Given function $$f:\mathbb{R}^n_{+} \rightarrow \mathbb{R}, \ f(x) = \sum_{k=1}^K c_k x_1^{a_{1k}} x_2^{a_{2k}} \cdots x_n^{a_{nk}}$$

Show that for any $x, y \in \text{dom} \ f, \theta \in [0,1]$, and let $z : z_i = x_i^{\theta} y_i^{1-\theta}$, the following holds $$ f(z) \leq f(x)^{\theta} f(y)^{1-\theta}$$

I know I need to to make the transformation $$ w_i = \log (x_i), \\ \ f(x) = \sum_{i=1}^K c_k (e^{w_1})^{a_{1k}} (e^{w_2})^{a_{2k}} \cdots (e^{w_n})^{a_{nk}} \\ = \sum_{i=1}^K e^{a_k^T w + b_k} \quad \text{where $b_k = \log c_k$} $$ Then $f(x)$ is convex. But I'm not sure how to proceed from here.


In general, your function $f$ is nonconvex! For example, on $\mathbb R_+^2$ consider $f :(x_1, y_2) \mapsto x_1x_2$. However, the inequality your trying to proof holds for arbitrary $f$ defined as you did. When such an inequality holds, we say $f$ is multiplicatively convex.

So then, let's show that your $f$ is multiplicatively convex. Indeed, $$f(x) = \sum_k c_kx_1^{a_{1,k}}\ldots x_n^{a_{n,k}} = \sum_k e^{a_k^Tw(x) + b_k}, $$ where $w(x) := (\log x_1, \ldots, \log x_n)$, and $b_k := \log c_k$.

Now, $w(z) = (\log(x_1^\theta y_1^{1-\theta}), \ldots, \log(x_n^\theta y_n^{1-\theta})) = \ldots = \theta w(x) + (1-\theta) w(y)$. Thus $$ \begin{split} f(z) &= \sum_k e^{a_k^T(\theta w(x) + (1-\theta) w(y)) + b_k} = \sum_k e^{\theta (a_k^Tw(x) + b_k)}e^{(1-\theta) (a_k^Ty(x) + b_k)}\\ &= \sum_k e^{\theta (a_k^Tw(x) + b_k)}\sum_k e^{(1-\theta) (a_k^Ty(x) + b_k)} \le \left(\sum_k e^{a_k^Tw(x) + b_k}\right)^\theta\left(\sum_k e^{a_k^Ty(x) + b_k}\right)^{1-\theta}\\ &\hspace{7.6cm}= f(x)^\theta f(y)^{1-\theta}, \end{split} $$ where the last inequality holds because $\theta, 1-\theta \in [0,1]$. Thus $f$ is multiplicatively convex.


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.