Fall 2012 - Problem 2

density argument

Suppose dμ\diff\mu is a Borel probability measure on the unit circle in the complex plane such that

limn{z=1}zndμ(z)=0.\lim_{n\to\infty} \int_{\set{\abs{z}=1}} z^n \,\diff\mu\p{z} = 0.

For fL1(dμ)f \in L^1\p{\diff\mu}, show that

limn{z=1}znf(z)dμ(z)=0.\lim_{n\to\infty} \int_{\set{\abs{z}=1}} z^nf\p{z} \,\diff\mu\p{z} = 0.
Solution.

Notice that if p=a0++amzmp = a_0 + \cdots + a_mz^m is a polynomial, then

limnznp(z)dμ(z)=limnk=0makzk+ndμ(z)=0.\lim_{n\to\infty} \int z^n p\p{z} \,\diff\mu\p{z} = \lim_{n\to\infty} \sum_{k=0}^m a_k \int z^{k+n} \,\diff\mu\p{z} = 0.

To generalize to fL1(S1)f \in L^1\p{S^1}, first observe that the unit circle S1S^1 is compact, so we may apply Stone-Weierstrass. Let A={k=0nakzknN,akC}A = \set{\sum_{k=0}^n a_kz^k \mid n \in \N, a_k \in \C}. It's clear that AA is closed under scaling by C\C, addition, and multiplication, so AA is an algebra. 1A1 \in A, so AA vanishes nowhere. Finally, zAz \in A, so AA separates points. Thus, AA is dense in C(S1)C\p{S^1} with respect to the uniform norm, hence with respect to the L1L^1 norm. Moreover, C(S1)C\p{S^1} is dense in L1(dμ)L^1\p{\diff\mu}, so given ε>0\epsilon > 0 and fL1(dμ)f \in L^1\p{\diff\mu}, there exists p=a0++amzmp = a_0 + \cdots + a_mz^m such that fpL1<ε\norm{f - p}_{L^1} < \epsilon. Thus, if nn is large enough, then znp<ε\abs{\int z^np} < \epsilon and so

znf(z)dμ(z)znp(z)dμ(z)+zn(f(z)p(z))dμ(z)ε+f(z)p(z)dμ(z)(we’re integrating over z=1)2ε.\begin{aligned} \abs{\int z^nf\p{z} \,\diff\mu\p{z}} &\leq \abs{\int z^np\p{z} \,\diff\mu\p{z}} + \int \abs{z^n\p{f\p{z} - p\p{z}}} \,\diff\mu\p{z} \\ &\leq \epsilon + \int \abs{f\p{z} - p\p{z}} \,\diff\mu\p{z} && (\text{we're integrating over } \abs{z} = 1) \\ &\leq 2\epsilon. \end{aligned}

Sending ε0\epsilon \to 0, we get the claim.