Let $f$ be a non-negative Riemann integrable function on $[a,b]$. If $f$ equals to zero except on an null set,then$\int_a^b f = 0$

Let $A$ be the null set and $M=\sup\left\{f(x):x\in[a,b]\right\}$.

For any $\epsilon>0$, there exists a sequence of intervals $(I_k)$ such that $A\subset \bigcup_{k=1}^\infty I_k$ and $\sum_{k=1}^\infty|I_k|<\epsilon$

My approach is 2-steps.

Step 1: To prove that there exists an interval $I$ such that $\{\bigcup_{k=N}^\infty I_N\} \cap A\subset I$ for some $N$

Step 2: After Step 1, $f$ on $[a,b]\setminus I$ is either zero or nonzero value but is covered by $I$ for some $K<N$, which means those nonzero values can be covered by finite numbers of intervals. Hence, it is easy to find suitable $\delta$ to make the Riemann sum in this step to be smaller than $0.5\epsilon$.

For step 1,

consider $\frac{\epsilon}{2M}$, there is a sequence of intervals $(I_k)$ such that $A\subset \bigcup_{k=1}^\infty I_k$ and $\sum_{k=1}^\infty|I_k|<\frac{\epsilon}{2M}$.

Define $I_k=[a_k,b_k]$, assume that $I_k$ is in the order that $\sup\{f(x):x\in I_k\}\le\sup\{f(x):x\in I_{k+1}\}$, otherwise we rearrange the sequence of interval.

Since $A$ is bounded from $a$ and$b$, $\sup_{x\in[a,b]}A$ is well defined, there exists $x_N\in A$, and thus $x_N\in I_N$ for some $N$ such that $x_N\gt \sup_{x\in[a,b]}A-\frac{\epsilon}{4M}$

Note that $|I_N|\lt \frac{\epsilon}{2M}$, because the infinite sum is less than$\frac{\epsilon}{2M}$, so $a_N\gt sup_{x\in[a,b]}A-\frac{\epsilon}{2M}$, so there exists an interval $I$ such that $I_N \cap A\subset I$. For $N+1$, since $x_N\le \sup\{f(x):x\in I_N\}\le \sup\{f(x):x\in I_{N+1}\}$, there exists $x\ge x_N \ge \sup_{x\in[a,b]}A-\frac{\epsilon}{2M}$, so again $I_{N+1} \cap A\subset I$. Then, we can do induction following the "N+1"argument and yield $\{\bigcup_{k=N}^\infty I_N\} \cap A\subset I$ for some $N$.

$|I|\lt \frac{\epsilon}{2M}$, so the Riemann sum on this interval is less than $\frac{\epsilon}{2}$, combining with setp 2, the proof is done.

If I am right, the assumption that $f$ is Riemann integrable is not needed, and the null set assumption can be weakened to be those intervals only need to be small than any $\epsilon\gt 0$?

If I am wrong, please teach me how to prove it.

Any help would be appreciated.


1 Answer 1


There must be a mistake in your proof if you nowhere used the assumption that $f$ is Riemann integrable (RI). (I assume we are working within the framework of the Riemann integral.) Otherwise you could let $f$ be the characteristic function of $\mathbb Q \cap [0,1].$ That $f$ satisfies the other hypotheses, but as is well known, it is not RI, hence its integral over $[0,1]$ is not even defined.

The result you're after has a short proof: Let $Z$ be the zero set of $f.$ Then $Z$ is dense in $[0,1].$ Otherwise there is an interval $I$ of positive length such that $Z\cap I = \emptyset.$ It follows that $I\subset A,$ which contradicts the assumption that $m(A)=0.$ Now let $P$ be a partition of $[0,1].$ Then $Z$ intersects each subinterval induced by $P.$ Thus, using the nonnegativity of $f,$ we have

$$0\le L(P,f)\le 0 \implies L(P,f)=0.$$

We then get $\int_0^1 f = \sup_P L(P,f) = 0$ as desired.

We can drop the assumption that $f\ge 0$ if we know the result

$$\int_0^1f = \lim_{\mu(P)\to 0} \sum_{k=1}^{n}f(c_k)\Delta x_k.$$

Here $\mu(P)$ is the mesh size of $P,$ the number of subintervals $I_k$ is $n,$ and the $c_k\in I_k$ can be chosen arbitrarily. In particular, we can choose all $c_k$ to lie in $Z.$ Such Riemann sums would then all be $0,$ showing $\int_0^1f=0.$

  • $\begingroup$ Why$L(P, f)\le0$? $\endgroup$
    – Belive
    Jan 1, 2017 at 23:35
  • $\begingroup$ Because $Z$ intersects each subinterval. So the inf over each subinterval is $\le 0.$ $\endgroup$
    – zhw.
    Jan 1, 2017 at 23:38
  • $\begingroup$ If I adopt the result in your last paragraph, it seems that I do not need to use the concept of dense set to finish the proof. For each$I_k$ containing an element from the null set, there exists an element not belongs to the null set, as the sum of the intervals is always less than $min |I_k|$ for $I_k$ containing an element from the null set. Hence I can use the elements not belongs to the null set as a tag for each $I_k $, and thus the riemann sum is zero. $\endgroup$
    – Belive
    Jan 1, 2017 at 23:49
  • $\begingroup$ I like the simpler proof based on Riemann sum given in your last paragraph. +1 $\endgroup$
    – Paramanand Singh
    Jan 2, 2017 at 8:06

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.