8
$\begingroup$

I need to calculate a limit of a complex expression (had it in a physics research) that contains a term $(r-b)^p$ for $r\rightarrow b+$ where $r,b$ are reals, and $p$ is complex, let's suppose for simplicity $p=i$.

At the beginning, I put it equals to $0$ as an obvious thing, but strangely enough, "Mathematica" gives me something strange: $e^{2i\operatorname{Interval}[0,\pi]}$, then I thought maybe I was wrong with "obviousity" of the result, because we can actually put (formally) $0=e^{-\infty+i\phi}$, then if we use $a^b=e^{b\log a}$ we have $0^i=e^{-i\infty-\phi}$ (where all $2\pi k$ included in the infinity or $\phi$).

This result seems to be very strange for me (it will introduce a new parameter $\phi$ into my theory), and I found no example for this, especially that I know one needs to be careful with branches when making such tricks, but I have no clue if that is correct or wrong, any help will be appreciated.

$\endgroup$
7
$\begingroup$

The problem is that $x^i$ is only defined as $e^{i\log x}$.

When $x$ gets close to zero, $\log x$ is of the form $a+bi$ where $a$ is a very large negative number and $b$ can be restricted to $(-\pi,\pi]$. Then $e^{i\log x}$ is $e^{-b} e^{ia}$.Note then that if we let $x$ approach zero along a line from one side at angle $\theta$, that means the modulous of $e^{i\log x}$ will be $e^{-\theta}$, which is constant, so it does not approach zero (and it does not even converge - $x^i$ basically whizzes around a circle when $x\to 0$ along a straight line.)

Exponentiation $x^u$ near $x=0$ just is not well-behaved.

And this doesn't even take into account that $\log x$ is more naturally a multi-valued function.

Even restricting $x$ to $\mathbb R^+$ with $\log x$ the normal real natural logarithm, $$x^i=e^{i\log x}$$

So $|x^i|=1$, and as $x\to 0+$, $x^i$ spins around the unit circle clockwise and certainly does not have a limit.

$\endgroup$
  • $\begingroup$ Maybe it was my mistake, I didn't mentioned that $r\rightarrow b+$ and $r,b$ are reals, then your $\theta=0$, then dose my result is right? $\endgroup$ – TMS Feb 20 '13 at 17:45
  • $\begingroup$ sorry for the delay, please note that in my question, I mentioned that I included multi-value of $\log$ in the angels for purpose (for that I didn't put the restriction $(-\pi,\pi]$ as you did, also this means that $\theta!=0$ even for real $x$ ), moreover, if you calculate something like $e^{i(\log x +i2\pi)}$ for $x=0.0001$ as example, you will find that for this brunch $x^i\rightarrow 0$ really! also we get undefined limit for something like $\sin \infty$ which has no meaning, but here we get a parametrized limit, and I'm note sure that we can consider it as undefined! $\endgroup$ – TMS Feb 21 '13 at 17:52
  • $\begingroup$ No, it doesn't. Basically, if you choose a continuous value of $\log x$ for $x$ real - that is $\log x = a + 2\pi n i$ for some fixed $n$ - then $e^{i\log x} = e^{-2\pi n} e^{ia}$. Now, $e^{-2\pi n}$ is constant, and $|e^{ia}|=1$. So you again are circling around a circle, just a circle with a different radius, as $x\to 0$. Now, if you let $n$ vary for each $x$, making $\log x$ discontinous, you could get $x^i\to 0$, but you could also get $x^i\to\infty$. @TMS $\endgroup$ – Thomas Andrews Feb 21 '13 at 18:17
  • $\begingroup$ I found my problem, it was Mathematica bug that mislead me, you have been right! thank you for your help. $\endgroup$ – TMS Feb 21 '13 at 18:32
1
$\begingroup$

The complex exponential $(x+iy)^p$ is not defined uniquely. It is $(x+iy)^p = \exp(p\log(x+iy))$ where the logarithm is defined up to a multiple of $2\pi i$. Hence your formula is not uniquely determined and you cannot say that it tends to zero...

$\endgroup$
  • $\begingroup$ interesting point ! Since log(x) has many sheets, you can't walk around the origin without jumping from one sheet to another. $\endgroup$ – user28921 Feb 21 '13 at 18:55
  • $\begingroup$ @user28921: $x$ does not walk around the origin. It approaches $0$ from the positive side. $\endgroup$ – robjohn Feb 21 '13 at 21:36
1
$\begingroup$

To make $\ln z$ unique, one needs to slit the complex plane and it is common to use only the main branch on $\mathbb C\setminus\mathbb R_{\le0}$. As in your specific problem $r,b$ are always reals and $r>b$, we can write $r-b=e^t$ with $t\in \mathbb R$ and $t\to-\infty$. Then $(r-b)^p=e^{pt}$. If $\Re p=0$ (but $\Im p\ne 0$) this indeed keeps rotating around on the unit circle so that $\lim_{r\to b^+}(r-b)^p$ does not exist. If $\Re p>0$, however, then $|(r-b)^p|=|e^{pt}|=e^{t\Re p}\to 0$. And if $\Re p<0$, then $|(r-b)^p|\to\infty$. In both these latter cases we still have that rotation phenomenon if $\Im p\ne 0$.

$\endgroup$

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.