Let $$ f(x)=\sin(x)\sqrt{1+\cos^{2}(x)+\cos^{4}(x)}. $$ In my study of almost commuting unitary matrices, $U$ and $V$, I have need for a bound like $$ \left\Vert \tilde{f}(V)U-U\tilde{f}(V)\right\Vert \leq C\left\Vert VU-UV\right\Vert , $$ where $\tilde{f}(e^{\pi ix})=f(x)$ so my functional calculus makes sense. (Matrix function to any applied listeners). The norm is the operator norm.

This is all in the context spin Chern numbers and the Pfaffian-Bott index, as in my paper with Hastings, Topological insulators and $C^{*}$- algebras: Theory and numerical practice'' which came out in 2011. I want precise bounds on how small a commutator I need to ensure a derived matrix remains invertible.

One way to estimate $C$ is to writing $f$ and $f^{\prime}$ as Fourier series and working term by term. Let $g=f^{\prime},$ so $$ g(x)=\frac{3\cos^{5}(x)}{\sqrt{1+\cos^{2}(x)+\cos^{4}(x)}}. $$ I need to know, or get a good estimate on $\left\Vert \hat{g}\right\Vert _{1}$, meaning the $\ell^{1}$-norm of the fourier series of $g(x)$.

What is the $\ell^{1}$-norm of the Fourier series of $\frac{3\cos^{5}(x)}{\sqrt{1+\cos^{2}(x)+\cos^{4}(x)}}$, or what is a good upper bound?

I can prove this is less than $3\pi$ but I want a tighter estimate. Perhaps someone has seen this before. Is there a stategy to bound this I should follow? Better still, what is the Fourier series for $g$?

Any hints on finding $C$ by another route will be welcome.

A little more about $f$: The Bott index and its relatives require three functions with $f^2 + g^2 + h^2 = 1$ and $gh=0$ and these are to be continuous, real valued and periodic. And non-trivial, so $h=0$ is not allowed. For theoretical work these are all equally valid, but to use these in index studies of finite systems we need to pick carefully to control a bunch of commutators. This $f$ is designed to be similar to $\sin$ but will big "flat spots." Notice $$ \left(f(x)+\cos^{3}(x)\right)^{2}=1. $$


1 Answer 1


Let $\|\cdot \|_F $ denote the $\ell^1$ norm of the Fourier series of a function on $[0,2\pi]$. This is a Banach algebra norm on functions whose Fourier series converge absolutely. Note that $$1 + \cos^2(x) + \cos^4(x) = \frac{15}{8} + \frac{1}{2} e^{2ix} + \frac{1}{2} e^{-2ix} + \frac{1}{16} e^{4ix} + \frac{1}{16} e^{-4ix}$$ Write this as $(15/8)(1 + W(x))$, where $\|W(x)\|_F = 3/5$. So $$g(x) = 3 (15/8)^{-1/2} \cos^5(x) (1+W(x))^{-1/2}= 3 (15/8)^{-1/2} \cos^5(x)\sum_{k=0}^\infty {{-1/2} \choose k} W(x)^k$$ Thus $$\|g\|_F \le 3 (15/8)^{-1/2} \sum_{k=0}^\infty \left| {{-1/2} \choose k}\right| (3/5)^k = 2 \sqrt{3}$$ Moreover, by explicitly evaluating the norm of a partial sum of the series and bounding the norm of the remainder you can get arbitrarily good approximations to the actual value of $\|g\|_F$.

  • $\begingroup$ I get so uncomfortable working in a norm that is not a $C^*$-norm. I really appreciate the help. I worked harder for a constant that was twice the size. $\endgroup$ Sep 16, 2012 at 21:19

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.