I want an expression to approximate the dottie number. The reason I want to use a Taylor series rather than, say, Newton's Method is because I want to guarantee that the approximation is within a certain percentage of error which is something that is easy to do with Taylor polynomials.

The problem is that the only way I seem to be able to do this is by setting the Taylor series for $\cos x$ equal to $x$. This leaves me with the following expression:

$$-x+\sum_{n=0}^{\infty} \frac{(-1)^nx^{2n}}{(2n)!}=0$$ Expanded this looks like: $$1-x-\frac{x^2}{2!} + \frac{x^4}{4!} - \frac{x^6}{6!} \cdots=0$$

I can solve for $x$ at $n=1,2$ but of course when $n>2$ the degree of the taylor polynomial is higher than a quartic and I can't solve the equation in general (Abel's Impossibility Theorem). To approximate those roots I would have to use something like Newton's method and we are back at square one.

Is it even possible to approximate the Dottie number to an arbitrary precision? Am I missing something?

  • 2
    $\begingroup$ Pedantic nitpick, don't know if it's relevant: "no closed form solution exists" does not follow from Abel's theorem, and I don't even know if it's true. Abel's theorem states, when applied to $n=5$, that there is no closed-form algebraic solution to $ax^5+bx^4+cx^3+dx^2+ex+f = 0$. It does not rule out an algebraic solution to $\frac{x^5}{6!} - \frac{x^3}{4!} + \frac{x}{2!} = 0$. $\endgroup$ – Patrick Stevens Apr 7 '18 at 16:05
  • 2
    $\begingroup$ Another thing: $\cos x - x$ does not have the Taylor series you gave. Its Taylor series is $1-x-\frac{x^2}{2!}+\frac{x^4}{4!}-\dots$. $\endgroup$ – Patrick Stevens Apr 7 '18 at 16:07
  • $\begingroup$ My bad, I'll edit the question to reflect this. Thanks! $\endgroup$ – Ozaner Hansha Apr 7 '18 at 16:10
  • $\begingroup$ For what it is worth, one of the solutions to $1-x-\frac{x^2}{2!}=0$ is about $0.7320508$, and with a lot of effort one of the real solutions to $1-x-\frac{x^2}{2!}+\frac{x^4}{4!}=0$ is about $0.739219$, while the real solution to $\cos(x)=x$ is about $0.739085$ $\endgroup$ – Henry Apr 7 '18 at 16:20
  • $\begingroup$ You can do the same with Newton's method. If $f(x)$ is a convex, increasing function over the interval $[a,b]$ with a root $\zeta\in(a,b)$, Newton's method with starting point $b$ converges to $\zeta$ quadratically and monotonically. $\endgroup$ – Jack D'Aurizio Apr 7 '18 at 17:12

You can find a zero of $f(x)=\cos(x)-x$ to arbitrary precision, and have an upper bound of your error estimate, using the bisection method.

We know $f(0)=1>0$, $f(1)<0$, so we evaluate $f(1/2)$. We find $f(1/2)>0$, so we change our interval of interest from $[0,1]$ to $[1/2,1]$ and continue by evaluating $f(3/4)$. It turns out $f(3/4)<0$, so we are now in $[1/2,3/4]$.

  • $\begingroup$ Thanks, I didn't realize there was a root finding method this simple that could be calculated to an arbitrary precision! $\endgroup$ – Ozaner Hansha Apr 7 '18 at 16:15

First of all if you plan to use Taylor series you need to modify your equation to $$\sum_{n=0}^{\infty} \frac{(-1)^nx^{2n}}{(2n)!}= 1-\frac{x^2}{2!} + \frac{x^4}{4!} - \frac{x^6}{6!} \cdots=x$$ Which is not easy to solve for x.

You may try other methods such as bisection method.


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.