Note: The following questions are from the third question of the 2010 Chinese Graduate Entrance Examination Mathematics (first set):

Suppose m, n are positive integers, which of the following conclusions on the convergence of the abnormal integral $\int_{0}^{1} \frac{\sqrt[m]{\ln ^{2}(1-x)}}{\sqrt[n]{x}} d x$ is correct?

$$\begin{array}{c} &(A)&\text{It is Only related to the value of m} \\ &(B)& \text{It is Only related to the value of n }\\ &(C)& \text{It is related to the values of m and n}\\ &(D)&\text{It has nothing to do with the value of m, n}\\ \end{array}$$

I used Mathematica to verify this problem, but I didn't get results:

Integrate[Power[Log[1 - x]^2, (m)^-1]/Power[x, (n)^-1], {x, 0, 1}, 
 Assumptions -> Element[m | n, PositiveIntegers]]

The answer is $D)$ for the following reason. First, expand the series around $x=0$. This gives you

$$\frac{\sqrt[m]{ \log ^2(1-x)}}{\sqrt[n]{x}}$$

$$=x^{\frac{2}{m}-\frac{1}{n}}+\frac{x^{\frac{2}{m}-\frac{1}{n}+1}}{m}+\frac{5 x^{\frac{2}{m}-\frac{1}{n}+2}}{12 m}+\frac{x^{\frac{2}{m}-\frac{1}{n}+3}}{4 m}+\frac{x^{\frac{2}{m}-\frac{1}{n}+3}}{6 m^3}+\frac{x^{\frac{2}{m}-\frac{1}{n}+2}}{2 m^2}+\frac{5 x^{\frac{2}{m}-\frac{1}{n}+3}}{12 m^2}+\cdots$$

Now, it is not too hard to prove that the terms in the exponent will always be of the form

$$k+\frac{2}{m}-\frac{1}{n}\text{ for }k\in\{0,1,2,...\}$$

(I leave the details of this as an exercise). In general, the integral

$$\int_0^1 x^\beta dx$$

converges unless $\beta\leq -1$. Thus, our question becomes, is it ever possible for

$$k+\frac{2}{m}-\frac{1}{n}\leq -1$$

The answer is no. This is obvious as

$$-1=0-1<\frac{2}{m}-1\leq \frac{2}{m}-\frac{1}{n}\leq k+\frac{2}{m}-\frac{1}{n}$$

We conclude that no matter which $n,m\in\mathbb{N}$ we choose, the given integral will always converge.

EDIT: I was going to formally prove the radius of convergence for the power series above. However, I got bogged down in the details so instead here is a more rigorous proof of an entirely different flavor:

First, note that


(and that $\log(1-x)<0$ for $x\in(0,1)$). We also know that around $x=\frac{1}{2}$ we have

$$-\log(1-x)=\log(2)+\sum_{i=1}^\infty \frac{2^i}{i}\left(x-\frac{1}{2}\right)^i$$

$$\frac{x}{1-x}=1+\sum_{i=1}^\infty 2^{i+1}\left(x-\frac{1}{2}\right)^i$$

It is a standard exercise to compute both these series and show that the radius of convergence of both is $r=\frac{1}{2}$. Since $\log(2)\approx0.69<1$, we can compare term by term and see that for $x\in (0,1)$


This implies

$$\int_0^1 \frac{\sqrt[m]{ \log ^2(1-x)}}{\sqrt[n]{x}}dx\leq \int_0^1 \frac{\left(\frac{x}{1-x}\right)^{2/m}}{\sqrt[n]{x}}dx$$

For $m\geq 3$, this integral evaluates to

$$=\frac{\Gamma \left(\frac{m-2}{m}\right) \Gamma \left(-\frac{1}{n}+1+\frac{2}{m}\right)}{\Gamma \left(2-\frac{1}{n}\right)}$$

For $m=2$ and $n\geq 2$, the integral evaluates to

$$\frac{n H\left(\frac{n-1}{n}\right)}{n-1}$$

where $H(x)$ is the function given by


This converges for all $x>-1$. For $m=2$ and $n=1$ we get $\frac{\pi^2}{6}$. For $m=1$ and $n=1$, the integral becomes $2\zeta(3)$ where

$$\zeta(x)=\sum_{i=1}^\infty \frac{1}{i^x}$$

This converges for $x>1$. For $m=1$ and $n=2$, the integral becomes

$$2 \left(8+\log ^2(4)-8 \log (2)\right)-\frac{2 \pi ^2}{3}$$

Finally, for $m=1$ and $n\geq 3$ we will use Holder's Inequality. For positive functions, the inequality states

$$\int f(x)g(x)dx\leq \left(\int f(x)^pdx\right)^{1/p}\left(\int g(x)^qdx\right)^{1/q}$$

where $q,p\in [1,\infty)$ and $\frac{1}{q}+\frac{1}{p}=1$. In our case, we will use




Then the inequality states

$$\int_0^1 \frac{ \log ^2(1-x)}{\sqrt[n]{x}}=\int_0^1 \frac{(-\log(1-x))^2}{x^{1/n}}\leq \left(\int_0^1(-\log(1-x))^4dx\right)^{1/2}\left(\int_0^1\frac{1}{x^{2/n}}dx\right)^{1/2}$$


Having exhausted all cases, we conclude the integral converges for all $m,n\in\{0,1,2,...\}$.

  • $\begingroup$ Your answer is great, thank you very much. $\endgroup$ – Please correct GrammarMistakes Aug 12 '20 at 1:48
  • $\begingroup$ I have another question. You have not proved that the convergence region of Taylor series contains [0,1], especially at point 1. How to solve this problem? $\endgroup$ – Please correct GrammarMistakes Aug 12 '20 at 2:44
  • 1
    $\begingroup$ It is easy enough to show that the radius of convergence includes $(0,1)$ (just start with the natural log term and simplify from there). For the endpoints, note that the integral is not even defined with these endpoints, and in fact is a limit as $\epsilon\to 0$ (where $\epsilon$ is the lower limit) and $\alpha\to 1$ (where $\alpha$ is the upper limit) of the integral respectively $\endgroup$ – QC_QAOA Aug 12 '20 at 3:04
  • $\begingroup$ Thank you very much. I hope you can attach the proof of convergence region containing (0,1) to the above answer in detail, so that it is easier for others to understand. $\endgroup$ – Please correct GrammarMistakes Aug 12 '20 at 3:23

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.