Consider the following transcendental equation:

$$\phi = \beta \sin \phi . \qquad (*)$$

How does one generate a description of how $\phi$ depends on $\beta$?

My attempt

From inspection (i.e. drawing pictures) one finds that for $\beta<1$ there is a single solution: $\phi=0$. At $\beta=1$ the solution at $\phi=0$, $\beta=0$ splits into three solutions, and then as $\beta$ increases additional solutions continue to appear in pairs.

The only technique I know is to assume that $\phi$ is implicitly given as a function of $\beta$. One then writes this function as a Taylor series and solves for the coefficients up to desired order.

Supposing $\beta$ is a function of $\phi$ we write a Taylor series $$\beta(\phi) = a\phi + b\phi^2 + \cdots.$$ Using this and expanding the $\sin$ function in $(*)$ we find

$$ \begin{align} \phi &= (a\phi + b\phi^2 + \cdots)(\phi - \phi^3/6 \cdots) \\ \phi &= a\phi^2 + b\phi^3 - \frac{a}{6}\phi^4 - \frac{b}{6}\phi^5 + \cdots \end{align} $$

There are no solutions other than $\phi=0$, $\beta=0$, so we have not captured the new pair of solutions which appears as $\beta$ increases past 1. This is strange. The Taylor series could have failed because $d\beta/d\phi$ is diverging at our expansion point. To check this, differentiate $(*)$ with respect to $\phi$ to find $$\frac{d\beta}{d\phi} = \frac{1 - \beta(\phi) \cos\phi}{\sin\phi}.$$ As $\phi$ approaches 0 from the right, we have $d\beta/d\phi = (1-\beta(\phi))/\phi$. This may or may not diverge; it depends on the functional form of $\beta(\phi)$.

The same ambiguity shows up if one instead regards $\phi$ as an implicit function of $\beta$.

From a numerical solution I can see that in fact $d\phi/d\beta = \infty$ at $\beta=1$. That suggests that the failure of the Taylor series for $\beta$ in terms of $\phi$ was not due to diverging derivative. This suggests that $\beta$ cannot be written as a Taylor series in $\phi$ for a different reason, e.g. fractional powers are needed. How then, does one proceed?


You are essentially looking for an inverse to the $\operatorname{sinc}$ function ($\operatorname{sinc}(z)=\frac{\sin z}{z}$), since then you could write $\frac{1}{\beta}=\frac{\sin\phi}{\phi}\implies\phi=\operatorname{sinc}^{-1}\left(\frac{1}{\beta}\right)$. But see this SE post for people's thoughts on that.

You can generally find solutions to equations like this to arbitrary precision using numerical methods though.

  • $\begingroup$ Any idea why the Taylor series didn't work? $\endgroup$ – DanielSank Oct 11 '14 at 16:44
  • $\begingroup$ For several reasons. Starting small, you aren't using the correct series for $\sin$ (alternating terms). But that doesn't matter because there are bigger issues. The graph of $\operatorname{sinc}$ should make it clear that $\operatorname{sinc}$ is not invertible on $(-a,a)$. It is, however, invertible on $[0,a)$ if $a$ doesn't get too large. But it's derivative at $0$ is $0$, so it's inverse at $y=1$ would be vertical. So if you set out to find a Taylor series for $\operatorname{sinc}^{-1}$ centered at $0$, you will not be able to find one that converges beyond radius $1$. $\endgroup$ – alex.jordan Oct 11 '14 at 18:19
  • $\begingroup$ Actually, perhaps the biggest issues is that you will only have one branch of a function that has infinitely many branches. $\endgroup$ – alex.jordan Oct 11 '14 at 18:38
  • $\begingroup$ Sure, but suppose I only care about following one branch's evolution as $\beta$ increases. For $\beta<1$ there is one solution, $\phi=0$. Then at $\beta=1$ this bifurcates into three branches. One remains at $\phi=0$ and the other two are mirrored about the origin. Suppose I just care about following one of those branches... $\endgroup$ – DanielSank Oct 11 '14 at 18:57
  • $\begingroup$ How do you get from $\phi=a\phi^2 + b\phi^3 - \frac{a}{6}\phi^4 - \frac{b}{6}\phi^5 + \cdots$ to "There are no solutions other than $\phi=0$"? Depending on $a,b,$ etc. there could be more solutions. And for $a,b,$ etc. that do give a local Taylor series for other branches of $\operatorname{sinc}^{-1}$, there would be other solutions. Also, those other branches' Taylor series would have a nonzero constant term. $\endgroup$ – alex.jordan Oct 11 '14 at 19:41

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.