Let $f(x)$ be a twice-differentiable function on $(a,b)$,show that

there exsit $\xi\in(a,b)$ ,such $$\int_{a}^{b}f(x)dx=\dfrac{1}{2}(b-a)[f(a)+f(b)]-\dfrac{1}{12}(b-a)^3f''(\xi)$$

if this problem condition is Amuss that $f(x)$ is a three-differentiable function,

$$f(x)=f(a)+f'(a)(x-a)+\frac{f''(\xi_{1})\cdot (x-a)^2}{2}$$ $$f(x)=f(b)+f'(b)(x-b)+\frac{f''(\xi_{2})\cdot (x-b)^2}{2}$$ so $(1)+(2)$ $$\Longrightarrow 2f(x)=f(a)+f(b)+f'(a)(x-a)+f'(b)(x-b)+\dfrac{1}{2}[f''(\xi_{1})+f''(\xi_{2})][(x-a)^2+(x-b)^2]$$ since $f(x)$ is a three-differentiable function,so $f''(x)$ is continuous,so there exsit $\xi\in(\xi_{1},\xi_{2})$,such $$\dfrac{1}{2}[f''(\xi_{1})+f''(\xi_{2})]=f''(\xi)$$ $$\int_{a}^{b}f(x)dx=\dfrac{1}{2}(b-a)[f(a)+f(b)]-\dfrac{1}{12}(b-a)^3f''(\xi)$$ But if $f(x)$ have twice-differentiable,this methods is not usefull

Then I use this methods can't prove it.


Quoting china math

"since $f(x)$ is a three times differentiable function hence $f''(x)$ is continuous, so there exist $\xi\in(\xi_{1},\xi_{2})$, such that $$\dfrac{1}{2}[f''(\xi_{1})+f''(\xi_{2})]=f''(\xi) \tag{1}$$"

This is also true for a twice differentiable function!

Darboux theorem says that let $I$ be an open interval and $f:I\rightarrow \mathbb{R}$ a real valued function. Then $f'$ has the intermediate value property.

In your proof, modify your assumption ($f$ is a twice derivable function) and complete the proof using Darboux theorem and you have one nice proof. By the way, there is a similar problem I have answered previously here (there is a hint on how to invoke Darboux theorem to prove $(1)$).


Since $(b-a)\frac{f(a)+f(b)}{2}$ is the integral over $[a,b]$ of the linear function whose values in $a,b$ are $f(a),f(b)$, we just need to show that for any twice differentiable function over $[0,1]$, such that $f(0)=f(1)=0$, $$\int_{0}^{1}f(x)\,dx = -\frac{1}{12} f''(\xi),\qquad \xi\in(0,1).\tag{1}$$ By exploiting the condition $f(0)=f(1)=0$ and using integration by parts we have:

$$\int_{0}^{1}f(x)\,dx = -\int_{0}^{1}x\,f'(x)\,dx = \int_{0}^{1}\left(\frac{1}{2}-x\right)\,f'(x)\,dx=\int_{0}^{1}\frac{x^2-x}{2}\,f''(x)\,dx\tag{2}$$ hence the claim follows by applying the mean value theorem in the (weighted) form: $$\int_{a}^{b}u(x) v(x)\,dx = u(\xi)\int_{a}^{b}v(x)\,dx$$ holding for $v(x)=\frac{x-x^2}{2}\geq 0$ and $u(x)=f''(x)$.


For fun, I plugged in $f(x)=|x|$ and $a=-1, b=1$, which does not have second derivative at $x=0$.

$$ 1 = \int_{-1}^1 |x| \; dx = \tfrac{1}{2} \big(1 - (-1)\big) \big[ f(1)+f(-1)\big] - \tfrac{1}{12}\big(1 - (-1)\big)\, f''(\xi)$$

After simplification, the linear term vanishes and you get the strange relation that

$$ 1 = 2 - \tfrac{1}{6} f''(\xi)$$

for some $-1 < \xi < 1$. Presumably $\boxed{\xi=0}$ The result is instructive.

Working out @Semiclassical's example. Let $x = \epsilon \tan \theta$ and $t = \sin \theta$.

$$ \int_{-1}^1 dx \; \sqrt{\epsilon^2 + x^2 } = \underbrace{\int_{-1+\epsilon}^{1-\epsilon} dx \; \sqrt{\epsilon^2 + x^2 }}_{I} + \underbrace{\int_{-\epsilon}^{+\epsilon} dx \; \sqrt{\epsilon^2 + x^2 }}_{II} $$

Let's estimate I and II.

$$ \int_{-1+\epsilon}^{1-\epsilon} dx \; \sqrt{\epsilon^2 + x^2 } \approx \int_{-1+\epsilon}^{1-\epsilon} dx \; |x| = 1 - \epsilon^2 $$

What really matters is around $x=0$.

$$ \int_{-\epsilon}^{+\epsilon} dx \; \sqrt{\epsilon^2 + x^2 } \approx \int_{-\epsilon}^{+\epsilon} dx \; \epsilon\sqrt{1 + (x/\epsilon)^2 } \approx \int_{-\epsilon}^{+\epsilon} dx \; \epsilon\left(1 + \frac{x^2}{2\epsilon^2}\right) = 2\epsilon^2 + \frac{\epsilon}{6}$$

This was a no-frills computation may irritate some analysts. The idea is to get a convincing answer quickly. And it even looks like a local version of the 1st part above.

Or try a Fourier series, $f(\theta) = \sum c_n e^{2\pi i n \theta}$. Then

$$ \int_0^1 f(x) dx = 0 = \frac{1}{12} f''(\xi)$$

So there is at least 1 (in fact, many) point(s) of inflection $0 < \xi < 1$ for any Fourier series.

  • $\begingroup$ It might also be instructive to use the nascent function $f_\epsilon(x)=\sqrt{\epsilon^2+x^2}$. That way it's twice-differentiable for all $\epsilon>0$ but not for $\epsilon=0$. $\endgroup$ – Semiclassical Sep 30 '14 at 2:06

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.