MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Floer (in "The unregularized gradient flow of the symplectic action") defined a dense subspace of $C^\infty$ with the structure of a Banach space, with norm $\Vert f \Vert = \sum_{k \ge 0} \epsilon_k \Vert f \Vert_{C^k}$ for some constants $\epsilon_k$.

Question: How do the constants $\epsilon_k$ in Floer's norm behave as $k \to \infty$? He just shows that some constants exist for which the resulting space is dense in $C^\infty$. Do the constants go to infinity, to zero, or neither?

Subquestion: Is Floer's space closed under products of functions?

share|cite|improve this question
In case $\epsilon_k\sim (k!)^s$ this essentially gives Gevrey classes, see my other answer. More general (log-convex) sequences give rise to the so called Mandelbrojt classes. Not that Mandelbrojt, an older one. – Piero D'Ancona Aug 20 '10 at 15:25
er.. I mean $\epsilon_k \sim (k!)^{-s}$ – Piero D'Ancona Aug 20 '10 at 15:37

Floer chooses $\epsilon_k$ so that this $\epsilon$ space is dense in $L^2$ (this should be equivalent to dense in $C^\infty$, since the latter is dense in $L^2$, and Floer's $\epsilon$ space sits in $C^\infty$).

To show that this subspace is dense in $L^2$, it suffices to show that one can approximate indicator functions.
For this, he needs the $\epsilon_k$ to go to zero very fast. His explicit construction in Lemma 5.1 (of the "unregularized gradient flow" paper) is to take a fixed cut-off function $\beta$, and approximate the characteristic function of a rectangle. The approximation to a characteristic function is going to be a product of terms that behave like $\beta(x/\delta)$ (with a better approximation as $\delta \rightarrow 0$). Thus, the behaviour of the $\epsilon$-norm is going to be roughly: \[ \sum_{k=0}^\infty \epsilon_k \delta^{-k} a_k, \] where $a_k = \sup | D^k \beta |$. We need this to converge for each $\delta > 0$. Floer takes $\epsilon_k = (a_k k^k)^{-1}$. In particular, then, these constants are going to $0$. Following this argument, it seems we can take the $\epsilon_k$ to be on the order of $1/k!$.

Note that if the sequence $\epsilon_k$ is not summable, we expect the space to be very small. In particular, consider this norm on a compact interval, say $[-\pi, \pi]$. Then, cos(x) is not in the space.

The Floer $\epsilon$ space forms a Banach algebra if $\epsilon_k$ decays faster than $1/k!$.
Then, \[ \sum \epsilon_k |D^{(k)}(fg)| \le \sum_{k=0}^\infty \sum_{l=0}^{k} \epsilon_k | D^{(l)} f| |D^{(k-l)} g | \binom{k}{l} = \sum_{l=0}^\infty |D^{(l)} f| \sum_{p=0}^{\infty} \binom{l+p}{p} \epsilon_{p+l} | D^{(p)}g| \] When $\epsilon_{p+l} \binom{l+p}{p} \le \epsilon_p \epsilon_l$, we are then in business. In particular, this works for Floer's original construction.

share|cite|improve this answer
Sam, do you have a lower bound on $a_k$? – Tim Perutz Nov 3 '10 at 14:54
If we follow Floer's definition of the cut-off functions, we can crudely bound them below by 1. To recall, Floer's definition is that $\beta : \mathbb{R} \rightarrow [0,1]$ with $\beta(x) = 0$ near $0$ and $\beta(x) = 1$ near 1, $\beta'(x) \ge 0$. He then uses products of functions of the form $\beta(|x-y|/\delta)$ to approximate characteristic functions of rectangles. Actually, we might be able to do better and get something like $1/d^k$, where d is the diameter of the support of $\beta$. (The exact bound doesn't really matter because of the $1/\delta^k$ term that shows up.) – Sam Lisi Nov 3 '10 at 23:18
Ah yes - by the mean-value theorem and induction. (For the uniform lower bound, I mean.) Missed that. Thanks! – Tim Perutz Nov 4 '10 at 13:46

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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