MA3422 (Functional Analysis 2) Tutorial sheet 5 [February 20, 2015] Name: Solutions 1. Let E be a normed space and α ∈ E ∗ . Show (a) ker α = {x ∈ E : α(x) = 0} is a closed linear subspace of E Solution: One way to do this is to say that α : E → K is continuous and ker α = {x ∈ E : α(x) = 0} = α−1 ({0}) is the inverse image of a closed set under a continuous function — hence closed. Another way would be to take a sequence (xn )∞ n=1 in ker α which converges (in E) to some x ∈ E. By continuity of α we must have limn→∞ α(xn ) = α(x). But α(xn ) = 0 for each n and so α(x) = 0 also, or x ∈ ker α. This is enough to show that ker α is closed in (the metric space) E. (b) If α 6= 0, there is a continuous linear bijection E → ker α ⊕ K with a continuous inverse [Hint: Choose x0 ∈ E with α(x0 ) 6= 0 and define α(x) x0 , α(x) ] T (x) = x − α(x0 ) Solution: Define T as in the hint. It does make sense since α(x) α(x) α x− x0 = α(x) − α(x0 ) = 0 α(x0 ) α(x0 ) (using linearity of α) and so the first component is in ker α (and the second is a scalar of course). We can check fairly easily that T must be linear: α(x + y) T (x + y) = x+y− x0 , α(x + y) α(x0 ) α(x) + α(y) = x+y− x0 , α(x) + α(y) α(x0 ) α(x) = x− x0 , α(x) α(x0 ) α(y) + y− x0 , α(y) α(x0 ) = T (x) + T (y) α(λx) T (λx) = λx − x0 , α(λx) α(x0 ) λα(x) = λx − x0 , λα(x) α(x0 ) α(x) = λ x− x0 , α(x) = λT (x) α(x0 ) Also T is bounded (let us use the norm k · k1 on E ⊕ F ) since α(x) + |α(x)| kT (x)k1 = x − x 0 α(x0 ) kx0 k + kαkkxk ≤ kxk + |α(x)| |α(x0 )| kx0 k ≤ kxk + kαkkxk + kαkkxk |α(x0 )| kx0 k = 1+ kαk + kαk kxk |α(x0 )| To show it is a bijection, one way would be to write down the inverse map and show that it works. Define S : ker α ⊕ K → E by S(x, λ) = x + λ x0 α(x0 ) (for x ∈ ker α, λ ∈ K). We have α(S(x, λ)) = α(x) + and so λ α(x0 ) = 0 + λ = λ α(x0 ) T (S(x, λ)) = S(x) − λ x0 , λ = (x, λ) α(x0 ) On the other hand, for x ∈ E, S(T (x)) = x is clear. So T has an inverse S and is a bijection. Finally S is easily seen to be linear and bounded since |λ| |λ| kS(x, λ)k ≤ kxk + kx0 k ≤ 1 + (kxk + |λk) α(x0 ) α(x0 ) So T and its inverse s are continuous. 2 2. In E = CP [0, 1] = {f ∈ C[0, 1] : f (0) = f (1)} show that there is α ∈ E ∗ with α(f ) = 0 for f (x) ≡ 1 and α(g) = 1 for g(x) ≡ x(x − 1). [Hint: Hahn-Banach.] Solution: Let M = span{f, g} = {λ1 f + λ2 g : λ1 , λ2 ∈ K} and define α : M → Ik by α(λ1 f + λ2 g) = λ2 Since f and g are linearly independent (if λ1 f + λ2 g = 0, then λ1 f (0) + λ2 g(0) = 0 ⇒ λ1 1 + λ2 0 = 0 ⇒ λ1 = 1 and then evaluation at x = 1/2 [or any x ∈ (0, 1)] gives λ2 g(1/2) = 0 ⇒ λ2 = 0). So α is well defined on M and clearly linear. Since M is finite dimensional, α must be bounded and there is a constant C > 0 so that |α(h)| ≤ Ckhk holds for all h ∈ M . By the Hahn-Banach theorem there is a linear extendsion β : CP [0, 1] → K of α which satisfies |β(h)| ≤ Ckhk for all h ∈ CP [0, 1]. But then β ∈ E ∗ and β(f ) = α(f ) = 0, β(g) = α(g) = 1. Richard M. Timoney 3