So, it would appear I have forgotten the basic requirements of Lebesgue integration!

Let $(\Omega,\Sigma,\mu)$ be an arbitrary $\sigma$-finite measure space. I'm particularly interested in the real numbers equipped with the Borel measure, which we could denote $(\mathbb{R},\mathcal{B},\beta)$. Similarly, let $(\overline{\mathbb{R}},\overline{\mathcal{B}},\beta)$ denote the set of extended reals equipped with the Borel measure.

Question 1. What measurability conditions need to be true of a function $f:\Omega\to[-\infty,\infty]$ in order for its Lebesgue integral $\int_\Omega f\;d\mu$ to be well-defined?

Definition. Let $(\Omega_1,\Sigma_1)$ and $(\Omega_2,\Sigma_2)$ be measure spaces, i.e. nonempty sets equipped with $\sigma$-algebras. A function $f:\Omega_1\to\Omega_2$ is called $(\Sigma_1,\Sigma_2)$-measurable if and only if $f^{-1}(A)\in\Sigma_1$ for all $A\in\Sigma_2$.

I have been laboring for some time under the impression that $f$ needs to be $(\Sigma,\overline{\Lambda})$-measurable, where $(\overline{\mathbb{R}},\overline{\Lambda},\lambda)$ denotes the extended real numbers equipped with the Lebesgue measure. However, I've just been reading both Bogachev and Royden, and they seem to be requiring only that $f$ be "measurable" in the following sense: that $\{x:f(x)<c\}\in\Sigma$ for every $c\in\mathbb{R}$, and also $f^{-1}(\infty),f^{-1}(-\infty)\in\Sigma$. This turns out to be equivalent to saying that $f$ is $(\Sigma,\overline{\mathcal{B}})$-measurable.

But there are functions $f:\mathbb{R}\to[-\infty,\infty]$ which are $(\mathcal{B},\overline{\mathcal{B}})$-measurable but not $(\mathcal{B},\overline{\Lambda})$-measurable. For instance, take the usual example of the homeomorphism $g:[0,2]\to[0,1]$ built out of the Cantor "devil's staircase" function, and extend it to all of $\mathbb{R}$ by letting $g(x)=g(0)$ whenever $x<0$ and $g(x)=g(2)$ whenever $x>2$. Then $g$ is continuous and hence $(\mathcal{B},\overline{\mathcal{B}})$-measurable, but it is not $(\mathcal{B},\overline{\Lambda})$-measurable. Is $g$ really integrable?

So, I guess Question 1 can be made more specific by breaking it up into the following:

Question 2. Does the definition of the Lebesgue integral of a function $f:\Omega\to[-\infty,\infty]$ require that $f$ be $(\Sigma,\overline{\Lambda})$-measurable, or is $(\Sigma,\overline{\mathcal{B}})$-measurability good enough?

I've also noticed that some of the Lemmas and Propositions involved with the Lebesgue integral require that $\Sigma$ be complete, i.e it contains all subsets of sets of measure zero. The Borel $\sigma$-algebra is the one I'm most interested in, but it is not complete. So:

Question 3. Does the definition of the Lebesgue integral of a function $f:\Omega\to[-\infty,\infty]$ require that $(\Omega,\Sigma)$ be complete? In particular, is $(\mathbb{R},\mathcal{B})$ allowed to be used for the domain of $f$?

According to my reading of the Royden and Bogachev texts, the answer to question 2 appears to be that we only need $(\Sigma,\overline{\mathcal{B}})$-measurability, not $(\Sigma,\overline{\Lambda})$-measurability. And the answer to question 3 appears to be that no, $(\Omega,\Sigma)$ need not be complete, and that $(\mathbb{R},\mathcal{B})$ is allowed to be used. But this seems suspicious, and I am not convinced I have understood the texts correctly.

Any help would be much appreciated. Thanks in advance!

  • $\begingroup$ take a look at the definition of the Bochner integral, what generalize and simplifies the Lebesgue integral $\endgroup$ – Masacroso Dec 10 '18 at 14:53

To define the Lebesgue integral you need the domain to be a measure space. You don't need (and never use) a measure on the codomain. What you need of the codomain is that it is a topological space, so that you can take limits (my mantra: an integral is a limit of sums of "value of the function times size of the region").

So, initially what you want for the integral of $f$ to make sense is that the preimage of open sets is measurable. This can be seen if you write for instance $$ \int_\Omega f\,d\mu=\lim_{n\to\infty}\sum_{k=1}^{n^2}\tfrac kn\,\mu(f^{-1}((\tfrac kn,\tfrac{k+1}n]) $$ (which is a great way to see why one needs measurability to define the integral).

Because the codomain has a topology, you can consider the Borel $\sigma$-algebra on it, and as you mention measurability in the above sense is equivalent to $(\Sigma,\mathcal B)$ in your notation.

In summary: you need a $\sigma$-algebra in the codomain. You don't need a topology in the domain.

On a separate note, you mention integration of non-positive value functions. That's a different beast. All the above applies when $f\geq0$. When $f$ takes both positive and negative values, you need both $f^+$ and $f^-$ to have finite integral, and then you can define $$ \int_\Omega f\,d\mu=\int_\Omega f^+\,d\mu-\int_\Omega f^-\,d\mu. $$

| cite | improve this answer | |
  • $\begingroup$ Thank you! And just one more thing: Does the domain measure space need to be complete? $\endgroup$ – Ben W Dec 10 '18 at 14:35
  • $\begingroup$ Just to define the Lebesgue integral, no. $\endgroup$ – Martin Argerami Dec 10 '18 at 15:05

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.