MA3422 (Functional Analysis 2) Tutorial sheet 3 [February 6, 2015] Name: Solutions 1. For H a Hilbert space and y, z ∈ H, define T : H → H by T (x) = hx, yiz. Verify that T ∈ B(H) and kT kop = kykkzk. Solution: To show that T is linear, T (x1 + λx2 ) = hx1 + λx2 , yiz = (hx1 , yi + λhx2 , yi)z = hx1 , yiz + λhx2 , yiz = T (x1 ) + λT (x2 ) To show kT kop ≤ kykkzk, we use Cauchy-Schwarz kT (x)k = |hx, yi|kzk ≤ kxkkykkzk = (kykkzk)kxk Then if y = 0 we have kT kop = 0. If y 6= 0, take x = y/kyk and we get (since kxk = 1) |hy, yi| y kyk2 = , y z kzk == kzk = kykkz| kT kop ≤ kT (x)k = kyk kyk kyk Also compute T 2 (x). Solution: T 2 (x) = T (T (x)) = hT (x), yiz = hhx, yiz, yiz = hx, yihz, yiz = hz, yiT (x) 2. Let H = Kn be the n-dimensional Hilbert space (with the k·k2 or euclidean norm). For g = (g1 , g2 , . . . , gn ) ∈ Kn , define Mg ∈ B(Kn ) by Mg (x1 , x2 , . . . , xn ) = (g1 x1 , g2 x2 , . . . , gn xn ). Show that kMg kop = max1≤i≤n |gi |. Solution: Recall the notation kgk∞ = max1≤i≤n |gi |. We have v uX u n kMg (x1 , x2 , . . . , xn )k = t |gj xj |2 j=1 v uX u n = t |gj |2 |xj |2 j=1 v uX u n ≤ t kgk2∞ |xj |2 j=1 v uX u n = kgk∞ t |xj |2 j=1 = kgk∞ k(x1 , x2 , . . . , xn )k2 This shows kMg kop ≤ kgk∞ . To show the reverse inequality, take x = ej (a standard basis vector for Kn ) and note that Mg (ej ) = gj ej , kMg (ej )k2 = |gj |kej k2 = |gj | and (since kej k2 = 1) we then have kMg kop ≥ |gj | for each j (1 ≤ j ≤ n). Thus kMg kop ≥ kgk∞ . This completes the proof. 3. For H = L2 [0, 1], define Mt : H → H by (Mt f )(t) = tf (t). Show that kMt kop = 1. [Hint: First check kMt kop ≤ 1 and then choose f supported close to 1.] Solution: Since |t| ≤ 1 for 0 ≤ t ≤ 1 we have s s Z 1 Z 2 |tf (t)| dt ≤ kMt f k2 = 0 1 |f (t)|2 dt = kf k2 0 and so kMt kop ≤ 1. If we take f = χ[1−δ,1] (where 0 < δ < 1) then s sZ Z 1 kf k2 = (χ[1−δ,1] )(t))2 dt = 0 s Z √ δ r q 1 − (1 − δ)3 t2 dt = [t3 /3]11−δ = 3 1−δ (tχ[1−δ,1] )(t))2 dt = kMt f k2 = 1 dt = 1−δ sZ 1 1 0 1 So kMt kop kMt f k2 ≥ = kf k2 r 1 − (1 − δ)3 = 3δ r 3δ − 3δ 2 + δ 3 p = 1 − δ + δ 2 /3 3δ Let δ → 0 to get kMt kop ≥ 1 and conclude the proof. 2 Remark: Perhaps it is almost easier to show the more general result that for g ∈ L∞ [0, 1] the operator Mg : L2 [0, 1] → L2 [0, 1] given by Mg (f ) = gf has kMg kop = kgk∞ . (Q3 was the special case where g(t) = t.) For the inequality kMg kop ≤ kgk∞ , use the fact that |g(t)| ≤ kgk∞ for almost all t ∈ [0, 1] to justify s s s Z 1 Z 1 Z 1 2 2 2 kMg (f )k2 = |g(t)f (t)| dt = |g(t)| |f (t)| dt ≤ kgk2∞ |f (t)|2 dt 0 0 0 s Z 1 |f (t)|2 dt = kgk∞ kf k2 = kgk∞ 0 For the reverse inequality kMg kop ≥ kgk∞ , it is certainly true if kgk∞ = 0. If kgk∞ > 0 then for 0 < δ < kgk∞ , Eδ = {t ∈ [0, 1] : |g(t)| > kgk∞ − δ} has positive measure (otherwise we would have |g(t)| ≤ kgk∞ − δ almost everywhere, contradicting the definition of the essential supremum kgk∞ ). We can take f = χEδ and compute s sZ Z 1 p kf k2 = χEδ (t)2 dt = 1 dt = µ(Eδ ) 0 Eδ (where µ is Lebesgue measure). Also s sZ Z 1 kMg (f )k2 = |g(t)|2 χEδ (t)2 dt = 0 sZ ≥ |g(t)|2 1 dt Eδ p (kgk∞ − δ)2 dt = (kgk∞ − δ) µ(Eδ ) Eδ So kMg kop ≥ kMg (f )k2 = kgk∞ − δ kf k2 Let δ → 0 to get kMg kop ≥ kgk∞ . Combining that with the reverse inequality, we have kMg kop = kgk∞ . This way we have only actually integrated multiples of characteristic functions. By choosing f to be supported where g has nearly its maximum modulus, we see that we can’t improve on the rather easy estimate that gave kMg kop ≤ kgk∞ . Richard M. Timoney 3