# Understand uncertainty principle from Fourier transform - Interpretation of Fourier transform on a restricted periodic function -

I have been just starting on Fourier transforms and I decided to try one of my own, and so decided to find the Fourier transform of the aperiodic function:

$$x(t)=\sin (\omega_1t)$$

that is only defined in a finite domain $$t\in[A,B]$$ and is $$0$$ everywhere else. So I came to this expression:

$$X(j\omega)=\int_{-\infty}^{+\infty}x(t)e^{-j\omega t}dt$$ $$=\int_{A}^{B}\sin (\omega_1)e^{-j\omega t}dt$$ $$=-\frac{e^{k(\omega_1-\omega)B}}{2(\omega_1-\omega)}-\frac{e^{k(\omega_1+\omega)B}}{2(\omega_1+\omega)}+\frac{e^{k(\omega_1-\omega)A}}{2(\omega_1-\omega)}+\frac{e^{k(\omega_1+\omega)A}}{2(\omega_1+\omega)}$$

And so I decided to plot for $$A=-1$$,$$B=1$$,$$\omega_1 =2\pi$$ and setting the domain of $$\omega$$ from -10 to 10 and got these plots:

Where the first plot is $$x(t)$$ only where it has non 0 values. However I was expecting two things:

1. there would be two spikes on the absolute value of the transform: one at $$\omega =2\pi$$ and other as $$\omega =-2\pi$$;
2. the spike would get wider and wider as $$A$$ got closer to $$B$$

but none is verified, am I expecting false results or was the math poorly done?

Also, the code used is this:

import matplotlib.pyplot as plt
import matplotlib.gridspec as gridspec
import numpy as np

w = np.linspace(-10, 10, 1000)
w1 = 2 * np.pi
A = -1
B = 1
domain = np.linspace(A, B, 1000)
func = np.sin(w1 * domain)
im = -np.sin((w1 - w) * B) / (2 * (w1 - w)) - np.sin((w1 + w) * B) / (2 * (w1 + w)) + np.sin((w1 - w) * A) / (
2 * (w1 - w)) - np.sin((w1 + w) * A) / (2 * (w1 + w))
re = -np.cos((w1 - w) * B) / (2 * (w1 - w)) - np.cos((w1 + w) * B) / (2 * (w1 + w)) + np.cos((w1 - w) * A) / (
2 * (w1 - w)) - np.cos((w1 + w) * A) / (2 * (w1 + w))

mod = np.sqrt(im ** 2 + re ** 2)
phase = np.arctan(im / re)

fig = plt.figure()
grid = gridspec.GridSpec(nrows=3, ncols=1, figure=fig)

ax1.plot(w, mod)
ax1.set_title(r'$$|X(j\omega)|$$')
ax1.grid(True)
ax1.set_xlabel(r'$$\omega$$')

ax2.plot(w, phase)
ax2.set_title(r'$$\phi (X(j\omega)$$')
ax2.grid(True)
ax2.set_xlabel(r'$$\omega$$')
ax2.hlines(np.pi / 2, -10, 10, linestyles="dashed")
ax2.hlines(-np.pi / 2, -10, 10, linestyles="dashed")

ax3.plot(domain, func)
ax3.grid(True)

plt.tight_layout()
plt.show()

• Check your integral result again. You have 3 "B"'s and only 1 "A". That seems wrong given the symmetry of the situation. Also, if your numerical simulation code doesn't match the mathematical results, the problem is with your code. You have not provided your code here, so no one can comment on what might be wrong with it. Apr 1, 2019 at 13:05
• @AndyWalls Yes, that was a mere typo on writing it, really sorry. Fixed it Apr 1, 2019 at 13:49

I am answering my own question because I found the error. It was something in the code. I'm leaving the results here if anyone finds it useful or maybe wants to have a bit of fun with it:

import matplotlib.pyplot as plt
import matplotlib.gridspec as gridspec
import numpy as np

w = np.linspace(-10, 10, 1000)
w1 = 2 * np.pi
A = np.array([-10, -5, -2, -1, -0.5])
B = -A

domain1 = np.linspace(A[0], B[0], 1000)
domain2 = np.linspace(A[1], B[1], 1000)
domain3 = np.linspace(A[2], B[2], 1000)
domain4 = np.linspace(A[3], B[3], 1000)
domain5 = np.linspace(A[4], B[4], 1000)

func1 = np.sin(w1 * domain1)
func2 = np.sin(w1 * domain2)
func3 = np.sin(w1 * domain3)
func4 = np.sin(w1 * domain4)
func5 = np.sin(w1 * domain5)

im1 = (np.sin((w1 - w) * A[0]) - np.sin((w1 - w) * B[0])) / (2 * (w1 - w)) + (
np.sin(-(w1 + w) * A[0]) - np.sin(-(w1 + w) * B[0])) / (2 * (w1 + w))
im2 = (np.sin((w1 - w) * A[1]) - np.sin((w1 - w) * B[1])) / (2 * (w1 - w)) + (
np.sin(-(w1 + w) * A[1]) - np.sin(-(w1 + w) * B[1])) / (2 * (w1 + w))
im3 = (np.sin((w1 - w) * A[2]) - np.sin((w1 - w) * B[2])) / (2 * (w1 - w)) + (
np.sin(-(w1 + w) * A[2]) - np.sin(-(w1 + w) * B[2])) / (2 * (w1 + w))
im4 = (np.sin((w1 - w) * A[3]) - np.sin((w1 - w) * B[3])) / (2 * (w1 - w)) + (
np.sin(-(w1 + w) * A[3]) - np.sin(-(w1 + w) * B[3])) / (2 * (w1 + w))
im5 = (np.sin((w1 - w) * A[4]) - np.sin((w1 - w) * B[4])) / (2 * (w1 - w)) + (
np.sin(-(w1 + w) * A[4]) - np.sin(-(w1 + w) * B[4])) / (2 * (w1 + w))

re1 = (np.cos((w1 - w) * A[0]) - np.cos((w1 - w) * B[0])) / (2 * (w1 - w)) + (
np.cos(-(w1 + w) * A[0]) - np.cos(-(w1 + w) * B[0])) / (2 * (w1 + w))
re2 = (np.cos((w1 - w) * A[1]) - np.cos((w1 - w) * B[1])) / (2 * (w1 - w)) + (
np.cos(-(w1 + w) * A[1]) - np.cos(-(w1 + w) * B[1])) / (2 * (w1 + w))
re3 = (np.cos((w1 - w) * A[2]) - np.cos((w1 - w) * B[2])) / (2 * (w1 - w)) + (
np.cos(-(w1 + w) * A[2]) - np.cos(-(w1 + w) * B[2])) / (2 * (w1 + w))
re4 = (np.cos((w1 - w) * A[3]) - np.cos((w1 - w) * B[3])) / (2 * (w1 - w)) + (
np.cos(-(w1 + w) * A[3]) - np.cos(-(w1 + w) * B[3])) / (2 * (w1 + w))
re5 = (np.cos((w1 - w) * A[4]) - np.cos((w1 - w) * B[4])) / (2 * (w1 - w)) + (
np.cos(-(w1 + w) * A[4]) - np.cos(-(w1 + w) * B[4])) / (2 * (w1 + w))

mod1 = np.sqrt(im1 ** 2 + re1 ** 2)
mod2 = np.sqrt(im2 ** 2 + re2 ** 2)
mod3 = np.sqrt(im3 ** 2 + re3 ** 2)
mod4 = np.sqrt(im4 ** 2 + re4 ** 2)
mod5 = np.sqrt(im5 ** 2 + re5 ** 2)

phase1 = np.arctan(im1 / re1)
phase2 = np.arctan(im2 / re2)
phase3 = np.arctan(im3 / re3)
phase4 = np.arctan(im4 / re4)
phase5 = np.arctan(im5 / re5)

fig = plt.figure()
grid = gridspec.GridSpec(nrows=3, ncols=2, figure=fig)

ax1.plot(w, mod1, label=r'$$x\in [-10;10]$$')
ax1.set_title(r'$$|X(j\omega)|$$')
ax1.grid(True)
ax1.vlines(w1, 0, 1, linestyles="dashed")
ax1.vlines(-w1, 0, 1, linestyles="dashed")
ax1.set_xlabel(r'$$\omega$$')
ax1.legend()

ax2.plot(w, mod2, label=r'$$x\in [-5;5]$$')
ax2.set_title(r'$$|X(j\omega)|$$')
ax2.grid(True)
ax2.vlines(w1, 0, 1, linestyles="dashed")
ax2.vlines(-w1, 0, 1, linestyles="dashed")
ax2.set_xlabel(r'$$\omega$$')
ax2.legend()

ax3.plot(w, mod3, label=r'$$x\in [-2;2]$$')
ax3.set_title(r'$$|X(j\omega)|$$')
ax3.grid(True)
ax3.vlines(w1, 0, 1, linestyles="dashed")
ax3.vlines(-w1, 0, 1, linestyles="dashed")
ax3.set_xlabel(r'$$\omega$$')
ax3.legend()

ax4.plot(w, mod4, label=r'$$x\in [-1;1]$$')
ax4.set_title(r'$$|X(j\omega)|$$')
ax4.grid(True)
ax4.vlines(w1, 0, 1, linestyles="dashed")
ax4.vlines(-w1, 0, 1, linestyles="dashed")
ax4.set_xlabel(r'$$\omega$$')
ax4.legend()

ax5.plot(w, mod5, label=r'$$x\in [-0.5;0.5]$$')
ax5.set_title(r'$$|X(j\omega)|$$')
ax5.grid(True)
ax5.vlines(w1, 0, 1, linestyles="dashed")
ax5.vlines(-w1, 0, 1, linestyles="dashed")
ax5.set_xlabel(r'$$\omega$$')
ax5.legend()

plt.tight_layout()
plt.show()


All this work follows from 3Blue1Brown video linking fourier transform and the quantum uncertainty principle. In the figure the legend tells us the time domain of x(t). Notice that the wider the time domain the more acute the maximums of the fourier transform get. That is because, if we are certain of where a particle is then its wave function will collapse to a dirac (I think so at least) and so its fourier transform will be a constant, that is, if we know exactly where a particle is, we have no idea what its frequency, and thus energy, is, it could be everywhere with the same likelihood in the frequency domain. As for the other limit, if we have no idea where a particle is, than its wave function spreads to infinity and the fourier transform converges to two diracs (positive and negative frequencies) giving us the exact frequency of the particle! Therefore, the the more certain we are of where a particle is, the thiner it's time domain has to be and its frequency domain is ever larger. We will never know for certain where and what its energy are at the same time. The fourier transform explains this concept beautifully.