Fourier Sine Series — Question 9

PDF ↗

Question 9

For 0<ε<π/20<\varepsilon<\pi/2, define a unit-area narrow pulse centered at π/2\pi/2: fε(x)={(2ε)−1,|x−π/2|<ε,0,otherwise,0≤x≤π.f_\varepsilon(x)=\begin{cases}(2\varepsilon)^{-1},&|x-\pi/2|<\varepsilon,\\ 0,&\text{otherwise},\end{cases}\qquad 0\leq x\leq\pi. Use bn(ε)=(2/π)∫0πfεsin⁡nxdxb_n(\varepsilon)=(2/\pi)\int_0^\pi f_\varepsilon\sin nx\,dx. You may use Parseval and Bessel’s inequality for square-integrable functions.

Tasks

  1. Verify the area and derive all sine coefficients. Interpret the value of sin⁡z/z\sin z/z at z=0z=0 by its limit.

  2. Find the limit of each fixed coefficient as ε↓0\varepsilon\downarrow 0. Prove that the limiting coefficient sequence cannot be the sine coefficients of any L2(0,π)L^2(0,\pi) function.

  3. Compute ∥fε∥22\|f_\varepsilon\|_2^2 and use Parseval to find ∑n≥1bn(ε)2\sum_{n\geq 1}b_n(\varepsilon)^2. Explain why a point-mass limit is not an ordinary square-integrable function.

  4. Use |sin⁡z/z−1|≤z2/6|\sin z/z-1|\leq z^2/6 to bound the largest coefficient error among 1≤n≤N1\leq n\leq N. Derive this inequality from sin⁡z/z=∫01cos⁡(zt)dt\sin z/z=\int_0^1\cos(zt)dt and explain why the resulting estimate is not uniform over all mode numbers.

Original worksheet page 1: question and worked solution for 8-4-009
Show solutionHide solution

Question 9 – Solution

Strategy. Separate a limit at each fixed frequency from a limit of the complete coefficient sequence in squared norm.

Step 1: Integrate the narrow pulse. The height times the width is one. Writing a=π/2a=\pi/2, integration gives bn(ε)=cos⁡(n(a−ε))−cos⁡(n(a+ε))πεn=2πsin⁡(nπ/2)sin⁡(nε)nε.b_n(\varepsilon)=\frac{\cos(n(a-\varepsilon))-\cos(n(a+\varepsilon))} {\pi\varepsilon n} =\boxed{\frac 2\pi\sin(n\pi/2)\frac{\sin(n\varepsilon)}{n\varepsilon}}. The factor sin⁡z/z\sin z/z has limiting value one at zero. Even modes vanish.

Step 2: Examine the fixed-mode limit. For each fixed nn, bn(ε)→βn=(2/π)sin⁡(nπ/2)b_n(\varepsilon)\to\beta_n=(2/\pi)\sin(n\pi/2). Every odd-indexed βn\beta_n has magnitude 2/π2/\pi, so ∑βn2\sum\beta_n^2 diverges. Bessel’s inequality would bound this sum for any L2L^2 function. Hence no square-integrable function has all these limiting coefficients.

Step 3: Track the diverging energy. Direct integration gives ∥fε∥22=1/(2ε)\boxed{\|f_\varepsilon\|_2^2=1/(2\varepsilon)}. Parseval therefore implies ∑n≥1bn(ε)2=1πε.\boxed{\sum_{n\geq 1}b_n(\varepsilon)^2=\frac 1{\pi\varepsilon}.} The energies grow without bound while area stays one. Indeed, integration against a continuous test function tends to its value at π/2\pi/2, by continuity and averaging over the shrinking interval. This is the point-mass limit, not an L2L^2 function or an ordinary pointwise sine representation of one.

Step 4: Quantify finite-band convergence. From |1−cos⁡s|≤s2/2|1-\cos s|\leq s^2/2 and the supplied integral representation, |sin⁡z/z−1|≤∫01z2t2/2dt=z2/6|\sin z/z-1|\leq\int_0^1z^2t^2/2\,dt=z^2/6. Thus max1≤n≤N|bn(ε)−βn|≤N2ε23π.\boxed{\max_{1\leq n\leq N}|b_n(\varepsilon)-\beta_n| \leq\frac{N^2\varepsilon^2}{3\pi}.} For each fixed NN this tends to zero. It grows with NN, so it does not justify uniform convergence across all frequencies or interchange with the infinite energy sum.

Original worksheet page 2: question and worked solution for 8-4-009

Original worksheet layout. Use Enlarge or open the PDF for a closer view.