A problem from an old Advanced Calculus qualifying exam:

"Choose positive real numbers $\alpha_1, \dotsc ,\alpha_n$ such that $\sum_1^n \alpha_i = 1$ and let $f:[0,\infty)^n\to \mathbb{R}:(x_1, \dotsc , x_n)\mapsto x_1^{\alpha_1}x_2^{\alpha_2}\dotsb x_n^{\alpha_n}$.

Define also $C:=\{\mathbf{x}\in [0,\infty)^n \mid \sum_1^n \alpha_i x_i =1 \}$.

(a) Show that there exists $\mathbf{a}\in [0,\infty)^n$ such that $f(\mathbf{a})=\sup_{x\in C}f(x)$, and $a_i >0$ for all $i$.

(b) Find all points such that $f(\mathbf{a})=\sup_{\mathbf{x}\in C}f(\mathbf{x})$. Deduce that for any $x\in C$, $x_1^{\alpha_1}x_2^{\alpha_2}\dotsb x_n^{\alpha_n}\leq 1$

(c) Deduce that for any $x\in [0, \infty)^n$, $x_1^{\alpha_1}x_2^{\alpha_2}\dotsb x_n^{\alpha_n}\leq \sum_i^n \alpha_i x_i$. When does equality hold?"

I believe I've solved part (a). The relevant domain of $\mathbf{x}$ is actually compact, even though it may not look like it at first. First, if one or more $\alpha$, say $\alpha_{i_1}$ is zero, we are not concerned about $x_{i_1}$, because it does not contribute to the function, and we may regard the function as over only the other variables. Take $m=\min_{i}\{\alpha_i \mid \alpha_i \neq 0\}$, and we find that the $x_i$ that count must all be less than $1/m$. So it attains its maximum.

I have been less successful at parts (b) and (c). Here are some ideas I've been playing with: Cauchy-Schwarz says that $(\sum_1^n \alpha_i x_i)^2 \leq (\sum_{1}^n \alpha_i)^2(\sum_1^n x_i)^2$. Since $\sum_1^n \alpha_i =1$ and $\alpha_i >0$, we have that $\alpha_i^2 \leq \alpha_i$, so $\sum_1^n \alpha_i^2 \leq \sum_1^n \alpha_i$, and putting this with what we had before, we have $(\sum_1^n \alpha_i x_i)^2 \leq \sum_1^n x_i^2$. For $\mathbf{x}\in C$, $(\sum_1^n \alpha_i x_i)^2=1$, so we have that $\|x\|\geq 1\,\,\,\forall \mathbf{x}\in C$.

Also, since $\sum_1^n \alpha_i =1$, we have $\min_{i}x_i \leq x_1^{\alpha_1}x_2^{\alpha_2}\dotsb x_n^{\alpha_n} \leq \max_i x_i$.

If we consider Lagrange multipliers, we have to have $\nabla f - \lambda \nabla g = 0$, for $f = x_1^{\alpha_1}x_2^{\alpha_2}\dotsb x_n^{\alpha_n}$ and $g = \alpha_1x_1 + \dotsb + \alpha_nx_n$, that is, $(\alpha_1 x_1^{\alpha_1 -1}\dotsb x_n^{\alpha_n}, \dotsc, \alpha_n x_1^{\alpha_1}\dotsb x_n^{\alpha_n -1}) - \lambda \vec{\alpha} = 0$. But that's as far as I've gotten. Any ideas?


Hint : for b) you can analize the expression $\log(x_1^{\alpha_1}x_2^{\alpha_2}\dotsb x_n^{\alpha_n})$. After some simple algebraic work you can apply Jensen inequality.

Then b) implies c) and you can use the equality condition of Jensen inequality to analize the equality case.

  • $\begingroup$ Thanks! I'll give it a try. $\endgroup$ – Eric Auld Jul 20 '13 at 0:24
  • $\begingroup$ Great! Very simple once I used your hint. :) $\endgroup$ – Eric Auld Jul 20 '13 at 20:42
  • $\begingroup$ Happy if i could help (: $\endgroup$ – sigmatau Jul 20 '13 at 20:49

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.