EDIT: As there were no takers at all... I have added below a possible approach I came up with...

I would like to ask the following elementary but tricky question about the density of spaces of smooth functions.

We are given spaces $\mathcal{J}^{u,v}, \mathcal{K}^u, \mathcal{J}^v$ such that

$\bigcap_{(u,v) \in \mathbb{R}^2} \mathcal{J}^{u,v}(\mathbb{R}^{2l+2k}) = S(\mathbb{R}^{2l + 2k})$ and

$\bigcap_{u \in \mathbb{R}} \mathcal{K}^u(\mathbb{R}^{2l}) = S(\mathbb{R}^{2l})$,

$\bigcap_{v \in \mathbb{R}} \mathcal{J}^v(\mathbb{R}^{2k}) = S(\mathbb{R}^{2k})$.

QUESTION: The tensor product $\mathcal{K}^u \otimes \mathcal{L}^v \subset \mathcal{J}^{u,v}$ see below. Is this tensor product with the induced $C^{\infty}$ Fr\'echet topology dense?


Here $\mathcal{J}^{u,v}(\mathbb{R}^{2l + 2k}) \subset C^{\infty}(\mathbb{R}^{2l + 2k})$, $\mathcal{K}^u(\mathbb{R}^{2l}) \subset C^{\infty}(\mathbb{R}^{2l})$ and $\mathcal{L}^v(\mathbb{R}^{2k}) \subset C^{\infty}(\mathbb{R}^{2k})$ are classes of smooth functions fulfilling some estimates on partial derivates. Here $u, v$ are arbitrary real numbers.

These estimates being:

For an $f$ in the first class: For each multi-indices $\gamma_1, \gamma_2 \in \mathbb{N}_0^l$ and $\gamma_3, \gamma_4 \in \mathbb{N}_0^k$ we can find a $C > 0$ such that $|\partial_z^{\gamma_1} \partial_w^{\gamma_2} \partial_x^{\gamma_3}$ $\partial_y^{\gamma_4} f(z, w, x, y)|$ $\leq C (1 + |z| + |w|)^{u - |\gamma_1| - |\gamma_2|} (1 + |x| + |y|)^{v - |\gamma_3| - |\gamma_4|}.$

For a $g$ in the second class: For each multi-indices $\gamma_1, \gamma_2 \in \mathbb{N}_0^l$ we find $C > 0$ such that $|\partial_z^{\gamma_1} \partial_w^{\gamma_2} g(z, w)|$ $\leq C (1 + |z| + |w|)^{u - |\gamma_1| - |\gamma_2|}$.

For $h$ in the third class: For each multi-indices $\gamma_3, \gamma_4 \in \mathbb{N}_0^k$ we find $C > 0$ such that $|\partial_x^{\gamma_3} \partial_y^{\gamma_4} g(x, y)|$ $\leq C (1 + |x| + |y|)^{v - |\gamma_3| - |\gamma_4|}$.

Define the semi-norms $p_{\gamma_1, \cdots, \gamma_4}$ as follows: $p_{\gamma_1, \cdots, \gamma_4}(f) := \sup_{w \in \mathbb{R}^{2l + 2k}} \frac{|\partial_{w_1}^{\gamma_1} \cdots \partial_{w_4}^{\gamma_4} f(w)|}{(1 + |w_1| + |w_2|)^{-u + |\gamma_1| +|\gamma_2|} (1 + |w_3| + |w_4|)^{-v + |\gamma_3| + |\gamma_4|}}$.

Hence the first space $\mathcal{J}^{u,v}$ becomes a Fr\'echet space with regard to the topology induced by these seminorms. The other two spaces have semi-norm systems defined in the analogous way and are therefore Fr\'echet as well.

Observation: I) It is interesting to note that the Schwartz space $S(\mathbb{R}^{2l}) \otimes S(\mathbb{R}^{2k})$ is dense in $S(\mathbb{R}^{2l + 2k})$. Also $S(\mathbb{R}^{2l})$ is contained in $\mathcal{K}^u(\mathbb{R}^{2l})$ and a closed subset. As well as $S(\mathbb{R}^{2k})$ contained and closed in $\mathcal{L}^v$ and $S(\mathbb{R}^{2l+2k})$ in $\mathcal{J}^{u,v}$. Because these spaces are just closed and not dense this is seems not useful in resolving the density question. But for me it gives a hint that a counterexample might lie in looking at polynomials of different orders. At least a usual Stone-Weierstraß argument fails it seems.

II) The compactly supported smooth functions $C_c^{\infty}$ are contained in the Schwartz spaces and dense in $C^{\infty}$, at least in the $C^{\infty}$-topology of convergence on compact subsets. We have the inclusions:

$C_c^{\infty}(\mathbb{R}^{2l}) \otimes C_c^{\infty}(\mathbb{R}^{2k}) \subset \mathcal{K}^u \otimes \mathcal{J}^v$.

As well as $\mathcal{J}^{u,v} \subset C^{\infty}(\mathbb{R}^{2l + 2k})$.

The idea is to use that $C_c^{\infty} \otimes C_c^{\infty}$ is dense in $C_c^{\infty}(\mathbb{R}^{2l + 2k})$ combined with the density in $C^{\infty}(\mathbb{R}^{2k + 2l})$. Is there a mistake somewhere?

EDIT: Rearranged the question. Added a further observation. Small correction.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.