On uniform convergence of partial derivatives on a compact set

114 Views Asked by At

In Rudin's Functional Analysis, §1.46, he defines the space $C^\infty(\Omega)$, where $\Omega\subseteq \mathbb R^n$ is open. He defines a topology on this space, in part by introducing an increasing sequence of compact sets $K_i$ that covers $\Omega$. He argues that this topology turns $C^\infty(\Omega)$ into a Fréchet space by showing (among other things) that a Cauchy sequence of functions $f_i$ converges. To make the argument work, he appeals to the following fact (where $\alpha$ is a multi-index).

...each $D^\alpha f_i$ converges (uniformly on compact subsets of $\Omega$) to a function $g_\alpha$. In particular, $f_i(x) \to g_0(x)$. It is now evident that $g_0 \in C^\infty (\Omega)$, that $g_\alpha = D^\alpha g_0$, and that $f_i \to g$ in the topology of $C^\infty(\Omega)$.

I am familiar with the more basic result that if $f_n : [a, b]\to \mathbb R$ is a sequence of functions such that $f_n'$ converges uniformly to some $g$ and $f_n(x)$ converges for some $x\in[a, b]$, then $f_n$ converges to some $f$ and $f' = g$. However, I'm confused about how to generalize this to Rudin's case, where we must consider arbitrary partial derivatives and arbitrary compact sets in $\mathbb R^n$ rather than intervals in $\mathbb R$.

Is there a simple way to prove the more general result using the basic one? Or does the general case require a separate proof? What does such a proof look like?

I'm also not sure, when he says "uniformly on compact sets," if he means just those in the sequence $K_i$ or general compact sets. This seems to affect the claim Rudin is making, which adds to my confusion.

Edit. I think I was confused about the claim Rudin was making, and my confusion has been cleared up by the exposition in Trèves's Topological Vector Spaces, Distributions and Kernels, Ch. 10, Example 1.

1

There are 1 best solutions below

0
On

This is the relevant proof in the book by Rudin (Example 1.46):

Proof

Where $\Omega$ is a non-empty subset of $\mathbb{R}^k$ for some $k \in \mathbb{N}$.

To elaborate on

...each $D^\alpha f_i$ converges (uniformly on compact subsets of $\Omega$) to a function $g_\alpha$. In particular, $f_i(x) \to g_0(x)$. It is now evident that $g_0 \in C^\infty (\Omega)$, that $g_\alpha = D^\alpha g_0$, and that $f_i \to g$ in the topology of $C^\infty(\Omega)$.

Let $\alpha$ be an arbitrary multi-index. Then for all $N \in \mathbb{N}$ larger than $|\alpha|$ it follows that $(D^\alpha f_i|_{K_N})_{i \in \mathbb{N}}$ converges to a function $g_\alpha|_{K_N}$ in $C (K_N)$, because it is a Banach space and $\| D^\alpha f|_{K_N} \|_{C(K_N)} \leq p_N (f) $ for all $f \in C^\infty (\Omega)$.

Define $g_\alpha : \Omega \to \mathbb{K}$ by $g_{\alpha}(x) = \lim_{n \to \infty} D^\alpha f_n (x)$. Then the above shows that this limit always exists and that $f_i|_{K_N} \to g_\alpha |_{K_N}$ in $C(K_N)$ for all $N \geq |\alpha|$.

Now let $x \in \Omega$ arbitrary. Then there exists a $K_N$ with $N\geq |\alpha|$ so that $x$ is in the interior of $K_N$. Therefore $g_\alpha$ is continuous at $x$, because $g_\alpha|_{K_N}$ is continuous.

It is left to show that $g_\alpha = D^\alpha g_0$.

Let $h = g_0$ and $s = (g_{(1,0, \dots,0)}, g_{(0,1,0,\dots)}, \dots, g_{(0, \dots , 0,1)} )$. It suffices to show that $ h^\prime = s$. Since the desired result follows from this one by induction.

Let $x \in \Omega$ be arbitrary, then as above there exist some $N\in\mathbb{N}$ and $\delta>0$ so that $B_\delta (x) \subset K_N$ (ball with radius $\delta$ centered at $x$). For $y \in B_\delta (x) $ define $u : [0,1]\to B_\delta (x) $ by $u(t) = ty+(1-t)x$. By the fundamental theorem of calculus and the chain rule we have: $$ f_i(y)-f_i(x) = \int_0^1 (f_i\circ u)^\prime (t) dt = \int_0^1 f_i^\prime ( u (t)) (y-x) dt. $$

Therefore (using the above, the Cauchy-Schwartz inequality and $|\int | \leq \int | | $)

$$|f_i (y)-f_i(x) - f_i^\prime (x) (y-x) | \leq \int_0^1 \| f_i^\prime (u (t)) - f_i^\prime (x) \| dt \|y-x\| $$

Furthermore $$ \begin{align*} \| f_i^\prime (u (t)) - f_i^\prime (x) \| &= \| f_i^\prime (u(t) ) - s (u(t)) + s(u(t))- f_i^\prime (x) + s(x)-s(x)\| \\ &\leq \sup_{t \in [0,1]} \| s(u(t))-s(x) \| + 2 \sup_{x \in K_N}\| f^\prime_i (x) - s(x)\| \end{align*} $$ insertion into the above equation and attacking with $\lim_{i\to \infty}$ yields the desired result.