0
$\begingroup$

I would like to Maximise the function $\log\left(1+\sum_i a_i x_i\right) + \sum_i \log x_i$ subject to the constraint $\sum_i x_i = 1$, where $a_i > 0$ and $x_i>0$. My first hunch was to define the Lagrangian $$ L = \log(1+\sum_i a_i x_i) + \sum_i \log x_i + \lambda(\sum_i x_i - 1). $$ Taking the derivative and setting to zero gives $$ \frac{\partial L}{\partial x_i} = \frac{a_i}{1+\sum_j a_j x_j}+\frac{1}{x_i}+\lambda=0. $$ Multiplying through by the product of the denominators, we find $$ 0 = a_i x_i+1+\sum_j a_j x_j + \lambda x_i\left(\sum_j a_j x_j\right). $$ But now I am a bit stuck because the expression couples different $x_i$ and $x_j$. Any idea how to make progress with this?

$\endgroup$
1
$\begingroup$

Let $S=\sum_{j} a_j x_j$. Starting from $$\frac{a_i}{1+S} + \frac{1}{x_i}+\lambda=0, $$ one gets $$a_ix_i + (1+S)(1 +\lambda x_i)=0. $$ Summating on $i$ yields $$S + (1+S)(n+\lambda)=0, $$ i.e. $$S=-\frac{n+\lambda}{1+n+\lambda}, $$ where $n$ is the number of variables. Now starting again from the beginning we get $$x_i(a_i + \lambda(1+S))+1+S=0, $$ so $$x_i=-\frac{1+S}{a_i+\lambda(1+S)}, $$ and summating on $i$ gives you a condition on $\lambda$ : $$1=-\sum_i \frac{1}{\lambda+(1+n+\lambda)a_i}. $$ Such a $\lambda$ exists and is unique in the interval $(-(n+1),-n)$. Though i do not know if there is a nice way to express $\lambda$ as a function of the parameters $a_i$. Then plug in $\lambda$ in the expressions to find $S$ and $x_i$. I hope this is correct.

$\endgroup$

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.