I want to calculate the Fourier transform of the function $ f\left(t\right)=e^{-t^{2}} $

The answer should be $$ \mathscr{F}\left(f\right)[\omega]=\frac{1}{2\sqrt{\pi}}e^{-\frac{\omega^{2}}{4}} $$

I solved it in 2 different ways (and got the same result), but I doubt my second way because I cannot explain in 100%.

So the first way is to note that both $$ \intop_{-\infty}^{\infty}|f\left(t\right)|dt<\infty,\thinspace\thinspace\thinspace\intop_{-\infty}^{\infty}t|f\left(t\right)|dt<\infty $$

And thus by a well known theorem $$ \frac{d}{d\omega}\mathscr{F}\left(f\right)[\omega]=\mathscr{F}\left(g\right)[\omega] $$

Where $ g\left(t\right)=-itf\left(t\right) $.

Using integration by parts, we find that $$ \frac{d}{d\omega}\mathscr{F}\left(f\right)[\omega]=-\frac{\omega}{2}\mathscr{F}\left(f\right)[\omega] $$

Solving this differential equation we find: $$ \mathscr{F}\left(f\right)[\omega]=Ce^{-\frac{\omega^{2}}{4}} $$

And for $\omega=0 $ we have $$ \mathscr{F}\left(f\right)[0]=\frac{1}{2\pi}\intop_{-\infty}^{\infty}f\left(t\right)dt=\frac{1}{2\pi}\intop_{-\infty}^{\infty}e^{-t^{2}}dt=\frac{1}{2\sqrt{\pi}} $$

And hence we found $ C $.

This way require a lot of details, and I have much easier one.

The second way:

Note that :

$$ \mathscr{F}\left(f\right)[\omega]=\frac{1}{2\pi}\intop_{-\infty}^{\infty}e^{-t^{2}}e^{-i\omega t}dt=\frac{1}{2\pi}\intop_{-\infty}^{\infty}e^{-\left(t^{2}+i\omega t\right)}dt=\frac{1}{2\sqrt{\pi}}e^{-\frac{\omega^{2}}{4}}\intop_{-\infty}^{\infty}\frac{\sqrt{2}}{\sqrt{2\pi}}e^{-\left(t+\frac{i\omega}{2}\right)^{2}}dt $$

And note that the density of a normal distributed random variable is given by $ N\left(\mu,\sigma\right)=\frac{1}{\sqrt{2\pi}\sigma}e^{-\frac{\left(x-\mu\right)^{2}}{2\sigma^{2}}} $

So we got $ \frac{1}{2\sqrt{\pi}}e^{-\frac{\omega^{2}}{4}} $ multiplied by an integral (over the whole space) over a density of random distributed random variable with $ N\left(-\frac{i\omega}{2},\frac{1}{\sqrt{2}}\right) $.

We got the same result.

Finally, my question is:

When I learned probability theory, we never mentioned complex numbers, So my argument that this is an integral over a density of a normal distribution, may be not acceptable.

Is it legit?

Thanks in advance.

  • $\begingroup$ You can make your second method more legit by just using a complex substitution to turn the integral into an actual canonical Gaussian integral. In fact I would say that that's relatively easier to make completely rigorous than the argument based on properties of Fourier transforms. $\endgroup$ – Izaak van Dongen Mar 11 at 18:51

If you know some complex analysis, you can fill in the details of the second argument. For simplicity, I'm going to omit all constant factors. Here's the claim:

For any $\beta\in\Bbb{R}$, we have $\int_{\Bbb{R}}e^{-t^2}\,dt = \int_{\Bbb{R}}e^{-(t+i\beta)^2}\,dt$.

To prove this, consider for each $R>0$, the closed rectangular loop $\gamma_R$ oriented counter-clockwise passing through the four points $-R,R, R+i\beta, -R+i\beta$ (draw a picture, and maybe you would like to assume $\beta>0$ as well for ease of visualization). Then, the function $z\mapsto e^{-z^2}$ is holomorphic on $\Bbb{C}$ so by Cauchy's theorem, we have \begin{align} \int_{\gamma_R}e^{-z^2}\,dz &= 0. \end{align} Writing this out explicitly, we have \begin{align} \int_{-R}^{R}e^{-x^2}\,dx + \int_{0}^{\beta}e^{-(R+iy)^2}\,dy - \int_{-R}^Re^{-(x+i\beta)^2}\,dx -\int_{0}^{\beta}e^{-(-R+iy)^2}\,dy &= 0. \end{align} Here, the $+,-$ signs take the orientation the the curve $\gamma_R$ into account (again draw a picture). I leave it to you to justify why as $R\to \infty$, the two integrals $\int_0^{\beta}$ vanish, thereby leaving us with the equality \begin{align} \int_{\Bbb{R}}e^{-t^2}\,dt =\int_{\Bbb{R}}e^{-(t+i\beta)^2}\,dt. \end{align}


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.