2
$\begingroup$

In the text "Basic Complex Analysis" Third Edition by Jerrold E. Marsden and Michael J. Hoffman I'm inquiring if there's an alternate way through Complex-Analysis to evaluate $\text{Example (4.38)}$ ? For those who don't have the textbook on hand I've written the approach the authors have taken.

$\text{Example (4.3.8)}$

Let $\omega$ be a nonzero real constant and evaluate,

$$\int_{-\infty}^{\infty} \frac{e^{i \omega x}}{1 + ix}dx.$$

$\text{Solution}$

The author notes that this is an integral of the Fourier transform type with our choice of $f$ being $1/(1+iz)$. The author notes along the real axis one can note that

$(1)$

$$|g(x)| = |e^{-i \omega z} | / | 1 + ix| = 1 / \sqrt{1 + x^{2}}.$$

It's rather trivial to see that,

$$ \int_{-\infty}^{\infty} | 1 / \sqrt{1 + x^{2}} | dx < \infty.$$

$\text{Commentary (1.0)}$

When the author noted, that the integral was of a Fourier Transform type it seems our problem boils down to evaluating the respective integrals in the form,

$(2)$

$$\frac{1}{\sqrt{2 \pi}} \int_{-\infty}^{\infty}f(z)e^{-iwz}dz$$

Or alternatively of the latter integrals of the form,

$$\int_{-\infty}^{\infty}f(z)e^{-iwz}dz $$

$\text{Remark}$

An important detail one shouldn't miss is that our choice of $f$ is seen as $f(z) = q(z)/p(z)$

To obtain convergence of our integral, the author makes note that,

$(2)$

$$\int_{-\infty}^{\infty} \frac{e^{i \omega x}}{1 + ix}dx = \lim_{A \rightarrow \infty, B \rightarrow \infty} \int_{-A}^{B}\frac{e^{-i \omega x}}{1 + ix}dx.$$

$\text{Commentary (1.1)}$

This development rather is a key ingredient since convergence is a necessary condition also one can work out that,

\begin{align*} \lim_{A \rightarrow \infty, B \rightarrow \infty} \int_{-A}^{B}\frac{e^{-i \omega x}}{1 + ix}dx&= \lim_{A \rightarrow \infty} \bigg( \int_{-A}^{B} f \, \bigg) dx \, + \lim_{B \rightarrow \infty} \bigg( \int_{-A}^{B} f \bigg)dx\\ &= \int_{-\infty}^{B} \big(f\big) \, dx + \int_{-A}^{\infty} \big(f) \, dx \end{align*}

The author notes that since $|f(z)| = 1/|1+iz| \leq 1/ (|z|-1)$ for $|z| > 1$, this factor does not shrink towards $0$. For each $\epsilon > 0$,there is an $R(\epsilon)$ such that $|f(z)| < \epsilon$ whenever $|z| \geq R(\epsilon)$. Also the author notes the exponential factor will play nice on the half plane. But which plane in question depends on the sigh of $\omega$. If $z = x+iy$ with $x$ and $y$ real, then $|e^{i \omega z}| = e^{i \omega x + \omega y} = e^{wy}$. Therefore, $\omega < 0$ implies $|e^{i \omega z} | = e^{ \omega y} \leq 1 $ in the upper half plane $\mathcal{H.}$ In a similar fashion one can say that,$\omega < 0$ implies $|e^{i \omega z} | = e^{ \omega y} \leq 1$ in the lower half plane $\mathcal{L.}$

$\text{Commentary (1.2)}$

Witnessing these ocean of developments it seems the location of the behavior depending on the sign of the exponential factor of our choice in $f$ in the planes $\mathcal{L}$ or $\mathcal{H.}$ Also as a consequence of this some of the analytic properties of our choice $f$ become restricted to either $\mathcal{L}$ or $\mathcal{H.}$ Immediate properties that come to mind are the locations of $\operatorname{Res}(f)$, where $f$ is analytic, etc. This intuition seems to be valid since the author calculates the residue's as follows

$$\operatorname{Res}\bigg(\operatorname{\frac{e^{i \omega z}}{i(z-i)}; i}\bigg) \bigg|_{z = i} = \frac{e^{\omega}}{i}.$$ The furthermore sets that,

$$\Sigma_{\mathcal{H}} = \text{the sum of the residues in } \mathcal{H}= \frac{e^{\omega}}{i}$$

$$ \Sigma_{\mathcal{L}} = \text{the sum of the residues in } \mathcal{L}=0.$$

After dealing with the location of our Residue's considers that $A$ and $B$ are both larger that $R(\epsilon)$ and larger that $1$, we can consider the rectangular paths. The rectangular contours that are considered are displayed in $\text{Figure (4.3.2)}$. He then goes on to mention that $\gamma$ is the segment of the real axis from $-A$ to $B$. We have a two closed rectangular contours denoted by $\Gamma = \gamma + \mu_{1} + \mu_{2} + \mu_{3}$ counterclockwise through the upper half plane and as a rectangle $\wedge = \gamma + \nu_{1} + \nu_{2} + \nu_{3}.$ In each case the distance $C$ the real axis will be selected larger than $R(\epsilon)$ depend on $A$ and $B$. So one will have,

$(4)$

$$ \oint_{\Gamma} g = 2\pi i \Sigma_{\mathcal{H}} = 2 \pi e^{\omega} \text{and} \oint_{\wedge} g = -2\pi i \Sigma_{\mathcal{L}}= 0. $$

Thus,

$(5)$

$$\int_{-A}^{B} \frac{e^{i \omega x}}{1 + ix}dx = \oint_{\gamma} g = 2 \pi i \Sigma_{\mathcal{H}} \! \! \! \! \! \! \! \! \! \! \! \! \! \! \! \! \, \, \, \, \, \, \, \, \, -\bigg( \oint_{\mu_{1}} g + \oint_{\mu_{2}} g + \oint_{\mu_{3}} g\bigg) $$

$$ \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, = -2 \pi i \Sigma_{\mathcal{L}\quad} \! \! \! \! - \bigg( \oint_{\nu_{1}}g + \oint_{\nu_{2}}g + \oint_{\nu_{2}}g \bigg)$$ enter image description here

$$\text{Figure (4.3.2): Paths for Example (4.3.8)}$$

$\text{Commentary (1.3)}$

The author supposes that $\omega < 0$, in the case that we have favorable behavior in $\mathcal{H}\quad$ Along $\mu_{1}$, also see that $z = B + iy$, so we have that

$(6)$

\begin{align*} \lim_{\epsilon \rightarrow \infty} \bigg| \oint_{\mu_{1}}g\bigg| &= \bigg| \int_{0}^{C} f(z)e^{i \omega z}dz\bigg| \\ &= \bigg| \int_{0}^{C} f(-A+iy)e^{-i \omega(-A+iy)} idy\bigg| \leq \int_{0}^{C}\epsilon e^{\omega y}dy \leq \frac{r}{|\omega|} \rightarrow 0. \end{align*}

In order to get our desired bound over $\mu_{1}$, one has to note that,

$$\frac{e^{i \omega(B+iy)}}{1+iz} \leq \frac{i}{i}\bigg| \frac{e^{(i \omega \beta + i\omega y)}}{1 + z} \bigg| \leq \frac{i\epsilon}{i} \leq \epsilon.$$

The author notes a very similar process over $\mu_{3}$ hence,

$(7)$

\begin{align*} \lim_{\epsilon \rightarrow \infty} \bigg| \oint_{\mu_{3}}g\bigg| &= \bigg| \int_{C}^{0} f(z)e^{i \omega z}dz\bigg| \\ &= \bigg| \int_{C}^{0} f(B+iy)e^{-i \omega(B+iy)} idy\bigg| \leq \int_{C}^{0}\epsilon e^{\omega y}dy \leq \frac{\epsilon}{|\omega|} \rightarrow 0. \end{align*}

Now for the contribution over $\mu_{3}$ it's a very similar however some adjustments were made, the strict requirement is posed that $C > R(\epsilon)$ this is initially done since due to the location of $C$ it's contribution has to be small or else our estimates will be invalid. We have

$(8)$

\begin{align*} \lim_{\epsilon \rightarrow \infty} \bigg| \oint_{\mu_{3}}g\bigg| &= \bigg| \int_{B}^{-A} f(z)e^{i \omega z}dz\bigg| \\ &= \bigg| \int_{-A}^{B} f(x+iC)e^{-i \omega(x+iC)} idy\bigg| \leq \int_{C}^{0}\epsilon e^{C \omega}dy \leq \epsilon(A+B)e^{\omega C} \rightarrow 0. \end{align*}

$\text{Commentary $$(1.4)}$

The author remembers that $\omega > 0$, one can select $C$ such that $C > 1$, and also that $C > R(\epsilon)$ and large enough so that $(A + B)e^{- \omega C} < 1$. We find that for $A$ and $B$ larger then $(1)$ He discovers the bound by noting that $A$ and $B$ are respectively larger than $(1)$ and larger then $R(\epsilon)$ so finally we have,

$$\lim_{\epsilon \rightarrow \infty}\bigg| \int_{-A}^{B}g(x)dx + 2 \pi i \Sigma_{\mathcal{L}\quad} \bigg| \leq \sum_{\psi} \bigg|\oint_{\nu_{\psi}} g\bigg| \leq \bigg(\frac{2}{| \omega|} + 1 \bigg) \epsilon \rightarrow 0. $$

$\endgroup$
1
$\begingroup$

What's being done by the author here seems—to me at least—to be over-complicated. We can evaluate this integral in a much simpler manner through application of Jordan's lemma.

To evaluate

$$I(\omega)=\int_{-\infty}^\infty\frac{e^{i\omega x}}{1+ix}dx$$

we will consider a semicircular contour $\gamma_R$ of radius $R>1$ in the upper half of the complex plane like so:

                                                              Our Contour

We can think of our integral along $\gamma_R$ as the sum of two integrals

$$\oint_{\gamma_R}\frac{e^{i\omega z}}{1+iz}dz=\int_{C_R}\frac{e^{i\omega z}}{1+iz}dz+\int_{-R}^R\frac{e^{i\omega z}}{1+iz}dz\tag{1}$$

where $C_R$ is the curved portion of $\gamma_R$. If we take the limit $R\rightarrow\infty$ we see that

$$\lim_{R\rightarrow\infty}\int_{-R}^R\frac{e^{i\omega z}}{1+iz}dz=I(\omega).$$

But this is not enough for our purposes, if we are to evaluate $I(\omega)$ by the residue theorem we must also have

$$\lim_{R\rightarrow\infty}\int_{C_R}\frac{e^{i\omega z}}{1+iz}dz=0\tag{2}$$

as we want the contribution along the arc to tend to $0$.

 

To show that $(2)$ happens we may apply Jordan's lemma. We consider $\omega>0$, then applying the lemma gives

$$\begin{align*} \left | \int_{C_R}\frac{e^{i\omega z}}{1+iz}dz \right |&\leq\frac{\pi}{\omega}M_R\\ &=\frac{\pi}{\omega}\cdot\underset{\theta\in[0,\,\pi]}{\text{max}}\left | \frac{1}{1+iRe^{i\theta}}\right |\\ &=\frac{\pi}{\omega}\cdot\underset{\theta\in[0,\,\pi]}{\text{max}}\left | \frac{1-iRe^{-i\theta}}{1+R^2}\right |\\ &\leq \frac{\pi}{\omega}\cdot\frac{1+R}{1+R^2},\qquad\text{by the triangle inequality.}\tag{3} \end{align*}$$

Since the expression $(3)$ goes to $0$ for all $\omega>0$ as $R$ increases, we can back-track through our manipulations and conclude the limit in $(2)$ holds true.

 

We may now turn to the integral on the left hand side of $(1)$ and apply the residue theorem. Remembering that we specified $R>1$ we have:

$$\begin{align*} \oint_{\gamma_R}\frac{e^{i\omega z}}{1+iz}dz&=2\pi i\cdot\underset{z=i}{\text{Res}}\left[\frac{e^{i\omega z}}{1+iz}\right]\\ &=\frac{2\pi}{e^{\omega}}. \end{align*}$$

But the above holds for all $R>1$, so it holds in the limit, I.e.

$$\begin{align*} \frac{2\pi}{e^{\omega}}&=\lim_{R\rightarrow\infty}\oint_{\gamma_R}\frac{e^{i\omega z}}{1+iz}dz\\ &=\lim_{R\rightarrow\infty}\underset{\text{goes to 0}}{\underbrace{\int_{C_R}\frac{e^{i\omega z}}{1+iz}dz}}+\lim_{R\rightarrow\infty}\int_{-R}^R\frac{e^{i\omega z}}{1+iz}dz\\ &=I(\omega). \end{align*}$$

So

$$\int_{-\infty}^\infty\frac{e^{i\omega x}}{1+ix}dx=\frac{2\pi}{e^{\omega}}$$

for $\omega>0.$

$\endgroup$
  • $\begingroup$ I've gone back and reworked the answer also in the book I got this out of it seems like your proof can be generalized for a wide variety of cases. $\endgroup$ – Zophikel May 30 '19 at 20:23
  • 1
    $\begingroup$ It’s a common technique, so it generalizes very well. It’s good that you figured out whatever that book was saying haha. $\endgroup$ – dxdydz May 31 '19 at 5:48

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.