Analysis of the archetypal functional equation in the non-critical case
We study the archetypal functional equation of the form $y(x)=\iint_{\mathbb{R}^2} y(a(x-b))\,μ(\mathrm{d}a,\mathrm{d}b)$ ($x\in\mathbb{R}$), where $μ$ is a probability measure on $\mathbb{R}^2$; equivalently, $y(x)=\mathbb{E}\{y(α(x-β))\}$, where $\mathbb{E}$ is expectation with respect to the distribution $μ$ of random coefficients $(α,β)$. Existence of non-trivial (i.e., non-constant) bounded continuous solutions is governed by the value $K:=\iint_{\mathbb{R}^2}\ln|a|\,μ(\mathrm{d}a,\mathrm{d}b)=\mathbb{E}\{\ln|α|\}$; namely, under mild technical conditions no such solutions exist whenever $K<0$, whereas if $K>0$ (and $α>0$) then there is a non-trivial solution constructed as the distribution function of a certain random series representing a self-similar measure associated with $(α,β)$. Further results are obtained in the supercritical case $K>0$, including existence, uniqueness and a maximum principle. The case with $\mathbb{P}(α<0)>0$ is drastically different from that with $α>0$; in particular, we prove that a bounded solution $y(\cdot)$ possessing limits at $\pm\infty$ must be constant. The proofs employ martingale techniques applied to the martingale $y(X_n)$, where $(X_n)$ is an associated Markov chain with jumps of the form $x\rightsquigarrowα(x-β)$.