Suppose $f$ is continuous on $[a, b]$, if for every continuous function $g$ on $[a, b]$ with $g(a) = g(b) = 0, \int_{a}^{b}f(x)g(x) dx = 0$, Show $f(x) = 0, \forall x \in [a, b]$,

I want to prove by contradiction, and then find a continuous $g$ such that $g(a) = g(b) = 0$ but $\int_{a}^{b}f(x)g(x) dx \neq 0$

Proof: Suppose by contradiction that $f(x) > 0 $ for some $x_0 \in [a, b]$. Since $f$ is continuous, $\exists \delta$ such that $f(x) > 0, \forall x \in [x_0 - \delta, x_0 + \delta]$. Take $g(x) = \begin{cases} 0 & \text{ if } x \in (a, x_0 - \delta) \cup (x_0+\delta, b) \\ -(x - x_0 - \delta)(x - x_0 + \delta) & \text{ if } x \in (x_0 - \delta, x_0 + \delta) \end{cases}$

Now I want to show that since $g$ is continuous on $[a, b]$, then it is integrable. Thus $\int_{a}^{b} g(x) dx = sup L(g, p)$. However, since the lower sum is $> 0$, it follows that the supremum is also $> 0$. Therefore $\int_{a}^{b} > 0$. A contradiction. However, I do not know how to show that $L(g, p) > 0$

  • $\begingroup$ The negation of $f(x) =0 \forall x$ is not $f(x) \neq 0 \forall x$. $\endgroup$ – N. S. Apr 6 '15 at 0:35
  • $\begingroup$ It should be for $\exists x \in [a, b]$. Thanks for pointing that out. $\endgroup$ – Adrian Apr 6 '15 at 0:37
  • 3
    $\begingroup$ You cannot take $g = f$ because it might not be the case that $f(a) = f(b) = 0$. My hint: Suppose $f(x) > 0$ (say; similar if it were negative) for some $x \in [a, b]$. Then $f$ is positive in a tiny interval around $x$. Let $g$ be a kind of "bump" function that is positive near x but is 0 outside this interval. What can you conclude? $\endgroup$ – Pedro Apr 6 '15 at 0:43
  • $\begingroup$ @Pedro, yes that was what my professor suggested as well. I am having trouble coming up with a "bump" function. Could you give me a hint? $\endgroup$ – Adrian Apr 6 '15 at 0:57
  • $\begingroup$ @Pedro, I came up with a "bump" function. And I edited my original post. However, I am stuck at showing how the lower sum has to be > 0 $\endgroup$ – Adrian Apr 8 '15 at 2:20

Let $g(x)=f(x)(x-a)(b-x)$ Then $$\int_a^b f(x)^2 (x-a)(b-x) dx =0 $$ Note that $$f(x)^2 (x-a)(b-x)\geq 0,\ (x-a)(b-x) > 0\ (x\in (a,b))$$

If for some $x\in [a,b]$, $f(x)\neq 0$ then since $f$ is continuous, there exists a closed set $x\in [s,t]$ : $$ f(x)^2\geq c > 0\ {\rm on}\ [s,t] \subseteq [a,b]$$

Hence $$ \int_a^b f(x)^2 (x-a)(b-x) dx \geq \int_s^t c (x-a)(b-x) dx > 0 $$ So contradiction.

  • 1
    $\begingroup$ This beats me. Very nice choice of $g(x)$. +1. $\endgroup$ – Paramanand Singh Apr 11 '15 at 6:16
  • 1
    $\begingroup$ I like solutions that are more clever than my patchwork. $\endgroup$ – r12 Apr 11 '15 at 6:38
  • $\begingroup$ Could you comment on how you showed that the the above integral is 0? If I understood the question, the integral being 0 is a property that is required. $\endgroup$ – Kayle of the Creeks Apr 26 '15 at 13:41
  • $\begingroup$ It is an assumption of $f$. If $f$ satisfies such assumption, we must show that $f=0$. That is, when $f$ satisfies the assumption, if $f$ is not $0$, we complete the proof by showing a contradiction. $\endgroup$ – HK Lee Apr 26 '15 at 13:53

Elaborating on N. S.'s comment, if $f$ is non zero at some $x_0 \in [a,b]$, then it is non-zero in some region $[x_0- \delta , x_0 + \delta ]$. WLOG take $f > 0$ in this region.

Then find a $g$ s.t. $g(a) = g(b) = 0$ but $g(x) > 0$ $ \forall x \in [x_0-\delta,x_0+\delta]$ (think of defining $g$ piecewise).

Further, make $g$ non zero only in this region (i.e. $g \geq 0$ everywhere, but $g >0$ only in $[x_0-\delta,x_0+\delta]$.

If $ x-\delta < a$ or $ x+ \delta >b$, then we take a smaller value for $\delta$. The final part is to show the integral is non-zero, but this should not be hard to do (think finding the area or in terms of step functions).

  • $\begingroup$ One pedantic issue in your construction of g, it can't just be "non-zero" only in that region. For example, if g = -f on [x0 - d, x0 + d], then the integral is zero on [x0 -d, x0 + d], and thus everywhere. $\endgroup$ – r12 Apr 10 '15 at 23:34
  • 1
    $\begingroup$ Very true. I had in the line before stated that $g(x) > 0$, but for clarity I should have made that clear with my use of "Better yet". Having said that, g = -f doesn't work, as it gets a strictly negative integral, but some function that crosses the x-axis could of course work. $\endgroup$ – Krishan Bhalla Apr 11 '15 at 4:44
  • 1
    $\begingroup$ Oh, you're right. g=-f doesn't work. Also, I thought when you wrote "non-zero", you were letting it be implied that it couldn't cross the x-axis from meaning "non-zero everywhere" and intermediate value theorem. $\endgroup$ – r12 Apr 11 '15 at 6:33

Suppose there exists some $n$ such that $f(n) = a$, $a \neq 0$. Since $f$ is continuous, let $E$ be the open interval around $n$ where $E \subset [a,b] $.

Let $E1$ be the open interval around $n$ such that $sgn(e) = sgn(n) \forall e\in E1 \subset E$.

Let $E2 \subset E1$.

Let $E2 = (c_2, d_2)$, $E1 = (c_1, d_1)$.

Construct $g$ so that $g(x) = f(x)$ for $x \in E1$, $g(x) = 0$ for $[a,b] \cap E2'$, and $g(x)$ connects the parts continuously on $E1 \cap E2'$.

Once you have that, then it should be trivial. Your integrand will be all one sign on $E1$ and non-zero, and zero everywhere else, which means your integral will be non-zero.


I'll go ahead and attempt to construct $g(x)$ on $[c_1,c_2]$ and $[d_2, d_1]$ explicitly.

$g(x) = f(x)* |\frac{x-c_1}{c_2-c_1}|$ on $x \in [c_1,c_2]$, and $g(x) = f(x)*|\frac{x-d_1}{d_2-d_1}|$.


Counter example: $f(x)= const$, $g(x)=sin(x) $, $a=0, b=2 \pi$.

Direct computations shows that $\int_{0}^{2\,\pi }f(x) \; \mathrm{sin}\left( x\right) dx =0$.

So you need more constraints.


I'll rename the $x_0$ you defined to $c \in [a, b]$.

Let $p$ be the partition $(x_0 < x_1 < x_2 < x_3) = (a < c - \frac{\delta}{2} < c + \frac{\delta}{2} < b)$.

On the intervals $[x_0, x_1]$ and $[x_2, x_3]$, either $g$ vanishes or both $f$ and $g$ are positive, so $$ \inf_{x \in [x_0, x_1]} f(x)g(x) = \inf_{x \in [x_2, x_3]} f(x)g(x) = 0. $$

On the interval $[x_0 - \frac{\delta}{2}, x_0 + \frac{\delta}{2}]$, both $f$ and $g$ are strictly positive, so $$ \inf_{x \in [x_1, x_2]} f(x)g(x) > 0. $$ Summing over all subintervals gives $$L(fg, p) = \delta \inf_{x \in [x_1, x_2]} f(x)g(x) > 0$$

so necessarily $\sup_p L(fg, p)$ > 0.


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.