Which distributions are their own Fourier transform besides the normal distribution and the generalized arcsine distribution?


1 Answer 1


Suppose the Fourier transform of $x(t)$ is $X(f)$ where $$X(f) = \int_{-\infty}^{\infty} x(t) \exp(-i2\pi f t) \mathrm dt$$ where $i = \sqrt{-1}$. The inverse transform is $$x(t) = \int_{-\infty}^{\infty} X(f) \exp(i2\pi f t) \mathrm df$$

Some properties of the Fourier transform are as follows:

  • The Fourier transform of $X(t)$ is $x(-f)$

  • If $x(t)$ is a real-valued even function of $t$, then $X(f)$ is a real-valued even function of $f$.

Thus, if $x(t)$ is a real-valued even function of $t$, then the Fourier transform of the real-valued even function $X(t)$ is $x(f)$

Now suppose that $x(t)$ is an even probability density function (so that $x(t) \geq 0$ for all $t$) with the additional property that $x(0) = 1$. Suppose also that its Fourier transform $X(f)$ has the property that $X(f) \geq 0$ for all $f$. Then, since $$x(0) = 1 = \int_{-\infty}^{\infty} X(f) \mathrm df$$ $X(f)$ is a even non-negative real-valued function of $f$ with area $1$, that is, $X(f)$ is also a probability density function with the property that $X(0) = 1$. One example of such a pair of functions is the normal distribution cited by OP Neil G $$x_1(t) = \exp(-\pi t^2), ~~ X_1(f) = \exp(-\pi f^2)$$ and another example is $$x_2(t) = (1 - |t|)\mathbf 1_{[-1,1]}, ~~ X_2(f) = \operatorname{sinc}^2(f) = \begin{cases}\displaystyle \left(\frac{\sin(\pi f)}{\pi f}\right)^2,&f\neq 0,\\ 1,&f=0.\end{cases}$$

Now note that $\frac{1}{2} x_2(t) + \frac{1}{2}X_2(t)$ is a mixture density whose Fourier transform is $\frac{1}{2} X_2(f) + \frac{1}{2}x_2(f)$ which is the same mixture density.

Thus, if $x(t)$ is a density function whose Fourier transform $X(f)$ is a density function, then the mixture density function $\frac{1}{2} x(t) + \frac{1}{2}X(t)$ is its own Fourier transform.

Finally, given two densities that are their own Fourier transforms, e.g. $x_1(t)$ and $\frac{1}{2} x_2(t) + \frac{1}{2}X_2(t)$, any mixture density $$\alpha x_1(t) + (1-\alpha)\left[\frac{1}{2} x_2(t) + \frac{1}{2}X_2(t)\right]$$ where $\alpha \in [0,1]$ is a density function that is its own Fourier transform.

  • 8
    $\begingroup$ (+1) This is quite clever. It should be noted that to guarantee a valid transform pair, we need an integrability condition on $X(f)$. Namely, $\int_{-\infty}^\infty X(f) \,\mathrm{d}f < \infty$ will guarantee that the stated inversion will recover the appropriate density. In a sense you employ such a condition later on. (I've already assumed the nonnegativity constraint on $X(f)$ has been imposed, so it needs no modulus.) $\endgroup$
    – cardinal
    Jan 16, 2012 at 23:37

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.