3
$\begingroup$

The bilateral Laplace transform is defined as

$$X(s) = \int_{-\infty}^{\infty}x(t)e^{-st}dt$$where $s = \sigma +j\omega$. I'm trying to prove rigorously

The ROC(region of convergence) of $X(s)$ consists of strips parallel to the $j\omega$-axis in the $s$-plane.

In Oppenheim's book it's proved by noting that the ROC of $X(s)$ consists of those values of $s = \sigma +j\omega$ for which the Fourier transform of $x(t)e^{-\sigma t}$ converges and this means that the ROC of the Laplace transform of $x(t)$ consists of those values of $s$ for which $x(t)e^{-\sigma t}$ is absolutely integrable: $$\int_{-\infty}^{\infty}|x(t)|e^{-\sigma t}dt\lt\infty$$Because this condition depends only on $\sigma$, the real part of $s$, we have proved the mentioned statement.

My problem with this proof is that $$\text{convergence of the Fourier transform} \implies \text{absolute integrability }$$In fact the converse of this implication is true and absolute integrability is a sufficient condition for the convergence of the Fourier transform. So I think this proof is wrong. Is my understanding correct? Also is the mentioned statement correct? How can we prove that rigorously?

$\endgroup$
3
$\begingroup$

The first thing to get straight is that convergence of the Fourier transform of $f$ does not imply absolute integrability, i.e., that $f \in L^1(\mathbb{R})$.

For a counterexample, take $f(x) = \sin (x^2)$ where

$$\int_{-\infty}^\infty\sin(x^2)e^{ikx} \, dx = 2\int_0^\infty\sin(x^2) \cos(kx) \, dx = \int_0^\infty(\sin(x^2+kx)+ \sin(x^2-kx)) \, dx$$

Making the change of variables $x = u \mp \frac{k}{2}$ we get

$$\begin{align} \int_0^\infty \sin(x^2\pm kx)\, dx &= \int_0^\infty \sin\left(\left(u\mp \frac{k}{2}\right)^2 \pm k\left(u \mp \frac{k}{2}\right)\right)\, du \\ &= \int_0^\infty\sin\left(u^2 - \frac{k^2}{2}\right)\, du \\ &= \cos \left(\frac{k^2}{4}\right)\int_0^\infty \sin (u^2) \, du - \sin \left(\frac{k^2}{4}\right)\int_0^\infty \cos (u^2) \, du \end{align}$$

The well-known Fresnel integrals on the RHS are $\displaystyle \int_0^\infty \cos (u^2) \, du= \int_0^\infty \sin (u^2) \, du = \sqrt{\frac{\pi}{8}}$.

Substituting above, we get

$$\hat{f}(k) = \int_{-\infty}^\infty\sin(x^2)e^{ikx} \, dx = \sqrt{\frac{\pi}{2}}\left( \cos \left(\frac{k^2}{4}\right)-\sin \left(\frac{k^2}{4}\right)\right), $$

and the Fourier transform converges. However, $f \not\in L^1(\mathbb{R})$. This can either be shown directly or inferred from the fact that

$$\lim_{k \to \infty} \left( \cos \left(\frac{k^2}{4}\right)-\sin \left(\frac{k^2}{4}\right)\right)\,\, \text{DNE},$$

If $f \in L^1(\mathbb{R})$ then by the Riemann-Lebesgue lemma we would have $\lim_{k \to \infty}\hat{f}(k) = 0$


ROC Proof

Suppose $x$ is locally integrable and for some $s_0 \in \mathbb{C}$ we have convergence of the unilateral Laplace transform

$$X_+(s_0) = \int_0^\infty x(t) e^{-s_0t} \, dt$$

Let $\alpha(t) = \int_0^t x(u)e^{-s_0 u} \, du$. For any $T>0$ it follows from a basic property of the Riemann-Stieltjes integral (proved here) that

$$\int_0^T x(t) e^{-st}\, dt = \int_0^T e^{-(s-s_0)t} x(t) e^{-s_0t}\, dt = \int_0^T e^{-(s-s_0)t} \, d\alpha(t)$$

Integrating by parts on the RHS, we get

$$\tag{*}\int_0^T x(t) e^{-st}\, dt = \alpha(T)e^{-(s-s_0)T} + (s-s_0)\int_0^T e^{-(s-s_0)t} \alpha(t) \, dt$$

Since $X_+(s_0)$ converges, it follows that $|\alpha(t)|$ is bounded for all $t \geqslant 0$. This implies that for all $s$ such that $\Re(s) > \Re(s_0)$, we have

$$\lim_{T \to \infty}\alpha(T)e^{-(s-s_0)T} = 0,$$

and the integral on the RHS of (*) is absolutely convergent.

Thus,

$$X_+(s) = \lim_{T \to \infty}\int_0^T x(t) e^{-st}\, dt= \lim_{T\to \infty}\int_0^T e^{-(s-s_0)t} \, d\alpha(t) = \int_0^\infty e^{-(s-s_0)t} \, d\alpha(t),$$

where $X_+(s)$ converges for all $s$ such that $\Re(s) > \Re(s_0)$.

In a similar way we can show that if we have convergence of

$$X_-(s_1) = \int_{-\infty}^0 x(t) e^{-s_1t} \, dt,$$

then $X_-(s)$ converges for all $s$ such that $\Re(s) < \Re(s_1)$.

$\endgroup$
9
  • $\begingroup$ Thank you so much. Can we prove the mentioned statement rigorously? $\endgroup$
    – S.H.W
    Feb 7 at 9:05
  • $\begingroup$ @S.H.W: Since $x(t)e^{-s t} = [x(t)e^{-\sigma t}]e^{-j \omega t}$, the Laplace transform converges if and only if the Fourier integral $\int_{-\infty}^\infty [x(t)e^{-\sigma t}]e^{-j \omega t} \, dt$ converges. There is nothing deep here as the author is just using the sufficient condition of absolute integrability for the convergence of a Fourier transform. This presumably is an applications-oriented book and it is common to just reference that strong sufficient condition and move on. $\endgroup$
    – RRL
    Feb 7 at 18:05
  • $\begingroup$ Engineering books as far as I have seen never bring up the distinction between conditional versus absolute convergence of the Fourier integral. $\endgroup$
    – RRL
    Feb 7 at 18:07
  • $\begingroup$ You are right but my mean was how can we prove the statement rigorously using other methods? $\endgroup$
    – S.H.W
    Feb 7 at 19:01
  • 1
    $\begingroup$ We would have $X(s) = X_+(s) + X_-(s)$ converge for $\Re(s_0) < \Re(s) < \Re(s_1)$. The lower transform $X_-$ converges to the left of $s_1$. So convergence is always in a strip where the strip could have finite width, be semi-infinite in either direction, or possibly be the entire plane. $\endgroup$
    – RRL
    Feb 7 at 22:45

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.