High-dimensional version of Borel's Lemma as a corollary of Whitney's extension theorem

216 Views Asked by At

A version of Borel's Lemma states that, given a sequence $(f_0, f_1, f_2, \ldots)$ of functions in $C^\infty(\mathbb{R}^{n-1})$, there always exists a function $F$ in $C^\infty(\mathbb{R}^{n})$ such that $$ \frac{\partial^k F}{\partial t^k}(0,\cdot) = f_k \quad \forall \ k \geq 0.$$ Note that, clearly, $\{ 0 \} \times \mathbb{R}^{n-1}$ is a closed subset of $\mathbb{R}^{n}$.

On the other hand, the much more general Whitney extension theorem deals with arbitrary closed subsets $X$ of $\mathbb{R}^n$. In its $C^\infty$ formulation, it gives necessary and sufficient conditions for the existence of a function $f$ in $C^\infty(\mathbb{R}^n)$ whose all partial derivatives take prescribed values $f^\alpha$ on $X$, with $\alpha \in \mathbb{N}_0^n$ an arbitrary multi-index. According to the theorem, such an extension $f$ of the candidate derivative data $\{f^\alpha\}$ exists if and only if the $\{f^\alpha\}$ (are continuous and) satisfy the same estimates, restricted to the set $X$, which the collection $\{\partial^\alpha g\}$ of partial derivatives of any $g \in C^\infty(\mathbb{R}^n)$ would satisfy as a result of Taylor's theorem. Let me call a collection $\{f^\alpha\}$, of functions defined and continuous on $X$ and satisfying such estimates, a $C^\infty$-Whitney field on $X$. More precisely, what is required of the collection is that, for any $m \geq 0$ and any compact subset $K$ of $X$, the "remainders" $$f^\alpha(\boldsymbol{x})-\sum_{|\beta|\le m}\frac{f^{\alpha+\beta}(\boldsymbol{y})}{\beta!}\cdot(\boldsymbol{x}-\boldsymbol{y})^\beta,\qquad \boldsymbol{x},\boldsymbol{y}\in K \tag I$$ be $o\big(||\boldsymbol{x}-\boldsymbol{y}||^{m}\big)$ uniformly as $\boldsymbol{x},\boldsymbol{y} \in K,$ $||\boldsymbol{x}-\boldsymbol{y}||\to 0$.

It is obvious (and frequently pointed out in the literature) that if $X = \{a\}$ then Whitney's extension theorem states that given any collection $\{c^\alpha\}_{\alpha \in \mathbb{N}_0^n}$ of real constants we can find a $f \in C^\infty(\mathbb{R}^n)$ such that $\partial^\alpha f (a) = c^{\alpha}$ for all multi-indices $\alpha$. In particular, if $n=1$ we get a particular case of the version of Borel's Lemma I mentioned in the first paragraph.

Now, to my question: How does one show that Borel's Lemma as above is a simple corollary of Whitney's extension theorem when $n>1$?

Let me try to show the origin of my confusion by illustrating the case $n=2$. I.e., we are given a sequence $(f_0, f_1, f_2, \ldots)$ of functions in $C^\infty(\mathbb{R})$, which we ultimately want to promote to be the derivatives of all orders and transverse to $X = \{ 0\} \times \mathbb{R}$ of an $F \in C^\infty(\mathbb{R}^2)$. Given that we'd like to use the Whitney theorem to find $F$, we must first complete the collection $$\{ f^{(0,0)} := f_0 , \, f^{(1,0)} := f_1, \, f^{(2,0)} := f_2, \ldots \}$$ (having identified the $f_i$'s with functions on $X$ in the obvious way). In view of what we want to achieve, the way to do so is by setting, for any $\alpha=(i,j) \in \mathbb{N}_0^2$, $$ f^{\alpha} = f_i^{(j)}, $$ where the superscripts denotes the ordinary $j$-th order derivative. We must now verify that the thus defined collection $\{ f^\alpha\}$ is a $C^\infty$-Whitney field on $X$, for then Borel's Lemma would immediately follow.

So let me test (I) for $m=1$, and with $\alpha=(0,1)$. Then $f^\alpha(0,x) = f_0'(x)$, while since the multi-indices of length less than or equal to $1$ are $(0,0), (1,0), (0,1)$, we have, setting $\boldsymbol{x}=(0,x)$ and $\boldsymbol{y}=(0,y)$: $$ f^\alpha(\boldsymbol{x}) - \sum_{|\beta|\le 1}\frac{f^{\alpha+\beta}(\boldsymbol{y})}{\beta!}\cdot(\boldsymbol{x}-\boldsymbol{y})^\beta = f_0'(x) - \sum_{k+l\le 1}\frac{f^{(k,l+1)}(0,y)}{k! l!}\cdot(x-y)^l \\= f_0'(x) - f^{(0,1)}(0,y) - f^{(1,1)}(0,y) - f^{(0,2)}(0,y)(x-y) \\ = f_0'(x) - f_0'(y) - f_1'(y) - f_0''(y)(x-y). $$ Now, $f_0'(x) - f_0'(y) - f_0''(y)(x-y)$ is clearly $o(||\boldsymbol{x} - \boldsymbol{y}||^m) = o(|x-y|)$ uniformly on compact sets as $|x-y| \to 0$, because $f_0'$ is smooth. My problem is with the $- f_1'(y)$ term above. As $|x-y| \to 0$ it is just $-f_1'(x)$, so unless $f_1$ is zero we seem to have a problem! Also, since $f_1$ was prescribed completely independently from the other $f_i$'s, we don't seem to have estimates linking it to $f_0$ in a way which could obviously save the day.

I have been stuck on this quite a while. Clearly I must have either messed up my calculations, or have badly misunderstood the statement of the (I'd guess) Whitney extension theorem. Either way, I'd really appreciate your help.

1

There are 1 best solutions below

1
On BEST ANSWER

The answer is simple.

For $\mathbf{x} = (0,x)$ and $\mathbf{y} = (0,y)$, and $\beta = (1,0)$, the term

$$ (\mathbf{x} - \mathbf{y})^{\beta} = (0-0)^1 \cdot(x-y)^0 = 0 $$

and not $1$, which you assumed. So when you are doing your expansion, the $f'_1(y)$ term should be multiplied by $0$ and disappear.