8
$\begingroup$

I am seeking solution verification for the following problem.

Suppose $f$ is an integrable function. We wish to show that there exists a simple function $\phi$ such that $$ \int{|f-\phi|} \> d\mu < \epsilon.$$ Attempt: Write $f = f^{+} - f^{-}$. The functions $f^{+}$ and $f^{-}$ are non-negative measurable functions, and our assumption implies that there integrals are finite. These functions can be approximated by monotonically increasing sequences of simple functions. In particular, there exists a sequence of simple functions $(\psi_{n})$ that monotonically increases to $f^{+}$ and there exists a sequence of simple functions $(\psi^{'}_{n})$ that monotonically increases to $f^{-}$. The Monotone Convergence Theorem implies $$\lim \int{\psi_{n}} \> d\mu = \int{f^{+}} \> d\mu$$ and $$\lim \int{\psi^{'}_{n}} \> d\mu = \int{f^{-}} \> d\mu.$$

Now, take $N$ large enough so that $n \geq N$ implies $$\left| \int{f^{+}} \> d\mu - \int{\psi_{n}} \> d\mu \right| < \frac{\epsilon}{2}.$$ Likewise, take $N'$ large enough so that $n \geq N'$ implies $$\left|\int{f^{-}} \> d\mu - \int{\psi^{'}_{n}} \> d\mu \right| < \frac{\epsilon}{2}$$ Observe that $|\int{f^{+}} \> d\mu - \int{\psi_{n}} \> d\mu| = \int{f^{+}} \> d\mu - \int{\psi_{n}} \> d\mu = \int{f^{+}-\psi_{n}} \> d\mu$ because $\psi_{n} \leq f^{+} $ for each $n$ and the linearity of the integral. The analogue is true for $f^{-}$ and $\psi^{'}_{n}$.

Put $\phi = \psi_{N} - \psi^{'}_{N'}$. Then,

\begin{align} \int{|f-\phi|} \> d\mu &= \int{|f^{+}-f^{-}-\psi_{N} + \psi^{'}_{N'}|} \> d\mu \\ &\leq \int{|f^{+}-\psi_{N}|+|-f^{-} + \psi^{'}_{N'}|} \> d\mu \\ &= \int{|f^{+}-\psi_{N}|+|-1||f^{-} - \psi^{'}_{N'}|} \> d\mu \\ &=\int{|f^{+}-\psi_{N}|+|f^{-} - \psi^{'}_{N'}|} \> d\mu \\ &= \int{f^{+}-\psi_{N}} \> d\mu +\int{f^{-} - \psi^{'}_{N'}} \> d\mu \\ &< \frac{\epsilon}{2} + \frac{\epsilon}{2} = \epsilon. \end{align} The proof is complete.

$\endgroup$
2
  • $\begingroup$ So you are assuming the statement in the case $f \geq 0$ and you wish to extend it to all $f \in L^1$. The proof seems correct. $\endgroup$ – snar Jul 30 '14 at 3:37
  • $\begingroup$ Yes. The $\psi$'s should be $\phi$'s. Fixed. :) $\endgroup$ – dgc1240 Jul 30 '14 at 3:58
3
$\begingroup$

Actually, you are a little bit mistaken. Your proof fails when you claim that $\int|f^+ - \psi_N | d\mu < \frac{\epsilon}{2} $. What you get is $ | \int f^+ - \psi_N d\mu | < \frac{\epsilon}{2} $. To solve this problem you have to use the Dominated Corvenge Theorem as follows:

First let $f$ be a nonnegative function. Then there exists a monotone increasing sequence of simple functions $(\phi_n)$ such that $\phi_n \rightarrow f$. Hence $|\phi_n| \leq |f|$. Not only that, but $\phi_n - f \rightarrow 0 $. Now, notice that: \begin{equation} |\phi_n - f| \leq |\phi_n| + |f| \leq 2|f|. \end{equation}

But one has that $2|f|$ is integrable. Then, $|\phi_n - f|$ is integrable. But notice that $|\phi_n - f| \rightarrow 0 $. Being $|\phi_n - f|$ dominated by an integrable function and converging to $0$, you can apply the Dominated Convergence Theorem to see that:

\begin{equation}0 = \int 0 d\mu = lim \int |\phi_n - f| d\mu. \end{equation}

Now you can see that, if $f$ is a nonnegative function, there exists a simple function that is a good approximation for $f$.

The exercise follows observing that any function $f$ can be written as the difference between its postive and its negative parts.

$\endgroup$
2
$\begingroup$

One can just use the definition of the Lebesgue integral without having to rely on the dominated convergence theorem.

Indeed, there exists a simple function $s_1$ such that $0 \leq s_1 \leq f^{+}$ and $\int f^+ d\mu < \int s_1 d\mu+\frac{\epsilon}{2}$. Similarly there exists a simple function $s_2$ such that $0\leq s_2\leq f^-$ and $\int f^- d\mu < \int s_2 d\mu + \frac{\epsilon}{2}$. Since $f^+-s_1 \geq 0$ and $f^--s_2\geq0$ we get $\int |f-s_1+s_2| d\mu \leq \int |f^+-s_1| d\mu +\int |f^--s_2|d\mu = \int f^+ -s_1 d\mu + \int f^--s_2 d\mu < \epsilon$.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.