Consider the following expression of Gamma function

$$\frac{\Gamma(z)}{p^z}=\int_{0}^{\infty}e^{-pt}t^{z-1}dt \ \ \ \ \ \ \ \ \ (1)$$ where $Re(z)>0$ and $Re(p)>0$.

In Lebedevs book "special functions and their applications" he uses that integral to calculate

$$\int_{0}^{\infty}e^{2izs}s^{\alpha-1}ds=\frac{\Gamma(\alpha)}{(-2iz)^{\alpha}} \ \ \ \ \ \ \ \ (2)$$

but what about case when $z=x$ is real? This means $Re(p)=0$ in $(1)$ expression.

P.S. The Formula $(2)$ must be valid for Real $z$ because of final result of calculation (asymptotic expression of Hankels' function $H_{\nu}^{(1)}(z)$).

I just want to know how to integrate (2) integral if z=x is real variable.

  • $\begingroup$ Hint: integration by parts and analytic continuation. $\endgroup$ – Fan Zheng Nov 14 '15 at 20:41

The identity can be deduced by the Euler integral by homotopy invariance of path integrals.

The function $e^{-z}z^{\alpha}$ is holomorphic on $\mathbb{C}\setminus \{\operatorname{Re}z\le0,\operatorname{Im}z=0 \}$, so the value of the path integral along the boundary of the domain $\{z: \operatorname{Re}z>0,\operatorname{Im}z>0, \epsilon< |z|< \rho \}$ is zero. Hence we get $$\int_\epsilon^{\rho}i^\alpha e^{-it}t^{\alpha-1}dt=\int_\epsilon^{\rho}e^{-t}t^{\alpha-1}dt+ \int_0^{\pi/2}ie^{-\rho e^{it} }(\rho e^{it})^{\alpha}dt- \int_0^{\pi/2}ie^{-\epsilon e^{it}}(\epsilon e^{it} )^{\alpha}dt $$

We consider separately the three integrals on the RHS.

  • The first integral, of course, converges to the Euler integral for $\Gamma(\alpha)$ as $\epsilon\to0$ and $\rho\to+ \infty$, provided $\operatorname{Re}\alpha>0$.

  • The second integrand has absolute value $ e^{-\rho\cos(t) -\operatorname{Im}\alpha t}\; \rho^{ \operatorname{Re}\alpha} $, and it is a bit singular at $t=\pi/2$; to evaluate the corresponding integral, it is convenient to spit it further in the integrals over $[0,x]$ and $[x,\pi/2]$ with a free $x$, optimizing then the bound over the choice of $x$. This way one gets a bound for this integral of order $O\Big(\rho ^{\operatorname{Re}\alpha -1} \log(\rho) \Big)$, which is $o(1)$ as $\rho\to+ \infty$, provided $\operatorname{Re}\alpha <1$.

  • The third integrand has absolute value $ e^{-\cos(t) \epsilon-\operatorname{Im}\alpha t}\; \epsilon^{ \operatorname{Re}\alpha}=O(\epsilon^{ \operatorname{Re}\alpha}) $.

We conclude that for $0< \operatorname{Re}\alpha<1$ the integral on the LHS converges, and

$$\int_0^{+\infty} e^{-it}t^{\alpha-1}dt=\frac{\Gamma(\alpha)}{i^\alpha},$$

which is the wanted identity for $z=-1/2$. For any real $z<0$, with a linear change of variable, $t=(-2z)s$ we plainly get

$$\int_0^{+\infty} e^{2izs}s^{\alpha-1}dt=\frac{\Gamma(\alpha)}{(-2iz)^\alpha}.$$

Finally, for a real positive $z>0$, we take complex conjugate to both sides to the latter identity, written for $-z$ and $\overline \alpha$, which yields to the identity for $z$ and $\alpha$.


let $z$ be a real number, $0<\alpha<1$, $L>0$ and start from the proper integral


use the asymptotics for the incomplete Gamma function: $$\quad\lim_{L\rightarrow\infty}\frac{\Gamma(\alpha,-2iLz)}{(-2iLz)^{\alpha-1}e^{2iLz}}=1\Rightarrow \lim_{L\rightarrow\infty}\Gamma(\alpha,-2iLz)=0$$



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.