1
$\begingroup$

Definitions:

We say a smooth Riemannian metric on $\mathbb R^n$ is smoothly equivalent to Lebesgue measure if the Radon Nikodym derivative of the associated Riemannian volume measure with respect to Lebesgue measure is smooth.

Given a smooth Riemannian metric $g$ on $\mathbb R^n$, and a point $x \in \mathbb R^n$, denote by $\gamma_v$ the unit speed geodesic starting at $x$ in the direction of $v$.

For a scalar function $F: \mathbb R^n \to \mathbb R$, denote by $\partial_v {F|}_g$ the scalar function given by $$ \partial_v {F|}_g (x) := \lim_{t \to 0} \frac{F(\gamma_v (t)) - F(x)}{t} $$

Question: Let $f: \mathbb R^n \to \mathbb R$, $h: \mathbb R^n \to \mathbb R$ be $C^1$ functions. For any $\varepsilon > 0$, does there always exist a smooth Riemannian metric $g$ that is smoothly equivalent to Lebesgue measure such that $$ \sup_{v \in S^{n-1}} \left\|\partial_v {h|}_g - \frac{\partial f}{\partial v}\right\|_{C^0} < \varepsilon $$ for each $1 \leq i \leq n$?

Note: Here $\|\cdot\|_{C^0}$ denotes the sup norm, and $S^{n-1}$ the unit sphere in $\mathbb R^n$ under the Euclidean metric.

$\endgroup$
4
  • $\begingroup$ Defining the derivative of a function does not require a metric; recall for example the definition of tangent vectors to a manifold. The scalar function $\partial_v h$ that you define is just $Dh(v)$, it seems to me. $\endgroup$
    – Leo Moos
    Jun 14, 2021 at 9:05
  • $\begingroup$ The difference I think is the requirement that the geodesic be unit speed with respect to $g$. $\endgroup$
    – Nate River
    Jun 14, 2021 at 9:08
  • $\begingroup$ I must confess I don't follow your point. Any curve $\gamma$ with $\gamma(0) = x$ and $\gamma'(0) = v$ will produce the same derivative, namely $Dh(v)$. $\endgroup$
    – Leo Moos
    Jun 14, 2021 at 9:09
  • $\begingroup$ This curve has $\gamma’(0) = \frac{v}{||v||_g}$ I believe.. $\endgroup$
    – Nate River
    Jun 14, 2021 at 9:14

1 Answer 1

1
$\begingroup$

Edit. For an even simpler example, let $f \in C^1(\mathbf{R}^n)$ be non-constant and $h = 0$. Such a metric cannot exist when the constant is so small that $\epsilon < \lvert Df \rvert_\infty$.

Let $f \in C^1(\mathbf{R}^n)$ be an arbitrary, non-constant function, and $h = -f$. Let moreover the metric $g$ be arbitrary. Then for all euclidean unit vectors $v \in \mathbf{R}^n$, that is vectors with $\sum_i (v^i)^2 = 1$, one has $D_v h/g(v,v)^{1/2} - D_v f = (1/g(v,v)^{1/2} + 1) D_v h$. Taking absolute values, the supremum of this taken over points $x \in \mathbf{R}^n$ is at least $\lvert D_v h \rvert_{\infty} = \lvert D_v f \rvert_{\infty}$. In particular if $\epsilon < \sup_{v \in \mathbf{S}^{n-1}} \lvert D_v f \rvert_{\infty}$ then the inequality cannot hold, regardless of the metric.

$\endgroup$
1
  • $\begingroup$ Ah.. I guess changing the metric can only alter the magnitude of the derivative, not the direction, so if they happen to be of opposite signs then we can’t do anything. $\endgroup$
    – Nate River
    Jun 14, 2021 at 10:11

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.