5
$\begingroup$

Let $F:[0,1]\to\mathbb{R}$ be a measurable function such that $$ \mu_n(F)=\int_0^1F(t)t^ndt\sim\frac1{(\ln n)^n}\quad\mbox{as}\quad n\to\infty. $$ More precisely, $$ 0<c<|\mu_n(F)|(\ln n)^n<C<\infty,\quad\forall n\in\mathbb{N}. $$ Note that in this case the series $$ \sum_{n=0}^\infty\mu_n(F)z^n $$ represents an entire function. The generating function $$ \sum_{n=0}^\infty\frac{\mu_n(F)z^n}{n!} $$ is an entire function of minimal exponential type. I am not aware of any explicit description of such a function, so this is already a sign.

Question: Can we find an explicitly given such function $F$? Here by "explicitly given" I mean in a "closed form", e.g., no series.

I think that by convexity arguments it can be shown that $F$ cannot be non-negative (non-positive). In fact, I do not need $F$ be a function; $F(x)dx$ may be a signed measure on $[0,1]$, just given explicitly.

Thank you.

$\endgroup$
1
  • 3
    $\begingroup$ Indeed it is clear that these can't be the moments of a positive measure since then $\int t^n \, d\rho(t) \gtrsim c^n$ for some $c>0$ (unless $\rho = \delta_0)$. $\endgroup$ Apr 26, 2019 at 4:02

1 Answer 1

5
$\begingroup$

$\newcommand{\R}{\mathbb{R}} \newcommand{\si}{\sigma} \newcommand{\supp}{\operatorname{\mathrm supp}} \newcommand{\cch}{\operatorname{\mathrm cch}} $ If $F\in L^2$, then the condition \begin{equation*} |\mu_n(F)|(\ln n)^n<C<\infty\quad\forall n\in\mathbb{N} \tag{1} \end{equation*} implies that $F=0$ almost everywhere (a.e.) on $[0,1]$.

Indeed, let \begin{equation*} f(z):=\sum_{n=0}^\infty\frac{\mu_n(F)(iz)^n}{n!} =\int_0^1 F(t)\,dt \sum_{n=0}^\infty\frac{(itz)^n}{n!} =\int_0^1 F(t)e^{itz}\,dt \end{equation*} for all complex $z$. By (1), for any natural $k\ge2$ \begin{equation*} |f(z)|\le O(1+|z|^{k-1})+C\sum_{n=k}^\infty\frac{(|z|/\ln n)^n}{n!} \le O(1+|z|^{k-1})+C\sum_{n=0}^\infty\frac{(|z|/\ln k)^n}{n!} \le c_k e^{|z|/\ln k} \end{equation*} for some real $c_k>0$ and all complex $z$. So, $f(z)$ is an entire function of exponential type $a$ for any real $a>0$.

Hence, by a Paley--Wiener theorem (more specifically, see e.g. Theorem 19.3 on page 375), for each real $a>0$ there is an $L^2$ function $F_a$ such that for all complex $z$ \begin{equation*} f(z)=\int_{-a}^a F_a(t)e^{itz}\,dt \end{equation*} Taking the inverse Fourier transform, we see that for each $a\in(0,1)$ and all complex $z$ \begin{equation*} f(z)=\int_0^a F(t)e^{itz}\,dt. \end{equation*} Thus, for all complex $z$ \begin{equation*} 0=f(z)=\int_0^1 F(t)e^{itz}\,dt. \end{equation*} So, indeed $F=0$ a.e. on $[0,1]$.


Consider now the more general setting when $F(t)\,dt$ is replaced by $\rho(dt)$, where $\rho$ is a signed measure over $\R$ with support $\supp\rho\subseteq[0,1]$. Then condition (1) is replaced by \begin{equation*} |\mu_n(\si)|(\ln(n+2))^n<C<\infty\quad\forall n\in\mathbb{N}_0:=\{0,1,\dots\}, \tag{1a} \end{equation*} where \begin{equation*} \mu_n(\si):=\int_0^1 t^n\si(dt)=\int_0^1 t^{n+2}\rho(dt)\quad\text{and}\quad\si(dt):=t^2\rho(dt). \end{equation*} So, $\supp\si\subseteq[0,1]$. Also, (1a) implies $\mu_0(\si)<\infty$, that is, the signed measure $\si$ is finite. So, by reasoning quite similar to that in the above $L^2$ case, we see that under condition (1a) \begin{equation*} f(z):=\int_\R e^{itz}\,\si(dt) \end{equation*} is an entire function of exponential type $a$ for any real $a>0$.

For real $b>0$ and $t\in\R$, let \begin{equation*} G_b(t):=(g_b*d\si)(t):=\int_\R g_b(t-s)\,\si(ds) =\frac{\si([t-b,t+b])}{2b}, \end{equation*} where $g_b:=\frac1{2b}\,1_{[-b,b]}$. Then $G_b\in L^2(\R)$, since the signed measure $\si$ is finite with $\supp\si\subseteq[0,1]$. Also, \begin{equation*} \hat g_b(z):=\int_\R e^{itz}g_b(t)\,dt=\frac{\sin bz}{bz} \end{equation*} for $z\ne0$, and hence \begin{equation*} f_b(z):=\int_\R e^{itz}\,G_b(t)\,dt =\int_\R e^{itz}\,(g_b*d\si)(t)\,dt=\hat g_b(z)f(z) \end{equation*} is of exponential type $b+a$ for all real $a>0$. Thus, by the cited Paley--Wiener theorem, $\supp(g_b*d\si)\subseteq[-b-a,b+a]$ for all real $a>0$ and hence \begin{equation*} \supp(g_b*d\si)\subseteq[-b,b]. \end{equation*}

On the other hand, because $\supp g_b$ and $\supp\si$ are both compact, by Theorem 4.3.3 on page 117, $\cch\supp(g_b*d\si)=\cch\supp g_b+\cch\supp\si$, where $\cch$ denotes the closed convex hull. Since $\cch\supp g_b\ni0$, we conclude that $\cch\supp\si\subseteq\cch\supp(g_b*d\si)\subseteq[-b,b]$, whence $\supp\rho\subseteq\supp\si\cup\{0\}\subseteq[-b,b]$, for any real $b>0$. Thus, condition (1a) implies \begin{equation*} \supp\rho\subseteq\{0\}. \tag{2} \end{equation*}

Vice versa, trivially (2) implies (1a). Thus, (1a) holds iff $\supp\rho\subseteq\{0\}$.


Response to the second comment by the OP: Let us try to sort all this out.

  1. Your condition \begin{equation} 0<c<|\mu_n(F)|(\ln n)^n<C<\infty \quad\forall n\in\mathbb{N} \tag{3} \end{equation} can actually never hold as stated, because the inequalities $0<c<|\mu_n(F)|(\ln n)^n$ will always be false for $n=1$. So, of course, I assumed (3) with $\mathbb{N}$ replaced by $\{2,3,\dots\}$.

  2. In your post here, you did not even mention "analytical distributions" or Ehrenpreis. Instead, you wrote: "Let $F:[0,1]\to\mathbb{R}$ be a measurable function" and then you also wrote "In fact, I do not need $F$ be a function; $F(x)dx$ may be a signed measure on $[0,1]$".

  3. If $F(x)dx$ is replaced by a signed measure $\rho(dx)$, then the corrected version of your condition (3) can be rewritten as the conjunction of my condition (1a) and the condition \begin{equation} 0<c<|\mu_n(\si)|(\ln n)^n \quad\forall n\in\{2,3,\dots\}. \tag{4} \end{equation}

  4. It was shown in this answer that, under condition (1a), the support of the signed measure $\rho$ must be contained in the set $\{0\}$. (Method-wise, this was first done in the case when $\rho(dx)=F(x)dx$ with $F\in L^2$, and then extended to the general case of any signed measure $\rho$ on $[0,1]$.)

  5. It then immediately follows that there is no signed measure $\rho$ on $[0,1]$ satisfying (1a) and (4) -- or, equivalently, satisfying the corrected version of your condition (3). In other words, your conditions do result in a contradiction. This completely answers your posted question.

$\endgroup$
6
  • $\begingroup$ I have added an extension to the more general setting when $F(t)\,dt$ is replaced by $\rho(dt)$, where $\rho$ is a signed measure. $\endgroup$ Apr 29, 2019 at 14:53
  • $\begingroup$ Paley-Wiener theorem (e.g., Theorem 19.3 in Ruding you refer to) requires $f(z)$ be $L^2$ on the real line, which is not true, I think (there are reasons why it shouldn't). Same is true for $f_b(z)$ in your second application of Paley-Wiener, I think. But I can see your point. If $f$ is of minimal exponential type and $F$ (or $\rho$) its Fourier transform, then $F$-s support is $\{0\}$. Instead of classical Paley-Wiener you can use other things (f.i., a la Ehrenpreis). Thank you. $\endgroup$
    – Bedovlat
    Apr 30, 2019 at 4:53
  • $\begingroup$ @Bedovlat : Since $F\in L^2(\mathbb R)$, by the Plancherel theorem $f=\hat F$ is also $L^2$ on $\mathbb R$. See e.g. mathworld.wolfram.com/PlancherelsTheorem.html or en.wikipedia.org/wiki/Plancherel_theorem . So, I see no problem here. (You may have confused my $f(z)$ with your (moment) generating function $f(-iz)$.) $\endgroup$ Apr 30, 2019 at 14:49
  • $\begingroup$ Dear Prof. Pinelis, your $f(-iz)$ is also an entire function of minimal exponential type, and thus cannot be $L^2$ on the real line. No one cancelled Plancherel theorem, of course, but what I am saying is that your assumption $F\in L^2$ is already in contradiction with your (1). In fact, $F$ cannot be even a distribution in the classical sense. It can be viewed as an "analytical distribution" in the sense of Ehrenpreis, as I wrote before. $\endgroup$
    – Bedovlat
    Apr 30, 2019 at 16:42
  • $\begingroup$ @Bedovlat : In my answer, I have added a detailed response to your latter comment. I hope this will help. Here I wanted to add that whatever may be said about $f(-iz)$ is irrelevant to my answer, dealing with $f(z)$. More importantly, we indeed get a contradiction. But this contradiction results from your own assumptions, not mine. In the answer, I did not make any assumptions in addition to yours. $\endgroup$ Apr 30, 2019 at 19:27

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.