134 CHAPTER 5. FUNCTIONS ON NORMED LINEAR SPACES

Definition 5.11.4 A norm is called strictly convex if whenever x ̸= y,∥∥∥∥x+ y2

∥∥∥∥< ∥x∥2 +∥y∥2

Theorem 5.11.5 Let E,F be Banach spaces with E having a strictly convex norm.Also suppose that A ⊆ E,B ⊆ F are compact and convex sets and that H : A×B→ R issuch that

x→ H (x,y) is convex

y→ H (x,y) is concave

Assume that x→ H (x,y) is lower semicontinuous and y→ H (x,y) is upper semicontin-uous. Then minx∈A maxy∈B H (x,y) = maxy∈B minx∈A H (x,y) . This condition is equivalentto the existence of (x0,y0) ∈ A×B such that

H (x0,y)≤ H (x0,y0)≤ H (x,y0) for all x,y (5.8)

called a saddle point.

Proof: One part of the main equality is obvious.

maxy∈B

H (x,y)≥ H (x,y)≥minx∈A

H (x,y)

and so for each x, maxy∈B H (x,y)≥maxy∈B minx∈A H (x,y) and so

minx∈A

maxy∈B

H (x,y)≥maxy∈B

minx∈A

H (x,y) (5.9)

Next consider the other direction.Define Hε (x,y)≡ H (x,y)+ ε ∥x∥2 where ε > 0. Then Hε is strictly convex in the first

variable. This results from the observation that∥∥∥∥x+ y2

∥∥∥∥2

<

(∥x∥+∥y∥

2

)2

≤ 12

(∥x∥2 +∥y∥2

),

By Lemma 5.11.3 there exists a unique x ≡ g(y) with Hε (g(y) ,y) ≡ minx∈A Hε (x,y) andalso, whenever y,z ∈ A, limt→0+ g(y+ t (z− y)) = g(y). Thus

Hε (g(y) ,y) = minx∈A

Hε (x,y) .

But also this shows that y→ Hε (g(y) ,y) is the minimum of functions which are uppersemicontinuous and so this function is also upper semicontinuous. Hence there exists y∗

such thatmaxy∈B

Hε (g(y) ,y) = Hε (g(y∗) ,y∗) = maxy∈B

minx∈A

Hε (x,y) (5.10)

Thus from concavity in the second argument and what was just defined, for t ∈ (0,1) ,

Hε (g(y∗) ,y∗)≥ Hε (g((1− t)y∗+ ty) ,(1− t)y∗+ ty)

≥ (1− t)Hε (g((1− t)y∗+ ty) ,y∗)+ tHε (g((1− t)y∗+ ty) ,y)

≥ (1− t)Hε (g(y∗) ,y∗)+ tHε (g((1− t)y∗+ ty) ,y) (5.11)

134 CHAPTER 5. FUNCTIONS ON NORMED LINEAR SPACESDefinition 5.11.4 4 norm is called strictly convex if whenever x # y,“yt|< lel lyTheorem 5.11.5 Let £E ,F be Banach spaces with E having a strictly convex norm.Also suppose that A C E,B C F are compact and convex sets and that H : A x B > R issuch thatx — H (x,y) is convexy > H (x,y) is concaveAssume that x — H (x,y) is lower semicontinuous and y — H (x,y) is upper semicontin-uous. Then minyea Maxyeg H (x,y) = Maxyeg Minye, H (x,y). This condition is equivalentto the existence of (xo,yo) € A x B such thatH (x0,y) SH (x0, Yo) SH (x, yo) for all x,y (5.8)called a saddle point.Proof: One part of the main equality is obvious.max H (x,y) > H (x,y) > min H (x, y)yeBand so for each x, maxyeg H (x,y) > maxyeg Minye, H (x,y) and soH (x,y) > H 5.9minmax (x,y) 2 max min (x,y) (5.9)Next consider the other direction.Define H, (x,y) =H (x,y) +€ ||x||? where € > 0. Then H, is strictly convex in the firstvariable. This results from the observation that2 2Ill] + Ill 1 ( 2 ”)ao a < _<( 5 S5 IIx[/" + lly"),By Lemma 5.11.3 there exists a unique x = g(y) with He (g(y),y) = minye, He (x,y) andalso, whenever y,z € A, lim,.9+ g(y+t(z—y)) =g(y). Thusx+y2He (g(y),y) = min He (x,y).But also this shows that y + He (g(y),y) is the minimum of functions which are uppersemicontinuous and so this function is also upper semicontinuous. Hence there exists y*such thatmax He (g(y) ,y) = He (g (y*) ,y*) = max min Ag (x,y) (5.10)yeB yEB xcAThus from concavity in the second argument and what was just defined, for t € (0,1),He (g(y"),y") 2 He(g((l—t)y" +1y),I—a)y" +49)> (1-1) He (g((1—t) y* +1ty) ,y*) +tHe (g ((1 —t) y* +1ty),y)> (1—t) He (g(y") ,y") +tHe (g ((1—-t) y* +ty),y) (5.11)