The following comes from Bass' book on Real Analysis: (Here $dy$ is Lebesgue measure)

Exercise 7.6
Suppose $f:\mathbb{R}\to\mathbb{R}$ is integrable, $a\in \mathbb{R}$, and we define $F(x):=\int_a^x f(y)\, dy$. Show that $F$ is a continuous function.

Another answer on stack exchange covers the case where $f$ is assumed to be Riemann integrable, with $f$ and $F$ being functions from some $[a,b]$. The proof there uses the fact that $f$ is Riemann integrable if and only if it is bounded and continuous almost everywhere.

Folland's book on analysis appears to define integration for functions $f:X\to[-\infty,\infty]$ on the extended real numbers. ("The integral defined in the previous section [for functions $f:\mathbb{R}\to[0,\infty]$] can be extended to real-valued measurable functions $f$..." Except, real-valued means $\mathbb{R}$ but extending the old definition forces the inclusion of infinities. However, as far as I can tell the actual definition as integrating the positive and negative parts agrees with taking $[-\infty,\infty]$-valued functions.

I only bring this up because it seems exercise 7.6 should hold for: $f:\mathbb{R}\to[-\infty,\infty]$. $$f(x):=\begin{cases} x^{-1/2}& x>0\\ (-x)^{-1/2} &x<0 \\ \infty & x=0\end{cases}$$

In trying a few things I also wondered if the $F$ from 7.6 is always measurable. (this should be an easy yes?)

It's worth noting that chapter 7 of Bass covers the usual monotone, dominated, etc. theorems.

The question then is if the modified exercise is correct, how different are the lebesgue measure -vs- riemann measure proofs, how to go about either version (Lebesgue preferred for the modified exercise)

  • $\begingroup$ Is $\displaystyle \int f \, \mathrm{d} y$ defined as $\displaystyle \int f^+ \, \mathrm{d} y + \displaystyle \int f^- \, \mathrm{d} y$ and $\displaystyle \int f^+ \, \mathrm{d} y = \sup \left\{ \int s \, \mathrm{d} y : s \leqslant f \text{ is a simple function, i.e. has finite range} \right\}$ ? $\endgroup$
    – Adayah
    May 20, 2015 at 6:33
  • $\begingroup$ Yes, that's what I'm using :) $\endgroup$
    – ttt
    May 20, 2015 at 6:35
  • $\begingroup$ I could ask the same question except defining the function to be zero at zero. Though, the function is only infinity on a null set so it shouldn't affect F. (I'm saying modifying f on a null set doesn't change F) $\endgroup$
    – ttt
    May 20, 2015 at 6:53

2 Answers 2


I believe the proof is somewhat different than the case of Riemann integration. Also, as a comment noted, allowing functions to take values $\pm \infty$ is harmless (an integrable function can only take these values on a measure zero set). All integrable functions are measurable; this may be easy or may take some work, depending on exact definitions.

One proof of the main exercise uses the dominated convergence theorem. Fix $x_0$. For convenience, let $\chi_{(x_0,x)}$ denote the characteristic function of the interval when $x_0 < x$, and denote $-\chi_{(x,x_0)}$ otherwise. Then $$\lim_{x \to x_0} F(x) - F(x_0) = \lim_{x \to x_0} \int_{x_0}^x f(y)dy = \lim_{x\to x_0} \int_{-\infty}^{\infty} \chi_{(x_0,x)}f(y)dy = \int_{-\infty}^{\infty}\lim_{x\to x_0} \chi_{(x_0,x)}f(y)dy = \int_{-\infty}^{\infty} 0 dy = 0\text{.}$$

  • $\begingroup$ As far as I know, the implication $\text{integrable} \implies \text{measurable}$ is quite lengthy to show with all detail. $\endgroup$
    – Adayah
    May 20, 2015 at 7:36
  • $\begingroup$ @Adayah, you're right, it's not totally immediate using the definition given in your comment to the OP. It's easy for some others, though. I edited my post. $\endgroup$
    – user231101
    May 20, 2015 at 7:43
  • $\begingroup$ I was asking if $F$ (not $f$) was measurable provided $f$ was integrable. But this is also good to know. $\endgroup$
    – ttt
    May 20, 2015 at 15:42
  • $\begingroup$ So just to be clear: When you write $\int_{x_0}^x$, as far as Lebesgue integrals go thats code for $\int_\mathbb{R} \chi_{(x_0,x)}$ or $-\int_\mathbb{R}\chi_{(x,x_0)}$ depending on if $x_0<x$ or the other way around. Since its just a +/- you can let the notation "handle" the cases. $\endgroup$
    – ttt
    May 20, 2015 at 16:43
  • $\begingroup$ @ttt Exactly. The point is that it should be $F(x) - F(x_0)$ in either case. $\endgroup$
    – user231101
    May 20, 2015 at 20:22


$$\int f \, \mathrm{d} y = \int f^+ \, \mathrm{d} y + \int f^- \, \mathrm{d} y,$$

it suffices to show that

$$G(x) = \int \limits_a^x f^+ \, \mathrm{d} y$$

is continuous and the other part will follow analogously. Fix $x_0 \in \mathbb{R}$ and $\varepsilon > 0$. We will search for such $\delta > 0$ that $|G(x_0) - G(x)| < \varepsilon$ i.e. $\displaystyle \left| \, \int \limits_x^{x_0} f^+ \, \mathrm{d} y \, \right| < 3\varepsilon$ whenever $|x - x_0| < \delta$.

Since by definition

$$\int \limits_{x_0-1}^{x_0+1} f^+ \, \mathrm{d} y = \sup \left\{ \int \limits_{x_0-1}^{x_0+1} s \, \mathrm{d} y : s \leqslant f \text{ is a simple function} \right\},$$

there is such partition $(E_i)_{i=1}^n$ of $[x_0-1, x_0+1]$ and a simple nonnegative function $\displaystyle s(x) = \sum_{i=1}^n v_i \cdot \mathbf{1}_{E_i} \leqslant f(x)$ that

$$\sum_{i=1}^n v_i \cdot \lambda(E_i) \leqslant \int \limits_{x_0-1}^{x_0+1} f^+ \, \mathrm{d} y < \varepsilon + \sum_{i=1}^n v_i \cdot \lambda(E_i).$$

Let $M = \max\{ v_i : i = 1, 2, \ldots, n \}$ and $\delta = \min \{ \frac{\varepsilon}{M}, 1 \}$. For $|x - x_0| < \delta$ clearly

$$\left| \, \int \limits_x^{x_0} f^+ \, \mathrm{d} y \, \right| \leqslant \int \limits_{x_0-\delta}^{x_0+\delta} f^+ \, \mathrm{d} y \qquad \text{ and } \qquad \int \limits_{x_0-\delta}^{x_0+\delta} ( f^+ - s) \, \mathrm{d} y \leqslant \int \limits_{x_0-1}^{x_0+1} ( f^+ - s) \, \mathrm{d} y < \varepsilon$$


$$\begin{align} \int \limits_{x_0-\delta}^{x_0+\delta} f^+ \, \mathrm{d} y & < \varepsilon + \int \limits_{x_0-\delta}^{x_0+\delta} s \, \mathrm{d} y = \varepsilon + \sum_{i=1}^n v_i \cdot \lambda( E_i \cap [x_0-\delta, x_0+\delta]) \\[1ex] & \leqslant \varepsilon + \sum_{i=1}^n M \cdot \lambda( E_i \cap [x_0-\delta, x_0+\delta]) \\[1ex] & = \varepsilon + M \cdot \lambda( [x_0-\delta, x_0+\delta] ) = \varepsilon + 2 M \delta \leqslant 3 \varepsilon \end{align}$$

  • $\begingroup$ Thanks for this additional approach! Cool. $\endgroup$
    – ttt
    May 20, 2015 at 21:05

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.