3
$\begingroup$

Suppose $ g_i: [0, 1] \to \Bbb R$, $i\in\Bbb N$, are $C^1$ functions and that there is some $c > 0$ such that for every $0 < \epsilon < c$, the functions $$ s(\epsilon)_i := \sum_{k=0}^i {\epsilon}^k g_k $$ converge uniformly to a $C^1$ function $s(\epsilon)$.

As $\epsilon \to 0$ does

i) $s(\epsilon) \to s(0)$ uniformly?

ii) $s(\epsilon)’ \to s(0)’$ uniformly?

$\endgroup$
5
  • $\begingroup$ Is the formulation of the problem correct? The answer seems trivial yes since $|g_k|$ are bounded and $\epsilon^k \to 0$ for $\epsilon \to 0$. Am I missing something? $\endgroup$ Commented Feb 25, 2019 at 21:27
  • $\begingroup$ The g_k aren’t bounded uniformly is the problem I think. Also a bunch of other stuff like the rate of convergence not being uniform in i, and the g_k being positive/negative. $\endgroup$ Commented Feb 26, 2019 at 1:04
  • $\begingroup$ Do you have this result for constant functions $g_k$? Of course uniform convergence then simply is convergence in $\mathbb{R}$. $\endgroup$ Commented Feb 27, 2019 at 11:29
  • $\begingroup$ Yes, I have it haha. Also for any strictly positive or strictly negative g_k. $\endgroup$ Commented Feb 27, 2019 at 13:06
  • 1
    $\begingroup$ You can look at $s(\varepsilon)=\sum\limits_{k=0}^\infty g_k \varepsilon^k$ as a power series with coefficents in the Banach space of continuous function. As in the scalar case, as soon as the radius of convergence is positive, this function is continuously differentiable on the open disc where it converges. This answers i) and ii) if $s(\varepsilon)'$ means the derivative with respect to $\varepsilon$. Or do you mean the power series with coefficients $g'$ in ii)? $\endgroup$ Commented Mar 31, 2020 at 12:12

1 Answer 1

0
$\begingroup$

Proof of i): First let $(B,\|.\|) := (C^0,\|.\|_\infty)$. For the following reasoning the Banach space $(B,\|.\|)$ may be arbitrary. Let $(g_k)_{k \in N_0}$ be any sequence in $B$, such that for some $\epsilon > 0$ and $s(\epsilon)_n := \sum_{k=0}^n \epsilon^k g_k$ $s(\epsilon) := \lim_{n \to \infty} s(\epsilon)_n$ exists. Put $u_k := s(\epsilon)_k - s(\epsilon)_{k-1} = \epsilon^k g_k$. Since $s(\epsilon)_k$ is converging, necessarily $\lim_{k \to \infty} u_k = 0$, in particular $U := \sup_{k \in N_0} \|u_k\| < \infty$. Now for $0 < \delta < \epsilon$ $s(\delta)_n = \sum_{k=0}^n \left(\frac{\delta}{\epsilon}\right)^k u_k$, $\|s(\delta)_m - s(\delta)_n\| \leq \frac{(\delta/\epsilon)^{m+1}}{1-\delta/\epsilon}U$ for $0 \leq m < n$, thus $s(\delta) = \lim_{n \to \infty} s(\delta)_n$ exists with $\|s(\delta)_0 - s(\delta)\| \leq \frac{\delta/\epsilon}{1-\delta/\epsilon}U$. But $s(\delta)_0$ is nothing else but $s(0)$.

$\endgroup$
2
  • $\begingroup$ Sorry, can you elaborate a bit on how (i) follows? $\endgroup$ Commented Feb 28, 2019 at 1:25
  • $\begingroup$ @James Baxter: I've made the proof more explicit. ii) seems to be much more interesting, since $C^1$ is a dense subspace of $C^0$ and we still have convergence in $C^0$, It would be nice if you could present a proof of ii) now. $\endgroup$ Commented Feb 28, 2019 at 10:18

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .