# Determinant of the Hessian, what is it good for?

For functions on $\mathbb{R}^2$, it lets you classify extreme points. When function is log-likelihood, it gives volume of the region where maximum likelihood estimates tend to fall. Are there other places it comes up?

-
Browsing the Wikipedia article leads you to en.wikipedia.org/wiki/… . – Qiaochu Yuan Sep 15 '10 at 18:55
I believe the determinant of the Hessian of f(x,y) gives you the Gaussian curvature of the graph of f. – John D. Cook Sep 15 '10 at 19:01

Yep. The condition that the Hessian is nonsingular at every critical point means that the function in question is suitably "nice," and is called a Morse function. Morse functions are useful because their inverse images $f^{-1}((-\infty, a])$ behave very nicely. Basically, as $a$ varies, you get things that are (up to homotopy equivalence) obtained by attaching cells when $a$ passes through a critical value (the type of cell being determined by the type of the critical point). So since every compact manifold admits a Morse function (the "typical" function is, in fact, Morse--I believe this is a standard transversality idea in differential topology), it follows that every compact manifold is homotopy equivalent to a CW complex.
Precisely, let $S$ be a surface, $p\in S$ a point and $h: S \longrightarrow \mathbb{R}$ be the height function of $S$ relative to $T_p S$; that is, $h(q) = (q-p)\cdot N(p)$, where $N(p)$ is the normal vector to the surface $S$ at the point $p$. Then the Hessian of the function $h$ at $p$ is the second fundamental form of $S$ at $p$.