Does function $f(x) = kf(x-a) + (1-k)f(x-b)$ for $a / b \notin \mathbb{Q}$ always have a limit?

137 Views Asked by At

(This is my attempt to simplify or generalize my previous unanswered question Proving technique for exchanging lim and (infinite) sum where the sum converges very slowly.)

Consider a function defined on positive real numbers $f: \mathbb{R_+}\to\mathbb{R}$ with the following properties

  • for all $x>a$, $f(x) = kf(x-a) + (1-k) f(x-b)$, for some constants $k$, $a$, and $b$ where $0<k<1$ and $0<b<a$
  • $a / b \notin \mathbb{Q}$ (otherwise a counter-example would be $f(x) = \sin(2\pi x/\gcd(a,b))$))
  • $f(x)$ is bounded
  • $f(x)$ is continuous (otherwise a counter-example would be the indicator function $f(x) = \mathbb{1}_{a\mathbb{Z}+b\mathbb{Z}}$)(can we weaken this to "discontinuity is nowhere dense"?)

Can we conclude that the limit $$\lim_{x\to\infty} f(x)$$ exists? If not, what other condition we need to add to ensure the limit exists?

Intuitively, this is true to me. The first functional equation means that each function value is a weighted average of some previous values, so overall it should converge to some sort of average of the "seeding region" $x\in(0,a]$. But I can't really convert this intuition to a concrete proof.


Another attempt I had was to try solving the functional equation. I can find the following solution basis $$f(x) = e^{tx},$$ where $t$ is a (potentially complex) solution to the equation $$ke^{-at} + (1-k) e^{-bt} = 1.$$ I don't know if this is the entire solution space, nor how to solve the coefficient for each term. Nevertheless, this equation always has a solution $t = 0$, whose corresponding term in $f(x)$ would be a constant, which is the limit if it exists. All other solutions $t$ have negative real components (but can be arbitrarily close to 0), meaning their corresponding term in $f(x)$ shrinks and has individual limit of 0. However, I don't think I can directly say the limit of the function is the sum of the limit of each term in this case.


Another path I tried to explore is to consider the function on a grid $F: \mathbb{Z}^2\to\mathbb{R}$, $F(x, y) = f(ax+by)$. It would be a great first step to prove that $\lim_{\mathbf{v}\to(+\infty,+\infty)}F(\mathbf{v})$ exists, and then we can "fill in" other values for $f(x)$ using continuity.

On the grid, it can be seen that what contributes to $F(x, y)$ is a structure similar to binomial expansion: \begin{aligned} F(x, y) &= kF(x-1, y) + (1-k) F(x,y-1)\\ &= k^2 F(x-2, y) + 2k(1-k) F(x-1,y-1) + (1-k)^2 F(x,y-2)\\ &= k^3 F(x-3, y) + 3k^2(1-k) F(x-2,y-1) + 3k(1-k)^2 F(x-1, y-2) + (1-k)^3 F(x,y-3)\\ ... \end{aligned} As good as it looks, it doesn't help me with the question.