Suppose $(M,g)$ is a three dimensional smooth compact simply connected Riemannian manifold with boundary and suppose that $\Sigma$ is a smooth simply connected hypersurface in $M$ with a smooth boundary $\partial \Sigma \subset \partial M$. Let $D_X Y$ denote the Levi-Civita connection on $(M,g)$. Let $Z$ be a smooth vector field on $\Sigma$ and let $f:\Sigma \to \mathbb R$ be a smooth function on $\Sigma$. Does there exist a smooth function $\phi$ in $M$ such that $\Sigma=\{\phi=0\}$ and such that $$ (D_{\nabla \phi} \nabla \phi)\big|_{\Sigma} = f\, Z.$$ Note that the left hand side is just the restriction of $D_{\nabla \phi}\nabla \phi$ to $\Sigma$ and that $\nabla \phi$ is the gradient of $\phi$ with respect to $g$.

If the answer is no, would it make a difference if the latter condition is replaced with $$(D_{\nabla \phi} \nabla \phi)\big|_{\Sigma} - f\, Z \in \textrm{Span}\{(\nabla \phi)|_{\Sigma}\}$$

  • 1
    $\begingroup$ The Whitney extension theorem could be relevant. (Also, do you need $f$ at all?) $\endgroup$
    – Leo Moos
    Commented Jul 12, 2021 at 20:13
  • $\begingroup$ Indeed $f$ is redundant :) $\endgroup$
    – Ali
    Commented Jul 12, 2021 at 20:18
  • $\begingroup$ Why is this question tagged with lorentzian and semi-riemannian geometry? All the objects in sight seems to be Riemannian. Is there some connection to semi-Riemannian stuff? If so, can you make it explicit? $\endgroup$ Commented Jul 13, 2021 at 0:26
  • $\begingroup$ Additionally, when $Z$ is said to be a smooth vector field on $\Sigma$, do you mean $Z$ is tangent to $\Sigma$, or a vector field of $M$ restricted to $\Sigma$? And when you say "restriction of $D_{\nabla \phi} \nabla \phi$ to $\Sigma$" do you mean the projection, to $T\Sigma$ or just the restriction? $\endgroup$ Commented Jul 13, 2021 at 0:28
  • $\begingroup$ $Z$ can be thought of as a vector field on $M$ that when restricted to $\Sigma$, is tangent to $\Sigma$. By the restriction of $D_{\nabla \phi}\nabla \phi$ to $\Sigma$ I mean the restriction, so the equation should be considered on $\Sigma$ only. $\endgroup$
    – Ali
    Commented Jul 13, 2021 at 0:55

1 Answer 1


Not a full answer, but there definitely should be some integrability constraint.

Take the simplest case where $M = \mathbb{R}^3$ (the boundary is unimportant for the discussion here) and $\Sigma$ is the $x$-$y$ plane.

If $\phi$ is a defining function of $\Sigma$, then restricted to $\Sigma$ we have $\nabla \phi|_{\Sigma} = \eta \partial_z$ for some function $\eta: \Sigma \to \mathbb{R}$.

Symmetry of the Hessian requires then $$ \nabla^2 \phi|_{\Sigma} = \begin{pmatrix} 0 & 0 & \partial_x \eta \\ 0 & 0 & \partial_y \eta \\ \partial_x \eta & \partial_y \eta & * \end{pmatrix} $$ This shows that $$ D_{\nabla\phi} \nabla\phi = \eta (\partial_x \eta, \partial_y \eta, *) $$

So a necessary condition for your equation to hold (setting $f \equiv 1$ since as you agreed it is unimportant) in the flat case is that $Z$ is the gradient of some scalar function on $\Sigma$, and that allowing there to be normal components (as in your modified question) has no impact.

The conclusion is unchanged in the Lorentzian setup, if you require $\Sigma$ to be non-degenerate. In the case where $\Sigma$ is null, however, since $\nabla\phi$ now lies in the tangent space of $\Sigma$ and $\Sigma$ is two dimensional, it may in fact be possible for the second variant to hold (but I don't have the energy to run through the analysis now).


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.