We all know that if $f\leq{}g$ in $[a,b]$ then $$ \int_a^bf\,dx\leq\int_a^bg\,dx $$ now, imagine that we have $f<g$, is it true that

$$ \int_a^bf\,dx<\int_a^bg\,dx $$

  • $\begingroup$ Hint: make it one integral in your inequality. $\endgroup$ – Steve D May 22 '16 at 1:38
  • $\begingroup$ @SteveD's hint : this reduces to the question of whether or not $h >0 \implies \int h > 0$ $\endgroup$ – Chill2Macht May 22 '16 at 1:39
  • 1
    $\begingroup$ This is indeed true, but I am not sure if we have any proof that avoids ideas from measure theory. $\endgroup$ – Sangchul Lee May 22 '16 at 1:46
  • 1
    $\begingroup$ @SangchulLee: huh? $(g-f)$ has a positive minimum ($[a,b]$ is compact)... $\endgroup$ – Steve D May 22 '16 at 2:10
  • 1
    $\begingroup$ @SteveD If it's not continuous, the infimums may be $0$. $\endgroup$ – MathematicsStudent1122 May 22 '16 at 2:11

Here is an argument which borrowed ideas from measure theory, but does not assume any direct knowledge on it.

Proof. It suffices to prove the following claim:

Claim. If $h \in \mathscr{R}([a, b])$ satisfies $h \geq 0$ and $\int_{a}^{b} h \, \mathrm{d}x = 0$, then $h(c) = 0$ for some $c \in [a, b]$.

Step 1. To this end, assume that $h$ satisfies the assumptions of the claim. Then we have the following observation:

Observation. For any $\epsilon > 0$ and $\delta > 0$, there exists a relatively open subset $U \subseteq [a, b]$ such that

  1. $U$ is the union of finitely many relatively open subintervals of $[a, b]$,
  2. the lengths of $U$ is less than $\delta$, and
  3. $\{ x \in [a, b] : h(x) > \epsilon \} \subseteq U$.

We first check that this indeed implies the claim. For each $n \geq 1$, choose $U_n$ as in Observation with $\epsilon = 1/n$ and $\delta = 3^{-n}(b-a)$, so that

  1. the length of $U_n$ is less than $3^{-n}(b-a)$, and
  2. $\{ x \in [a, b] : h(x) > 1/n \} \subseteq U_n$.

Then we find that

$$ \{ x \in [a, b] : h(x) > 0 \} = \bigcup_{n=1}^{\infty} \{ x \in [a, b] : h(x) > 1/n \} \subseteq \bigcup_{n=1}^{\infty} U_n. $$

Now assume otherwise that $h > 0 $ on all of $[a, b]$. Then it follows that $\bigcap_{n=1}^{\infty} U_n = [a, b]$ and thus $\{ U_n : n \geq 1 \}$ is an open cover of $[a, b]$. So we can pick a finite subcover, say $\{ U_{n_1}, \dots, U_{n_K} \}$. This implies that

$$ [a, b] = U_{n_1} \cup \cdots \cup U_{n_K}. $$

This is a contradiction since the right-hand side has length at most

$$\sum_{n=1}^{\infty} 3^{-n}(b-a) < b-a. $$

Step 2. It now remains to prove the observation. (The proof is essentially a variant of the Markov's inequality.)

Choose a partition $P$ such that $U(P, h) < \delta \epsilon$. Write $P = \{a = x_0 < \cdots < x_N = b\}$ an define $M_j = \sup_{[x_{j-1}, x_j]} h$ and $\Delta x_j = x_j - x_{j-1}$. Then we know that $U(P, h) = \sum_{j=1}^{N} M_j \Delta x_j < \delta \epsilon$. On the other hand, let $J$ be the set of indices $j$ for which $M_j > \epsilon$. Then

$$ \sum_{j \in J} \Delta x_j \leq \frac{1}{\epsilon} \sum_{j \in J} M_j \Delta x_j \leq \frac{1}{\epsilon} U(P, h) < \delta $$

and that $\cup_{j \notin J} [x_{j-1}, x_j]$ is a finite union of closed intervals on which $h \leq \max_{j \notin J} M_j \leq \epsilon$ holds. Therefore the observation follows by taking $U$ as the complement of $\cup_{j \notin J} [x_{j-1}, x_j]$.


Suppose on the contrary that $\int f=\int g$, we can assume $f,g$ vanishes outside the interval, then they are both $\mathscr L^1$, then we have $$\int_\Bbb R(f-g)dx=0. $$ since $f- g\ge 0$, indeed $f=g$ a.e. on $\Bbb R$, contradiction.

I'd appreciate an elementary proof without any appeal to Lebesgue theory.

  • 1
    $\begingroup$ It suffices to show that there is at least one point where $f-g$ is continuous. This answer gives a proof which does not formally require measure theory, although it has a measure-theoretic flavor. It would be nice to see a more elementary proof, but i haven't found one. $\endgroup$ – Bungo May 22 '16 at 6:04
  • $\begingroup$ @Bungo fair observation. Reduces the problem to a considerably more trivial one. $\endgroup$ – Vim May 22 '16 at 6:25

Assuming $$f+h=g$$

and $$h>0$$

$$\int_a^b f dx+\int_a^b h dx=\int_a^b g dx$$


$$\int_a^b h dx>(b-a)\times\min(h(x))>0$$ We can write

$$\int_a^b f dx < \int_a^b g dx$$

but what if $h$ has no minimum? it is enough to find any piece where $h$ has a minimum to prove $\int_a^b h dx >0$. Unless $h$ has no minimum at any neighborhood. But if $f$ and $g$ are integrable then $h$ must be integrable too.

  • $\begingroup$ Might be quite tricky to show that in at least one neighbourhood $h$ attains a minimum. The idea is inspiring, though. $\endgroup$ – Vim May 22 '16 at 3:25
  • $\begingroup$ @Vim Not that tricky. Riemann integrable functions are continuous almost everywhere. $\endgroup$ – MathematicsStudent1122 May 22 '16 at 3:25
  • 3
    $\begingroup$ @MathematicsStudent1122 that's an appeal to measure theory, which I think Arashium wants to avoid using here. $\endgroup$ – Vim May 22 '16 at 3:28
  • $\begingroup$ @Vim, Arashium is an engineer and not a mathematician. I have no idea about measure theory :) $\endgroup$ – Arashium May 22 '16 at 3:31
  • $\begingroup$ @Vim Can we prove the weaker statement $\text{discontinuous everywhere} \Longrightarrow \text{not Riemann integrable}$ without measure theory? If so, we're done. Intuitively, it seems clear; in such a situation, regardless of how fine one makes a partition, the $\sup$ and $\inf$ of $f$ in every subinterval will still be "far away" from each other. $\endgroup$ – MathematicsStudent1122 May 22 '16 at 5:38

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.