Lecture #16 MATH 321: Real Variables II University of British Columbia Lecture #16: Instructor: Scribe: February 11, 2008 Dr. Joel Feldman Peter Wong Remark. (Not part of this course) 1. Reference for Arzelà’s theorem: Tom Apostol, Mathematical Analysis. pp.405-406 2. (Lebesgue dominated convergence theorem) Assume • µ is a measure on R • for each n ∈ N, fn : R → R is measurable • g : R → R is measurable • for each x ∈ R, limn→∞ fn (x) = f (x) exists • |fn (x)| ≤ g(x) for all x ∈ R R • g 0 (x) dµ(x) < ∞ Then f is measurable and limn→∞ Special case: R fn (x) dµ(x) = R f (x) dµ(x). • µ is a Lebesgue measure on [a, b] • fn is Riemann integrable • g(x) = M for some constant M • for each x ∈ R, limn→∞ fn (x) = f (x) exists Convergence and Differentiation Question: If fn is differentiable and converges to f , then is it true that f 0 exists and lim fn0 = f 0 ? Counterexamples. 1. Problem Set 5 Question 2(h): fn differentiable and converges uniformly to f , fn0 converges pointwise to g, but f 0 does not exists. 2. Problem Set 5 Question 2(i): fn differentiable and converges uniformly to a differentiable f , fn0 converges pointwise to g, but f 0 6= g. Theorem. If • for each n ∈ N, fn : [a, b] → R differentiable • { fn0 }n∈N converges uniformly on [a, b] • for some x0 ∈ [a, b], { fn (x0 ) }n∈N converges (to prevent, say, fn (x) = n, which diverges). Then (a) { fn }n∈N converge uniformly on [a, b]. Call the limit f (x) (b) f (x) is differentiable on [a, b], and f 0 (x) = limn→∞ fn0 (x). 2 MATH 321: Lecture #16 Proof. (a) Use the Cauchy test |fn (x) − fm (x)| ≤ |fn (x) − fm (x) − fn (x0 ) + fm (x0 )| + |fn (x0 ) − fm (x0 )| = |(fn − fm )(x) − (fn − fm )(x0 )| + |fn (x0 ) − fm (x0 )| = |(fn − fm )0 (c)(x − x0 )| + |fn (x0 ) − fm (x0 )| for some c between x and x0 0 ≤ kfn0 − fm k∞ (b − a) + |fn (x0 ) − fm (x0 )| Let ε > 0 Then ∃N1 such that n, m ≥ N1 =⇒ |fn (x0 ) − fm (x0 )| < ∃N2 0 such that n, m ≥ N2 =⇒ kfn0 − fm k∞ < =⇒ (b) ε 2 ε 2(b−a) n, m ≥ max{N1 , N2 } =⇒ kfn − fm k∞ < ε. Step 1: Fix x ∈ [a, b]. Define 0 fn (x), φn (t) = fn (x)−fn (t) x−t if t=x φ(t) = , otherwise 0 (x), limm→∞ fm f (x)−f (t) x−t , if t = x otherwise Since fn ’s are differentiable, φn (t)’s must be continuous. Thus, it suffices to prove that φ(t) is continuous: lim φ(t) = φ(x) =⇒ t→x lim t→x f (x) − f (t) 0 = lim fm (x) m→∞ x−t It suffices to prove that φn ’s converge uniformly to φ: 1 |x−t| |fn (x) − fn (t) − f (x) + f (t)|, |φn (t) − φ(t)| = 0 0 |fn (x) − limm→∞ fm (x)|, which does exist if t 6= x if t = x For t 6= x |φn (t) − φ(t)| = ≤ 1 |x−t| |(fn − fm )(x) − (fn − fm )(t)| 0 0 0 1 |x−t| lim sup kfn − fm k∞ |x − t| ≤ lim sup kfn m→∞ m→∞ 0 − fm k∞ Lecture #17 MATH 321: Real Variables II University of British Columbia Lecture #17: Instructor: Scribe: February 13, 2008 Dr. Joel Feldman Peter Wong Proof (Continued): Step 2: We know that f is differentiable and f 0 (x) = limn→∞ fn0 (x). Fix any x ∈ [a, b]. Define ( fn (x)−fn (t) , for t 6= x x−t φn (t) = fn0 (x), for t = x • φn (t) is continuous at t = x because limt→x φn (t) = φn (x) • φn (t) is continuous at t 6= x because the denominator x − t 6= 0 so φn (t) is continuous everywhere for all n ∈ N. Define ( f (x)−f (t) , x−t φ(t) = limn→∞ fn0 (x), for x 6= t for x = t • Need to show φ(t) is continuous: limx→t φ(t) = φ(x) • This means that: lim φ(t) = lim fn0 (x) x→t 0 n→∞ f (x) = lim fn0 (x) n→∞ Let ε > 0. We are assuming { fn0 } converges uniformly; that is, 0 (t)| < ε, for all t ∈ [a, b] ∃N such that n, m ≥ N =⇒ |fn0 (t) − fm Now show that for all t ∈ [a, b], n, m ≥ N =⇒ |φn (t) − φ(t)| ≤ ε. For t = x, 0 (x)| ≤ ε |φn (x) − φ(x)| = lim |fn0 (x) − fm m→∞ For t 6= x, fn (x) − fn (t) f (x) − f (t) |φn (x) − φ(x)| = − x−t x−t 1 |fn (x) − fn (t) − fm (x) + fm (t)| = lim m→∞ |x − t| 1 = lim |(fn − fm )(x) − (fn − fm )(t)| m→∞ |x − t| 1 = lim |(fn − fm )0 (c)||x − t| m→∞ |x − t| 0 = lim |fn0 (c) − fm (c)| ≤ ε when m ≥ N m→∞ Theorem. (Arzelà-Ascoli, See Rudin pp.154-158) Let K be a compact metric space and C(K) be the space of real (or complex) valued continuous functions on K. If { fn } is a sequence in C(K) obeying (H1) for each p ∈ K, there is a ϕ(p) > 0 such that |fn (p)| < ϕ(p) for all n ∈ N (then the sequence { fn } is called pointwise-bounded), and 2 MATH 321: Lecture #17 (H2) ∀ε > 0, ∃δ > 0 such that for all n ∈ N, dK (p, p0 ) < δ =⇒ |fn (p) − fn (p0 )| < ε. Then (a) there is an M > 0 such that |fn (p)| < M for all p ∈ K and all n ∈ N (b) the sequence { fn } has a uniformly convergent subsequence Proof. Step 1 (of 4): Lemma 1. There is a countable subset { p` | ` ∈ N } ⊂ K that is dense in K. Proof of Lemma 1: K is compact =⇒ K is totally bounded, that is, ∀ε > 0, K can be covered by a finite union of balls of radius ε. In particular, for each m ∈ N, there are a finite number 1 of one of of points pm,1 , pm,2 , . . . , pm,nm such that every point in K is within a distance of M pm,1 , pm,2 , . . . , pm,nm . We can then choose { p1 , p2 , . . . } = { p1,1 , . . . , p1,n1 , p2,1 , . . . , p2,n2 , . . . } | {z } | {z } m=1 m=2 Lecture #18 MATH 321: Real Variables II University of British Columbia Lecture #18: Instructor: Scribe: February 15, 2008 Dr. Joel Feldman Peter Wong Step 1 (of 4): Lemma 1. There is a countable subset { p` | ` ∈ N } ⊂ K that is dense in K. Proof of Lemma 1: Done last class. Step 2 (of 4): Lemma 2. Let { fn }n∈N be any pointwise bounded sequence of functions on the countable set { p` }`∈N . Then there is a subsequence { fnk }k∈N of { fn }n∈N that converges at every p` , ` ∈ N. (i.e., limk→∞ fnk (p` ) exists for each ` ∈ N.) Proof. • Concentrate on p1 . By hypothesis, { fn (p1 ) }n∈N is bounded =⇒ ∃M1 such that |fn (p1 )| ≤ M1 for alln ∈ N Since { y ∈ R | |y| ≤ M1 } is compact, there is a convergent subsequence fnj (p1 ) j∈N =⇒ Rename nj to N1 (j). • Concentrate on p2 . By hypothesis, =⇒ ∃M2 fN1 (j) (p2 ) n∈N is bounded such that |fN1 (j) (p2 )| ≤ M2 for allj ∈ N Since { y ∈ R | |y| ≤ M2 } is compact, =⇒ there is a convergent subsequence fNj (jk ) (p2 ) k∈N Rename N1 (jk ) to N2 (k). • Repeat: For each ` ∈ N, we construct a subsequence fN` (m) m∈N of fN`−1 (p) p∈N and hence of { fn }n∈N for which fN` (m) (p` ) m∈N converges. • Choose the subsequence in the conclusion of this Lemma to be fNn (n) n∈N . This subse quence converges at every p` because, for each ` ∈ N, fNn (n) (p` ) n≥` is a subsequence of fn` (j) (p` ) j∈N , because for n ≥ `, Nn (n) is in the range of Nn , R(Nn ), and R(Nn ) ⊂ R(Nn−1 ) ⊂ · · · ⊂ R(N` ). Step 3 (of 4): Proof of Arzelà-Ascoli (a), i.e., (H1) Pointwise bounded + (H2) equicontinuity =⇒ (a) uniformly bounded. Proof. We have to find an M ≥ 0 such that |fn (p)| ≤ M for all n ∈ N and p ∈ K. Apply equicontinuity with ε = 1. So ∃δ such that |fn (p) − fn (q)| < 1 for all n ∈ N, p, q ∈ K with dK (p, q) < δ Since K is compact, there are finitely many points p1 , . . . , pr in K such that every p ∈ K is within a distance of δ of at least one of the points. For all n ∈ N, p ∈ K, |fn (p)| ≤ |fn (p) − fn (pi )| + |fn (pi )| < 1 + φ(pi ) ≤ 1 + max φ(pi ) = M 1≤i≤r with pi chosen such that dK (p, pi ) < δ 2 MATH 321: Lecture #18 Step 4 (of 4): Proof of Arzelà-Ascoli (b) (H1) Pointwise bounded + (H2) equicontinuity =⇒ (b) there is a subsequence which converges uniformly. Proof. • Apply Lemma 1 to give a countable dense subset { p` }`∈N of K. • Apply Lemma 2 to give a subsequence fnj j∈N which converges at each p` , ` ∈ N.