The integral $$\displaystyle\int_{-\infty}^\infty e^{-\pi (a+z)^2}\, da$$ should converge for every $z\in\mathbb C$, but in my proof, it seems that it diverges when $\operatorname{Re}z=\alpha=0$: $$\begin{align*}\left|e^{-\pi (a+z)^2}\right|&=\left|e^{-\pi a^2}e^{-2\pi a\overbrace{z}^{\alpha +\beta i}}e^{-\pi \overbrace{z^2}^{(\alpha +\beta i)^2}}\right|\\&=e^{-\pi a^2-2\pi a\alpha -\pi \alpha ^2+\pi\beta ^2}\underbrace{\left|e^{-\beta i(2\pi a+2\pi \alpha)}\right|}_{1}\end{align*}$$ $$\begin{align*}\left|\displaystyle\int_{-\infty}^\infty e^{-\pi (a+z)^2}\, da \right|&\le \displaystyle\int_{-\infty}^\infty \left|e^{-\pi (a+z)^2}\right|\, da\\&=e^{-\pi \left(\alpha ^2-\beta ^2\right)}\displaystyle\int_{-\infty}^\infty e^{-\pi \left(a^2+2a\alpha\right)}\, da\\&\overset{\alpha ^2+2a\alpha =t^2}{=}e^{-\pi \left(\alpha ^2-\beta ^2\right)}\dfrac{\sqrt{\alpha (\alpha +2a)}}{\alpha}\displaystyle\int_{-\infty}^\infty e^{-\pi t^2}\, dt\\&\overset{t=m/\sqrt{\pi}}{=}\dfrac{e^{-\pi \left(\alpha ^2-\beta ^2\right)}\sqrt{\alpha (\alpha +2a)}}{\alpha}\displaystyle\int_{-\infty}^\infty e^{-m^2}\, \dfrac{dm}{\sqrt{\pi}}\\&=\dfrac{e^{-\pi \left(\alpha ^2-\beta ^2\right)}\sqrt{\alpha (\alpha +2a)}}{\alpha}.\end{align*}$$ There must be a mistake somewhere, but I can't find it.


The change of variables $$\alpha^2+2a\alpha=t^2$$ is meaningless when $\alpha=0$ so you can't reach any conclusions in that case. You'll need to handle it separately. Fortunately the integral is easily seen to be convergent then.

  • $\begingroup$ I tried to find the antiderivative, and then I evaluated it at the limits: $\int_{-\infty}^\infty e^{-\pi (a+z)^2}\, da=\frac{1}{2}\left[\operatorname{erf}\sqrt{\pi}(a+z)\right]_{-\infty}^\infty=\lim_{a\to\infty}\operatorname{erf}\sqrt{\pi}(a+z)=1$, though I don't know how would I rigorously prove the result of the limit. $\endgroup$ – Tu1 Aug 27 '19 at 12:20
  • $\begingroup$ Riemann sphere came to my mind, namely that $z+\infty =\infty$. Can this be used in this context to conclude that $\lim_{a\to\infty}\operatorname{erf}\sqrt{\pi}(a+z)=\lim_{a\to \infty}\operatorname{erf}a=1?$ $\endgroup$ – Tu1 Aug 27 '19 at 12:28

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.