Lebesgue integration of step function

933 Views Asked by At

Let $J$ be a finite subinterval of the real line and $f:J\rightarrow\mathbb{R}$ a simple function taking on values $c_1,...,c_n$. The function $f$ is called a step function if $f^{-1}(c_i)$ is a finite union of intervals for each $i$. Given a simple function $s:J\rightarrow\mathbb{R}$ and a positive number $\varepsilon$, show that there exists a step function $f$ such that $$\int_J|s-f|\,d\mu_L<\varepsilon$$ Hint: Show that, if $A$ is a measurable subset of $J$, there exists a finite union of intervals $B$ such that $d(A,B)=\mu(S(A,B))<\varepsilon$. Now prove the above for $s=1_A$. Proceed.

I'm not sure how to do this problem, even with the hint. I don't know how to show the statement in the hint or how to get from there to the question itself.

1

There are 1 best solutions below

3
On

Assuming that we are talking about the Lebesgue measure, let us use the letter $\lambda$, we remind its definition, for every $\lambda-$measurable set $A$: $$\lambda(A):=\inf\left\{\sum_{i=1}^\infty l(I_i):I_i\text{ is an open interval, }i=1,2,\dots,\ A\subseteq\bigcup_{i=1}^\infty I_i\right\}$$ where $l(I)$ is the length of an iterval $I$ - so, if the interval has boundaries $a,b\in\mathbb{R}$ with $a<b$ then $l(I)=b-a$.

Note now that, if $I_i$, $i=1,2,\dots$ are all open intervals, and $I=\bigcup_i I_i$ is their union, we can see that the following sequence of sets: $$\begin{align*} B_1&=I_1\\ B_2&=I_1\setminus B_1\\ &\vdots\\ B_n&=I_n\setminus\bigcup_{i=1}^{n-1}B_i\\ &\vdots \end{align*}$$ It is clear that $B_k\cap B_l=\varnothing$ for every $k,l\in\mathbb{N}$ and that: $$\bigcup_{i=1}^\infty B_i=\bigcup_{i=1}^\infty I_i$$ Moreover, for every $k=1,2,\dots$, we can see, from the definition of $B_k$, that $B_k$ is a finite union of (not necessarily open) intervals - it is an open interval "minus" a finite number of open intervals. Also, note that: $$\sum_{i=1}^\infty l(I_i)\geq\lambda\left(\bigcup_{i=1}^\infty I_i\right)=\lambda\left(\bigcup_{i=1}^\infty B_i\right)=\sum_{i=1}^\infty\lambda(B_i)=\sum_{i=1}^\infty l(B_i)\tag{$\star$}$$ Taking infima to $(\star)$, we can easily see that: $$\lambda(A)=\inf\left\{\sum_{i=1}^\infty l(B_i):B_i\text{ is a finite union of intrv., }B_i\cap B_j=\varnothing,\ \forall\ i,j\in\mathbb{N},\ A\subseteq\bigcup_{i=1}^\infty B_i\right\}$$ Can you now prove the hint, after this sub-hint? I'll come up with a complete solution, later on, if you cannot get it till the end! :)

Edit: Note at first that, if $\lambda(A)=0$, then is is easy to find such a set - e.g. a subinterval of $J$ with length lesser than $\epsilon$. So let us assume that $\lambda(A)>0$.

Let $\epsilon>0$. By definition of the Lebesgue measure - the "alternative" definition - we have that there exists a sequence $B_n$, where $B_n$ is a finite union of intervals, $B_k\cap B_l=\varnothing$, $\forall\ k,l=1,2,\dots$ such that: $$\sum_{n=1}^\infty l(B_n)-\lambda(A)<\frac{\epsilon}{2}$$ Also, since $B_n$ are not intersecting, we have that: $$\lambda\left(\bigcup_{n=1}^\infty B_n\setminus A\right)<\frac{\epsilon}{2}$$ Consider now the set sequence $$S_n=\bigcup_{i=1}^nB_i\setminus A$$ It is clear that $S_n$ is increasing and, hence: $$\lim_{n\to\infty}\lambda(S_n)=\lambda\left(\bigcup_{i=1}^\infty B_i\setminus A\right)$$

So, for every $n\in\mathbb{N}$ $$\lambda(S_n)<\frac{\epsilon}{2}$$

Also, note that, if $$a_n=\lambda\left(A\setminus\bigcup_{i=1}^nB_i\right)$$ then $a_n$ is decreasing and, since it is bounded, it is convergent. Since $a_1<\infty$, we have that: $$\lim_{n\to\infty}a_n=\lambda\left(A\setminus\bigcup_{i=1}^\infty B_i\right)=\lambda(\varnothing)=0$$ So, there exists a $n_0\in\mathbb{N}$ such that, for every $n\geq n_0$: $$0\leq a_{n}<\frac{\epsilon}{2}\Rightarrow0\leq\lambda\left(A\setminus\bigcup_{i=1}^n B_i\right)<\frac{\epsilon}{2}$$ Now, let $$B=\bigcup_{i=1}^{n_0}B_i$$ so: $$\lambda(S(A,B))=\lambda(A\setminus B)+\lambda(B\setminus A)<\frac{\epsilon}{2}+\frac{\epsilon}{2}=\epsilon$$ Since $B_i$ are all finite unions of intervals, it is obvious that $B$ is also a finite union of intervals.

Now, if $s$ is a simple function with values $c_1,\dots,c_n$, and $A_i=s^{-1}(\{c_i\})$ for every $i=1,2,\dots,n$, then: $$s(x)=\sum_{i=1}^nc_i1_{A_i}(x)$$ Due to the triangular inequality and the linearity of Lebesgue's integral, we shall prove the requested only in the case when $s=1_A$, where $A$ is some measurable set - note that a set $A$ is measurable if and only if $1_A$ is a measurable function.

Let $\epsilon>0$. If $s=1_A$, and $s$ is measurable, then $A$ is also measurable. So, there exists a union of intervals $B$, such that: $$\lambda(S(A,B))<\epsilon$$ Let $f=1_B$. Note now that, on every $x\in A\cap B$ we have $s=f=1\Rightarrow s-f=0$. On every $x\in J\setminus(A\cup B)$ we have that $s=f=0\Rightarrow s-f=0$. Now, on every $x\in S(A,B)$ we have that $|s-f|=1$ (explain why), so $$|s-f|=!_{S(A,B)}$$ Now it follows that: $$\int_J|s-f|d\lambda=\int_J1_{S(A,B)}d\lambda=\int_{S(A,B)}d\lambda=\lambda(S(A,B))<\epsilon$$.

Try to complete this proof - all the details missing. Ask whatever you may need! :)