Let $f:X \to X$ ($X = R^d$ for some $d$) be a mapping such that $f^n (x) \to x^\ast$ for all $x \in X$ as $n \to \infty$ ($x^\ast$ is unique). Can we say anything about the spectral structure of the gradient matrix $\nabla f (x^\ast)$ ? Do we know that the spectral or the operator norm of this matrix is less than one?

  • 3
    $\begingroup$ Of course, a priori, there is no reason why $f$ should be differentiable at $x^*$ $\endgroup$ – Dick Palais Feb 9 '11 at 5:43

I think this is a counter-example to what you are asking. Choose a smooth function $a : R \to R$ with $a(0) = 0$, $a(-x) = a(x)$, $a(x) = x^2$ near $x = 0$, $a(x)$ monotonically increasing for $x > 0$, and $a(x)$ everywhere less than $1$. Then if we let $f(x) := (1 - a(x)) x$ we have $f'(0) = 1$, but $f^n(x)$ converges to $0$ for all $x$. This is because $f^n(x)$ is monotonic and bounded, and so approaches a limit, which must be a fixed point and so can only be $0$.

  • $\begingroup$ An example of such an $f$ is $\frac{x}{1+x^2}$. But what if we demand that $f_n(x)$ converges to $x^*$ uniformly? $\endgroup$ – Vít Tuček Feb 9 '11 at 13:32
  • $\begingroup$ @Vít Tuček: for instance the iterates of your $f$ do converge uniformly to $0$; $\|f^{n}\|_\infty=f^{n}(1)=o(1)$. $\endgroup$ – Pietro Majer Oct 27 '13 at 8:47

Your Answer

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

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