[Math] Conjugate function of negative logarithm

convex optimizationfunctions

I'm trying to understand how conjugate function is built for example like

$f(x) = -\log x$ where $\operatorname{dom} f = \mathrm{R_{++}}$

One of the lectures which I'm checking

By definition, $f^{*}(y) = \sup_x(yx + \log x)$

Questions:

  • what does it mean that the difference is bounded?
  • why the difference is bounded if and only if $y < 0$?
  • why is it reaching its maximum at $x = \frac{-1}{y}$?

Best Answer

You want to compute $\sup_x f(x)$ with $f : \mathbb{R}_{++} \to \mathbb{R}$ defined by $f(x) = yx+\log(x)$. For computing the supremum, it is useful to know the derivative:$$\frac{df}{dx} = y+\frac{1}{x}.$$ Since $x>0$, $df/dx>0$ when $y\geq 0$. That means that when $y\geq 0$:$$\sup_x f(x) = \lim_{x \to \infty} f(x) = \infty.$$ On the other hand, when $y<0$, the derivative is positive for $x<-1/y$, and negative for $x>-1/y$. That means that the supremum is attained at $x=-1/y$.

Related Question