2
$\begingroup$

I am currently reading volume 2 of "Generalized Functions" by Gelfand and $\mathcal{S}^{2}_0(\mathbb{R})$ is defined to be the collection of $C^\infty$ functions $f$ on $\mathbb{R}$ such that \begin{equation} \sup_{x \in \mathbb{R}}\lvert x^k f^{(q)}(x) \rvert \leq CA^kB^q q^{2q} \end{equation} where $k$ and $q$ are arbitrary nonnegative integers and $A,B,C>0$ are constants depending on $f$.

I also figured out that any function in $\mathcal{S}^{2}_0(\mathbb{R})$ must have a compact support.

My question is this: does there exist a certain $f$ for prescribed $A,B,C>0$? For example, if $A \in (0,1]$ is given, can we find some function $f \in \mathcal{S}^{2}_0(\mathbb{R})$ satisfying the above bounds with this given $A$?

The function space of $\mathcal{S}$-type are quite new to me. Could anyone please help me understand them?

Add : Here, $2$ can be replaced by any bigger number. That is, could anyone provide a nontrivial element for some $\mathcal{S}^b_0$ where $b$ can be as large as one wants?

$\endgroup$
8
  • 1
    $\begingroup$ $C$ can always be made small by dividing a given $f \in S^2_0$ by the $C$ that it came with. You can also make one (but not both) of $A$ and $B$ small by scaling (replacing $f(x) \mapsto \tilde{f}(x) = f(\lambda x)$ changes $B \mapsto \lambda B$ and $A \mapsto \lambda^{-1} A$). So for the "example" in the question it is doable. I am not sure whether $A$ and $B$ can be both made small; there may be some uncertainty principle obstructions, since $A$ is somewhat like the spatial support size and $B$ is somewhat like frequency support size. $\endgroup$ Commented Jan 18, 2023 at 13:30
  • $\begingroup$ I do not need $B$ to be small. Only $A$ has to be small. Thank you for your answer. One last thing : how do you know that $S^2_0$ contains nontrivial functions? $\endgroup$
    – Isaac
    Commented Jan 18, 2023 at 13:33
  • $\begingroup$ Inferring from Willie Wong's comment, it does seem to have nontrivial elements... $\endgroup$
    – Isaac
    Commented Jan 18, 2023 at 20:08
  • $\begingroup$ @Isaac: no, all I wrote is that given any $f\in S^2_0$ you can make an element with arbitrarily small $C$ and $A$ values. I do not assert the existence of any element of $S^2_0$. $\endgroup$ Commented Jan 18, 2023 at 20:10
  • $\begingroup$ The problem is the growth in $q$. $\endgroup$ Commented Jan 18, 2023 at 20:10

1 Answer 1

2
$\begingroup$

$\newcommand\S{\mathcal S}\newcommand\R{\mathbb R}$For $x\in(0,1)$, let $g(x):=\frac1x$, and then let $$f(x):=e^{-g(x)-g(1-x)}$$ for $x\in(0,1)$, with $f(x):=0$ for real $x\notin(0,1)$.

Then $f\in\S_0^3(\R)$ -- see the details on this at the end of the answer.

As noted in Willie Wong's comment, you can rescale $f$ vertically and horizontally to get desired $C$ and $A$ (and you said you do not need $B$ to be small).

Details: Everywhere here, $x\in(0,1)$. We have $$f'(x)=P_1(x)f(x),$$ where $P_1(x):=\frac1{x^2}-\frac1{(1-x)^2}$. It is now easy to see by induction that for any $r=0,1,\dots$ $$f^{(r)}(x)=P_r(x)f(x),$$ where $P_r(x)$ is a polynomial in $\frac1x\,,\frac1{1-x}$ of total degree $\le2r$, containing $\le C^r$ monomials with coefficients $\le C^r r!$ in absolute value; here and elsewhere $C$ denotes various universal positive real constants. Also, for any nonnegative integers $m$ and $p$ such that $m+p\le2r\le2q$ we have $$\dfrac1{x^m(1-x)^p}\,e^{-g(x)-g(1-x)}\le e^{-m-p}m^m p^p \le q^{m+p}\le q^{2q}.$$ Thus, $$|f^{(q)}(x)|\le C^q q^{3q}.$$


By a more accurate accounting on the coefficients of the monomials of $P_r(x)$, one can get \begin{equation*} |f^{(q)}(x)|\le C^q q^{2q} \tag{10}\label{10} \end{equation*} and thus $f\in\S_0^2(\R)$.

Indeed, for $r=0,1,\dots$, \begin{equation*} f^{(r+1)}(x)=(P_r(x)f(x))'=(P'_r(x)+P_r(x)(\tfrac1{x^2}-\tfrac1{(1-x)^2}))f(x), \end{equation*} so that \begin{equation*} P_{r+1}(x)=P'_r(x)+P_r(x)(\tfrac1{x^2}-\tfrac1{(1-x)^2}). \end{equation*} Consider any monomial $m(x)=c(\frac1x)^i(\frac1{1-x})^{p-i}$ of the polynomial $P_r(x)$ in $\frac1x\,,\frac1{1-x}$; this monomial is of total degree $p$ and with coefficient $c$. Then $(m(x)f(x))'/f(x)$ is the sum of four monomials of $P_{r+1}(x)$. Two of these four monomials are monomials of $P'_r(x)$ (say they are of the 1st kind) and the other two monomials are monomials of $P_r(x)(\tfrac1{x^2}-\tfrac1{(1-x)^2})$ (say they are of the 2nd kind).

The total degree of each of the two monomials of the 1st kind is $p+1$ and the absolute value of its coefficients is $\le p|c|\le2r|c|$. The total degree of each of the two monomials of the 2nd kind is $p+2$ and the absolute value of its coefficient is $\le |c|$.

So, if a monomial of $P_q(x)$ is obtained by $k$ steps of the 1st kind (and, hence, $q-k$ of the 2nd kind), then it is of the form \begin{equation*} M(x)=c(\tfrac1x)^i(\tfrac1{1-x})^{p-i} \end{equation*} with \begin{equation*} p=k+2(q-k)\quad\text{and}\quad |c|\le (2q)^k\le2^q q^k. \end{equation*} Recalling now the inequality $u^s e^{-u}\le s^s$ for $u>0$ and $s\ge0$ (with $0^0:=1$), we see that \begin{equation*} |M(x)f(x)|\le2^q q^k i^i (p-i)^{p-i}\le2^q q^k p^p=2^q h(k), \end{equation*} where $h(k):=q^k (k+2(q-k))^{k+2(q-k)}$. Note that $h$ is log convex, $h(0)=(2q)^{2q}$, and $h(q)=q^{2q}$. So, $h(k)\le(2q)^{2q}$ for all $k\in[0,q]$. So, for any monomial $M(x)$ of $P_q(x)$ we have \begin{equation*} |M(x)f(x)|\le2^q (2q)^{2q}=8^q q^{2q}. \end{equation*} Since $P_q(x)$ contains $\le4^q$ monomials, we conclude that \eqref{10} holds. $\quad\Box$

$\endgroup$
7
  • $\begingroup$ Thank you so much!!! $\endgroup$
    – Isaac
    Commented Jan 18, 2023 at 16:01
  • $\begingroup$ Oh, could you cite any reference where this example is given? $\endgroup$
    – Isaac
    Commented Jan 18, 2023 at 17:27
  • 1
    $\begingroup$ I don't know if this example is given anywhere in the existing literature, and I actually have never encountered $\mathcal{S}^{2}_0(\mathbb{R})$ before. If you have a difficulty in verifying this example, please let me know, and I will then provide details. $\endgroup$ Commented Jan 18, 2023 at 17:32
  • $\begingroup$ Multiplying $x^k$ to the derivatives does not change the bound since the function $f$ is supported in $[0,1]$. Right? $\endgroup$
    – Isaac
    Commented Jan 18, 2023 at 20:35
  • 1
    $\begingroup$ @Isaac : I see. The factor $x^k$ will only play a role when rescaling horizontally, to accommodate $A$. $\endgroup$ Commented Jan 18, 2023 at 23:04

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .