### Author Topic: Week 13 -- BP3, 4, 5, 6  (Read 1429 times)

#### Victor Ivrii

• Elder Member
• Posts: 2511
• Karma: 0
##### Week 13 -- BP3, 4, 5, 6
« on: April 04, 2018, 12:34:15 PM »
$\renewcommand{\Re}{\operatorname{Re}}$
There are several connected bonus problems:
a.
Let $(x\pm i0)^{\nu}= \lim _{\varepsilon\to +0} (x\pm \varepsilon i)^{\nu}$. Prove, it exists for $\Re\nu >-1$.

b. Using equality

[(x\pm \varepsilon i)^{\nu}]'= \nu (x\pm \varepsilon i)^{\nu-1},
\label{eq-11.1.12}

prove that these limits exist (in the sense of distributions) for  $\nu \ne -1,-2,\ldots$.

c. Using

[\ln (x\pm \varepsilon i) ]'=
(x\pm \varepsilon i)^{-1},
\label{eq-11.1.13}

prove that these limits exist (in the sense of distributions) for  $\nu =-1$. Then, using (\ref{eq-11.1.12}), prove that these limits exist (in the sense of distributions) for $\nu=-2,-3,\ldots$.

d. Prove that
\begin{align}
&(x-i0)^{-1} + (x+i0)^{-1}=2 x^{-1},
\label{eq-11.1.14}\\
&(x-i0)^{-1}- (x+i0)^{-1}=2\pi i \delta(x)
\label{eq-11.1.15}
\end{align}
with integral understood in the sense of principal value, use (\ref{eq-11.1.13}) and
\begin{align}
&\ln (x+i0)+\ln (x-i0)= 2\ln |x|,\\
&\ln (x+i0)-\ln (x-i0)= -2\pi i\theta(-x).
\end{align}
You need to justify these formulae.
« Last Edit: April 08, 2018, 01:55:23 PM by Victor Ivrii »

#### Andrew Hardy

• Full Member
• Posts: 34
• Karma: 10
##### Re: Week 13 -- BP3, 4, 5, 6
« Reply #1 on: April 07, 2018, 04:10:16 PM »
a) Consider
$$(x\pm i0)^{\nu}= f'$$
Now via definitions $$(f',\theta) = - (f,\theta')$$ and then f is defined as
$$\frac{1}{\nu+1}(x \pm i0)^{\nu+1} = f$$ This  exists for  $\Re\nu >-1$. Therefore the limit exists for this condition.
b) Extending the proof from part a, define $$(x\pm i0)^{\nu}= f^{(n)}$$ then with the same rules of distribution derivatives I can write $$[(x\pm \varepsilon i)^{\nu}]^{(n)}= (\nu ! )(x\pm \varepsilon i)^{\nu-n}$$ which is then defined for corresponding $\Re\nu \neq -n$ where  $n = 2, 3,\ldots$
« Last Edit: April 09, 2018, 09:20:18 AM by Andrew Hardy »

#### Victor Ivrii

• Elder Member
• Posts: 2511
• Karma: 0
##### Re: Week 13 -- BP3, 4, 5, 6
« Reply #2 on: April 07, 2018, 04:30:05 PM »
One does not need derivatives to prove that the limit (in $L^1$) $\lim \varepsilon\to +0 (x\pm i\varepsilon)^\nu$ exists and equals
|x|^\nu \left\{\begin{aligned} &1 &&x>0,\\ &e^{\pm i\pi \nu} && x<0, \end{aligned}\right.
provided $\Re\nu >-1$ because then this function belongs to $L^1$. The fin starts from (b)
« Last Edit: April 08, 2018, 01:51:16 PM by Victor Ivrii »

#### Zhongnan Wu

• Jr. Member
• Posts: 5
• Karma: 2
##### Re: Week 13 -- BP3, 4, 5, 6
« Reply #3 on: April 08, 2018, 12:50:30 PM »
I think there is a type of question c). Should be [ln(x+epsilon i)]', here the power of v is 1 and should not be shown.
Also for the second half part of this question, should we use the equality from b) rather than from a)?

#### Jingxuan Zhang

• Elder Member
• Posts: 106
• Karma: 20
##### Re: Week 13 -- BP3, 4, 5, 6
« Reply #4 on: April 08, 2018, 01:00:33 PM »
Zhongnan,
I think c) is alright as the way it stands. After all \eqref{eq-11.1.12} is deduced from \eqref{eq-11.1.13} so that is also fine...

#### Victor Ivrii

• Elder Member
• Posts: 2511
• Karma: 0
##### Re: Week 13 -- BP3, 4, 5, 6
« Reply #5 on: April 08, 2018, 01:54:42 PM »
Andrew, you right with powers but losing numerical factor.
Zhongnan  is right (copy-paste-not-clening-up error). I corrected
« Last Edit: April 08, 2018, 01:58:22 PM by Victor Ivrii »

#### Andrew Hardy

• Full Member
• Posts: 34
• Karma: 10
##### Re: Week 13 -- BP3, 4, 5, 6
« Reply #6 on: April 09, 2018, 09:21:53 AM »
Dr. Ivrii? Are Part a and b correct now? In Gel'fand's "Normalized Functions" he has what seems the same argument, but rather does the integrals explicitly. I can go that in depth if I haven't earned the points yet. I'm stilling looking for c and d.

#### Andrew Hardy

• Full Member
• Posts: 34
• Karma: 10
##### Re: Week 13 -- BP3, 4, 5, 6
« Reply #7 on: April 09, 2018, 09:37:19 AM »
d) *from Gel'fand*
Equations 5 and 6 come from the natural log of  the generalized expression
$$(x+iy)^{\nu} = \exp[\nu \log(x+iy) + iArg(x+iy)] \tag{A}$$  which is just the representation on the complex plane
Then we can find
$$(x+i0)^{-\nu} = x^{-\nu} - \frac{i\pi(-1)^{\nu-1}}{(\nu -1)!}\delta^ {(nu -1)}(x) \tag{B}$$
$$(x-i0)^{-\nu} = x^{-\nu} + \frac{i\pi(-1)^{\nu-1}}{(\nu -1)!}\delta^ {(nu -1)}(x) \tag{C}$$
and by combining these and setting $\nu =1$ we arrive at
$$(x-i0)^{-1} + (x+i0)^{-1}=2 x^{-1} \tag{D}$$
$$(x-i0)^{-1}- (x+i0)^{-1}=2\pi i \delta(x) \tag{E}$$
« Last Edit: April 09, 2018, 10:08:43 AM by Victor Ivrii »

#### Victor Ivrii

• Elder Member
• Posts: 2511
• Karma: 0
##### Re: Week 13 -- BP3, 4, 5, 6
« Reply #8 on: April 09, 2018, 10:54:40 AM »

(A): multiple errors: it should be $=\exp \bigl[\nu \bigr( \log|x+iy|+i\arg (x+iy)\bigr)\bigr]$. First factor $\nu$ is applied to everything second, we do not allow $(x+iy)$ to cross $(-\infty,0]$; so angle is defined uniquely and it rans from $-\pi$ to $\pi$.
Anyway: $[(x\pm i \varepsilon)^\nu]' =\nu (x\pm i \varepsilon)^{\nu-1}$ from complex variables and here $\nu$ could be even complex.

As $\varepsilon\to +0$, there is a limit in $\mathscr{D}'$ (even in $L^1_{loc}$) of $(x\pm i \varepsilon)^\nu$ provided $\Re\nu >-1$; but then there exists a limit of its derivative, i.e. $\nu (x\pm i \varepsilon)^{nu-1}$ and we can divide by $\nu\ne 0$. So we defined $(x\pm i \varepsilon)^{\nu}$ as long as $\Re \nu >-2$ and $\nu \ne -1$.

Repeating, we define $(x\pm i \varepsilon)^{\nu}$ as long as $\Re \nu >-3$ and $\nu \ne -1,-2$. ... and so on... as $\nu\ne -1,-2,\ldots$.

Remark. To mitigate the latter restriction, $f_\nu ^\pm :=\frac{(x\pm i0)^\nu}{\Gamma(\nu+1)}$ could be considered where $\Gamma$ is Euler's $\Gamma$-function; it has simple poles at $0,-1,-2,\ldots$ and $\Gamma(\nu+1)=\nu\Gamma(\nu)$.

(B), (C) are out of the window: they do not follow; also what is $x^{-\nu}$ for $x<0$ and $\nu\notin\mathbb{Z}$? What is $\delta^{\nu-1}$ for $\nu\ne 1,2,\ldots$? We can define those but a posteriori.

I suggested a simple way: look at $\log (x\pm i0)$ as $x>0$ and $x<0$; obviously $\log (x\pm i0)=\log|x|$ as $x>0$ and $\log (x\pm i0)=\log|x|\pm i\pi$ as $x<0$. In other words $\log (x\pm i0)=\log |x| \pm i\pi \theta(-x)$. Differentiating in $\mathscr{D}'$ we get
$$(x\pm i0)^{-1}= (\log |x|)' +\pm i \pi (\theta (-x))'=x^{-1} \mp i\pi \delta(x)$$
where (see other of bonus problems, http://forum.math.toronto.edu/index.php?topic=1167.0) $(\log |x|)' =x^{-1}$ in vp sence, and
$(\theta (-x))'=-\delta(x)$.

PS Gelfand--Shilov 1--6 (+coauthors in higher volumes) is a truly remarkable book, but IMHO, sometimes they go too far. F.e. considering F.T. of distributions not in $\mathscr{S}'$ they get distributions over some classes of the entire analytic functions , in particular they get $\delta (x-c)$ with any $c\in \mathbb{C}$ (which is definitely a perversion). Unfortunately, G.Shilov died too young (at 58) and I never met him.  I.Gelfand was a great mathematician.

« Last Edit: April 09, 2018, 10:57:23 AM by Victor Ivrii »