Show that the eigenvalues of the Hessian of $$f(x_1,x_2) := x_1^2+x_1x_2+x_2^2+\ln (1+2e^{x_2})$$ are bounded, i.e., $$1 \leq \lambda_{\min} \left(\nabla^2 f(x)\right) \leq \lambda_{\max}\left(\nabla^2 f(x)\right) \leq 4$$

What I tried is finding gradient and then Hessian

$$ \nabla f(x) = \begin{bmatrix} 2x_1+x_2 \\ x_1+2x_2+\frac{2e^{x_2}}{1+2e^{x_2}} \end{bmatrix} $$

and Hessian is:

$$ \nabla^2 f(x) = \begin{bmatrix} 2 & 1 \\ 1 & 2 +\frac{2e^{x_2}}{(1+2e^{x_2})^2} \end{bmatrix} $$

Let $y=\frac{2e^{x_2}}{(1+2e^{x_2})^2}$. Then

$$ \nabla^2 f(x) = \begin{bmatrix} 2 & 1 \\ 1 & 2 +y \end{bmatrix} $$


$$ \lambda_{1,2} = \frac{4+y\pm \sqrt{y^2+4}}{2} $$

My questions are:

  1. Am I mistaken?

  2. Is this function a well-known function?

  • $\begingroup$ The Hessian seems to depend on $x_2$. Must the boundedness of the eigenvalues hold for all $x_2 \in \mathbb R$? $\endgroup$ – Rodrigo de Azevedo May 12 at 17:22
  • $\begingroup$ First of all, show that $y$ is bounded! $\endgroup$ – Fakemistake May 12 at 17:29

Hint: Show that $y$ is bounded for all $x_2\in\mathbb{R}$:

Part 1:

\begin{align}y=\frac{2e^{x_2}}{(1+2e^{x_2})^2}< \frac{1+2e^{x_2}}{(1+2e^{x_2})^2}=\frac{1}{1+2e^{x_2}}<1\end{align}

Part 2: Observe that \begin{align} \frac{2e^{x_2}}{(1+2e^{x_2})^2}=\frac{-1+1+2e^{x_2}}{(1+2e^{x_2})^2}=-\frac{1}{(1+2e^{x_2})^2}+\frac{1}{1+2e^{x_2}}\end{align} but since $(1+2e^{x_2})^2>1+2e^{x_2}$ we have $$\frac{1}{1+2e^{x_2}}>\frac{1}{(1+2e^{x_2})^2}$$ and thus $$-\frac{1}{(1+2e^{x_2})^2}+\frac{1}{1+2e^{x_2}}>0$$ Consequently

\begin{align}y=\frac{2e^{x_2}}{(1+2e^{x_2})^2}=-\frac{1}{(1+2e^{x_2})^2}+\frac{1}{1+2e^{x_2}}>0 \end{align}

Now you can do the final part with $0<y<1$

  • $\begingroup$ For the part two you have an extra exponent for the first denominator. My question: how would you conclude the result using $0<y<1$ and closed form formula of $\lambda_{1,2}$ ? $\endgroup$ – Saeed May 12 at 18:53
  • $\begingroup$ In other words why $1 \leq \lambda \leq 4$? $\endgroup$ – Saeed May 12 at 19:07
  • $\begingroup$ @Saeed I've edited my answer and deleted the wrong part. Maybe it helps a little bit. $\endgroup$ – Fakemistake May 19 at 8:52

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.