Let $(M,g)$ be a Riemannian manifold with the corresponding $LC$ connection $\nabla$. For a smooth function $f:M \to \mathbb{R}$, the Hessian of $f$ is defined as a 2-linear map on the tangent space as follows $$Hess(f)(x)(V,W)=g(\nabla_V \nabla f,W)$$ where $V,W \in T_x M$.

A linear function is a function whose Hessian is identically zero.

What is an example of a non constant linear function on each of the following manifolds

1)On $S^2$ with standard metric

2)On $\mathbb{C}P^2$ with Fubini Study metric

Furthermore, what can be said about the dimension of the space of linear functions?

  • $\begingroup$ Please show your efforts as this post has a few closure votes now. $\endgroup$ – Saad Sep 29 '18 at 1:24

If $M$ is assumed to be connected and complete, then the only such function is the zero function unless $M$ is isometric to a Riemannian product of the form $\mathbb R \times N$ for some complete Riemannian manifold $N$. In particular, if $M$ is compact (for example, $S^2$ or $\mathbb C\mathbb P^2$), then there is only the zero function.

The full proof would involve too many details to write down here, but here's a sketch. Suppose $M$ is complete and $f$ is a smooth function satisfying $\text{Hess}(f) \equiv 0$. Since $\nabla f$ usually represents a $1$-form rather than a vector field in Riemannian geometry, I prefer to write this as $$ 0 = \text{Hess}(f)(V,W) = g(\nabla _V (\text{grad}f),W), $$ for any smooth vector fields $V,W$, where $\text{grad} f = (\nabla f)^\sharp= (df)^\sharp$, with $\sharp$ denoting the musical isomorphism from $T^*M$ to $TM$.

Then for any vector field $V$, we have $$ \nabla_V |\text{grad} f|^2 = 2 g (\nabla_V (\text{grad}f ), \text{grad} f) = 2\text{Hess}(V,\text{grad}f) = 0, $$ so $|\text{grad}f|^2$ is constant. Suppose now that it is nonzero. After multiplying $f$ by a constant, we can assume that $|\text{grad} f| \equiv 1$.

Let $X$ denote the vector field $\text{grad} f$. The fact that $X$ is bounded implies that it's complete. Along any integral curve $\gamma$ of $X$, $$(f\circ\gamma)'(t) = df(\gamma'(t)) = g(X, \gamma'(t) ) = g(X,X) = 1,$$ so $f(\gamma(t)) = f(\gamma(0)) + t$. In particular, $f\colon M\to \mathbb R$ is surjective. Moreover, the fact that $\nabla X \equiv 0$ implies that $X$ is a Killing vector field, i.e., $\mathscr L_X g \equiv 0$.

Let $N = f^{-1}(0)$. Because $df$ never vanishes, this is a nonempty embedded codimension-$1$ submanifold. A standard computation in Riemannian geometry using $|\text{grad} f|\equiv 1$ shows that $f(x)$ is equal to the Riemannian distance from $x$ to $N$, and the integral curves of $X$ are geodesics that intersect $N$ orthogonally.

Let $\Phi\colon \mathbb R\times M\to M$ be the flow of $X$, and let $\phi\colon \mathbb R \times N\to M$ be the restriction of $\Phi$. I claim that $\phi$ is a diffeomorphism. Surjectivity follows from the fact that every point of $M$ can be connected to $N$ by a minimizing geodesic, which intersects $N$ orthogonally and therefore is the image of an integral curve of $X$. Injectivity is proved as follows: if $\phi(t,x) = \phi(t',x')$, then $t = f(\phi(t,x)) = f(\phi(t',x')) = t'$. Since $x\mapsto \phi(t,x)$ is a diffeomorphism with inverse $x\mapsto \phi(-t,x)$, this implies $x=x'$. Then it can be shown that $d\phi$ never vanishes, so $\phi$ is a bijective local diffeomorphism and hence a diffeomorphism.

By pulling back $g$ by the diffeomorphism $\phi$, we might as well think of $g$ as a metric on $\mathbb R\times N$, and think of $X$ as the vector field $\partial/\partial t$. Both $g$ and the product metric are invariant under the flow of $X$ (since $X$ is a Killing vector field for both metrics), and they agree along $\{0\}\times N$, so they are equal everywhere.

(If you omit the hypothesis that $M$ is complete, then the argument above can be adapted to show that $M$ is locally isometric to a Riemannian product with $\mathbb R$.)

  • $\begingroup$ Dear Prof. Lee Thank you very much for this very interesting answer. $\endgroup$ – Ali Taghavi Sep 29 '18 at 7:15
  • $\begingroup$ @AliTaghavi: You’re welcome! $\endgroup$ – Jack Lee Sep 29 '18 at 13:34

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.