My task was to find the critical points of the function $f(x,y) = x^2+y^2$, to then compute the Hessian, and to use the second derivative test to determine whether the critical points are local maxima or minima.

I think I did it correctly, but I'm new to multivariate calculus, so I'd like someone to check my work.


$f$ has a critical point when all the partial derivatives are $0$. The partials are $$\frac{\partial}{\partial x} (x^2+y^2) = 2x$$ and $$\frac{\partial}{\partial y} (x^2+y^2) = 2y$$ So $Df(x,y) = [2x \quad 2y]$. Clearly $2x=2y=0$ iff $x=y=0$. Now we need to compute the second-order partial derivatives: $$\frac{\partial}{\partial x}\left(\frac{\partial}{\partial x}\right) = \frac{\partial}{\partial x} 2x = 2 = D_{xx}f(x,y)$$ $$\frac{\partial}{\partial x}\left(\frac{\partial}{\partial y}\right) = \frac{\partial}{\partial x} 2y = 0 = D_{yx}f(x,y)$$ $$\frac{\partial}{\partial y}\left(\frac{\partial}{\partial x}\right) = \frac{\partial}{\partial y} 2x = 0 = D_{xy}f(x,y)$$ $$\frac{\partial}{\partial y}\left(\frac{\partial}{\partial y}\right) = \frac{\partial}{\partial y} 2y = 2 = D_{yy}f(x,y)$$

The Hessian is then given by: $$\textbf{H}(x,y) = \left[ \begin{matrix} D_{xx}f(x,y) & D_{xy}f(x,y) \\ D_{yx}f(x,y) & D_{yy}f(x,y) \end{matrix}\right] = \left[ \begin{matrix} 2 & 0 \\ 0 & 2 \end{matrix}\right] $$ By the second derivative test, $(0,0)$ is a local minimum if $\textbf{H}(0,0)$ is positive definite, and a local maximum if $\textbf{H}(0,0)$ is negative definite. $\textbf{H}$ is positive definite if $z^T\textbf{H}z$ is positive for every non-zero column vector $z \in \Bbb R^n$. Thus, take $z = [a \quad b]$. We have that $$[a \quad b]\left[ \begin{matrix} 2 & 0 \\ 0 & 2 \end{matrix}\right]\left[\begin{matrix} a \\ b\end{matrix}\right] = [a \quad b] \left[ \begin{matrix} 2a + 0b \\ 0a + 2b \end{matrix} \right] = [a \quad b] \left[ \begin{matrix} 2a \\ 2b \end{matrix} \right] = [2a^2 + 2b^2] $$ Which is clearly positive definite since $x^2 \geq 0 \forall x\in\Bbb R$. So $(0,0)$ is a local minimum.

  • 2
    $\begingroup$ When checking for positive/negative definiteness, it is more efficient (and more common, in this context) to do so using Sylvester's criterion $\endgroup$ – Omnomnomnom May 6 '15 at 12:13

Perfect answer, but a small comment with regard to the last matrix computation: \begin{equation} \begin{bmatrix} a & b \end{bmatrix} \begin{bmatrix} 2a \\ 2b \end{bmatrix} = 2a^2+2b^2 \end{equation} Remember that the quadratic form $z^T H z$ is a real number, not a matrix. We call the graph of the function $z=x^2+y^2$ an elliptic paraboloid, btw.

  • $\begingroup$ Why is it a real number (scalar), rather than a $1\times 1$ matrix? $\endgroup$ – Newb May 6 '15 at 12:03
  • $\begingroup$ Every 1x1 matrix is a real number, so you were correct. I just want to draw your attention on it. $\endgroup$ – Steven Van Geluwe May 6 '15 at 12:05
  • $\begingroup$ Ah, thanks. By the way, it appears there is some controversy regarding the equivalence of $1\times 1$ matrices and scalars: see math.stackexchange.com/questions/65002/… $\endgroup$ – Newb May 6 '15 at 12:08
  • $\begingroup$ I don't know linear orders that respect multiplication on matrices, but I do know this for real numbers. Now, you can call this $1\times 1$ matrix, but what does positive matrix mean? As is pointed in the link you just gave, this is unnecessary notation overloading. There is a reason why we think of vectors and scalars separately. $\endgroup$ – Ennar May 6 '15 at 12:12
  • $\begingroup$ Maybe for logicians or programmers, but not in analytic geometry or multivariable analysis. $\endgroup$ – Steven Van Geluwe May 6 '15 at 12:13

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.