1

I was thinking about infinite fractions of the form $1+\frac{a}{1+\frac{a}{1+...}}$ but realised it would be much more useful and satisfying and only a little bit harder to solve fractions of the form $a+\frac{b}{a+\frac{b}{a+...}}$. One thing that I realised is that solving that fraction should be equivilant to solving for $x$ in $x=a+\frac{b}{x}$. Working that out I got $x=\frac{1}{2}(a\pm\sqrt{4b+a^2})$, but this formula is not defined for all and $a$ and $b$ values because of the square root, and disagrees with the infinite fraction for many values of $a$ and $b$ but works for others. Is there another formula that I missed online, or is there no general formula (which I doubt), or is it just a simple mistake that I made. One thing that I noticed is that most of the inconsistencies are when $16a^2>b^4$.

  • How can you have a formula for a number that doesnt terminate? There is a formula for a continued fraction of length $k$ – XRBtoTheMOON Jan 17 '18 at 21:35
  • If such a formula for fraction length $k$ exists, could you not look at the limit as k approaches infinity, assuming the limit exists? – Aaron Quitta Jan 17 '18 at 21:43
  • @AaronQuitta Normally, $a^2 + 4b >0$. You only run into trouble when $a^2 + 4b < 0$. – Math Lover Jan 17 '18 at 21:46
  • 3
    If $4b+a^2\lt 0$ so that the square root is not a real number, $b$ is relatively large and negative - have you tested that case for convergence? Can you provide some examples of where you say the fraction and the formula disagree so we can see what you mean. – Mark Bennet Jan 17 '18 at 21:46
  • @MathLover Yes, thats what I was trying to convey in my not, I guess I messed up with that, thank you. – Aaron Quitta Jan 17 '18 at 22:13
  • @MarkBennet Whoops, it appears I made a mistake with my implementation, that caused the problem that I was going to show. But as mentioned above, the is the problem with the square root. Is there anyway around the square root problem? Also is there any way to find out whether to use the positive or negative square root? – Aaron Quitta Jan 17 '18 at 22:24
  • @AaronQuitta Assuming that you take $a,b\gt 0$, then note that the continued fraction is always $\gt a$; only one of the two solutions of the quadratic is. – Steven Stadnicki Jan 17 '18 at 22:26
  • A quick check with a computer program suggests that the values do converge to the stated limit, provided a is not 0 and 4b + a^2 > 0. The larger root if a > 0, the smaller if a < 0. I will try to prove this. – Michael Behrend Jan 17 '18 at 22:36
  • @MichaelBehrend Thank you, any idea if there is a formula for when 4b+a^2=0? – Aaron Quitta Jan 17 '18 at 22:56
  • Also @btcgrl what formula are you referring to, can you link to or explain it? – Aaron Quitta Jan 17 '18 at 22:56

1 Answers1

1

A partial answer, in which the series of fractions is proved convergent if $4b + a^2 \ge 0$ and $a \neq 0$, but not proved divergent if $4b + a^2 < 0$ (though this is probably true). For given $a$ and $b$, the successive terms are defined by $x_0 = 1$, $x_{n+1} = a + b/x_n$ (provided $x_n \neq 0$). If $a = 0$, the terms are alternately $1$ and $b$, and hence converge iff $b = 1$. If $a + b = 1$ then all terms are $1$. From now on assume $a \neq 0$ and $a + b \neq 1$. First consider $4b + a^2 = 0$. We can put $a = 2c$ and $b = -c^2$ where (by the above assumptions) $c \neq 0, 1$. It's easily proved by induction that $$x_n = {{(n + 1)c - nc^2}\over{n - (n - 1)c}} = {{c(1 - c + 1/n)}\over{1 - c + c/n}},$$ from which $x_n \rightarrow c$ as $n \rightarrow \infty$ (but the inductive definition fails if $c = 1 + 1/n$ for some $n$). Now consider $4b + a^2 > 0$. Define $$r = {1\over2}(a + \sqrt{4b + a^2}),\quad s={1\over2}(a - \sqrt{4b + a^2}).$$ By induction, $x_n = p_{n+1}/p_n$ where $p_n$ is defined by the Fibonacci-like recurrence $$p_0 = 1,\quad p_1 = 1,\quad p_{n+1} = ap_n + bp_{n-1}.$$ This is solved in the usual way as $p_n = Ar^n + Bs^n$, where $A = (1 - s)/(r - s)$, $B = (r - 1)/(r - s)$. Hence $$x_n = {{(1 - s)r^{n+1} + (r - 1)s^{n+1}}\over{(1-s)r^n + (r - 1)s^n}}.$$ The above assumption $a + b \neq 1$ implies $r \neq 1$ and $s \neq 1$. If $a > 0$ then $|r| > |s|$, hence $x_n \rightarrow r$. If $a < 0$ then $|s| > |r|$, hence $x_n \rightarrow s$.

  • (Later) For the case 4b + a^2 < 0, one can argue by contradiction. Assuming the limit exists, it's not hard to see that it must be a root of x^2 - ax - b = 0; but it can't be if the roots are imaginary. – Michael Behrend Jan 18 '18 at 12:34
  • Where does that formula for "successive terms" come from, and what do you mean by that? I also unfamiliar with the usual way, and how you came to conclusion at the end based off of the magnitudes of r and s. And for the case when a+b=1, doesn't the fraction alternate between 1 and another number, and not just hold the value of one depending on whether your ending on an addition or a division? – Aaron Quitta Jan 18 '18 at 17:31