From Peressini, Sullivan, Uhl, the mathematics of nonlinear programming,

A function is coercive if

$\lim\limits_{\|x\| \to \infty} f(x) \to \infty$

and super-coercive if,

$\lim\limits_{\|x\| \to \infty} \dfrac{f(x)}{\|x\|} \to \infty$

I want to know if $f(x,y) = x\log(x)+y\log(y)$ is a (super) coercive function

If I apply the definition, I get,

$\lim\limits_{\|(x,y)\| \to \infty} x\log(x)+y\log(y)$


$\lim\limits_{\|(x,y)\| \to \infty} \dfrac{x\log(x)+y\log(y)}{\sqrt{x^2 + y^2}}$

But there is no obvious dependence on $\|(x,y)\|$ for $f(x,y)$. Is there any way I can argue that this function is or isn't coercive.


Let $x = r\cos t, y = r\sin t$ with $0\leq t\leq \frac \pi 2.$ Then: $$\dfrac{f(x,y)}{r} = \dfrac{r\cos t\log(r\cos t)+r\sin t\log(r\sin t)}{r}=$$ $$=\log(r)(\cos t+\sin t)+\cos t\log\cos t+\sin t\log\sin t.$$ Can you take it from here?

You can prove that: $$g(t) = \cos t\log\cos t+\sin t\log\sin t$$ is bounded below when $0<t<\pi/2$ by some simple calculus.


Since $f(x,y)=g(x)+g(y)$ where $g(x)=x\log(x)$ you can bound it with a one-dimensional function.

First, presume without loss of generality that $x\geq y\geq0$. Then $\left\|(x,y)\right\|=\sqrt{x^2+y^2}\leq\sqrt{2}x$.

Secondly, we have $g(x)$ is bounded below hence $f(x,y)=g(x)+g(y)\geq g(x)+C$ for some finite constant $C=\min_{x\geq0}g(x)$.

Now we can get $$\lim_{\stackrel{\left\|(x,y)\right\|\rightarrow\infty}{x\geq y}}\frac{f(x,y)}{\left\|(x,y)\right\|}\geq\lim_{x\rightarrow\infty}\frac{g(x)+C}{\sqrt{2}x}$$ and you can simply prove that $g(x)$ (hence $g(x)/\sqrt{2}$) is (super-)coercive in one dimension.

  • $\begingroup$ $g(y)$ can be negative for small $y,$ so the lower bound may not hold. $\endgroup$ – dezdichado Dec 3 '18 at 3:45
  • $\begingroup$ Thanks, need bounded below instead $\endgroup$ – obscurans Dec 3 '18 at 3:50

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.