_(0)K_{0} does not be KK bound regular. Then by the criterion (3.1) there exist a neighbourhood UU of 0 in EE and a K_(0)K_{0} increasing sequence ( x_(n)x_{n} ) in K_(0)K_{0} which is KK order bounded and for which x_(n+1)-x_(n)!in Ux_{n+1}-x_{n} \notin U for every nn. Put V={x_(n):n in N}V=\left\{x_{n}: n \in N\right\}. Define a K_(0)K_{0} metric rr on VV by putting r(x_(k),x_(n))=x_(m)-x_(n)r\left(x_{k}, x_{n}\right)=x_{m}-x_{n}, where m=max{h,k},n=min{h,k}m=\max \{h, k\}, n=\min \{h, k\}. Then VV is trivially rr complete since it is discrete. Let us define F:V rarr EF: V \rightarrow E by putting F(x)=-xF(x)=-x. Then FF is KK lower bounded (since V={x_(n)}V=\left\{x_{n}\right\} is KK order bounded). Because VV is discrete, FF is trivially KK submonotone.
Put epsi=1//2\varepsilon=1 / 2 and consider x_(n)x_{n} to be arbitrary in VV. Let m > nm>n. Then
F(x_(n))-F(x_(m))-(1)/(2)r(x_(n),x_(m))=-x_(n)+x_(m)-(1)/(2)(x_(m)-x_(n))=(1)/(2)(x_(m)-x_(n))in K.F\left(x_{n}\right)-F\left(x_{m}\right)-\frac{1}{2} r\left(x_{n}, x_{m}\right)=-x_{n}+x_{m}-\frac{1}{2}\left(x_{m}-x_{n}\right)=\frac{1}{2}\left(x_{m}-x_{n}\right) \in K .
We have in conclusion for every v(=x_(n))v\left(=x_{n}\right) in VV that there exists some w(=x_(m):}w\left(=x_{m}\right. with {:m > n)\left.m>n\right) in V\\{v}V \backslash\{v\} such that the relation (iii) in theorem 6.1 fails.
We observe the parallelism in form and in content between our theorem and the criterion (4.2). In fact, the nonconvex minimization principle can be considered itself a criterion for KK bound regularity.
7. A FIXED POINT THEOREM
In the papers [5-7], Eisenfeld and Lakshmikantham have succeeded in extending various important results in metric fixed point theory for matrices with values in regular cones in Banach spaces. In [7] (Lemma 3.3) they extend the Kirk-Caristi fixed point theorem for metrics with values in regular minihedral cones with nonempty interior in separable Banach spaces. Theorem 6.1 permits us to obtain an essentially extended form of this theorem. The quoted result of [7] is the key of obtaining the principal result of the cited paper, which also can be extended using our following theorem.
Theorem 7.1. Let EE be a locally convex Hausdorff space, KK a closed normal cone in EE and K_(0)K_{0} a complete KK bound regular subcone of KK. Let (V,r)(V, r) be a complete K_(0)K_{0} metric space and let FF be an operator from VV to EE which is KK submonotone and has the property that the set
(F(z)-K)nn F(V)(F(z)-K) \cap F(V)
is KK order bounded for some zz in VV. If f:V rarr Vf: V \rightarrow V satisfies the condition
for every uu in VV, then ff has a fixed point vv such that
F(v) <= F(z)-r(v,z)F(v) \leqslant F(z)-r(v, z)
where <=\leqslant stands for the KK order.
Proof. Put epsi=1\varepsilon=1 and apply theorem 6.1 to VV and FF in the above theorem. Then it follows the existence of a vv in VV such that F(v) <= F(z)-r(v,z)F(v) \leqslant F(z)-r(v, z) and
{:(7.2)F(v)-F(w)-r(w","v)!in K" whenever "w in V\\{v}.:}\begin{equation*}
F(v)-F(w)-r(w, v) \notin K \text { whenever } w \in V \backslash\{v\} . \tag{7.2}
\end{equation*}
On the other hand (7.1) implies
F(v)-F(f(v))-r(f(v),v)in KF(v)-F(f(v))-r(f(v), v) \in K
If we would have f(v)!=vf(v) \neq v, the obtained relation would contradict the relation (7.2).
REFERENCES
Antonovskij M. Ja., Boltjanskij V. G. & Sarymsakov T. E. Topological Semifields, Izd. SamGU, Tashkent (1960). (In Russian.)
BRØNDSTED A., Fixed points and partial orders, Proc. Am. math. Soc. 60, 365-366 (1976).
Day M. M., Normed Linear Spaces, Springer, Berlin (1973).
Dunford N. & Schwartz J. T., Linear Operators I, Interscience, New York (1958).
Eisenfeld J. & Lakshmikantham V., Comparison principle and nonlinear contractions in abstract spaces, JJ. math. Analysis Applic. 49, 504-511 (1975).
Eisenfeld J. & Lakshmikantham V., Fixed point theorems through abstract cones, J. math. Analysis Applic. 52, 25-35 (1975).
Eisenfeld J. & Lakshmikantham V., Fixed point theorems on closed sets through abstract cones, Technical Report No. 39, University of Texas at Arlington (March 1976).
Isbell J. R., Uniform Spaces, Am. Math. Soc. Surveys No. 12, Providence, RI (1964).
Kirk W. A. & Caristi J., Mapping theorems in metric and Banach spaces, Bull. Acad. Polon. Sci. Ser. Math. 23, 891-894 (1975).
Krasnosel'skij M. A., Positive Solutions of Operator Equations, Nordhoff, Groningen (1964).
McArthur C. W., In what spaces is every closed normal cone regular?. Proc. Edinb. Math. Soc. 17 (Series II), 121-125 (1970).
Németh A. B., Nonconvex minimization principle in ordered regular Banach spaces, Mathematica (Cluj) 23 (46), 43-48 (1981).
NEMETH A. B., Summation criteria for regular cones with applications, "Babeş-Bolyai" Univ. Faculty of Math. Research Semin., preprint No. 4, pp. 99-124 (1981).
Németh A. B., Normal cone valued metrics and nonconvex vector minimization principle, "Babeş-Bolyai" Univ. Faculty of Math. Research Semin., preprint No. 1 (1983).
NEmeth A. B., Nonconvex vector minimization principles with and without the axiom of choice, "Babeş-Bolyai" Univ. Faculty of Math. Research Semin., preprint No. 1 (1983).
Peressini A. L., Ordered Topological Vector Spaces, Harper & Row, New York (1967).