18
$\begingroup$

Throughout, let $f$ be a Lebesgue measurable function (or continuous if you wish, but this is probably no easier). (Questions with distributions etc. are possible also but I want to keep things simple here).

FINAL CLARIFICATION/REWRITE!!

Thanks to all who have commented so far. I will need some more time to digest it properly. The original forms of the questions are at the end; here I have rewritten the questions, hopefully more clearly; sorry for my poor explanation before!

Definition $\mathcal{M}_a$, the space of absolutely null moment functions, is the set of all measurable $f$ satisfying $$ \int_0^\infty t^n |f(t)| dt < \infty, \quad \int_0^\infty t^n f(t) dt = 0, \qquad \forall \, n=0,1,2,\ldots. $$ $\mathcal{M}$, the space of null moment functions, is the set of all measurable $f$ satisfying $$ \int_0^T |f(t)| dt < \infty \quad \forall \, T>0, \qquad \lim_{R \to \infty} \int_0^R t^n f(t) dt = 0, \quad \forall \, n=0,1,2,\ldots $$

Thus, trivially $0 \in \mathcal{M}_a \subseteq \mathcal{M}$, but $\mathcal{M}_a$ contains many other non-trivial functions. It seems certain that $\mathcal{M}_a \ne \mathcal{M}$ (I would be amazed if the spaces were equal), although constructing an explicit example seems tricky.

Definition Given a function $\psi \geq 0$, let $G(\psi)$ be the set of all $f$ such that $|f| \leq \psi$.

(Of course we're identifying functions equal a.e., so really we should consider equivalence classes etc. just as for $L^p$ spaces).

Rephrased Question 1 Find general simple necessary and/or sufficient conditions on $\psi$ with the property $$ G(\psi) \cap \mathcal{M}_a = \{ 0 \}. $$

Rephrased Question 2 The same as Question 1, but with $G(\psi) \cap \mathcal{M}$ instead.

Or, if $G(\psi)$ is not the appropriate space for these problems, consider $\int_T^{2T} |f| dt \leq g(T) $ or $\int_n^{n+1} |f| dt \leq A_n$ or something similar instead. Finding the correct kind of restrictions on $f$ is part of the problem.

Thus, $G(\psi) \cap \mathcal{M}_a = \{ 0 \}$ for $\psi(t) = \exp(-\delta t)$, by the discussion below; and also for any compactly supported $\psi \in L^1$.

But $G(\psi) \cap \mathcal{M}_a \ne \{ 0 \}$ for $\psi(t) = \exp(-t^{1/4}) |\sin(t^{1/4})|$.


Original QUESTION 1

If $$ \int_0^\infty t^n |f(t)| dt < \infty, \quad \int_0^\infty t^n f(t) dt = 0, \qquad \forall \, n=0,1,2,\ldots, $$ when must $f \equiv 0$ almost everywhere?


EDIT: CLARIFICATION: this is really about classes of functions, expressed in terms of a growth/decay rate function $\phi$, which give unique solutions to the moment problem. I am NOT asking how to solve the moment problem itself!

If possible, find necessary and sufficient conditions on functions $\phi \searrow 0$ with the property that $$ \int_R^\infty |f(t)| dt \leq \phi(R) \qquad \Rightarrow \qquad f \equiv 0. $$ So, $\phi(R) = \int_R^\infty \exp(-t^{1/4}) |\sin(t^{1/4})| dt$ is not enough (by Example 2 below); nor is $\phi(R) = \int_R^\infty |f(t)| dt$ with f given by "coudy" in his/her answer below.

But $\phi = \chi_{[0,b]}$ is enough (by Example 1 below); moreover $\phi(R) = \exp(-\delta R)$ would be enough for any fixed $\delta > 0$, by my discussion of Example 1 below, because the relevant Laplace transform $F = \mathcal{L}f$ is analytic on the half-plane $\{ \mathrm{Re}(z) > -\delta \}$. So we want to know about the gap between $\exp(-\delta R)$ and functions like that given by "coudy" below.

FURTHER CLARIFICATION: by analogy, maybe an example from PDEs will explain better (I'm not saying this is related to my problem; I'm saying that this is the same kind of result as what I want):

Definition A null temperature function is a continuous function $u = u(x,t) : \mathbb{R} \times [0, \infty) \to \mathbb{R}$ such that the heat equation is satisfied, i.e. $\frac{\partial u}{\partial t} = \frac{\partial^2 u}{\partial x^2}$ in $\{ t>0 \}$, and $u(x, 0) = 0$ for all $x$.

Theorem There exists a null temperature function satisfying $|u(x,t)| \leq \exp(A/t)$ with $A>0$, such that $u(x,t) \not\equiv 0$ for all $t>0$.

Theorem Let $u$ be a null temperature function satisfying $|u(x,t)| \leq A \exp(B t^{-\delta})$, for some $A,B>0$ and $\delta<1$. Then $u \equiv 0$.

So here the "critical" growth rate for null temperature functions is roughly $\exp(A/t)$. I am looking for a similar thing with "null moment functions".

Note that this is a totally different problem to: given $v$, find some $u$ satisfying the heat equation such that $u(x,0)=v(x)$.


Original QUESTION 2

If instead we have only $\int_0^R |f(t)| dt < \infty$ for each $R>0$, and $$ \lim_{R \to \infty} \int_0^R t^n f(t) dt = 0, \qquad n=0,1,2,\ldots $$ when must $f \equiv 0$ almost everywhere? (I have very little idea about this).


I think these questions are clearly very natural, interesting, and important, but Googling etc. didn't work well (I tried "vanishing moments" and other phrases, but there's just too much stuff out there). Standard known examples/methods follow.

Example 1: if $f$ is compactly supported on $[a,b]$, say, then $f \equiv 0$ a.e. because polynomials are dense in $C[a,b]$.

Example 2: by taking imaginary parts of $\int_0^\infty t^{4n+3} \exp(-(1+i)t) dt \in \mathbb{R}$, we get $$ \int_0^\infty t^{4n+3} e^{-t} \sin t \, dt = 0, $$ and so by substituting $t = x^{1/4}$, $$ \int_0^\infty x^n \exp(-x^{1/4}) \sin(x^{1/4}) dx = 0, \qquad n=0,1,2,\ldots $$

Alternative method for Example 1: consider the Laplace transform $F(z) = \int_0^\infty e^{-zt} f(t) dt$. In Example 1, $F$ is an entire function such that $F^{(n)}(0) = 0$ for all $n$, so $F \equiv 0$ and thus $f \equiv 0$ a.e. as required.

So, any condition on $f$ forcing $F$ to be analytic on some disc with centre $0$ is enough; but can we do better?

In Example 2, $f \in L^1(0,\infty)$ and so $F$ is bounded and analytic on $\{ \mathrm{Re}(z) > 0 \}$, and continuous on the boundary, with $\lim_{z \to 0} F^{(n)}(z) = 0$ for all $n$. But this is still not enough to force $F \equiv 0$.

$\endgroup$
7
  • 2
    $\begingroup$ See en.wikipedia.org/wiki/Stieltjes_moment_problem and the references therein: that should help. $\endgroup$ Commented Jul 10, 2010 at 13:44
  • 1
    $\begingroup$ Thanks very much! But although this is interesting, I don't see how it works for my problem. I am not asking "given a sequence of numbers, find a function f in a specific class with those moments" - all my moments are zero, so I do know one solution, trivially! I want to know how nasty the others must be; unfortunately I don't know which class of functions is relevant (that's the whole point of my question). $\endgroup$
    – Zen Harper
    Commented Jul 10, 2010 at 14:46
  • 1
    $\begingroup$ if you suppose a strong rate of decay (sub-gaussian for example), then your statement becomes true. See Terry Tao's blog for a discussion (section on the moment method) on related ideas: terrytao.wordpress.com/2010/01/05/… $\endgroup$
    – Alekk
    Commented Jul 10, 2010 at 14:49
  • $\begingroup$ @Zen. To see how your problem connects to the problem of moments in probability theory, just write f as $f_+-f_-$. Now saying that f has zero moments amounts to saying that the two measures $f_+dx$ and $f_ dx$ have the same moments. Does that imply that the two measures are equal ? The link provided by Mariano may lead you to the [[Carleman's condition][en.wikipedia.org/wiki/Carleman%27s_condition]], which traduces back into a condition on the decreasing rate of the moments of $|f|$. $\endgroup$
    – coudy
    Commented Jul 10, 2010 at 15:14
  • 2
    $\begingroup$ Some null temperature function references: S.-Y. Chung, D. Kim. An example of nonuniqueness of the Cauchy problem for the heat equation. Comm. Partial Differential Equations 19 (1994), no. 7-8, 1257–1261. MR1284810 (95c:35114) S.-Y. Chung. Uniqueness in the Cauchy problem for the heat equation. Proc. Edinburgh Math. Soc. (2) 42 (1999), no. 3, 455–468. MR1721765 (2000h:35060) Also discussed in my paper "LAPLACE TRANSFORM REPRESENTATIONS AND PALEY–WIENER THEOREMS FOR FUNCTIONS ON VERTICAL STRIPS" available on my website: zenharper.jimdo.com/research-papers-and-publications $\endgroup$
    – Zen Harper
    Commented Jul 10, 2010 at 16:51

1 Answer 1

21
$\begingroup$

The answer to the first question is no. The following example is standard in probability theory, see e.g. Billingsley "probability and measure", example 30.2.

$$f(x) = {1\over \sqrt{2\pi}\ x}\ e^{-{(\ln x)^2\over 2}}\ \sin(2\pi \ln x) \ {\bf 1}_{[0,\infty[}(x)$$

You can check that all the moments are zero using the change of variable $\ln x = s+k$.

This example is used to show that a probability measure is not always determined by its moments.

$\endgroup$
3
  • $\begingroup$ Thanks very much! This is interesting, but only a very partial answer to my first question: it shows that a decay rate $\int_R^\infty |f(x)| dx \leq \phi(R)$ with some particular $\phi$ is not enough to force $f=0$. If possible, I would really like a general necessary and sufficient condition on $\phi$. $\endgroup$
    – Zen Harper
    Commented Jul 10, 2010 at 14:54
  • $\begingroup$ Sufficient I can understand, and see Alekk's comment about sub Gaussianity, but what do you mean by necessary? If you mean something like "if $F$ is a class of functions such that only the trivial element $f = 0$ in $F$ has all moment zero, then $F$ must be dominated by some rate $\phi$", I think you won't get anything meaningful. Take for example the class of non-negative, continuous, Lebesgue integrable functions. There are definitely elements in there who decay slower than the Gaussian. $\endgroup$ Commented Jul 10, 2010 at 15:22
  • 1
    $\begingroup$ I'll accept this since it tells me something interesting and nontrivial I didn't know before, and I feel guilty since my original question is maybe a bit stupid. $\endgroup$
    – Zen Harper
    Commented Sep 19, 2010 at 10:10

You must log in to answer this question.

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