Consider a parameterization $r:U\rightarrow M$ for a regular surface $M$.Denote by $n$ the unit normal to the surface. I was given the following defition for the matrix $B$ that defines the second fundamental form: $$B= \begin{pmatrix} r''_{uu}\cdot n & r''_{uv}\cdot n \\ r''_{vu}\cdot n & r''_{vv}\cdot n \\ \end{pmatrix} $$ Also, I was given the following definition for the matrix $g$ that defines the Riemannian metric: $$g= \begin{pmatrix} r'_u\cdot r'_u & r'_u\cdot r'_v \\ r'_v\cdot r'_u & r'_v\cdot r'_v \\ \end{pmatrix} $$ The shape operator $S$ is $-dG$ (minus the differential of the Gauss map). It can be shown that $-dG=Bg^{-1}$.

What can be said about the eigenvalues of $B$? What can be said about the eigenvalues of $S$? Are they the same? Which of these eigenvalues are defined to be the principal curvatures of $M$? Is $\det B$ considered to be the Gauss curvature, or either $\det S$?

There is alot of information in Do Carmo's book and in other sources, but it is not consistent with the above definitions, which got me very cofused.

  • 1
    $\begingroup$ Your equation for the shape operator is actually wrong. You should have $g^{-1}$ on the left. (I know of one published book that gets this, along with other things, definitely wrong.) The eigenvalues of $S$ are the principal curvatures, and $\det S$ is the Gaussian curvature. You might want to look at my text, which is a bit more readable than doCarmo. It's freely available from the link in my profile. $\endgroup$ – Ted Shifrin Jan 9 '18 at 0:48
  • $\begingroup$ So the only case that $B$ and $S$ are the same is when $r'_u,r'_v$ are orthonormal? Is there any way, given a parameterization $r$, to "force" them into being orthonormal? $\endgroup$ – MasterJ Jan 9 '18 at 9:09
  • 1
    $\begingroup$ @MasterJ: if your coordinate vectors are orthonormal then your surface is flat, so this cannot be done in general. What you can do is find an orthonormal frame $e_1(u,v),e_2(u,v)$ which spans the tangent space at each point, but this will typically not be $r_u,r_v$ for any parametrization $r$. $\endgroup$ – Anthony Carapetis Jan 9 '18 at 9:13

I find the best way to resolve this kind of confusion is to take the perspective of abstract linear algebra: think of $B$ and $S$ as more than matrices. Really $B$ is a bilinear form on the tangent space $T_pM$, while $S$ is a linear operator $T_p M \to T_p M$. Without extra structure, it only makes sense to talk about eigenvalues and determinant of a linear operator; so the principal curvatures are the eigenvalues of $S$, and likewise $K = \det S.$

In a context where we have a preferred inner product, we often also talk about eigenvalues (or determinant, or...) of a bilinear form $b$; but what we mean by this is simply eigenvalues of the linear operator $L$ obtained from $b$ via the usual correspondence $b(X,Y) = g(L(X), Y).$ Since the second fundamental form and shape operator are related in this way (up to a sign), the eigenvalues will agree (up to sign).

| cite | improve this answer | |
  • $\begingroup$ You claim that the eigenvalues of $B$ and $S$ will agree (up to sign). But if we take for example the torus $((2+\cos u)\cos v,(2+\cos u)\sin v,\sin u)$, then using the above we get that the eigenvalues of $B$ are $\{1,(2+\cos u)\cos u \}$ and for $S$ are $\{1,\frac {\cos u}{2+\cos u}\}$, and they are not the same. $\endgroup$ – MasterJ Jan 9 '18 at 9:05
  • $\begingroup$ @MasterJ: When I say the eigenvalues of $B$, I'm talking about the eigenvalues of the bilinear form $B$ with respect to the matrix $g$. These are the eigenvalues of the matrix $g^{-1}B,$ not the matrix $B$. (The latter depend on the choice of coordinates, so they're geometrically meaningless.) $\endgroup$ – Anthony Carapetis Jan 9 '18 at 9:10

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.