Exercises for Section 1.1: Norm and Inner Product 1. Define the ℓ1 -norm on Rn by kxk1 = n X i=1 |xi |, and define the sup-norm on Rn by kxk∞ = sup |xi | . Show that these satisfy Theorem ??. Proof. (a) It should be clear that kxk1 ≥ 0, and kxk∞ ≥ 0, and that eqality holds in each case if and only if x = 0. (b) kx + yk1 = n X i=1 i i |x + y | ≤ n X i=1 |xi | + |y i| = kxk1 + kyk1. This relies on the triangle inequality on R: |a + b| ≤ |a| + |b|. kx + yk∞ = sup |xi + y i| . i But note that by the triangle inequality on R, |xi +y i| ≤ |xi |+|y i|, and therefore, sup |xi + y i| ≤ sup |xi | + |y i| i i ≤ sup |xi | + sup |y i| i i = kxk∞ + kyk∞ . (c) Note that in general, sup {ai + bi } ≤ sup {ai } + sup {bi }. kaxk1 = X |axi | = X |a| · |xi | = |a| X |xi | = |a| · kxk1 . kaxk∞ = sup |axi | = sup |a| · |xi | = |a| sup |xi | = |a|·kxk∞ . i i i 1 2. Prove that kxk ≤ than the ℓ1 -norm. n X i=1 |xi |. In other words, the usual norm is no greater Proof. First note that for a, b ≥ 0, square both sides). Then √ a+b ≤ √ a+ √ b (to see this, v u n n n q X X uX 2 2 i t i (x ) = |xi | = kxk1 . kxk = (x ) ≤ i=1 i=1 i=1 3. Prove that kx − yk ≤ kxk + kyk. (Compare this with part (2) of Theorem ??.) When does equality hold? Proof. kx − yk = kx + (−y)k ≤ kxk + k − yk = kxk + kyk. Equality holds if and only if y = −λx for some λ > 0. The “if” part is clear, let’s prove the “only if.” Suppose kx − yk = kxk + kyk. Then kx − yk2 = kxk2 + 2kxk · kyk + kyk2 But kx − yk2 = hx − y, x − yi = hx, xi − 2 hx, yi + hy, yi , and kxk2 + 2kxk · kyk + ky 2k2 = hx, xi + 2kxk · kyk + hy, yi . Comparing these, we see that |hx, yi| = kxk · kyk. But Cauchy-Schwarz says this can only happen if y = tx for some t ∈ R. We will now show that t is negative. − hx, yi = kxk · kyk, but − hx, yi = −t hx, xi, and kxk · kyk = |t| · kxk2 . Thus |t| = −t, so t is negative. 4. Prove that kxk − kyk ≤ kx − yk. 2 Proof. In proving this, we will see an alternative approach to Problem 3, above. Look at kx − yk2 . kx − yk2 = hx − y, x − yi = hx, xi − 2 hx, yi + hy, yi = kxk2 − 2 hx, yi + kyk2. But, by Cauchy Schwarz, hx, yi ≤ kxk · kyk. Therefore, we may conclude that kxk2 −2kxk·kyk+kyk2 ≤ kxk2 −2 hx, yi+kyk2 ≤ kxk2 +2kxk·kyk+kyk2. The left-hand side is (kxk − kyk)2 , the center is kx − yk2, and the right-hand side is (kxk + kyk)2 . So we have (kxk − kyk)2 ≤ kx − yk2 ≤ (kxk + kyk)2 . Take the square root throughout: kxk − kyk ≤ kx − yk ≤ kxk + kyk. 5. The quantity ky − xk is called the distance between x and y. Prove and interpret the “triangle inequality”: kz − xk ≤ kz − yk + ky − xk. Proof. kz − xk = kz − y + y − xk = k(z − y) + (y − x)k ≤ kz − yk + ky − xk. 6. Let f and g be integrable on [a, b]. (a) Prove the integral version of the Cauchy-Schwarz inequality: Z b Z b 1/2 Z b 1/2 ≤ f g f2 g2 . a a Hint: Consider separately the cases 0 = Rb and 0 < a (f − tg)2 for all t ∈ R. 3 a Rb a (f −tg)2 for some t ∈ R, Rb Proof. Consider the integral a (f − tg)2 , where t is any real number. This is integrating a non-negative function, so the integral itself must be non-negative. Furthermore, we can expand it: Z b 2 a (f − tg) = Z b 2 f − 2t a Z b 2 fg + t a Z b g2. a This, is a quadratic polynomial in t which is non-negative. Therefore it has either no real roots, or exactly one real root. Ruling out the possiblity of two distinct real roots means that its discriminant must be non-positive. Computing the discriminant for this polynomial, we get: 4 Z b fg a 2 −4 Z b f 2 a Z b g 2 a ≤ 0. . Therefore, Z a b fg 2 ≤ Z b f a 2 Z b g a 2 Taking the square root yields Z b Z b 1/2 Z b 1/2 . g2 f2 f g ≤ a a a (b) If equality holds, must f = tg for some t ∈ R? What if f and g are continuous? Solution. If f and g are continuous functions then f must be a multiple of g. But, if they are merely integrable, this is not so. The best we can say in that f = tg “almost everywhere.” (c) Show that the Cauchy-Schwarz inequality is a special case of (a). Proof. Let x, y ∈ Rn . Define f, g : [0, n] −→ R via step functions where the height of the ith step (where each step is width 1) of f 4 is the value of xi , and the height of the ith step of g is y i. Then Z b = hx, yi , f g a Z b 1/2 f2 = kxk, a Z b g 2 a 1/2 = kyk. Therefore, the inequality in part (a) yields the Cauchy-Schwarz inequality. 7. A linear transformation T : Rn −→ Rn is norm preserving if kT (x)k = kxk, for all x ∈ Rn , and inner product preserving if hT x, T yi = hx, yi, for all x, y ∈ Rn . (a) Prove that T is norm preserving if and only if it is inner product preserving. Proof. Suppose T is norm preserving. Then kT xk = kxk. Now use the Polarization Identity on hT x, T yi: 4 hT x, T yi = = = = kT x + T yk2 − kT x − T yk2 kT (x + y)k2 − kT (x − y)k2 kx + yk2 − kx − yk2 4 hx, yi . The other direction is even easier: if T is inner product preserving, then kT xk2 = hT x, T xi = hx, xi = kxk2 . (b) Prove that such a linear transformation is 1-1, and T −1 is norm preserving (and inner product preserving). Proof. Suppose T is norm preserving, and suppose T x = T y. Then 0 = kT x − T yk = kT (x − y)k = kx − yk, so x = y, and T 5 is 1-1. Thus we can talk about its inverse. Use the fact that T is norm preserving: kxk = T T −1 x = T −1 x , to see that T −1 is norm preserving. 8. If T : Rm −→ Rn is a linear transformation, show that there is a number M such that kT (h)k ≤ Mkhk for all h ∈ Rm . Hint: Estimate kT (h)k in terms of khk and the entries in the matrix for T . Proof. Let A = (aij ) be the matrix for T with respect to the standard bases for Rm and Rn . Also write h = (h1 , . . . , hm ) with respect to the standard basis. Let y = T (h) in Rn , and let ri denote the ith row of A. Then y i = hri , hi. By Cauchy-Schwarz, |y i| ≤ kri k · khk. Now look at kT (h)k: kT (h)k = ≤ = (y 1 )2 + · · · + (y n )2 1/2 (kr1 k · khk)2 + · · · + (krn k · khk)2 1/2 kr1 k2 + · · · + krn k2 khk 1/2 This completes the proof with M = kr1 k2 + · · · + krn k 2 1/2 = X i,j (aij )2 !1/2 . m 9. If x, y ∈ Rn , and pz, w ∈ R , show that h(x, z), (y, w)i = hx, yi + hz, wi, and k(x, z)k = kxk2 + kzk2 . Note that (x, z) and (y, w) denote points in Rn+m . 6 Proof. h(x, z), (y, w)i = (x1 , . . . , xn , z 1 , . . . , z m ), (y 1, . . . , y n , w 1, . . . , w m ) = x1 y 1 + · · · xn y n + z 1 w 1 + · · · + z m w m = x1 y 1 + · · · xn y n + z 1 w 1 + · · · + z m w m = hx, yi + hz, wi . p h(x, z), (x, z)i p = hx, xi + hz, zi p kxk2 + kzk2 . = k(x, z)k = 10. If x, y ∈ Rn , then x and y are called perpendicular (or orthogonal), and we write x ⊥ y, if hx, yi = 0. If x ⊥ y, prove that kx + yk2 = kxk2 + kyk2. Proof. kx + yk2 = = = = = = hx + y, x + yi hx, x + yi + hy, x + yi hx, xi + hx, yi + hy, xi + hy, yi hx, xi + 2 hx, yi + hy, yi hx, xi + 2(0) + hy, yi kxk2 + kyk2 . 7