- #1
psie
- 269
- 32
- TL;DR Summary
- I'm reading Schiff's book The Laplace Transform. On page 92-93 he proves the convolution theorem. It's a very self-contained proof, with no reference to any prior results. However, I have some integration-related questions that I struggle with.
Here follows the theorem and proof:
Questions:
1. I do not understand the following part "...and hence, in view of the preceding calculation, ##\int_0^\infty \int_0^\infty |e^{-st}f(\tau)g(t-\tau)|dtd\tau## converges".
We know that ##\mathcal{L}\big(f(t)\big)## and ##\mathcal{L}\big(g(t)\big)## converge absolutely. So does their product converge absolutely (I assume this is the statement he is making)? If yes, how come?
The definition of absolute convergence given in the book is that ##\int_0^\infty |e^{-st}f(t)|dt## converges for a given real or complex parameter ##s##.
2. Regarding the footnote, if ##I=\int_0^\infty\int_0^\infty f(x,y)dxdy##, can we then always write ##I## as a double series, i.e. ##I=\sum_{n=0}^\infty\sum_{m=0}^\infty c_{mn}## where ##c_{mn}=\int_n^{n+1}\int_m^{m+1} f(x,y)dxdy## (I assume this is what Schiff is doing)? If not, what justifies that we can in this case and how?
Theorem (Convolution Theorem). If ##f## and ##g## are piecewise continuous on ##[0,\infty)## and of exponential order ##\alpha##, then $$\mathcal{L}\left[(f*g)(t)\right]=\mathcal{L}\big(f(t)\big)\cdot\mathcal{L}\big(g(t)\big)\quad \Big(Re(s)>\alpha\Big).$$
Proof. Let us start with the product \begin{align} \mathcal{L}\big(f(t)\big)\cdot\mathcal{L}\big(g(t)\big)&=\left(\int_0^\infty e^{-s\tau}f(\tau)d\tau\right)\left(\int_0^\infty e^{-su}g(u)du\right) \nonumber \\
&=\int_0^\infty \left(\int_0^\infty e^{-s(\tau+u)}f(\tau)g(u)du\right)d\tau \nonumber .
\end{align}
Substituting ##t=\tau+u##, and noting that ##\tau## is fixed in the interior integral, so that ##du=dt##, we have $$\mathcal{L}\big(f(t)\big)\cdot\mathcal{L}\big(g(t)\big)=\int_0^\infty \left(\int_\tau^\infty e^{-st}f(\tau)g(t-\tau)dt\right)d\tau .\tag 1$$ If we define ##g(t)=0## for ##t<0##, then ##g(t-\tau)=0## for ##t<\tau## and we can write ##(1)## as $$\mathcal{L}\big(f(t)\big)\cdot\mathcal{L}\big(g(t)\big)=\int_0^\infty \int_0^\infty e^{-st}f(\tau)g(t-\tau)dtd\tau .$$ Due to the hypotheses on ##f## and ##g##, the Laplace integrals of ##f## and ##g## converge absolutely and hence, in view of the preceding calculation, $$\int_0^\infty \int_0^\infty |e^{-st}f(\tau)g(t-\tau)|dtd\tau$$ converges. This fact allows us to reverse the order of integration,* so that \begin{align} \mathcal{L}\big(f(t)\big)\cdot\mathcal{L}\big(g(t)\big)&=\int_0^\infty \int_0^\infty e^{-st}f(\tau)g(t-\tau)d\tau dt \nonumber \\
&=\int_0^\infty \left(\int_0^t e^{-st}f(\tau)g(t-\tau)d\tau\right)dt \nonumber \\
&=\int_0^\infty e^{-st} \left(\int_0^t f(\tau)g(t-\tau)d\tau\right)dt \nonumber \\ &=\mathcal{L}[(f*g)(t)]. \nonumber
\end{align}
*Let $$a_{mn}=\int_n^{n+1}\int_m^{m+1} |h(t,\tau)|dtd\tau,\quad b_{mn}=\int_n^{n+1}\int_m^{m+1} h(t,\tau) dtd\tau,$$ so that ##|b_{mn}|\leq a_{mn}##. If $$\int_0^\infty\int_0^\infty |h(t,\tau)|dtd\tau <\infty,$$ then ##\sum_{n=0}^\infty\sum_{m=0}^\infty a_{mn}<\infty##, implying ##\sum_{n=0}^\infty\sum_{m=0}^\infty |b_{mn}|<\infty##. Hence, by a standard result on double series, the order of summation can be interchanged $$\sum_{n=0}^\infty\sum_{m=0}^\infty b_{mn}=\sum_{m=0}^\infty\sum_{n=0}^\infty b_{mn},$$
i.e., $$\int_0^\infty\int_0^\infty h(t,\tau) dtd\tau =\int_0^\infty\int_0^\infty h(t,\tau) d\tau dt.$$
Questions:
1. I do not understand the following part "...and hence, in view of the preceding calculation, ##\int_0^\infty \int_0^\infty |e^{-st}f(\tau)g(t-\tau)|dtd\tau## converges".
We know that ##\mathcal{L}\big(f(t)\big)## and ##\mathcal{L}\big(g(t)\big)## converge absolutely. So does their product converge absolutely (I assume this is the statement he is making)? If yes, how come?
The definition of absolute convergence given in the book is that ##\int_0^\infty |e^{-st}f(t)|dt## converges for a given real or complex parameter ##s##.
2. Regarding the footnote, if ##I=\int_0^\infty\int_0^\infty f(x,y)dxdy##, can we then always write ##I## as a double series, i.e. ##I=\sum_{n=0}^\infty\sum_{m=0}^\infty c_{mn}## where ##c_{mn}=\int_n^{n+1}\int_m^{m+1} f(x,y)dxdy## (I assume this is what Schiff is doing)? If not, what justifies that we can in this case and how?
Last edited: