Methods of Mathematical Physics X1 Homework 3 Solutions

Transcription

1 Methods of Mathematical Physics X Homework 3 Solutions. (Problem 2.. from Keener.) Verify that l 2 is an inner product space. Specifically, show that if x, y l 2, then x, y x k y k is defined and satisfies the properties of an inner product. (Here we re assuming that our sequences are real, so no need for the complex conjugate.) Hint: Think about how we proved Bessel s Inequality in class. Solution. The hardest part of this will be to show that if x, y l 2, then x, y is finite and x+y l 2 ; verifying everything else will be straightforward. We have x k y k lim n x k y k. By the Cauchy-Schwarz inequality for the standard dot product on R n, we know that ( n 2 x k y k ) x 2 k yk 2 x 2 k yk 2. The right-hand side is finite, and moreover it is independent of n, and thus x ky k is a convergent sequence. Since we have taken absolute values, this means that the series x ky k is an absolutely convergent sequence, and thus converges as well. Moreover, note now that if x, y l 2, then (x k + y k ) 2 x 2 k + 2x ky k + yk 2 <, so x + y l 2. Of course αx l 2 for all α R as well, and this makes l 2 a vector space. Proving that the remaining axioms of an inner product are satisfied is straightforward at this point. 2. (Problem 2..3 from Keener.) Show that the sequence x n n k! is a Cauchy sequence. Since the reals are complete, this means it converges. To which number does this sequence converge? Solution. We need to check that for any ǫ >, there is an N such that n, m > N means that But we have x n x m x n x m < ǫ. km+ k!. k 2 km+ But we can replace a sum with an integral with only adding perhaps a constant (think of the Riemann sums, for example), and we have n k 2 x 2 dx m n n m nm < m. km+ m So if we choose N /ǫ, then if m, n > N then x n x m < ǫ. Therefore this is a Cauchy sequence and thus converges. (Of course, replacing the factorial with a square threw away a lot this sum converges much faster than k 2.)

2 Finally, we know from calculus that so e x k xk k!, k! k! e. 3. Show that the sequence x n n k is not a Cauchy sequence. Solution. Let us try to bound x n x m, where we get x n x m km+ k k n m+ ( ) n dx log. x m + Choose n α(m + ), then this difference is at least log α. This means that no matter how large m is, we can make this difference large by choose n large enough. Therefore the sequence is not Cauchy. 4. Consider the Hilbert space L 2. Prove that the list of vectors {cos(nx)} n {sin(nx)} n is an infinite orthogonal list in L 2 with respect to the inner product f, g f(x)g(x) dx. Hint: You will need some trigonometric identities to solve this problem, e.g. you will need to compute integrals like cos(mx) cos(nx) dx. Recall that we can use Euler s formula to get, for example, cos((m + n)x) Re(e i(m+n)x ) Re(e imx e inx ) Re((cos(mx) + i sin(mx))(cos(nx) + i sin(nx)) cos(mx) cos(nx) sin(mx) sin(nx). If you recombine these formulas in a clever way, you can do all of the integrals. Solution. We need to show that cos(mx)cos(nx)dx δ mn C m, sin(mx)sin(nx)dx δ mn S m, cos(mx)sin(nx)dx, where C m and S m are some constants, and then we are done. Use the formula above for cos((m + n)x), and note then that cos((m n)x) cos(mx) cos(nx) + sin(mx) sin(nx) (cosine is even and sine is odd!) and then we have cos((m + n)x) + cos((m n)x) 2 cos(mx)cos(nx). 2

3 Therefore cos(nx)cos(mx)dx 2 cos((m + n)x) + cos((m n)x)dx 2 (2πδ m+n, + 2πδ m n, ). Since m, n, we can only have m + n if m n. Then we have 2π, m n, cos(nx) cos(mx) dx π, m n,, m n. Similarly, we have sin(mx)sin(nx) (cos((m n)x) cos((m + n)x)), 2 and thus sin(nx)sin(mx)dx 2 Since m, n >, we cannot have m + n. Then cos((m n)x) cos((m + n)x)dx 2 (2πδ m n, 2πδ m+n, ). sin(nx) sin(mx) dx Finally, we need the other Euler s formula, namely sin((m + n)x) Im(e i(m+n)x ) Im(e imx e inx ) This gives so { π, m n,, m n. Im((cos(mx) + i sin(mx))(cos(nx) + i sin(nx)) sin(mx) cos(nx) + cos(mx) sin(nx). sin((m + n)x) + sin((m n)x) 2 sin(mx)cos(nx), sin(mx)cos(nx)dx 2 (recall that sin(x) for all x and thus its integral is as well). sin((m + n)x) + sin((m n)x)dx 5. (Problem 2.2. from Keener.) Find the best quadratic polynomial fit to the function f(x) x, where we choose as inner product f, g for each of the weights ω(x), x 2, ( x 2 ) /2. f(x)g(x)ω(x) dx, Hint: You might find it convenient to compute some orthogonal polynomials for each weight and then compute the answer in terms of these polynomials and we already did most of the work here on the last homework! Solution. See attached Mathematica notebook/pdf. 6. (Problem from Keener.) Suppose that {φ n (x)} n is a set of orthonormal polynomials, where we choose the inner product f, g b a f(x)g(x)ω(x) dx, (ω(x) > ) and assume that φ n (x) is a polynomial of degree n with leading coefficient (specifically, we mean that φ n (x) x n + (terms of power n or less). Then show: 3

4 (a) If f is a polynomial of degree less than n, then φ n, f. (b) Show that every polynomial of degree n can be written in the form for some numbers α i. α i φ i (c) the polynomials satisfy a recurrence relation of the form i φ n+ (x) (A n x + B n )φ n (x) C n φ n (x), for every n, where A n + /. Compute B n, C n in terms of A n, A n, φ n. Hint: What do we know about φ n+ (x) A n xφ n (x)? Use part (b), take the inner product with φ j, what do you get? Also, notice that for this inner product, xf, g f, xg. Solution. It s slightly more efficient to do (b) first and then (a). To prove (b), we will use induction. If n, then φ k, and if f is degree, then f β for some β R, so we have f β k φ. (I ll also work out the n case directly to give more of the idea.) Let n, so we have φ k, φ k x + C. Now, if f is degree one then Then if we have then f(x) β x + β. f(x) α φ + α φ, f(x) α k + α (k x + C) α k x + (α C + α k ), so we choose α, α to solve the two-by-two system α k β, α C + α k β. Now, we use induction. Let us assume that for any polynomial f of degree n, we can write f as a linear combination of {φ,..., φ n }. Now assume f is degree n +, where Then f(x) β n+ x n g(x) : f(x) β n+ + φ n+ is a polynomial of degree n (since we picked constants to kill off the first term). By the induction hypothesis, we can then write g(x) α i φ i, so then i f(x) β n+ + φ n+ + α i φ i, which is a linear combination of {φ,..., φ n+ }, and we are done. i 4

5 Now, to prove part (a), assume f is a polynomial of degree k < n. Then so f(x) k α i φ i, i k f, φ n α i φ i, φ n i k α i φ i, φ n. Finally, we do part (c). Basically, we use the ideas above, plus a clever trick or two. First of all, we know by assumption that From this, we know i φ n+ + x n+ + O(x n ), φ n x n + O(x n ). f(x) : φ n+ + xφ n is a polynomial of degree n. We then know, from part (b) above, that f(x) α i φ i for some constants α i. Note further that since the φ i are orthonormal, we know that α j f, φ j φ n+ k n+ xφ n, φ j. Since j n, φ n+, φ j, so that i α j + xφ n, φ j. Now, of course, we have no idea what xφ n, φ j is, in general. However, we have one trick up our sleeve: notice that for any functions f, g, we have xf, g f, xg, because of the way we ve defined our inner product. NB. Of course, we cannot do this for every inner product, but this one has a special form. So we then have xφ n, φ j φ n, xφ j, and if j < n, then the degree of xφ j is less than n, and by part (a) this is zero. Therefore we know f(x) α n φ n + α n φ n, so define B n α n, C n α n, and we have established the formula for some B n, C n. It remains to compute B n, C n. We will use the two equations From the first equation in (), we have φ n+, φ n, φ n+, φ n. () φ n+, φ n A n xφ n + B n φ n C n φ n, φ n A n xφ n, φ n + B n φ n, φ n C n φ n, φ n. (2) 5

6 Using the fact that φ n, φ n, φ n, φ n, equation (2) becomes B n A n xφ n, φ n. From the second equation in (), we have φ n+, φ n A n xφ n + B n φ n C n φ n, φ n A n xφ n, φ n + B n φ n, φ n C n φ n, φ n. (3) Similarly, this becomes C n A n xφ n, φ n. We would now like to write this in terms of only φ n. But notice that xφ n, φ n φ n, xφ n φ n, x( x n + O(x n 2 )) To compute the last term there, notice that φ n, x n + O(x n ) φ n, x n. φ n, φ n φ n, x n + O(x n ) φ n, x n, so that and thus φ n, x n, xφ n, φ n A n. 7. (Problem 2.2. from Keener.) Problem fixed! Consider the inner product (ω(x) > ). Show that f, g f(x)g(x)ω(x) dx d n P n (x) : ( ω(x)( x 2 ω(x) dx n ) n) is orthogonal to every polynomial of degree less than n. Hint: We proved this in class when ω. Adapt that argument to this case. Solution. Let f be a polynomial of degree k < n. Then we have f, P n f(x) dn dx n ( ω(x)( x 2 ) n) dx. Recall the two lemmas we proved in class. First of all, since ( x 2 ) is zero at ±, then if we define g(x) ( x 2 ) n, g, and its first n derivatives are as well, i.e. g(±) g (±) g (±) g (n ) (±). Now, the question is, does the function ω(x)g(x) also have the same property, i.e. are its first n derivatives zero as well? The answer is yes: recall the product rule from calculus, d p p ( p dx p (ω(x)g(x)) k k 6 ) ω (k) (x)g (p k) (x),

7 and if we replace p with any integer less than n, then all the derivatives on g which appear in the sum are less than n, and these are all zero at ±, so therefore we know ωg and its first n derivatives are zero at ±. Now we use the other lemma we proved in class, namely that if h(±) h (±) h (n ) (±), then f(x) dn d n h(x)dx ( )n dxn dx n f(x)h(x)dx. But notice that f is a polynomial of degree k < n, and so if we take n derivatives on f it is zero. 8. (Problem from Keener.) Suppose that f(t) and g(t) are 2π-periodic functions with Fourier series representations f(t) f k e ikt, g(t) g k e ikt. Now define Compute the Fourier series for h. Solution. We compute h(t) Now, if α is a non-zero integer, then h(t) k,l but if α then the integral is 2π, so Thus we have h(t) k,l f(t x)g(x)dx. f k e ik(t x) l f k g l e ik(t x) e ilx dx g l e ilx dx f k g l e ikt e i(l k)x dx. e iαx dx eiαx iα k,l k,l x2π x e iαx dx 2πδ α,., f k g l e ikt e i(l k)x dx f k g l e ikt 2πδ l,k 2πf k g k e ikt. So, the kth Fourier coefficient of h is 2πf k g k, i.e. forming the convolution of f and g is equivalent (up to a constant) to multiplying their Fourier series term-by-term. 7