0
$\begingroup$

I'm working on some problems involving Fourier transforms and convolution problems and there is one problem I cannot solve. In my situation we have a complex valued function $f(ix)$, with $x\in\mathbb{R}$ and two functions from which we know the following: $|g_2(ix)|\le|g_1(ix)|=1$.

The convolution problem is the following, I have two functions (spectra in my case) that are defined as $$ U_1(i\omega) = \int_{-\infty}^\infty f(ix)g_1(i(\omega-x))\mathrm{d}x $$ $$ U_2(i\omega) = \int_{-\infty}^\infty f(ix)g_2(i(\omega-x))\mathrm{d}x $$ assuming these exists. I want to show that $|U_2(i\omega)|\le|U_1(i\omega)|$ for all $\omega$. Intuitively, this is true, since the magnitude of $g_1$ is equal to $1$ and $g_2$ is less or equal to $1$. Also, when I perform some simulation experiments with random functions $f$, I obtain that this relationship holds, however I cannot prove it. sadface

What I have been able to show is this: $$|U_1(i\omega)| = \int_{-\infty}^\infty f(ix)g_1(i(\omega-x))\mathrm{d}x \le \int_{-\infty}^\infty |f(ix)g_1(i(\omega-x))|\mathrm{d}x =\int_{-\infty}^\infty |f(i(\omega-x))|\mathrm{d}x$$ $$|U_2(i\omega)| = \int_{-\infty}^\infty f(ix)g_2(i(\omega-x))\mathrm{d}x \le \int_{-\infty}^\infty |f(ix)g_2(i(\omega-x))|\mathrm{d}x \le\int_{-\infty}^\infty |f(i(\omega-x))|\mathrm{d}x$$ Hence, the upperbound on $|U_2|$ is less or equal than the upperbound on $|U_1|$, however, this does not guarantee that $|U_2(i\omega)|\le|U_1(i\omega)|$ for all $\omega$.

Do you maybe have any idea how to approach this further?

Edit: According to some comments here below, I can extend this question with; How could I formulate extra (nontrivial) assumptions on $f$, $g_1$ and $g_2$, such that I can prove this relationship. (It is quite common for example that the considered functions are rational functions with polynomials as numerator and denominator)

$\endgroup$
5
  • 3
    $\begingroup$ I think you need to provide more information about $g_1$ and $g_2$ for this to have a chance. If $g_1$ changes sign a lot and $g_2$ doesn't, the first integral could easily be 0 for some $\omega$ while the second is never 0. $\endgroup$ Commented Jun 2, 2020 at 14:27
  • 2
    $\begingroup$ This is not true without some additional assumptions on your functions. $\endgroup$ Commented Jun 2, 2020 at 14:33
  • $\begingroup$ @AlexandreEremenko I think than my question would be, how should I formulate these assumptions on the functions. I'm working on quite a fundamental topic, so having these assumptions to start with is already quite a win :P $\endgroup$
    – seaver
    Commented Jun 2, 2020 at 14:50
  • 2
    $\begingroup$ @seaver: I believe that the many concrete details stand in the way of understanding your problem. If $(X,m)$ is a measure space, it seems that you want conditions on $f_1$ and $f_2$ such that $|f_1| \le |f_2|$ should imply $| \int_X f_1 \ \mathrm d m | \le | \int_X f_2 \ \mathrm d m |$. I doubt that you will get anything useful (i.e. anything other than trivial conditions). $\endgroup$
    – Alex M.
    Commented Jun 2, 2020 at 15:06
  • 1
    $\begingroup$ @AlexM. I believe that is kind-of what I want indeed. I have no background in measure theory, so formulating the problem within this field is not my strongest point. The goal would be that with some (structural) knowledge, you can prove this implication. [regarding your edit]: that is unfortunate... :P I am hoping for some luck $\endgroup$
    – seaver
    Commented Jun 2, 2020 at 15:17

0

You must log in to answer this question.

Browse other questions tagged .