Evaluation of Mellin Transforms via Residue Calculus

Mellin transform of a function f:(0,\infty)\to\mathbb{R} is of the form

\displaystyle (Mf)(a)=\int_0^\infty x^{a-1}f(x)\,dx

for a>0.  In order to implement residue calculus to evaluate this, one needs to make some assumptions on f.  The following proposition will tell us what these assumptions should be.

Proposition.  Let f have finitely many poles in \mathbb{C} all lying in \tilde{\mathbb{C}}=\mathbb{C}-\{z:\mbox{Re}\,z\geq 0\},  a\in\mathbb{C}-\mathbb{Z}, and both \lim_{z\to 0}f(z)z^a=0 and \lim_{z\to\infty}f(z)z^a=0.  Then

\displaystyle\int_0^\infty x^{a-1}f(x)\,dx=\frac{2\pi i}{1-e^{2\pi ia}}\sum_{p\in\tilde{\mathbb{C}}}\mbox{res}_p(z^{a-1}f(z)).

Proof.  We first choose a nonprincipal branch of the logarithm compatible with this situation.  Let z=|z|e^{i\theta} for \theta\in[0,2\pi) and then set \ln z=\ln |z| +i\theta and z^a=\exp(a\ln z).  Then these functions are holomorphic on \tilde{\mathbb{C}}.  One other lemma we need is the following: if I is a compact interval on the positive real axis and \varepsilon>0, then \lim_{\varepsilon\to 0}(x+i\varepsilon)^a=x^a and \lim_{\varepsilon\to 0}(x-i\varepsilon)^a=x^ae^{2\pi ia}, and the convergence is uniform.

Note that we have \lim_{\varepsilon\to 0}\ln(x+i\varepsilon)=\ln z and \lim_{\varepsilon\to 0}\ln(x-i\varepsilon)=\lim_{\varepsilon\to 0}\ln|x-i\varepsilon|+i\theta=\ln (x)+2\pi i, which gives the result of the lemma.

Now to integrate the function, we consider the complex function z^{a-1}f(z) and the contour given bycontour

with \gamma=\gamma_1+\gamma_2+\gamma_3+\gamma_4.  Then by the residue theorem

\displaystyle\lim_{\varepsilon\to 0}\lim_{r\to 0}\lim_{s\to\infty}\int_\gamma z^{a-1}f(z)\,dz=2\pi i\sum_{p\in\tilde{\mathbb{C}}}\mbox{res}_p(z^{a-1}f(z)),

For \gamma_2 we have

\displaystyle\left|\int_{\gamma_2}z^{a-1}f(z)\,dz\right|\leq 2\pi s|z^{a-1}f(z)|=2\pi |z^af(z)|

since |z|=s on \gamma_2.  We have a similar bound for \gamma_4.  Now due to the assumption of the proposition, it follows that |z^af(z)|\to 0 and hence \int_{\gamma_2}z^{a-1}f and \int_{\gamma_4}z^{a-1}f go to 0 as \varepsilon\to 0 and r,s\to\infty.

Observe that \gamma_1(x)=x+i\varepsilon and -\gamma_3(x)=x-i\varepsilon.  Then by the lemma we have

\displaystyle\lim_{\varepsilon\to 0}\int_{\gamma_1}z^{a-1}f(z)\,dz=\int_r^sx^{a-1}f(x)\,dx

and

\displaystyle\lim_{\varepsilon\to 0}\int_{\gamma_3}z^{a-1}f(z)\,dz=-e^{2\pi ia}\int_r^sx^{a-1}f(x)\,dx.

Thus we have

\displaystyle\lim_{r\to 0}\lim_{s\to\infty}\lim_{\varepsilon\to 0}\int_\gamma z^{a-1}f(z)\,dz=(1-e^{2\pi ia})\int_0^{\infty}x^{a-1}f(x)\,dx,

which yields the result.

[1]  Remmert, Reinhold.  Theory of Complex Functions.  Springer-Verlag, 1991.