Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Let $a_i >0$ for $1\le i \le n$, and let $J=(0,1) \times \dots \times (0,1)$.

I want to prove : $$\int_J{1 \over x_1^{a_1}+x_2^{a_2}+ \dots+x_n^{a_n}}dx<\infty \Longleftrightarrow \sum^n_{i=1} {1 \over a_i}>1$$

It's not simple problem I think. How can I prove that? Should I use Fubini theorm?

share|improve this question

2 Answers 2

up vote 3 down vote accepted

Here is a proof that $\sum_{i=1}^n \frac{1}{a_i}>1$ implies the integral is finite: Denote $\sum_{i=1}^n \frac{1}{a_i}$ as $\frac{1}{a}$, then $\sum_{i=1}^n \frac{a}{a_i} = 1$. There is an arithmetic-geometric mean inequality, according to which $y_1^{s_1}y_2^{s_2}\cdots y_n^{s_n}\le s_1y_1+\cdots+s_ny_n$ when the $y_i$ are positive and the sum of the $s_i$ is 1. Apply this using $y_i=x_i^{a_i}$ and $s_i = \frac{a}{a_i}$. You find that $$y_1^{a/a_1}y_2^{a/a_2}\cdots\le\frac{a}{a_1}y_1+\frac{a}{a_2}y_2+\cdots$$ The left hand side is equal to $(x_1x_2\cdots)^a$, and the right side is less than $y_1+y_2+\cdots$. Therefore $$(x_1x_2\cdots)^a\le x_1^{a_1}+x_2^{a_2}+\cdots$$ The reciprocal gives $$\frac{1}{x_1^{a_1}+\cdots}\le\frac{1}{x_1^ax_2^a\cdots}$$ If $a<1$ then each integral $\int_0^1 \frac{1}{x_i^a}\,dx_i$ converges. Therefore, if $a<1$, then the integral converges.

share|improve this answer
    
Thanks for your perfect answer. –  japee Jun 11 '12 at 4:32

Here's a way to see both directions roughly at the same time. For any $1\le k\le n$, let $J_k$ be the subregion $\{ x \in J : x_k^{a_k} = \max(x_1^{a_1}, \ldots, x_n^{a_n})\}$. Equivalently, $x \in J_k$ iff $x_i \le x_k^{a_k/a_i}$ for each $1 \le i \le k$.

Notice that for any $x \in J_k$ we have $$x_k^{a_k} \le \sum_{i=1}^n x_i^{a_i} \le nx_k^{a_k}.$$ Let's define $I_k = \int_{J_k} x_k^{-a_k} dx.$ By the above inequalities, we clearly have $I \ge \frac1n I_k$ since $J_k \subset J$, and also $I \le I_1 + I_2 + \cdots + I_n$, because $J_1 \cup \cdots \cup J_n = J$.

Therefore $I$ converges if and only if each $I_k$ converges. It is now a routine matter to calculate $I_k$ by iterated integration, saving $x_k$ for last. The innermost $n-1$ integrals are just the volume of a box with dimensions $x_k^{a_k/a_i}$ (from the definition of $J_k$). For instance, when $k=1$ we get

$$I_1 = \int_0^1 \left(\prod_{i=2}^n x_1^{a_1/a_i} \right) x_1^{-a_1} dx_1 = \int_0^1 x_1^{-a_1 + a_1/a - 1} dx_1,$$ where $\frac1a = \frac1{a_1} + \cdots + \frac1{a_n}$ as in @BobTerrell's solution. This one-dimensional integral converges iff $-a_1 + a_1/a > 0$, in other words exactly when $a < 1$.

share|improve this answer

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.