Istances and Brø ndsted’s variational principle

Assume q(x,y)=0 and q(y,x)=0. Let us adopt x_{2k-1} = x and x_{2k} = y, k \in \mbox{N}. Clearly, \mbox{q(x_n,x_m)} = 0 holds for each m \neq n, m, n \in \mbox{N}, and \mbox{(x_n)_n N} is a Cauchy sequence in \mbox{(X,U)} (see (2)), i.e. x = y. $

Assume that C contains at least two elements. Let us adopt \mbox{C_y}=\{z \in C\colon q(z,y)=0 \} and suppose that there exists a U \in \mbox{U} such that \mbox{C_y} \not \subset U(y), for each y \in C. Then there exists a sequence \mbox{(x_n)_n N} in C such that q(\mbox{x_n+1},\mbox{x_n})=0 and \mbox{x_n+1} \not \in U(\mbox{x_n}). In view of (2), \mbox{(x_n)_n N} is a Cauchy sequence in \mbox{(X,q)} and consequently, \mbox{(x_n)_n N} is a Cauchy sequence in \mbox{(X,U)}. This contradicts the way \mbox{(x_n)_n N} was defined. Thus, for each U \in \mbox{U} there exists a y \in C such that \mbox{C_y} \subset U(y). The family \mbox{C} = \{\mbox{C_y}\colon y \in C \} of closed sets has the finite intersection property and \mbox{C} contains small sets. Therefore [Ke, Theorem 23, p. 193], there exists a unique x \in \bigcap \mbox{C}. From the fact that x \in \mbox{C_y} and the lower semicontinuity of q(\cdot,y) it follows that q(x,y)=0, \mbox{y C {x}}. On the other hand, x \in \{z \in X\colon \psi(z) \leq \psi(y)\}, for each \mbox{y C {x}}, and consequently, x \mbox{} y, \mbox{y C {x}}, i.e. x is the unique smallest element of C (see (3)). $

The relation \mbox{} is transitive and in view of Kuratowski’s lemma [Ke, p. 33] for any \mbox{x_0} \in X \setminus B (see (5)) there exists a maximal chain A containing \mbox{x_0}. Let us adopt \alpha = \inf \{ \psi(z)\colon z \in A \} and suppose \alpha < \psi(x), for each x \in A. Then there exists a sequence \mbox{(x_n)_n N} in A such that \mbox{((x_n))_n N} decreases to \alpha. Condition \mbox{x_m} \mbox{} \mbox{x_n} for m < n would mean that

which is impossible. Therefore, \mbox{x_n} \mbox{} \mbox{x_m} must be true (\mbox{x_n}, \mbox{x_m} \in A) and we obtain

i.e. \mbox{(x_n)_n N} is a Cauchy sequence in \mbox{(X,q)}, as \psi is bounded below. Consequently, (see (2)) \mbox{(x_n)_n N} is a Cauchy sequence in \mbox{(X,U)}. On the other hand, all \mbox{x_n} for large n are contained in \{z \in X\colon \psi(z) \leq \psi(\mbox{x_m}) \} which is complete (see (5)). Therefore, \mbox{(x_n)_n N} is convergent; let us say that v \in \mbox{_n } \mbox{x_n}. From the lower semicontinuity of \psi, \mbox{q(,x_m)} it follows that \psi(v) \leq \alpha and

On the other hand, for any y \in A there exists an \mbox{x_n} such that \psi(\mbox{x_n}) < \psi(y) and consequently, (\mbox{x_n}, y \in A)

must hold, as

is false. Now, we obtain (see (2))

This reasoning proves that there exists a v \in A such that \psi(v) = \alpha. If v is not a (unique) smallest element of A, then from y \mbox{} v for a y \in A \setminus \{v\} it follows that

and we obtain \psi(y) = \alpha, q(y,v) = 0. Consequently, if (5) is satisfied, then x = v is the unique smallest element of A. Assume that \mbox{(X,U)} is Hausdorff. Let us consider C = \{z \in A\colon q(z,v) = 0\}. It is a maximal chain satisfying the assumptions of Lemma 4 for \mbox{_} in place of X (see (5)). The smallest element x of C is the unique smallest element of the maximal chain A. Conditions (ii), (iii) follow. Now, (5) yields x \in B (otherwise x would have another predecessor), i.e. (i) is satisfied. $

For any maximal chain A containing \mbox{x_0} we follow the proof of Theorem 5. If x is the smallest element of A and precedes itself, then the following holds

The last sentence of the proof of Theorem 5 cannot be added, and we do not know if x \in B. $

In view of (8), Theorem 6 here applies to nonempty chains. Now, condition (8) and (iii) mean x \in F(x), and q(x,x) = 0. $

Suppose F has no fixed point. Then Theorem 6 applies (A is nonempty) and (iii) contradicts (9). $

The reasoning based on our Theorem 6 is similar to the one presented in the proof of [Ka, Theorem 3]. The set Y = \{z \in X\colon \psi(z) \leq \psi(\mbox{x_0})\} is complete. Suppose that for each x \in Y there exists a y \in Y \setminus \{x\} such that \psi(y) + q(y,x) - \psi(x) \leq 0. Then by Theorem 6 (iii) there exists an x \in Y such that 0 < \psi(y) + q(y,x) - \psi(x) - a contradiction, i.e. (i) is proved for Y in place of X. Condition (i) is trivial for y \in X \setminus Y. Now, let us consider \epsilon q in place of q. From \mbox{x_0} \mbox{} \mbox{x_0} and Theorem 6 it follows that there exists an x \in X such that \psi(x) + \epsilon q(x,\mbox{x_0}) - \psi(\mbox{x_0}) \leq 0 and 0 < \psi(y) + \epsilon q(y,x) - \psi(x), \mbox{y X {x}}. From \epsilon q(x,\mbox{x_0}) \leq \psi(\mbox{x_0}) - \psi(x) and the second assumption of (ii) we obtain \[ q(x,x_{0}) //1.