Partial Solutions to Homework XII Y. Zhou Section 6.5 6. k Proof. It is sufficient to show that Bik (x) = Bi−1 (x − 1). The claim of the problem follows from the iterative application of this equality untill getting B0k : k k Bik (x) = Bi−1 (x − 1) = Bi−2 (x − 2) = · · · = B0k (x − i) = B0k (x − ti ). k (x − 1). We use mathmatical induction to prove Bik (x) = Bi−1 k k For k = 0, Bi (x) = Bi−1 (x − 1) follows directly from the definition of Bi0 (x). Assume that k−1 Bik−1 (x) = Bi−1 (x − 1), we have Bik (x) x − ti ti+k+1 − x k−1 B k−1 (x) + B (x) ti+k − ti i ti+k+1 − ti+1 i+1 x − ti ti+k+1 − x k−1 k−1 = Bi−1 (x − 1) + B (x − 1) ti+k − ti ti+k+1 − ti+1 i ti+k − (x − 1) k−1 (x − 1) − ti−1 k−1 B (x − 1) + Bi (x − 1) = ti−1+k − ti−1 i−1 ti+k − ti = Bi−1 (x − 1), = where the first and the last equalities follow from the iterative relation of B-splines; the second eqnality follows from the assumption about Bik−1 (x), and the third equality holds true since ti = i and x − ti = (x − 1) − (ti − 1) = (x1 ) − ti−1 . 8. Proof. Since Bik (x) is compactly supported in (ti , ti+k+1 ), we know ∞ X 0= ci Bik (x) = i=−∞ j X ci Bik (x), ∀x ∈ (tj , tj+1 ). i=j−k k This equation indicates that cj−k , · · · , cj are all zeros because Bj−k (x), · · · , Bj are linearly independent on (ti , ti+k+1 ) (Lemma 8). Letting j vary over all integers we will get that ci = 0 for all i. 16. Proof. ∞ X k ci Bi (x) dx −∞ Z ∞ Z ∞ ≤ ∞ X ci Bik (x) dx −∞ −∞ ∞ ∞ X −∞ Z = |ci |Bik (x)dx −∞ −∞ = ∞ X Z = |ci | −∞ ≤ ∞ X Bik (x)dx −∞ −∞ ∞ X ∞ |ci | ti+k+1 − ti k+1 |ci |m −∞ ∞ X = c |ci |, −∞ where we use the fact that Bik (x) ≥ 0 in deriving the first inequality, and use the conclusion of problem 7 in deriving the third equality. In deriving the second inequality we use the relation that ti+k+1 − ti = (ti+k+1 − ti+k ) + (ti+k − ti+k−1 ) + · · · + (ti+1 − ti ), with m terms in total and each term less than m according to supi |ti+1 − ti | ≤ m. 1 2 6.6 2. Proof. For any x ∈ R, we can always find j such that x ∈ [tj , tj+1 ), and Sf = ∞ X f (xi )Bi2 (x) i=−∞ = j X f (xi )Bi2 (x) i=j−2 By using the definition Sf = of xi and Bi2 (x), 2 f (xj−2 )Bj−2 (x) we have 2 + f (xj−1 )Bj−1 (x) + f (xj )Bj2 (x) x − tj−2 1 tj+1 − x 1 Bj−2 (x) + Bj−1 (x) + tj − tj−2 tj+1 − tj−1 x − tj−1 tj+2 − x 1 1 Bj−1 (x) + Bj (x) + f (xj−1 ) tj+1 − tj−1 tj+2 − tj x − tj tj+3 − x 1 Bj1 (x) + Bj+1 (x) f (xj ) tj+2 − tj tj+3 − tj+1 x − tj−1 tj+1 − x 1 = Bj−1 (x) f (xj−2 ) + f (xj−1 ) + tj+1 − tj−1 tj+1 − tj−1 tj+2 − x x − tj Bj1 (x) f (xj−1 ) + f (xj ) tj+2 − tj tj+2 − tj tj + tj+1 x − tj−1 tj−1 + tj tj+1 − x 1 + = Bj−1 (x) + 2 tj+1 − tj−1 2 tj+1 − tj−1 tj+1 + tj+2 x − tj tj + tj+1 tj+2 − x + Bj1 (x) 2 tj+2 − tj 2 tj+2 − tj tj+1 − x tj−1 + tj tj+1 − x tj + tj+1 x − tj−1 = + + tj+1 − tj 2 tj+1 − tj−1 2 tj+1 − tj−1 tj + tj+1 tj+2 − x x − tj tj+1 + tj+2 x − tj + . tj+1 − tj 2 tj+2 − tj 2 tj+2 − tj = f (xj−2 ) This is a quadratic equation. The coefficient of x2 is tj−1 + tj tj + tj+1 1 tj + tj+1 tj+1 + tj+2 1 − + + + = 0, − tj+1 − tj 2(tj+1 − tj−1 ) 2(tj+1 − tj−1 ) tj+1 − tj 2(tj+2 − tj ) 2(tj+2 − tj ) the constant term is tj+1 [(tj−1 + tj )tj+1 − (tj + tj+1 )tj−1 ] tj [(tj + tj+1 )tj+2 − (tj+1 + tj+2 )tj ] − 2(tj+1 − tj )(tj+1 − tj−1 ) 2(tj+1 − tj )(tj+2 − tj ) tj+1 tj tj+1 tj − = 2(tj+1 − tj ) 2(tj+1 − tj ) =0, and it can be checked that coefficient of x is 1. 19. Proof. Lemma 2 in Section 6.5 says that Bik (x) > 0 for any x ∈ (ti , ti+k+1 ). Thus for any x ∈ (t0 , t1 ), Bik (x) > 0 for −k ≤ i ≤ 0, i.e., k +1 nonzeros spline functions in total. Similarly, for any x ∈ (t1 , t2 ), Bik (x) > 0 for −k + 1 ≤ i ≤ 1, adding one more nonzero function B1k (x). Repeating the argument for each of the remaining subintervals (t2 , t3 ), · · · , (tn−1 , tn ), we can have total n + k nonzero splines k B−k (x), · · · , Bnk (x). 6.8 3. Q Proof. Let g(x) be the best approximation of f (x) in n , we need to show that g(x) = g(−x). According to the definition of best approximation, we know Z 1 (f (x) − g(x))xk dx = 0, ∀0 ≤ k ≤ n, −1 3 Q as {xk }, k = 0, · · · , n are the basis functions of n . Since f (x) is an even function, we have for k even Z 1 0 = (f (x) − g(x))xk dx −1 1 Z (f (−x) − g(x))(−x)k dx = −1 1 Z (f (−x) − g(x))(−x)k dx = − −1 1 Z (f (−x) − g(x))(−x)k d(−x) = −1 Z −1 = (f (x) − g(−x))xk dx 1 −1 Z (f (x) − g(−x))xk dx = − 1 Z 1 (f (x) − g(−x))xk dx, = −1 suggesting that Z 1 (g(x) − g(−x))xk dx = 0, ∀k even. −1 Following the similarly procedure we can show that Z 1 (g(x) − g(−x))xk dx = 0, ∀k odd. −1 Q Q As g(x) ∈ n , g(−x) ∈ n , g(x) − g(−x) ∈ n and {xk } are the basis functions of n , it follows that g(x) − g(−x) = 0 and therefore g(x) is an even function. Q Q 8. Proof. By definition, an = hxPn−1 , Pn−1 i/hPn−1 , Pn−1 i, hence we only need to show that Z a hxPn−1 , Pn−1 i = 2 xPn−1 wdx = 0. −a 2 Note that Pn−1 is nonnegative and xw is odd, so a sufficient condition for above equality is that 2 Pn−1 is even. In other words, it is sufficient to show that Pn is an even or an odd function. We will prove this by using mathematical induction. By definition P0 = 1 is an even function, and thus a1 = 0 and P1 = x is an odd function. It follow that a2 = 0 and P2 = xP1 + b2 P0 is an even function. But then a3 = 0 and P3 = xP2 + b3 P1 is an odd function. Let Pn−2 be an odd function, Pn−1 be an even function, we have an = 0, Pn = xPn−1 + bn Pn−2 being an odd function, an+1 = 0, and Pn+1 = xPn + bn+1 Pn−1 being an odd function. Hence we proved that Pn is an even function if n is even and is an odd function if n is odd, as required. 11. 4 Proof. By defintion, bn = hxPn−1 , Pn−2 i/hPn−2 , Pn−2 i = hxPn−1 , Pn−2 i/kPn−2 k2 , thus it suffices to show that hxPn−1 , Pn−2 i = kPn−1 k2 . Note that hxPn−1 , Pn−2 i = hxPn−2 , Pn−1 i and Pn−1 = (x − an−1 )Pn−2 + bn−1 Pn−3 , or xPn−2 = Pn−1 + an−1 Pn−2 − bn−1 Pn−3 we conclude that hxPn−1 , Pn−2 i = hxPn−2 , Pn−1 i = hPn−1 + an−1 Pn−2 − bn−1 Pn−3 , Pn−1 i = hPn−1 , Pn−1 i + an−1 hPn−2 , Pn−1 i + bn−1 hPn−3 , Pn−1 i = hPn−1 , Pn−1 i = kPn−1 k2 as hPn−2 , Pn−1 i = 0, hPn−3 , Pn−1 i = 0 due to the orthogonality. This finishes the proof. 16. Proof. We need to show that hf1 , f2 i = 0 for any f1 , f2 are such that Af1 = λ1 f1 , Af2 = λ2 f2 with λ1 6= λ2 . For such f1 , f2 and self-adjoint A we have λ1 hf1 , f2 i = hλ1 f1 , f2 i = hAf1 , f2 i = hf1 , Af2 i = hf1 , λ2 f2 i = λ2 hf1 , f2 i. Since λ1 6= λ2 , this equality holds ture if and only if hf1 , f2 i = 0.