Fall 2009 - Problem 4

Lebesgue differentiation theorem, measure theory, regularity of Borel measures

Let μ\mu be a finite positive Borel measure on R\R, singular with respect to Lebesgue measure. Then for Lebesgue almost every xRx \in \R,

limε0μ([xε,x+ε])2ε=0.\lim_{\epsilon\to0} \frac{\mu\p{\br{x - \epsilon, x + \epsilon}}}{2\epsilon} = 0.
Solution.

Let mm be the Lebesgue measure on R\R. Because mm and μ\mu and singular, there exists a decomposition R=AAc\R = A \cup A^\comp, where μ(A)=0\mu\p{A} = 0, m(Ac)=0m\p{A^\comp} = 0, and AA is Borel. Since m(Ac)=0m\p{A^\comp} = 0, it suffices to only consider xAx \in A. Let ε>0\epsilon > 0 and set

Ek={xA|lim supδ0μ([xδ,x+δ])m([xδ,x+δ])>1k}.E_k = \set{x \in A \st \limsup_{\delta\to0}\,\frac{\mu\p{\br{x - \delta, x + \delta}}}{m\p{\br{x - \delta, x + \delta}}} > \frac{1}{k}}.

Since μ\mu is a finite Borel measure on R\R, it is outer regular. Thus, there exists an open set UAU \supseteq A such that μ(UA)<ε\mu\p{U \setminus A} < \epsilon. By measurability of AA,

μ(UA)=μ(U)μ(A)=μ(U)    μ(U)<ε.\mu\p{U \setminus A} = \mu\p{U} - \mu\p{A} = \mu\p{U} \implies \mu\p{U} < \epsilon.

Then for any xEkx \in E_k, there exists δx>0\delta_x > 0 such that I(x,δx)=[xδx,x+δx]UI\p{x, \delta_x} = \br{x - \delta_x, x + \delta_x} \subseteq U and

μ(I(x,δx))m(I(x,δx))=μ([xδx,x+δx])m([xδx,x+δx])>1k.\frac{\mu\p{I\p{x, \delta_x}}}{m\p{I\p{x, \delta_x}}} = \frac{\mu\p{\br{x - \delta_x, x + \delta_x}}}{m\p{\br{x - \delta_x, x + \delta_x}}} > \frac{1}{k}.

We also have

EkxEkI(x,δx5).E_k \subseteq \bigcup_{x \in E_k} I\p{x, \frac{\delta_x}{5}}.

By the Vitali covering lemma, there exist a countable collection {xn}nEk\set{x_n}_n \subseteq E_k which are pairwise disjoint and satisfy

EkxEkI(x,δx5)n=1I(xi,δi),E_k \subseteq \bigcup_{x \in E_k} I\p{x, \frac{\delta_x}{5}} \subseteq \bigcup_{n=1}^\infty I\p{x_i, \delta_i},

where δi=δxi\delta_i = \delta_{x_i}. Hence, by monotonicity,

m(Ek)m(i=1I(xi,δi))i=1m(I(xi,δi))i=1kμ(I(xi,δi))kμ(i=1I(xi,δi))kμ(U)<kε.\begin{aligned} m\p{E_k} \leq m\p{\bigcup_{i=1}^\infty I\p{x_i, \delta_i}} &\leq \sum_{i=1}^\infty m\p{I\p{x_i, \delta_i}} \\ &\leq \sum_{i=1}^\infty k\mu\p{I\p{x_i, \delta_i}} \\ &\leq k\mu\p{\bigcup_{i=1}^\infty I\p{x_i, \delta_i}} \\ &\leq k\mu\p{U} \\ &< k\epsilon. \end{aligned}

Since kk and ε\epsilon are independent, we may send ε0\epsilon \to 0 to see that m(Ek)=0m\p{E_k} = 0. Thus,

m(k=1Ek)k=1m(Ek)=0,m\p{\bigcup_{k=1}^\infty E_k} \leq \sum_{k=1}^\infty m\p{E_k} = 0,

and

x(k=1Ek)c    lim supδ0μ(xδ,x+δ)2ε1k k1,x \in \p{\bigcup_{k=1}^\infty E_k}^\comp \iff \limsup_{\delta\to0}\,\frac{\mu\p{x - \delta, x + \delta}}{2\epsilon} \leq \frac{1}{k}\ \forall k \geq 1,

i.e., the Lebesgue differentiate theorem holds Lebesgue almost everywhere.