It is a classical fact that a $L^p(R^d)$ ($1\leq p<+\infty$) function can be approximated by step functions with compact support, but my question will be, can we require that the step function is monotonically increasing (suppose that $f>0$ a.e. or we can do it for $f^\pm$ respectively)?

  • $\begingroup$ I googled and get no results. And my knowledge of proof of the classical fact can not modify easily to get a proof of my requirement. $\endgroup$ – van abel Jul 8 '14 at 9:02
  • 1
    $\begingroup$ What is your definition of a step function? Are we talking about simple functions (i.e. $\sum_{i=1}^n \alpha_i \chi_{A_i}$ with $A_i$ measurable) or about "Riemann" step functions, i.e. $\sum_{i=1}^n \alpha_i \chi_{A_i}$ with $A_i$ intervalls? $\endgroup$ – PhoemueX Jul 8 '14 at 9:26
  • $\begingroup$ @PhoemueX Surely, he means the latter. If anybody ever says step function when they mean simple function, I would assume they misspoke. $\endgroup$ – Harald Hanche-Olsen Jul 8 '14 at 9:45
  • $\begingroup$ I would not say that. For example Serge Lang uses the term "step map" to denote functions of the form $\sum_{i=1}^n \alpha_i \chi_{A_i}$ with $A_i$ measurable in his book "Real and Functional Analysis". $\endgroup$ – PhoemueX Jul 8 '14 at 9:48
  • $\begingroup$ @PhoemueX The step function usually means simple function on Cubes( intervals in $R^1$ case). $\endgroup$ – van abel Jul 8 '14 at 13:22

I expect you mean to ask if you can find a monotonically increasing sequence of step functions approximating the given function from below?

The answer is a resounding no. For a simple example, consider the characteristic function of a fat Cantor set.


Looking at non-negative unsigned Lebesgue integrable functions on $\mathbb{R}$,

  • step function: taking finitely many values on finitely many intervals.
  • simple function: taking finitely many values on finitely many Lebesgue measurable sets.

True: Both step functions and simple functions are dense in $L^1$. They can approximate any Lebesgue integrable function $f$ arbitrarily close using the integral norm $||\cdot||_{L^1}$.

True: For each unsigned Lebesgue integrable function $f$, there exists a sequence of monotone increasing simple functions $g_n$ such that $g_n\rightarrow f$ in $L^1$.

As Harald Olsen said, even step functions are dense in $L^1$, the statement about monotone is false. And here is the intuitive idea:

The pre-image of a step function has to be intervals. But given a Lebesgue measurable set $A$, if only using intervals, some times we can only approximate $A$ from the outside but not from the inside. That is $$\mu(A) = \inf\left\{ \sum_{n=1}^\infty |I_n| : A\subset \cup_n I_n \right\}$$ but $$\mu(A) \neq \sup\left\{ \sum_{n=1}^\infty |I_n| : A\supset \cup_n I_n \right\}.$$

  • $\begingroup$ I think you have some misunderstanding about the approximate, I think we should allow $g\leq f$ holds in the a.e. sense. In that case, we can take $f_k\equiv 1$ on $[0,1]$ as step function in your counterexample on $[0,1]$. $\endgroup$ – van abel Jul 8 '14 at 13:16
  • $\begingroup$ @vanabel Yes, you are right. The fat Cantor set is the correct example. I was only thinking in term of measure while writing down this example. $\endgroup$ – Xiao Jul 8 '14 at 13:22
  • $\begingroup$ By the way, Can you give some exactly reference or raise an answer here, of your second TRUE? I believe it will be useful to me, although the step function is replaced by simple function. $\endgroup$ – van abel Jul 8 '14 at 13:27
  • $\begingroup$ @vanabel by the definition of unsigned integral $$\int f = \sup_{0\leq g\leq f \text{ a.e. }; \text{ simple}} \int g,$$ we know $f$ can be approximated from below by a sequence of simple functions $g_n$ in $L^1$ and pointwise a.e. Taking the maximum of the first $k$ simple functions will give you the monotone sequence of simple functions $h_k$. $\endgroup$ – Xiao Jul 8 '14 at 14:05
  • $\begingroup$ That's it! I just forgot what I have learned in my mind. $\endgroup$ – van abel Jul 8 '14 at 14:39

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.