Fall 2010 - Problem 3

density argument, Riemann-Lebesgue lemma, weak convergence

Consider the following sequence of functions:

fn ⁣:[0,1]Rbyfn(x)=exp(sin(2πnx)).\func{f_n}{\br{0,1}}{\R} \quad\text{by}\quad f_n\p{x} = \exp\p{\sin\p{2\pi nx}}.
  1. Prove that fnf_n converges weakly in L1([0,1])L^1\p{\br{0,1}}.
  2. Prove that fnf_n converges weak-* in L([0,1])L^\infty\p{\br{0,1}}, viewed as the dual of L1([0,1])L^1\p{\br{0,1}}.
Solution.
  1. Because [0,1]\br{0, 1} is σ\sigma-finite with respect to the Lebesgue measure, we know that (L1([0,1]))L([0,1])\p{L^1\p{\br{0,1}}}^* \simeq L^\infty\p{\br{0,1}}. Thus, to prove weak convergence, it suffices to show that there exists fL1([0,1])f \in L^1\p{\br{0,1}} such that

    limnfngdx=fgdx\lim_{n\to\infty} \int f_ng \,\diff{x} = \int fg \,\diff{x}

    for any gL([0,1])g \in L^\infty\p{\br{0,1}}. Observe that for such a gg,

    gL1gLm([0,1])=gL<,\norm{g}_{L^1} \leq \norm{g}_{L^\infty}m\p{\br{0,1}} = \norm{g}_{L^\infty} < \infty,

    so L([0,1])L1([0,1])L^\infty\p{\br{0,1}} \subseteq L^1\p{\br{0,1}}. We simplify the problem as follows:

    • Decompose g=g+gg = g^+ - g^-, where f+=max{0,g}f^+ = \max\set{0, g} and g=max{0,g}g^- = \max\set{0, -g}. By approximating these separately, it suffices to approximate non-negative gg.
    • Since there exists a sequence of simple functions which increase to such a gg, we may reduce to approximating simple functions, in light of the monotone convergence theorem.
    • Simple functions are finite linear combinations of integrable characteristic functions, so it further suffices to prove the result for characteristic functions on a measurable set of finite measure.
    • Finally, by regularity of the Lebesgue measure, we may reduce to an integrable characteristic function on an interval. Hence, we may assume without loss of generality that g=χ[a,b]g = \chi_{\br{a,b}}, with <a<b<-\infty < a < b < \infty.

    To emphasize the fact that fnf_n is 1n\frac{1}{n}-periodic, we let

    an=sup{kn|akn, k=0,1,,n}bn=inf{kn|knb, k=0,1,,n}\begin{aligned} a_n &= \sup\,\set{\frac{k}{n} \st a \leq \frac{k}{n},\ k = 0, 1, \ldots, n} \\ b_n &= \inf\,\set{\frac{k}{n} \st \frac{k}{n} \leq b,\ k = 0, 1, \ldots, n} \end{aligned}

    so that aanbnba \leq a_n \leq b_n \leq b for all n1n \geq 1, annaa_n \xrightarrow{n\to\infty} a, and bnnbb_n \xrightarrow{n\to\infty} b. Notice also that bnanb_n - a_n is an integer multiple of 1n\frac{1}{n}, so let kn=(bnan)nk_n = \p{b_n - a_n}n also. Thus,

    01fngdx=abexp(sin(2πnx))dx=aanexp(sin(2πnx))dx+anbnexp(sin(2πnx))dx+bnbexp(sin(2πnx))dx(ana)exp1+(bnb)exp1+anan+knnexp(sin(2πnx))dx=e(ana)+e(bnb)+knn01exp(sin(2πx))dx=e(ana)+e(bnb)+(bnan)01exp(sin(2πx))dx.\begin{aligned} \int_0^1 f_ng \,\diff{x} &= \int_a^b \exp\p{\sin\p{2\pi nx}} \,\diff{x} \\ &= \int_a^{a_n} \exp\p{\sin\p{2\pi nx}} \,\diff{x} + \int_{a_n}^{b_n} \exp\p{\sin\p{2\pi nx}} \,\diff{x} + \int_{b_n}^b \exp\p{\sin\p{2\pi nx}} \,\diff{x} \\ &\leq \p{a_n - a}\exp{1} + \p{b_n - b}\exp{1} + \int_{a_n}^{a_n+\frac{k_n}{n}} \exp\p{\sin\p{2\pi nx}} \,\diff{x} \\ &= e\p{a_n - a} + e\p{b_n - b} + \frac{k_n}{n} \int_0^1 \exp\p{\sin\p{2\pi x}} \,\diff{x} \\ &= e\p{a_n - a} + e\p{b_n - b} + \p{b_n - a_n} \int_0^1 \exp\p{\sin\p{2\pi x}} \,\diff{x}. \end{aligned}

    We get the inequality since fnf_n is uniformly bounded above by exp1=e\exp{1} = e, and we transformed the third integral via the substitution xn(xan)x \mapsto n\p{x - a_n} and using the fact that fnf_n is 1n\frac{1}{n}-periodic. Sending nn\to\infty, we get

    limn01fngdx=(ba)01exp(sin(2πx))dx=(01exp(sin(2πx))dx)(01gdx).\begin{aligned} \lim_{n\to\infty} \int_0^1 f_ng \,\diff{x} &= \p{b - a} \int_0^1 \exp\p{\sin\p{2\pi x}} \,\diff{x} \\ &= \p{\int_0^1 \exp\p{\sin\p{2\pi x}} \,\diff{x}}\p{\int_0^1 g \,\diff{x}}. \end{aligned}

    Hence, if we set f=01exp(sin(2πx))dxf = \int_0^1 \exp\p{\sin\p{2\pi x}} \,\diff{x}, then fL1([0,1])f \in L^1\p{\br{0,1}} and fnff_n \to f weakly.

  2. Notice that in (1), we proved a stronger statement: we showed that fngdxfgdx\int f_ng \,\diff{x} \to \int fg \,\diff{x} for any gL1([0,1])g \in L^1\p{\br{0,1}} instead of just gL([0,1])g \in L^\infty\p{\br{0,1}}. This is precisely what it means for fnff_n \to f weakly-* in L([0,1])L^\infty\p{\br{0,1}}, so we are done.