Convergence of Fourier Series — Question 10

PDF ↗

Question 10

Let f∈L2(−π,π)f\in L^2(-\pi,\pi) be real and periodic with mean mm and Fourier coefficients an,bna_n,b_n. For t>0t>0, define the damped expansion Ut(x)=m+∑n=1∞e−n2t(ancos⁡nx+bnsin⁡nx).U_t(x)=m+\sum_{n=1}^{\infty}e^{-n^2t}(a_n\cos nx+b_n\sin nx). Use Parseval and L2L^2 convergence of Fourier partial sums. This problem concerns the series itself; no partial differential equation is assumed.

Tasks

  1. Prove that for each fixed t>0t>0 the series, and every series obtained by differentiating in xx finitely many times, converge uniformly.

  2. Prove Ut→fU_t\to f in L2L^2 as t↓0t\downarrow 0 using a finite-head and small-tail argument, with an exact coefficient formula for the error.

  3. If ∑n2(an2+bn2)≤M\sum n^2(a_n^2+b_n^2)\leq M, prove ∥Ut−f∥2≤πMt\|U_t-f\|_2\leq\sqrt{\pi Mt}.

  4. Show that the mean is preserved and that ∥Ut−m∥2≤∥f−m∥2\|U_t-m\|_2\leq\|f-m\|_2. Explain why the general L2L^2 assumption alone cannot guarantee uniform convergence to the assigned function ff as t↓0t\downarrow 0.

Original worksheet page 1: question and worked solution for 8-7-010
Show solutionHide solution

Question 10 – Solution

Strategy. Damping controls all high frequencies for positive tt; removing it requires a separate convergence argument.

Step 1: Prove smoothness for positive damping. Parseval gives E=∑(an2+bn2)<∞E=\sum(a_n^2+b_n^2)<\infty. After kk derivatives, the absolute value of the nnth harmonic is at most nke−n2tan2+bn2n^ke^{-n^2t}\sqrt{a_n^2+b_n^2}. Cauchy–Schwarz bounds the sum by E(∑n≥1n2ke−2n2t)1/2<∞.\sqrt E\left(\sum_{n\geq 1}n^{2k}e^{-2n^2t}\right)^{1/2}<\infty. The exponential dominates every power, so the Weierstrass test applies for each fixed t>0t>0 and each integer k≥0k\geq 0. Repeated use of the uniform derivative theorem justifies differentiation; UtU_t is smooth and periodic.

Step 2: Remove damping in the integral norm. The Fourier coefficients of the difference give ∥Ut−f∥22=π∑n≥1(1−e−n2t)2(an2+bn2).\boxed{\|U_t-f\|_2^2=\pi\sum_{n\geq 1} (1-e^{-n^2t})^2(a_n^2+b_n^2).} Given ε>0\varepsilon>0, choose KK so that π∑n>K(an2+bn2)<ε/2\pi\sum_{n>K}(a_n^2+b_n^2) <\varepsilon/2. Since 0≤1−e−n2t≤10\leq 1-e^{-n^2t}\leq 1, this bounds the error tail for every tt. The remaining finite sum tends to zero, hence is below ε/2\varepsilon/2 for sufficiently small tt. This proves the limit.

Step 3: Obtain a quantitative rate under extra information. For z≥0z\geq 0, 1−e−z≤min⁡(z,1)1-e^{-z}\leq\min(z,1), so (1−e−z)2≤z(1-e^{-z})^2\leq z. Applying this with z=n2tz=n^2t gives ∥Ut−f∥22≤πt∑n2(an2+bn2)≤πMt.\boxed{\|U_t-f\|_2^2\leq\pi t\sum n^2(a_n^2+b_n^2)\leq\pi Mt.} Taking square roots yields the requested rate; this uses a stronger hypothesis than mere square summability.

Step 4: Separate smoothing from uniform recovery. Uniform integration of the damped series preserves the mean mm. Parseval gives ∥Ut−m∥22=π∑e−2n2t(an2+bn2)≤∥f−m∥22\|U_t-m\|_2^2=\pi\sum e^{-2n^2t}(a_n^2+b_n^2)\leq\|f-m\|_2^2. Every UtU_t is continuous. Choose an L2L^2 periodic step function with a jump: it cannot be a uniform limit of these continuous functions. Even isolated redefinitions of ff are invisible to the coefficients. Thus smooth approximants and mean-square recovery do not alone guarantee uniform recovery of an arbitrarily assigned representative.

Original worksheet page 2: question and worked solution for 8-7-010

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