Brandon Behring Functional Analysis HW 4 Exercise 12.4 Give an example of a monotonic decreasing sequence of nonnegative functions converging point wise to a function f that the equality in Theorem 12.33 does not hold. Solution: Let fn = n1 χ0≤n , then fn → f = 0 but Z Z Z n 1 1 fn dµ = χ0≤n dµ = dµ = 1 n 0 n but Z Z fn dµ = so 1 χ0≤n dµ = n Z lim n→∞ Z 0 n 1 dµ = 1 n Z fn dµ = 1 6= f dµ = 0. Exercise 12.6 Use the dominated convergence theorem to prove Corollary 12.36 for differentiation under an integral sign. Proof: Let us take all the assumptions of Corollary 12.36. For a fixed t ∈ I, let n be big enough that t + 1/n ∈ I then define fn (x, t) ≡ f (x, t + n1 ) − f (x, t) 1/n By assumption (b), f (x, ·) is differentiable almost everywhere in I so fn (x, t) → ∂f (x, t)/∂t as n → ∞. We can now also use the mean value theorem to see that for some c ∈ (t, t + 1/n) such that |f (x, t + 1/n) − f (x, t)| ≤ | ∂f 1 1 (x, c)| · | | ≤ g(x) ∂t n n where the second inequality comes from assumption (c), that g(x) is an integrable function that dominates the partial derivatives with respect to t for every t ∈ I. Thus, |fn (x, t)| ≤ g(x) for all x ∈ X and we can apply the Lebesgue Dominated Convergence to bring the limit inside the integral Z Z lim fn dµ = f dµ. n→∞ 1 (note: assumption (a) tells us that each fn does exist) Noting that Z ρ(t) = f (x, t)dµ(x) X than ρ(t + 1/n) − ρ(t) = 1/n Z X f (x, t + n1 ) − f (x, t) dµ(x) = 1/n Z fn (x, t)dµ(x). X Taking the limits as n → ∞, we have dρ (t) dt ρ(t + 1/n) − ρ(t) 1/n Z fn (x, t)dµ(x) = lim n→∞ X Z = lim fn (x, t)dµ(x) n→∞ ZX ∂f (x, t) = dµ(x) ∂t X = lim n→∞ as claimed. Exercise 12.8 Let fn : X → C be a sequence of measurable functions converging to f pointwise almost everywhere. Suppose there exists g ∈ Lp (X) such |fn | ≤ g a.e. Prove that fn → in the Lp -norm. Proof: Note that for all n, |fn | ≤ g almost everywhere so we must also have |f | ≤ g almost everywhere. Thus |f − fn | ≤ |f | + |fn | ≤ g + g = 2g almost everywhere and thus |f − fn |p ≤ (2g)p = 2p g p . Since g ∈ Lp (X), we can use the dominated convergence theorem to bring the limit inside the integral Z lim n→∞ |f − fn |p dµ = Z lim |f − fn |p dµ = n→∞ 2 Z 0dµ = 0. Exercise 12.10 If x, y ≥ 0 and > 0 is any positive number, then 2 1 x + y2 . 2 2 xy ≤ Proof: If x, y ≥ 0 and > 0, then 0 ≤ = 2 1 √ x − y 1 x2 − 2xy + y 2 . √ Bringing 2xy to the left hand side and dividing both sides by two gives us 1 2 x + y2 . 2 2 xy ≤ Exercise 12.12 Prove the following generalization of Hölders inequality: if 1 ≤ pi ≤ ∞, where i = 1, . . . , n, satisfy n X 1 = 1, p i=1 i and fi ∈ Lpi (X, µ), then f1 · · · fn ∈ L1 (X, µ) and Z | f1 · · · fn dµ| ≤ ||f1 ||p1 · · · ||fn ||pn . Proof: Let us proceed by induction. For n = 1 this is basic and for n = 2 it is Holder’s inequality. Let us assume this is true for n functions and see this implies the inequality holds for n+1 functions. Assume fn+1 ∈ Lpn+1 , P n+1 1 i=1 pi = 1 and let g(x) = f1 f2 · · · fn . Then by Holder, Z | where (f1 · · · fn )fn+1 dµ| = ||g||Q ||fn+1 ||pn+1 1 1 + =1 Q pn+1 or Q= We also have 3 pn+1 . pn+1 − 1 n+1 X i=1 n X 1 1 1 = + =1 pi p pn+1 i=1 i so n X 1 pn+1 − 1 1 1 =1− = = p p p Q i n+1 n+1 i=1 This tells us that Pn Q i=1 pi = 1. 1 Q Noting that if ||g Q ||1 = ||g||Q we can apply our inductive hypothesis on Pn fiQ ∈ Lpi /Q noting that i=1 pQi = 1 Z | (f1 · · · fn )fn+1 dµ| = ||g||Q ||fn+1 ||pn+1 1 = ||g Q ||1Q ||fn+1 ||pn+1 1 = ||f1Q f2Q · · · fnQ ||1Q ||fn+1 ||pn+1 1 1 1 = ||f1Q ||pQ1 /Q · ||f2Q ||pQ2 /Q · · · ||fnQ ||pQn /Q ||fn+1 ||pn+1 = ||f1 ||p1 ||f2 ||p2 · · · ||fn ||pn ||fn+1 ||pn+1 Exercise 12.15 If f ∈ Lp Rn ∩ Lq (Rn ), where p < q, then f ∈ Lr (Rn ) for any p < r < q, and show that 1/r−1/q 1/p−1/r ||f ||r ≤ (||f ||p ) 1/p−1/q (||f ||q ) 1/p−1/q This result is one of the simplest examples of an interpolation inequality. Proof: Let λ = 1/p−1/r 1/p−1/q , then we can rewrite this as ||f ||r ≤ (||f ||p )1−λ (||f ||q )λ . Noting that if we define P = Q = then 4 p r(1 − λ) q rλ 1 1 + P Q = = = = = = = r(1 − λ) rλ + p q r(1 − λ) rλ + p q r 1/r − 1/q r 1/p − 1/r + p 1/p − 1/q q 1/p − 1/q 1 − r/q r/p − 1 + 1 − p/q q/p − 1 q−r r−p + q−p q−p q−r+r−p q−p q−p =1 q−p so P are Q are conjugate. Using Holder’s inequality Z Z |f |r dµ = |f |(1−λ)r |f |λr dµ ||f (1−λ)r ||P ||f λr ||Q Q1 Z P1 Z λQr (1−λ)P r |f | dµ = |f | dµ ≤ Z = = r(1−λ) Z rλ p q q |f | dµ |f | dµ p (||f ||p )(1−λ)r (||f ||q )λr . Taking the r-th roots of both sides gives us Z r1 r ||f ||r = |f | dµ ≤ (||f ||p )1−λ (||f ||q )λ . Exercise 12.16 A function f : Rn → C is said to be Lp continuous if τh f → f in Lp (Rn ) as h → 0 in Rn , where τh f (x) = f (x − h) is the translation of f by h. Prove that, if 1 ≤ p < ∞, every f ∈ Lp (Rn ) is Lp - continuous. Give a counter-example to show this result is not true when p = ∞. Proof: We wish to prove τh f → f in Lp , that is as h → 0 we have ||τh f − f ||p → 0. 5 Because Cc (Rn ) are dense in Lp (R)n , for any > 0 we can approximate f by a continuous function of compact support g such ||f − g||p < /3. Additionally, doing a simple substitution u = x − h ||τh f − τh g||pp Z |τh f (x) − τh g(x)|p dn x = Rn Z |f (x − h) − g(x − h)|p dn x = Rn Z = Rn |f (u) − g(u)|p dn u = ||f − g||pp so we also have ||τh f − τh g||pp < /3. Since g is uniformly continuous on a compact set, we clearly have Z ||τh g − g|| = |g(x − h) − g(x)|p dn x < /3 Rn for sufficiently small h and thus ||τh f − f ||p = ||τh f − τh g + τh g − g + g − f ||p ≤ ||τh f − τh g||p + ||τh g − g||p + ||g − f ||p + + = . < 3 3 3 This proof fails for p = ∞ because the set of continuous functions of compact support is not dense in L∞ . Exercise 12.17 Prove that the unit ball in Lp ([0, 1]), where 1 ≤ p ≤ ∞ is not strongly compact. Proof: We need to find a sequence fn in Lp ([0, 1]) such that ||fn ||p = 1 and fn does not have a convergent subsequence. n Let In = 21n , 2 21n and fn = 2 p χIn then for finite p Z ||fn ||p 1 n |2 p χIn |p dµ = 0 Z 2−n+1 2n dµ = 2−n = 2n · (2 = 2n · 6 1 1 − n) 2n 2 1 =1 2n and for n 6= m we note that χIm χIn = 0 and 1 Z ||fn − fm ||pp m n |2 p χIn − 2 p χIm |p dµ = 0 1 Z = n m 2 p χIn + 2 p χIm dµ 0 = ||fm ||p + ||fn ||p = 1 + 1 = 2. Thus, the distance between any two points is a constant and thus we can have no Cauchy sequence and no convergent subsequence. If p = ∞, let fn = χIn then clearly ||fn ||p = 1 and ||fn − fm ||∞ = 1 since we have disjoint characteristic functions. From here- we repeat the same argument to conclude we have no convergent subsqeuences. Exercise 12.18 Give an example of bounded sequence in L1 ([0, 1]) that does not have a weakly convergent subsequence. Why doesn’t this contradict the Banach-Alaoglu theorem. Proof: We use the same constructions from the previous problem but with p = 1. The sequence fn converges weakly if for any element of ρ(L1 )∗ then ρ(fn ) → ρ(f ). If we take 1 Z ρ(f ) = f 0 ∞ X (−1)k χInk dµ k=1 then Z ρ(fnj ) 1 = f nk 0 = 2 nj 2 nj Z 0 = (−1)k χInk dµ k=1 Z 1 χInj 0 = ∞ X ∞ X (−1)k χInk dµ k=1 1 χInj (−1)j dµ (−1)j 2nj Z 2−nj +1 dµ 2−nj = (−1)j 2nj 1 = (−1)j . 2nj so fnj does not converge weakly. This does not contradict the BanachAlaoglu theorem because L1 is not reflexive and weak-∗ and weak convergence are not the same. 7
© Copyright 2026 Paperzz