MATH 321: Real Variables II Lecture #19 University of British Columbia Lecture #19: Instructor: Scribe: February 25, 2008 Dr. Joel Feldman Peter Wong Theorem. (Arzelà-Ascoli) If K is a compact metric space and { fn }n∈N is a sequence of continuous functions on K that are (H1) pointwise-bounded, and (H2) equicontinuous, then (a) the sequence is uniformly bounded, and (b) it contains a uniformly convergent subsequence. Step 4 (of 4): Proof of (b). • By Lemma 1, there is a countable dense subset { p` }`∈N in K. • By Lemma 2, there is a subsequence { fnk }k∈N such that limk→∞ fnk (p` ) exists for each ` ∈ N. • Choose { fnk } as our subsequence. Rename fnk to gk . We shall prove the { gk } converges uniformly by verifying ∀ε > 0, ∃N ∈ N such that ∀p ∈ K, i, j ≥ N =⇒ |gi (p) − gj (p)| < ε. Let ε > 0. • By equicontinuity (H2), ∃δ such that if dK (p, p0 ) < δ then |gk (p) − gk (p0 )| < ε/3. • Now, we look for a finite number of p` ’s, p`1 , . . . , p`r such that every p ∈ K is less than a distance of δ away from at least one of p`1 , . . . , p`r . Since K is compact, K is totally bounded, which means that: Given any δ > 0, K is covered by finitely many open balls of radius δ/2. That is, there are finitely many points q1 , . . . , qr in K such that each p ∈ K is less than a distance of δ/2 away from at least one of q1 , . . . , qr . By the density of { p` }, for each 1 ≤ m ≤ r, there is a p`m within a distance of δ/2 away from qm . Thus, { p`1 , . . . , p`r } work. • For each 1 ≤ m ≤ r, { gk (p`m ) } converges. This means that ∃Nm such that i, j ≥ Nm =⇒ |gi (p`m ) − gj (p`m )| < ε . 3 • Choose N = max 1 ≤ m ≤ rNm . Then i, j ≥ N =⇒ |gi (p) − gj (p)| ≤ |gi (p) − gi (p`m )| + |gi (p`m ) − gj (p`m )| + |gj (p`m ) − gj (p)| < ε/3 + ε/3 + ε/3 =ε (equicontinuity) (convergence of p`m ) (equicontinuity) with m chosen so that dk (p, p`m ) < δ. Corollary. Let a < b, and c ∈ [a, b]. Let, for each n ∈ N, fn : [a, b] → R be a diferentiable function. Assume that (i) M = sup{ |fn (c)| : n ∈ N } < ∞ (ii) M 0 = sup{ |fn0 (x)| : x ∈ [a, b] n ∈ N } < ∞ Then { fn } has a uniformly convergent subsequence. 2 MATH 321: Lecture #19 Proof. Apply Arzelà-Ascoli with K = [a, b], which is compact. Step (1) Verification of equicontinuity (H2): Let ε > 0. Choose δ = ε/M 0 . Then x, y ∈ [a, b] obey |x − y| < δ =⇒ |fn (x) − fn (y)| = |fn0 (c̃)(x − y)| ≤ M 0 |x − y| < M 0 δ = M ε/M 0 = ε. Step (2) Verification of Pointwise boundedness (H1): Let x ∈ [a, b]. Then |fn (x)| ≤ |fn (c)| + |fn (x) − fn (c)| ≤ M + |fn0 (c̃)(x − c)| ≤ M + M 0 |x − c| = Φ(x) ≤ M + M 0 (b − a) Notes: No lecture on this Wednesday – Lecture #20 will be a midterm exam session. MATH 321: Real Variables II Lecture #21 University of British Columbia Lecture #21: Instructor: Scribe: February 29, 2008 Dr. Joel Feldman Peter Wong Notes: No lecture on Wednesday – Lecture #20 was a midterm exam session. Third Issue: How to approximate some possibly complicated functions by some simple functions? Theorem. (Weierstraß, 1885) If f : [a, b] → C is continuous, then there exists a sequence of polynomials { Pn }n∈N that converges uniformly to f . Further, if f is real-valued, then the Pn ’s may also be taken as real-valued. Idea of the proof: Z f (x) = f (t)δ(x − t) dt Of course, f cannot be so naı̈vely defined by a Dirac delta function this way. What weRwant to do is to find some polynomials Qn which approximated δ(t) very well. To do this, we will define Pn (x) = f (t)Qn (x − t) dt. Proof. Reduction #1: It suffices to consider [a, b] = [0, 1]. Otherwise, define h : [0, 1] → C by h(y) = f ((b−a)y+a). (Check: h(0) = f (a) and h(1) = f (b)) x−a x − a sup |h(y) − Pn (y)| = sup h − Pn b−a b−a y∈[0,1] x∈[a,b] Reduction #2: It suffices to consider the case in which f (0) = 0, f (1) = 0. Otherwise, we should write f (x) = f (0) + x [f (1) − f (0)] + {f (x) − f (0) − x [f (1) − f (0)]} Let g(x) = {f (x) − f (0) − x [f (1) − f (0)]}. Notice that g obeys g(0) = 0 and g(1) = 0. Also, f (0) + x [f (1) − f (0)] is already a polynomial. R1 Definition of Qn (x)(≈ δ(x)): Qn (x) = cn (1 − x2 )n , where cn is chosen such that −1 Qn (x) dx = 1 (i.e., cn = R1 { −1 (1 − x2 )n }−1 .) Also, notice that = 1, for x = 0, (1 − x2 )n = 0, for x = ±1, < 1, for 0 < |x| < 1. b b −1 0 Properties of Qn (x) for −1 ≤ x ≤ 1: 1 (1) Qn (x) ≥ 0 (2) decreases as |x| increases R1 (3) −1 Qn (x) dx = 1 √ (4) 0 < δ ≤ |x| ≤ 1 =⇒ Qn (x) ≤ Qn (δ) = cn (1 − δ 2 ) ≤ n(1 − δ 2 )n → 0 as n → ∞, since Z 1 Z 1 2 n 1 (1 − x ) dx = 2 (1 − x2 )n dx cn ≥ 0 −1 ≥2 Z ≥2 Z 0 √ 1/ n √ 1/ n 0 ≥ and since (1 − y)n = 1 − ny + (1 − x2 )n dx h (1 − nx ) dx = 2 x − 2 nx3 3 √1 n 1 d2 2! d2 y (1 − y)n y=c = 1 − ny + 1 2! n(n √ i1/ n = 0 4 √1 3 n − 1)(1 − c)n−2 y 2 ≥ 1 − ny. 2 MATH 321: Lecture #21 Definition of Pn (x): Extend f (x) to −∞ < x < ∞ by setting f (x) = 0 for x ∈ [0, 1]. Define Pn (x) = = = Z 1 f (x + t)Qn (t) dt, −1 Z 1+x∈[1,2] setting s = x + t f (s)Qn (s − x) ds −1+x∈[−1,0] Z 1 f (s)Qn (s − x) ds 0 where Qn is now a polynomial in s and x. Notice that Pn is a real-valued polynomial if f is real-valued. Midterm Results and Comment: x̄ = 18.2 30 #2(a) If you wrote |fn (xn ) − f (x)| ≤ |fn (xn ) − fn (x)| + |fn (x) − f (x)|, you ran into a wall. See solution.