• 沒有找到結果。

Let f, g be real-valued functions defined on X × Y with the following properties: (2)(0) f (x, y

N/A
N/A
Protected

Academic year: 2021

Share "Let f, g be real-valued functions defined on X × Y with the following properties: (2)(0) f (x, y"

Copied!
6
0
0

加載中.... (立即查看全文)

全文

(1)

2 Minimax theorems under η-connectedness

Let X be a nonempty set in a topological space, Y a nonempty η-connected set in a vector space for some η, and f : X × Y −→ R be a real-valued function. Then f is said to be lower semicontinuous on the η of Y if for each x ∈ X and y1, y2 ∈ Y , the function

t −→ f(x, η(y1, y2, t))

is a lower semicontinuous function of t on [0, 1]. If f is a lower semicontinuous function on Y , then f is clearly lower semicontinuous on the η of Y .

In this section, our main result involves only the η-connectedness instead of convexity.

Before proceeding this, we require the following technical lemma of Farka’s type.

Lemma 2.1. Let X be a nonempty compact set of a topological space and let Y be a nonempty η-connected set. Let f, g be real-valued functions defined on X × Y with the following properties:

(2)

(0) f (x, y) ≤ g(x, y) for all (x, y) ∈ X × Y ;

(i) f (·, y) and g(·, y) are upper semicontinuous on X for each y ∈ Y ; (ii) For any y0, y1 ∈ Y ,

g(x, η(y0, y1, t)) ≤ max{g(x, y0), f (x, y1)}

for all x ∈ X, for all t ∈ [0, 1];

(iii) f is lower semicontinuous on the η of Y ;

(iv) For any y1, · · · , yn ∈ Y , any λ ∈ R, the set ∩ni=1{x ∈ X; g(x, yi) ≥ λ} is either connected or empty.

Then for any λ ∈ R, we have the following alternative:

Either there exists x0 ∈ X such that g(x0, y) ≥ λ for all y ∈ Y or there exists y0 ∈ Y such that f(x, y0) < λ for all x ∈ X.

Proof. For each λ ∈ R, y ∈ Y , let

Ug(y) = {x ∈ X; g(x, y) ≥ λ}, and

Uf(y) = {x ∈ X; f(x, y) ≥ λ}.

Fixed λ ∈ R, if for some y0 ∈ Y, Uf(y0) = ∅, then f(x, y0) < λ for all x ∈ X. Thus, we may assume that

Uf(y) 6= ∅. (1)

for all y ∈ Y . We are going to show that

\

y∈Y

Ug(y) 6= ∅.

Since X is compact, we need only to show that the family {Ug(y)}y∈Y has the finite inter- section property, that is,

\

y∈F

Ug(y) 6= ∅ for all finite subsets F ⊂ Y.

If |F |=1. By (1), Uf(y) 6= ∅ for all y ∈ Y . From condition (0) , Ug(y) ⊃ Uf(y), so Ug(y) 6= ∅ for all y ∈ Y .

Let F = {y0, y1} ⊂ Y, y0 6= y1. We want to show that Ug(y0) ∩ Ug(y1) 6= ∅ By condition(0) and

Ug(y0) ∩ Uf(y1) = \

>0

(Ug(y0, λ − ) ∩ Uf(y1, λ − )), it is sufficent to show that for any  > 0,

Ug(y0, λ − ) ∩ Uf(y1, λ − ) 6= ∅ (2)

(3)

Fixed  > 0, for each t ∈ [0, 1], consider the following sets Sg(t) ≡ Ug(η(y0, y1, t), λ − )

= {x ∈ X; g(x, η(y0, y1, t)) ≥ λ − }, and

Sf(t) ≡ Uf(η(y0, y1, t), λ − )

= {x ∈ X; f(x, η(y0, y1, t)) ≥ λ − }.

Notice that, for all t ∈ [0, 1], Sg(t) is a nonempty compact connected subset of X by (1), conditions (i) and (iv), and Sf(t) is nonempty closed subset of X by (1) and condition (i).

Let

A0 ≡ {t ∈ [0, 1]; Sf(t) ⊂ Sg(0)}, and

A1 ≡ {t ∈ [0, 1]; Sf(t) ⊂ Sf(1)},

If Sg(0) ∩ Sf(1) were empty set, we claim (α), (β) and (γ) as follows.

(α) A0∩ A1 = ∅.

This is clear by the construction of these two sets.

(β) A0∪ A1 = [0, 1].

Indeed, by (ii), we have, for all x ∈ X, for all t ∈ [0, 1],

g(x, η(y0, y1, t)) ≤ max{g(x, y0), f (x, y1)}.

For each t ∈ [0, 1], let x ∈ Sf(t), so x ∈ Sg(t) by condition (0) . Then g(x, η(y0, y1, t)) ≥ λ−. It follows that g(x, y0) ≥ λ− or f(x, y1) ≥ λ− by condition (ii). This implies that x ∈ Sg(0) or x ∈ Sf(1), and hence Sg(t) ⊂ Sg(0) ∪ Sf(1). Since Sg(t) is connected, Sg(0) and Sf(1) are nonempty closed sets and Sg(0)∩Sf(1) = ∅, we have either Sg(t) ⊂ Sg(0) or Sg(t) ⊂ Sf(1) and hence Sf(t) ⊂ Sg(0) or Sf(t) ⊂ Sf(1). It follows that t ∈ A0 or t ∈ A1. This shows (β).

(γ) Both A0 and A1 are closed subsets of [0, 1].

Let {tn} be a sequence in A0 with tn−→ t0 ∈ [0, 1]. Let x ∈ Sf(t0).

Then

f (x, η(y0, y1, t)) ≥ λ − . (3)

From (iii), since f is lower semicontinuous on the η of Y , the function F defined by F (t) = f (x, η(y0, y1, t)) is lower semicontinuous on [0, 1]; thus lim infnF (tn) ≥ F (t0). It follows that there exits m ∈ N such that F (tm) ≥ F (t0). That is, f (x, η(y0, y1, tm)) ≥ f (x, η(y0, y1, t0)). By (3), this implies that x ∈ Sf(tm). Since tm ∈ A0, we have x ∈ Sg(0).

Therefore, Sf(t0) ⊂ Sg(0). It follows that t0 ∈ A0, so A0 is closed. Similarly, A1 is closed.

(4)

From (α), (β), (γ), we get a contradiction about the connectedness of [0, 1], since A0 and A1 are nonempty sets (0 ∈ A0 and 1 ∈ A1). Therefore, Sg(0) ∩ Sf(1) 6= ∅. Hence, (2) holds.

Now, we proceed the proof by induction on |F |, and suppose that ∩y∈FUg(y) 6= ∅ for all F ⊂ Y, 2 ≤ |F | ≤ k.

Choose {y0, · · · , yk} ⊂ Y, yi 6= yj if i 6= j. From the above argument, in order to show that ∩ki=0Ug(yi) 6= ∅, it is sufficent to show, for each  > 0,that

Ug(y0, λ − )\

Uf(y1, λ − )\

(∩ki=2Ug(yi, λ − )) 6= ∅. (4) Let  > 0 and H = ∩ki=2Ug(yi, λ − ). Notice that, by induction hypothesis, since

T

>0[H ∩ Ug(y, λ − )] = (∩ki=2Ug(yi))T

Ug(y) 6= ∅ for all y ∈ Y, H ∩ Ug(y, λ − ) 6= ∅ for all y ∈ Y . In particular,

H ∩ Sg(t) 6= ∅ for all t ∈ [0, 1].

Also, by (iv), H ∩ Sg(t) is connected for all t ∈ [0, 1]. For t ∈ [0, 1], we let B0 ≡ {t ∈ [0, 1]; H ∩ Sf(t) ⊂ H ∩ Sg(0)},

and

B1 ≡ {t ∈ [0, 1]; H ∩ Sf(t) ⊂ H ∩ Sf(1)}.

If (H ∩ Sg(0)) ∩ (H ∩ Sf(1)) were empty set, then by the same argument as above, we would have

0) B0∩ B1 = ∅.

0) B0∪ B1 = [0, 1].

0) Both B0 and B1 are nonempty closed subsets of [0, 1].

This again contradicts the connectedness of [0, 1]. Therefore, (4) holds and hence ∩ki=0Ug(yi) 6=

∅.

Remark 2.2. The condition (iv) of Lemma 2.1 is a weaker form introducted by Geraghty and Lin [6] than a connected condition used by Terkelsen [7].

Using the above technique Lemma, we now establish our main result as follow.

Theorem 2.3. Let X be a nonempty compact set of a topological space and let Y be a nonempty η-connected set. Let f, g be real-valued functions defined on X × Y with the following properties:

(0) f (x, y) ≤ g(x, y) for all (x, y) ∈ X × Y ;

(i) f (·, y) and g(·, y) are upper semicontinuous on X for each y ∈ Y ; (ii) For any y0, y1 ∈ Y ,

g(x, η(y0, y1, t)) ≤ max{g(x, y0), f (x, y1)}

(5)

for all x ∈ X, for all t ∈ [0, 1];

(iii) f is lower semicontinuous on the η of Y ;

(iv) For any y1, · · · , yn ∈ Y , any λ ∈ R, the set ∩ni=1{x ∈ X; g(x, yi) ≥ λ} is either connected or empty.

Then

inf

y∈Y sup

x∈Xf (x, y) ≤ sup

x∈X

inf

y∈Yg(x, y) Proof. Let λ ∈ R with

sup

x∈X

inf

y∈Yg(x, y) < λ.

Thus, there exists y0 ∈ Y such that g(x, y0) < λ for all x ∈ X. By Lemma 2.1, there exists y0 ∈ Y such that f(x, y0) < λ for all x ∈ X. This implies that infy∈Y supx∈Xf (x, y) < λ.

It follows that

y∈Yinf sup

x∈Xf (x, y) ≤ sup

x∈X

y∈Yinf g(x, y)

The η-connected property uniformly on Y can be replaced by pointwise η-connectedness defined below. Let Y be a nonempty set. Define

G(y0, y1) ≡



η : Y × Y × [0, 1] −→ Y ; η is continuous and η(y0, y1, 0) = y0 η(y0, y1, 1) = y1



We say Y is pointwise η-connected if for any y0, y1 ∈ Y , there exists η in G(y0, y1) such that

g(x, η(y0, y1, t)) ≤ max{g(x, y0), f (x, y1)}

for all x ∈ X, for all t ∈ [0, 1].

By a similar process of proof in Lemma 2.1, we can get the following theorem imme- diately.

Theorem 2.4. Let X be a nonempty compact set of a topological space, let Y be a nonempty pointwise η-connected set. Let f, g be real-valued functions defined on X × Y with the following properties:

(0) f (x, y) ≤ g(x, y) for all (x, y) ∈ X × Y ;

(i) f (·, y) and g(·, y) are upper semicontinuous on X for each y ∈ Y ; (ii) For any y0, y1 ∈ Y , there exists η in G(y0, y1) such that

g(x, η(y0, y1, t)) ≤ max{g(x, y0), f (x, y1)}

for all x ∈ X, for all t ∈ [0, 1], and

f is lower semicontinuous on the η of Y

(iii) For any y1, · · · , yn ∈ Y , any λ ∈ R, the set ∩ni=1{x ∈ X; g(x, yi) ≥ λ} is either connected or empty. Then for any λ ∈ R, we have the following alternative:

Either there exists x0 ∈ X such that g(x0, y) ≥ λ for all y ∈ Y or there exists y0 ∈ Y such that f(x, y0) < λ for all x ∈ X.

(6)

Theorem 2.5. Let X be a nonempty compact set of a topological space, let Y be a nonempty pointwise η-connected set. Let f, g be real-valued functions defined on X × Y with the following properties:

(0) f (x, y) ≤ g(x, y) for all (x, y) ∈ X × Y ;

(i) f (·, y) and g(·, y) are upper semicontinuous on X for each y ∈ Y ; (ii) For any y0, y1 ∈ Y , there exists η in G(y0, y1) such that

g(x, η(y0, y1, t)) ≤ max{g(x, y0), f (x, y1)}

for all x ∈ X, for all t ∈ [0, 1], and

f is lower semicontinuous on the η of Y

(iii) For any y1, · · · , yn ∈ Y , any λ ∈ R, the set ∩ni=1{x ∈ X; g(x, yi) ≥ λ} is either connected or empty.

Then

inf

y∈Y sup

x∈Xf (x, y) ≤ sup

x∈X

inf

y∈Yg(x, y)

參考文獻

相關文件

The best way to picture a vector field is to draw the arrow representing the vector F(x, y) starting at the point (x, y).. Of course, it’s impossible to do this for all points (x, y),

If x or F is a vector, then the condition number is defined in a similar way using norms and it measures the maximum relative change, which is attained for some, but not all

In this paper, we build a new class of neural networks based on the smoothing method for NCP introduced by Haddou and Maheux [18] using some family F of smoothing functions.

(a) The magnitude of the gravitational force exerted by the planet on an object of mass m at its surface is given by F = GmM / R 2 , where M is the mass of the planet and R is

when certain additional symmetry Y is present, topological invariants of TCIs protected by symmetry X can be inferred by the Y -symmetry eigenvalues of energy

Figure 6 shows the relationship between the increment ∆y and the differential dy: ∆y represents the change in height of the curve y = f(x) and dy represents the change in height

Theorem (Comparison Theorem For Functions) Suppose that a ∈ R, that I is an open interval that contains a, and that f,g are real functions defined everywhere on I except possibly at

Then g is defined on [a, b], satifies (11), and is continuous on [a, b] by the Sequential Characterization of Limits.. Thus, f