2

How can I linearize for the continuous function $f$ the functional equation $f^2(x)+1=2f(x+1)$

I' ve been studying recently the book for functional equations by Christopher G. Small and I' ve encountered the term "Linearization" of functional equations. I cannot figure out how to apply this method in this equation but if I had the equation $f^2(x)=2f(x+1)$ instead I know I could apply logarithms in both sides (thought I don't know if $f(x)>0$) such that $2\ln f(x)=\ln f(x+1)+ \ln2$.

Is it possible to do something like this in the first functional equation?

1 Answers1

1

I am not sure that linearization is applicable to this functional equation, but I will depict another way to solve this. First, lets consider trivial solutions. $$f^2(x)+1=2f(x+1)$$ $$f(x)=c$$ $$c^2+1=2c$$ $$c^2-2c+1=0$$ $$(c-1)^2=0$$ $$c=1$$ $$f(x)=1$$ I assume that the function is meant to be continuous and defined everywhere. Lets find some restrictions we can put on $f(x)$. The domain must be all real numbers. $$f(x+1)=\frac{f^2(x)}{2}+\frac{1}{2}$$ $$f(x-1)= \pm \sqrt{2f(x)-1}$$ Both of these operations should be valid if we want $f(x)$ to have a domain of all real numbers. There is no problem scaling a value of $f(x)$ up, but there may be a problem scaling down due to the domain of square roots. $$2f(x)-1 \geq 0$$ $$2f(x) \geq 1$$ $$f(x) \geq \frac{1}{2}$$ This means that we only need to consider the positive branch of the square root. $$f(x-1)= \sqrt{2f(x)-1}$$ We will further bound $f(x)$ by analyzing a sequence constructed from the equation. $$a_{n+1}= \sqrt{2a_n-1}$$ We must ensure that $a_n$ never goes below $\frac{1}{2}$. We will first establish that $a_n$ is a monotonically decreasing sequence. $$a_n>a_{n+1}$$ $$a_n>\sqrt{2a_n-1}$$ $$a_n^2>2a_n-1$$ $$a_n^2-2a_n+1>0$$ $$(a_n-1)^2>0$$ This is true for all $a_n$ except at $a_n=1$, which is the only solution to $a_n=a_{n+1}$. Since $a_n$ is monotonically decreasing, all $a_n<1$ can be ruled out as they will eventually go below $\frac{1}{2}$. We have tightened the bound on $f(x)$ to: $$f(x) \geq 1$$ For all $a_n>1$, we must prove that $a_n$ always stays above $1$. $$\sqrt{2a_n-1}>1$$ $$2a_n-1>1$$ $$2a_n>2$$ $$a_n>1$$ Since $a_n=1$ is the only solution to $a_n=a_{n+1}$, $a_n$ is monotonically decreasing, and $a_n$ stays above $1$ for all $a_n>1$, $a_n$ converges to $1$ for all $a_0>1$. Convergence of this sequence to $1$ and the bound on $f(x)$ means that: $$\lim_{x \to -\infty} f(x)=1$$ $$\text{From the main identity:}$$ $$f^2(x)+1=2f(x+1)$$ It is clear that the entirety of $f(x)$ can be defined if $f(x)$ is known over a domain interval of length $1\text{.}$ To find the other values $f(x)$ can be shifted using the identity. Let this interval be $x \in [0, 1]\text{.}$ Let $S(x)$ be a continuous function defined on this interval. $S(x)$ can be used contrive other values of $f(x)\text{.}$ For $f(x)$ to be continuous at $x=1$, S(x) must satisfy: $$S^2(0)+1=2S(1)$$ Generally, any continuous seed function on the $[0, 1]$ interval will allow for computation of a continuous function $f(x)$ given that the seed function satisfies: $$S(x) \geq 1$$ $$S^2(0)+1=2S(1)$$ Conclusions that can be drawn about the resulting continuous function $f(x)$ include: $$f^2(x)+1=2f(x+1)$$ $$f(x) \geq 1$$ $$\lim_{x \to -\infty} f(x)=1$$ Additionally, there are a few things that can be said about $f(x)$ as $x$ tends to $\infty\text{.}$
If $S(x)=1$, then $\lim_{x \to \infty}f(x)=1$
If $S(x)>1$, then $\lim_{x \to \infty}f(x)=\infty$
If $S(x)=1$ for some $x$ in the interval but not all $x$, then: $$\liminf_{x \to \infty}f(x)=1 \quad\quad \limsup_{x \to \infty}f(x)=\infty$$ This occurs because the sequence $b_n$ is monotonically increasing for all $b_n>1$ where: $$b_{n+1}=\frac{b_n^2}{2}+\frac{1}{2}$$ $$b_{n+1}>b_n$$ $$\frac{b_n^2}{2}+\frac{1}{2}>b_n$$ $$b_n^2+1>2b_n$$ $$b_n^2-2b_n+1>0$$ $$(b_n-1)^2>0$$ Note that using the seed function method to generate $f(x)$ can produce all possible $f(x)$ that are continuous everywhere. The next step would be to determine if any solutions are differentiable everywhere. I will come back to edit this answer if I find any.

cvogt8
  • 430
  • +1 Very cool! If I may ask, where did you learn this method from? Or solving functional equations in general? – Ovi Jan 04 '18 at 03:01
  • @Ovi Another question I answered requested the use of the Stolz-Cesaro theorem. Learning about that theorem made me think about the relationship between sequences and recursive functional equations. I learned most of it from just messing around. The link that Will Jagy provided in the question comments looks like an interesting resource on functional equations though. The seed function idea came from thinking about uniqueness theorems for the harmonic number function. Also, I mean non-trivial differentiable solutions in the last part. – cvogt8 Jan 04 '18 at 03:28