Fall 2020 - Problem 1

measure theory
  1. Suppose f ⁣:[0,1]×[0,)[0,1]\func{f}{\br{0,1} \times \pco{0,\infty}}{\br{0,1}} is continuous. Prove that F ⁣:[0,1][0,1]\func{F}{\br{0,1}}{\br{0,1}} defined by

    F(x)=lim supyf(x,y)F\p{x} = \limsup_{y\to\infty} f\p{x, y}

    is Borel measurable.

  2. Show that for any Borel set E[0,1]E \subseteq \br{0,1} there is a choice of continuous function f ⁣:[0,1]×R[0,1]\func{f}{\br{0,1} \times \R}{\br{0,1}} so that FF agrees almost everywhere with the indicator function of EE.

Solution.
  1. Observe that by continuity,

    F(x)=lim supnf(x,n)=infNsupq>N;qQf(x,q),F\p{x} = \limsup_{n\to\infty} f\p{x, n} = \inf_N \sup_{q > N; q \in \Q} f\p{x, q},

    so because supremums and infimums of a countable family of measurable functions is measurable, it follows that FF is Borel measurable.

  2. Let F\mathcal{F} be the set of Borel sets which satisfy this condition. Since 0=χ0 = \chi_\emptyset and 1=χ[0,1]1 = \chi_{\br{0,1}}, we see that ,[0,1]F\emptyset, \br{0, 1} \in \mathcal{F}. If EE is an open set, then χE\chi_E is lower semi-continuous, so there exists a sequence {fn}n\set{f_n}_n of continuous functions which converge pointwise to χE\chi_E, and if for ny<n+1n \leq y < n + 1 we set

    f(x,y)=(yn)fn+1(x)+((n+1)y)fn(x)f\p{x, y} = \p{y - n}f_{n+1}\p{x} + \p{\p{n+1} - y} f_n\p{x}

    then we see that ff is piecewise continuous and continuous at each endpoint, since each fnf_n was. Thus, F\mathcal{F} contains open sets. If EFE \in \mathcal{F}, then let f(x,y)f\p{x, y} be the associated continuous function. Observe that if g(x,y)=1f(x,y)g\p{x, y} = 1 - f\p{x, y}, then gg has range in [0,1]\br{0, 1}, is continuous, and satisfies

    lim supyg(x,y)=1lim supyf(x,y)=1χE=χEc,\limsup_{y\to\infty} g\p{x, y} = 1 - \limsup_{y\to\infty} f\p{x, y} = 1 - \chi_E = \chi_{E^\comp},

    so F\mathcal{F} is closed under complements. Finally, suppose {En}nF\set{E_n}_n \subseteq \mathcal{F} with corresponding continuous functions {fn}n\set{f_n}_n. Let E=k=1EkE = \bigcup_{k=1}^\infty E_k and {gn}n\set{g_n}_n be the sequence f1,f1,f2,f1,f2,f3,f_1, f_1, f_2, f_1, f_2, f_3, \ldots, i.e., each fkf_k appears infinitely often. For each k1k \geq 1, by regularity of the Lebesgue, there exists an open set UkEU_k \supseteq E such that m(UkE)<1km\p{U_k \setminus E} < \frac{1}{k}. Let φk\phi_k be a piecewise linear approximation to χUk\chi_{U_k} such that φk\phi_k and χUk\chi_{U_k} only differ on an interval of length 1k\frac{1}{k} such that φk=1\phi_k = 1 on UkU_k. As before, set for ny<n+1n \leq y < n + 1

    f(x,y)=(yn)gn+1(x,y)φn+1(x)+((n+1)y)gn(x,y)φn(x)f\p{x, y} = \p{y - n}g_{n+1}\p{x, y}\phi_{n+1}\p{x} + \p{\p{n+1} - y} g_n\p{x, y}\phi_n\p{x}

    as before. Then by construction, if xEkx \in E_k, then because fkf_k appears infinitely often, it follows that lim supyg(x,y)=1\limsup_{y\to\infty} g\p{x, y} = 1. If xEkx \notin E_k for every kk, then there exists nn large enough so that φn(x)=0\phi_n\p{x} = 0, so by construction, f(x,y)=0f\p{x, y} = 0 for all yny \geq n, and so lim supyg(x,y)=χE\limsup_{y\to\infty} g\p{x, y} = \chi_E almost everywhere, as required.