1
$\begingroup$

Fix compact intervals $X, P \subseteq \mathbb{R}$.

Let $f_n : X \times P \to \mathbb{R}$ be a sequence of $C^2$ functions converging uniformly to a $C^2$ function $f$. The first and second derivatives of $f_n$ also converge uniformly to those of $f$.

For any $p \in P$, the function $f(\cdot,p) : X \to \mathbb{R}$ has a unique zero $x^*(p)$, and $f_1(x^*(p),p) \neq 0$. Therefore there is a sequence $x_n^* (p) \to x^*(p)$ such that $f_n(x_n^*(p),p)=0$ for large $n$.

When can we also say that $x_n^*(p) \to x(p)$ uniformly in $p$? Is there a standard reference for such results?

$\endgroup$
2
  • $\begingroup$ How about $f_n(x,p)=(x-p)^2+1/n$? This converges to $f(x,p)=(x-p)^2$, but $f_n(x,p)\neq 0$ for all $x,p$ and $n$. Therefore, your therefore does not seem right, unless I am misinterpreting something. $\endgroup$ Apr 24, 2012 at 21:01
  • $\begingroup$ I think " $f_1(x,p)$ " denotes the partial derivative wrto the first variable, x. In other words, he's in the hypotheses of the implicit function thm. $\endgroup$ Apr 24, 2012 at 22:04

1 Answer 1

1
$\begingroup$

It's a quick proof by contradiction.

Assume that $ x _ n ^ * $ does not converge uniformly to $x^*$. Then, there exists $\epsilon > 0$ and, for all $n\in \mathbb{N}$, a point $p_n\in P$ such that $|x_n ^ *(p _ n)-x ^ *(p _ n)|\ge\epsilon$.

A subsequence $\big( p_{n_k}\, ,\, x_ {n_k} ^ *(p _{n_k})\big)$ converges to some $( p, y ^ * ) \in P \times X$, a zero of $f$ different from $(p,x^*(p))$, contradiction.

$\endgroup$
3
  • 1
    $\begingroup$ Note that this requires nothing about the derivatives, just uniform convergence of $f_n$ to $f$ on $X \times P$, continuity of $f$ on $X \times P$, and the fact that $f(\cdot,p)$ has a unique zero in $X$ for each $p \in P$. $\endgroup$ Apr 26, 2012 at 7:36
  • $\begingroup$ (and compactness of $X$ and $P$) $\endgroup$ Apr 26, 2012 at 7:37
  • $\begingroup$ Indeed. We may say it's an instance of a kind of general situation: "uniqueness plus compactness gives continuous dependence". $\endgroup$ Apr 26, 2012 at 7:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.