2
$\begingroup$

Find expansions of all the real roots of $$x\tan(x)=\epsilon?$$ (You have to consider the first root separately)

It is really bothering me. If I assume $x=x_0+x_1\epsilon +x_2\epsilon^2$ and do the taylor expansion of tan(x). Then I end up with $$(x_0+x_1\epsilon +x_2\epsilon^2)[(x_0+x_1\epsilon +x_2\epsilon^2)+1/3 (x_0+x_1\epsilon +x_2\epsilon^2)^3+...]=\epsilon$$.

From the above equation $x_0=0$, then no matter what value $x_1$ is, the coefficient of $\epsilon$ on the LHS is always $0$.

Could anyone kindly give me some hints? Thanks!

$\endgroup$

2 Answers 2

6
$\begingroup$

Your method is valid for the roots away from $x=0$, but you missed a bunch of roots of $\tan$ that lead to non-trivial solutions. If $\epsilon=0$, then the solutions are $x=k\pi$ for any integer $k$.

The case $x=0$ is special, because $x\tan(x)$ has slope 0 here, and hence a single solution for $\epsilon=0$, but as soon as $\epsilon$ is not zero, that root splits into two roots, one in $(-\pi/2,0)$ and the other in $(0,\pi/2)$, so it's a singular perturbation problem. I'll do this case after the $x\neq0$ case.

For the $x\neq 0$ case, you have $x=k\pi+\epsilon x_1+\epsilon^2x_2+\cdots$, and you know that $\tan(x)$ will be small because $\tan(k\pi)=0$ and $\tan$ is continuous. So, expand $\tan(k\pi+\epsilon x_1+\epsilon^2x_2+\cdots)$ around $k\pi$:

$$ \tan(k\pi+\epsilon x_1+\epsilon^2x_2+\cdots)=\tan(k\pi)+\frac{\epsilon x_1+\epsilon^2x_2+\cdots}{\cos^2(x_0)}-(\epsilon x_1+\epsilon^2x_2+\cdots)^2\frac{2\tan(x_0)}{\cos^2(x_0)}+O(\epsilon^3) $$ where I've gone up to $O(\epsilon^2)$. Putting in $x_0=k\pi$ and truncating $O(\epsilon^3)$ terms gives $$ \tan(k\pi+\epsilon x_1+\epsilon^2x_2+\cdots)=\epsilon x_1+\epsilon^2x_2+O(\epsilon^3) $$ and now we are looking to solve $$ (k\pi+\epsilon x_1+\epsilon^2x_2)(\epsilon x_1+\epsilon^2x_2)=\epsilon $$ the $O(\epsilon)$ equation is $$k\pi x_1=1\Rightarrow x_1=\frac{1}{k\pi}. $$ Then the $O(\epsilon^2)$ equation is $$x_1^2+k\pi x_2=0\Rightarrow x_2=-\frac{x_1^2}{k\pi}=-\frac{1}{(k\pi)^3}. $$

So for the roots away from $x=0$ you get $$x=k\pi+\frac{\epsilon}{k\pi}-\frac{\epsilon^2}{(k\pi)^3}+O(\epsilon^3).$$

For the roots near 0, we already determined it was a singular perturbation problem, so the standard expansion $x=x_0+\epsilon x_1+\epsilon^2x_2+\cdots$ wont work. Instead, let's search for an expansion of the for $x=\epsilon^\alpha(x_0+\epsilon x_1+\epsilon^2x_2+\cdots)$. Use the Taylor series expansion of $tan$ to give $$ x\tan(x)=\epsilon^\alpha(x_0+\epsilon x_1+\epsilon^2x_2+\cdots)\left(\epsilon^\alpha(x_0+\epsilon x_1+\epsilon^2x_2+\cdots)+\frac{1}{3}\epsilon^{3\alpha}(x_0+\epsilon x_1+\epsilon^2x_2+\cdots)^3+O(\epsilon^{5\alpha})\right) $$ which can be simplified to $$ x\tan(x)=\epsilon^{2\alpha}\left((x_0+\epsilon x_1+\epsilon^2x_2+\cdots)^2+\frac{1}{3}\epsilon^{2\alpha}(x_0+\epsilon x_1+\epsilon^2x_2+\cdots)^4+O(\epsilon^{4\alpha})\right) $$ Now since $x\tan(x)=\epsilon$, we write $$ \left((x_0+\epsilon x_1+\epsilon^2x_2+\cdots)^2+\frac{1}{3}\epsilon^{2\alpha}(x_0+\epsilon x_1+\epsilon^2x_2+\cdots)^4+O(\epsilon^{4\alpha})\right)=\epsilon^{1-2\alpha} $$ now the largest terms of the left hand side are $O(1)$, so to ensure the leading order solution is non-trivial we need the left hand side to be $O(1)$, and so $1-2\alpha=0\Rightarrow\alpha=1/2$.

Now our equation to solve is $$ \left((x_0+\epsilon x_1+\epsilon^2x_2+\cdots)^2+\frac{1}{3}\epsilon(x_0+\epsilon x_1+\epsilon^2x_2+\cdots)^4+O(\epsilon^{2})\right)=1 .$$

The $O(\epsilon^0)$ equation is $$x_0^2=1\Rightarrow x_0=\pm1.$$ The $O(\epsilon)$ equation is $$2x_0x_1+\frac{1}{3}x_0^4\Rightarrow x_1=\pm\frac{1}{6}.$$

So the first two terms in solution for the roots near zero are $$ x=\epsilon^{1/2}-\frac{1}{6}\epsilon^{3/2},\quad\text{and}\quad x=-\epsilon^{1/2}+\frac{1}{6}\epsilon^{3/2}.$$

So we have $$x=k\pi+\frac{\epsilon}{k\pi}-\frac{\epsilon^2}{(k\pi)^3}+O(\epsilon^3),\quad k\in\mathbb Z,\quad k\neq0$$ and $$ x=\epsilon^{1/2}-\frac{1}{6}\epsilon^{3/2}+O(\epsilon^{5/2}),\quad\text{and}\quad x=-\epsilon^{1/2}+\frac{1}{6}\epsilon^{3/2}+O(\epsilon^{5/2}).$$

(Barring any arithmetic errors I have probably made! Although what I've posted seems to be alright looking at some plots.)

$\endgroup$
5
  • $\begingroup$ @Why can we assume $x=\epsilon^\alpha(x_0+\epsilon x_1+\epsilon^2x_2+\cdots)$? Then the case $x=x_0+\epsilon^{1/2} x_1+\epsilon x_2+\cdots$ is not considered? $\endgroup$
    – Sherry
    Apr 18, 2015 at 1:44
  • $\begingroup$ Narasimham's answer deals with that issue very neatly. Basically, the largest term on the LHS (that is, the largest term in the expansion of $x\tan(x)$), should balance with the RHS, so the lowest power of $\epsilon$ on the LHS should be 1 (the same as the lowest power of $\epsilon$ on the RHS). $\endgroup$
    – David
    Apr 18, 2015 at 4:20
  • $\begingroup$ I mean the case $x=\epsilon^{1/2} x_1+\epsilon x_2+\cdots$, where the ratio of second and first is $\epsilon^{1/2}$ instead of $\epsilon$? $\endgroup$
    – Sherry
    Apr 18, 2015 at 5:35
  • $\begingroup$ Sorry for the inconvenience. I mean why the ratio of first and second must be $\epsilon$? Can the expansion be $\epsilon ^{1/2}x_1+\epsilon x_2+ \epsilon^{3/2}x_3$? While the form $\epsilon^{\alpha}(x_0+\epsilon x_1+\epsilon^2 x_2+ \epsilon^{3}x_3)$ doesn't consider the above case? $\endgroup$
    – Sherry
    Apr 18, 2015 at 5:50
  • $\begingroup$ Yes you can't assume that each term will differ by $\epsilon$, but since it works I went with it. I'm not aware of any good way to pick the ratio either, I'd be interested to know too! $\endgroup$
    – David
    Apr 18, 2015 at 8:36
2
$\begingroup$

For small x, i.e., $ x << \pi/2 , x \tan x \approx x^2 \rightarrow ,x =\sqrt \epsilon $.

Considering co-terminal angles the roots are $ x =\sqrt \epsilon + k\; \pi . $

Please note that only small angle approximation near $x= k \pi$ is involved here

but not asymptotic behavior valid nearer $x= (2 k-1) \pi/2 \;! $

$\endgroup$

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.