5
$\begingroup$

Let $M$ be the set of smooth divergence-free vector fields $u$ on $\mathbb{R}^3$ with $$|\partial_x^{\alpha} u(x)| \leq C_{\alpha K}(1+|x|)^{-K}$$ on $\mathbb{R}^3$ for any $\alpha,K$. Further, we consider the subset of analytic functions $M_0 \subset M$. The question is now, whether for each $u \in M$ and $\varepsilon >0$, there is an $v \in M_0$, such that $$|u(x)-v(x)| < \varepsilon$$ for each $x \in \mathbb{R}^3$.

As a second question consider $M$ to be the set of smooth divergence-free vector fields $u$ on $\mathbb{R}^3$ with $$u(x+e_j) = u(x)$$ for all $x \in \mathbb{R}^3$ and $1 \leq j \leq 3$. Again, consider the subset of analytic functions $M_0 \subset M$. The question here is now again, whether for each $u \in M$ and $\varepsilon >0$, there is an $v \in M_0$, such that $$|u(x)-v(x)| < \varepsilon$$ for each $x \in \mathbb{R}^3$.

These questions are fairly similar to the Stone-Weierstrass theorem, but it seems it is not possible to derive it directly from it.

$\endgroup$
4
$\begingroup$

I believe the most natural approach to this particular question is via Fourier analysis. In the periodic case we have the series $$u(x)=\sum_{k\in\mathbb{Z}^3}u_k e^{2\pi i (k,x)},$$ and the condition $\nabla\cdot u=0$ simply means $(u_k,k)=0$. Taking $$v(x)=\sum_{|k|<K}u_k e^{2\pi i (k,x)}$$ for sufficiently large $K>0$ we can approximate $u$ as close as we want because the Fourier series of a smooth function converges uniformly and very quickly. Obviously, a polynomial $v(x)$ is analytic.

The same trick works in the first case too except the Fourier series has to be replaced by the Fourier integral, and the fact that $v$ is analytic may be a little less obvious (but still true).

$\endgroup$
1
  • 1
    $\begingroup$ In fact in problem1, one can also approximate by convolution with the Weierstrass kernel $C\exp(-||x||^2/\epsilon)$ $\endgroup$ – Pietro Majer Aug 12 '20 at 13:00
3
$\begingroup$

Here is how to derive both results from the Stone-Weierstras theorem. As you say, it's not direct, but not a long way either. Recall these simple applications of the S-W theorem, to be used in PB1 resp. in PB2

  • The algebra $A$ of rapidly decreasing real analytic functions on $\mathbb{R}^3$ is uniformly dense in the space of continuous functions vanishing at infinity on $\mathbb{R}^3$. This follows from the S-W theorem applied to the one-point compactification $\mathbb{R}^3\cup\{\infty\}$, the $3$-sphere. To check that the above algebra separates points it is sufficient to consider the function $\exp(-\|x\|^2)$ and its translates.

  • The algebra $A$ of $\mathbb{Z}^3$-periodic real analytic functions on $\mathbb{R}^3$ is uniformly dense in the space of $\mathbb{Z}^3$-periodic continuous functions on $\mathbb{R}^3$. This follows from the S-W theorem applied to the quotient $\mathbb{R}^3/\mathbb{Z}^3$, the $3$-torus. To check that the above algebra separates points it is sufficient to consider the functions $\sin(2\pi x_1)$, $\sin(2\pi x_2)$, $\sin(2\pi x_3)$ and their translates.

Now, given $u\in M$ and $\epsilon>0$, we find $w_i\in A$, such that $\|u_i-w_i\|_{\infty}\le\epsilon$ (for $i=1,..,3$). To define a divergence-free approximation $v=(v_1,v_2,v_3)\in M_0$, we may then take $v_1:=w_1$, $v_2=w_2$ and for all $(x,y,z)\in\mathbb{R}^3$ $$v_3(x,y,z):=w_3(x,y,0)-\int_0^z\big\{\partial_1w_1(x,y,s)+\partial_2w_2(x,y,s)\big\}ds.$$ It is easy to check that, both in problem 1 and 2, $v_3\in A$, that ${\rm div\, } v=0$, and that $v$ is still uniformly close to $u$.

$\endgroup$

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.