Here is a problem that am I stuck and I appreciate any help. In essence, I am trying to show that the only solutions for the described problem are the ones provided below. Best..

Setup: In what follows, we consider the functions defined over the half-closed symmetric interval $\mathcal{I}\triangleq [-\frac{1}{2},\frac{1}{2}]$ with the usual inner product defined as \begin{align} \langle f \mid g \rangle = \int_{x\in\mathcal{I}} f(x)\, g(x) \, \mathrm{d}x \end{align} Consider orthonormal basis for the the even-symmetric functions $\{\varphi_k\}_{k=0}^\infty$ that are defined for all $x\in\mathcal{I}$ as \begin{align} \varphi_0(x)&=1\\ \varphi_k(x)&=\sqrt{2}\cos(2\pi x),\quad k=1,2,\ldots. \end{align}

Conjecture: Among all real valued even-symmetric functions defined over the interval $\mathcal{I}$, the following functional equation (for the given fixed real pair ($\alpha_0\neq 0, \alpha_1$), \begin{align} f(x) \circledast \ln(f(x)) = \alpha_0+\alpha_1\varphi_1(x) \end{align} is satisfied either by \begin{align} f_1 (x) = \beta_0+\beta_1\varphi_1(x), \end{align} or \begin{align} f_2 (x) = \gamma_0 \exp\left(\gamma_1 \varphi_1(x)\right), \end{align} for some appropriately selected constants $(\beta_n,\gamma_n)$, for $n=0,1$. Here, the convolutions are in the circular sense over the interval $\mathcal{I}$.

  • $\begingroup$ Dear @user64076, please use markdown instead of mathjax to format text. For example, instead of $mathbf{some text}$, you can use **some text** to produce bold text. Mathjax is meant for mathematical displays alone. Thank you. $\endgroup$ – Ricardo Andrade Dec 17 '14 at 19:14
  • $\begingroup$ Do you conjecture that $f_1$ and $f_2$ are solutions to the functional equation? Or do you conjecture that $f_1$ and $f_2$ are the only solutions to the functional equation? In any event you need to assume more properties of $f$ than that it is real-valued and even: if it takes negative values, is not measurable, or is measurable but very far from being integrable, the convolution written above does not make sense. $\endgroup$ – Ian Morris Dec 18 '14 at 10:45
  • $\begingroup$ Ok, fair enough.. The functions belong to the probability simplex defined over the interval $\mathcal{I}$, i.e., they are non-negative Lebesgue absolute-integrable functions. And by the way, yes, I am conjecturing that $f_1$ and $f_2$ are the only solutions.. I hope this clarifies the question a bit.. @IanMorris $\endgroup$ – user64076 Dec 18 '14 at 16:58

Well, I would say that at least for some (degenerate) $\alpha_0, \alpha_1$ there will be more solutions. Consider $f_2$ of the form $f_2(x)=\gamma_0+\gamma_1 \cos x + \gamma_2 \cos 2x$. If your equation to never has a solution of a form other than the two above, then the convolution $f_2* \log f_2$ would never be a degree one trigonometric polynomial.

Though, this convolution is surely a degree at most two trigonometric polynomial, as $f_2$ is. Hence, to construct an example that becomes of degree one, you need only to eliminate the second degree term. To do so, you need $\log f_2$ to have zero $\cos 2x$-Fourrier coefficient. And it is just one relation on $\gamma_0$, $\gamma_1$ and $\gamma_2$. For instance, once you check that this coefficient can be (for different values of $\gamma_i$'s) both positive and negative, you are done.

It seems to me that this is easy to be checked: for instance, taking $\gamma_0=1$, $\gamma_1$ to be small positive and $\gamma_2$ to be small negative, one gets $$ \log(1+\gamma_1\cos x +\gamma_2 \cos 2x)= \gamma_1 \cos x - \frac{1}{4} \gamma_1^2 (1+\cos 2x) + \gamma_2 \cos 2x + o(\gamma_1^2+|\gamma_2|).$$ So one can both ensure positivity and negativity of the $\cos 2x$-coefficient: the former by taking $\gamma_1=0$, $\gamma_2$ small, the latter by taking $\gamma_2=0, \gamma_1$ small.

This leads, surely, to a degenerate example. But it shows that even if for generic $\alpha$'s your conjecture holds, it should be handled by more elaborate arguments.

| cite | improve this answer | |

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.