MathOverflow is a question and answer site for professional mathematicians. It's 100% free, no registration required.

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

Let $u(t) = \Sigma_{k=1}^n c_k e^{i \lambda_k t} (c_k \in \mathbb C, \lambda_k \in \mathbb R) $ be an exponential polynomial of order $n$ with purely imaginary exponents. We can assume that the exponents are $\delta$-separated.

Define $E_n$ to be the collection of all exponential polynomial of order $n$. i.e.,

$$ E_n:= \{ u : u(t) = \sum_{k=1}^n c_k e^{i \lambda_k t}, c_k \in \mathbb C, \lambda_k \in \mathbb R, and |\lambda_i - \lambda_j| > \delta\\ when\\ i \neq j \}. $$

Of course $u$ is an almost periodic function i.e., given $\epsilon >0$ there exists $T_\epsilon$ such that every interval of length $T_\epsilon$ contains an almost period of $u$. i.e., $\forall x \in \mathbb R,$ $\exists \tau \in (x,x+T_\epsilon)$ such that

$$ \sup_{t\in \mathbb R} |u(t)-u(t+\tau)| < \epsilon$$

Is it possible to find a bound on the $T_\epsilon$ for such an exponential polynomial, I added the separation condition hoping that it would lead to an affirmative answer.

I have this suspicion that the almost period may have to do something with the number theoretical properties of the set of exponents. After all if all the exponents are in a lattice then the function is periodic.

and may I even dare hope to find a bound which will work for the entire class? Some nice subclass maybe ?

If that is too much to ask then what is a good question to ask ?

Has some one studied related questions or variations of it ? I would be glad to know.

share|cite|improve this question
I added the analytic number theory tag because I have this suspicion that the almost period may have to do something with the number theoretical properties of the set of exponents. – Vagabond Sep 10 '10 at 13:04
up vote 1 down vote accepted

You are basically interested in what is called $\epsilon$-dual Characters.

For a set $\Lambda \subset \R^d$ we define

$$\Lambda^\epsilon := \{ t \in \R^d | \left| e^{2 \pi x \cdot t} -1 \right| < \epsilon \, \forall x \in \Lambda \}\,.$$

In your case $\Lambda := \{ \lambda_1, .., \lambda_n \}$ and for all $t \in \Lambda^\epsilon$ you get

$$\| T_tu - u\|_{\infty} < (|c_1|+...+|c_n|) \epsilon \,.$$

Here is a short review of what is known about $\Lambda^\epsilon$:

1) If $\Lambda \subset \R^d$ is finite, then $\Lambda^\epsilon$ is relatively dense for all $0< \epsilon <2$. , that is $\Lambda^\epsilon +K_\epsilon = \R^d$ for some compact $K_\epsilon$. How to finding this $K$ is exactly waht you need. I will address this question below (*).

2) If $\Lambda \subset \R^d$ is relatively dense and $\Lambda-\Lambda := \{ x-y | x,y \in \R^d \}$ is uniformly discrete (i.e. $|z_1-z_2| > \delta$, for all $z_1, z_2 \in \Lambda- \Lambda$) then $\Lambda^\epsilon$ is relatively dense for all $0< \epsilon <2$. Such a set is called a {\bf Meyer set}.

(*) How do we actually find $K$ so that $\Lambda^\epsilon +K= \R^d$? This is covered by Meyer in the book mentioned below, when he studies the connection between $\Lambda^\epsilon $ relatively dense and $\Lambda$ being harmnious.

I will explain directly what happens in your case, since in general things are a little more complicated.

First lets observe that if $t= \frac{2pi n}{ \lambda}\, n \in \Z$ then $e^{2 \pi \lambda \cdot t} =1 $, which is the reson the problem is easy for latices.

If we only have 2 $\lambda$'s, here is what we need to do: First if $\lambda_1/\lambda_2$ is rational, then the intersection of the latices $ \frac{2pi}{ \lambda_1}\Z \cap \frac{2pi}{ \lambda_2}\Z$ is non-empty and any $t$ in here will do.

If $\lambda_1/\lambda_2$ is irrational, then the latices $ \frac{2pi}{ \lambda_1}\Z$ and $\frac{2pi}{ \lambda_2}\Z$ come arbitrarily close within a fixed gap, and that will do.

More exactly, fix a $\alpha$ so that if $|x-y| < \alpha $ then $ \left| e^{2 \pi x \cdot \lambda_i} -e^{2 \pi y \cdot \lambda_i} \right| < \epsilon$, and then the dirichclet theorem tells you that there exists number $N$ so that you can find integers $m,n$, with $m$ in any interval of length $N$ so that:

$$ | m\frac{2pi}{ \lambda_1} - n\frac{2pi}{ \lambda_2} | < \alpha \,.$$

Then $m\frac{2pi}{ \lambda_1}$.

If you have more than two $\lambda$'s, exactly as above all you have to do is prove that there exists a number $N$ so that within any interval of lenght $N$ there exists a $t$ so that for all $i$ we have:

$$ | t- n_i\frac{2pi}{ \lambda_i} | < \alpha \,.$$

This is what Meyer calls a Harmnious set, and a stronger version of Dirischlet Theorem proves it.

I recommend to you the following book(s):

-Y. Meyer, Algebraic numbers and Harmonic Analysis.

-R. V. Moody - Meyer Sets and their duals.

share|cite|improve this answer

Unfortunately, not. Take $\lambda_1=\delta$, $\lambda_2=2\delta+\gamma$ with very small $\gamma$ that is not a rational multiple of $\delta$. Then we can choose $t$ such that $e^{\lambda_1 t}\approx 1$ and $e^{\lambda_2t}\approx -1$. Now, there will be no $\varepsilon$-almost period in $[t,t+0.1\gamma^{-1}]$ (for a polynomial, an almost-period should be an almost-period of each exponent).

share|cite|improve this answer
But then the question remains given some fixed exponents $ \{\lambda_1 , \lambda_2 , \dots, \lambda_k\}$ and some $\epsilon$ how does one go about finding a bound on $T_\epsilon$ ? – Vagabond Sep 10 '10 at 13:52
Do you mean absolutely nothing can be done at all ? Can we not put some kind of condition on the set of exponents so that we can say something reasonable ? – Vagabond Sep 10 '10 at 14:31
We can certainly put a condition on the possible rate of decay of non-zero linear combinations of $\lambda_j$ with integer coefficients (which will hold for "generic" choice of $\lambda_j$), or express $T_\varepsilon$ in terms of this rate, but will that be of any use to you? – fedja Sep 10 '10 at 17:34
If I have understood correctly then given exponents $\{\lambda_1, \lambda_2,\dots, \lambda_k \}$ and an $\epsilon > 0$ when we try to find $T_\epsilon$ what we need to study is simultaneous solution of $|e^{2 \pi i \lambda_j}-1| < \epsilon$. The solution set should turn out to be union of infinitely many intervals and we are looking for a bound on the maximum distance between two consecutive intervals ? Now how does $T_\epsilon$ varies with $\epsilon$ are there jumps' change in the rate' etc ? This surely seems like a number theory question to me. It must have been studied somewhere no ? – Vagabond Sep 10 '10 at 17:56
@fedja absolutely anything which says something affirmative will restore my faith that there is some order/hope in this world. – Vagabond Sep 10 '10 at 18:02

Here is a first attempt:

For each $\lambda_k$ find a rational number $\frac{p_k}{q_k}$ such that $\lambda_k\approx\frac{p_k}{q_k}$.

Then an approximation pestimistic bound for $T_{\epsilon}$ would be $(q_1\ldots q_n)2\pi$ since $\frac{p_k}{q_k}(q_1\ldots q_n)2\pi$ is a multiple of $2\pi$ for every $k$.

share|cite|improve this answer
Not really. $T_\varepsilon$ is not the bound for the least almost-period (that one is easy) but the bound for the length of the longest interval between almost-periods. – fedja Sep 10 '10 at 13:34

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.