WEBVTT
00:00:03.560 --> 00:00:07.160
You’ve probably heard of the Heisenberg uncertainty principle from quantum mechanics.
00:00:07.480 --> 00:00:12.400
That the more you know about a particle’s position, the less certain you can be of its momentum and vice versa.
00:00:12.400 --> 00:00:17.680
My goal here is for you to come away from this video feeling like this is utterly reasonable.
00:00:17.680 --> 00:00:18.600
It’ll take some time.
00:00:18.600 --> 00:00:20.760
But I think you’ll agree that digging deep is well worth it.
00:00:21.440 --> 00:00:30.920
You see, the uncertainty principle is actually one specific example of a much more general trade-off that shows up in a lot of everyday totally nonquantum circumstances involving waves.
00:00:31.640 --> 00:00:35.840
The plan here is to see what this means in the context of sound waves, which should feel reasonable.
00:00:36.160 --> 00:00:41.000
And then Doppler radar, which should again feel reasonable and a little bit closer to the quantum case.
00:00:42.360 --> 00:00:49.280
And then for particles which, if you’re willing to accept one or two premises of quantum mechanics, hopefully feels just as reasonable as the first two.
00:00:50.440 --> 00:00:55.520
The core idea here has to do with the interplay between frequency and duration.
00:00:55.520 --> 00:01:01.320
And I bet you already have an intuitive idea of this principle before we even get into the math or the quantum.
00:01:02.160 --> 00:01:10.560
If you were to pull up behind a car at a red light and your turn signals were flashing together for a few seconds, you might kind of think that they have the same frequency.
00:01:10.920 --> 00:01:17.400
But at that point, for all you know, they could fall out of sync as more time passes, revealing that they actually had different frequencies.
00:01:18.160 --> 00:01:23.760
So an observation over a short period of time gave you low confidence over what their frequencies are.
00:01:24.560 --> 00:01:33.200
But if you were to sit at that red light for a full minute and the signals continued to click in sync, you would be a lot more confident that the frequencies are actually the same.
00:01:33.880 --> 00:01:39.640
So that certainty about the frequency information required an observation spread out over time.
00:01:40.480 --> 00:01:49.000
And this trade-off right here between how short your observation can be and how confident you can feel about the frequency is an example of the general uncertainty principle.
00:01:50.080 --> 00:01:51.520
Similarly, think of a musical note.
00:01:51.840 --> 00:01:56.400
The shorter it lasts in time, the less certain you can be about what its exact frequency is.
00:01:57.000 --> 00:02:03.160
In the extreme, I could ask you what the pitch of a clap or a shock wave is and even someone with perfect pitch would be unable to answer.
00:02:04.080 --> 00:02:08.680
And on the flip side, a more definite frequency requires a longer duration signal.
00:02:09.760 --> 00:02:17.960
Or, rather than talking about definiteness or certainty, it would be a little more accurate here to say that the short signal correlates highly with a wider range of frequency.
00:02:21.480 --> 00:02:26.880
And that the signal correlating strongly with only a narrow range of frequencies must last for a longer time.
00:02:29.640 --> 00:02:33.040
Here, that’s the kind of phrase that’s made a little bit clearer when we bring in the actual math.
00:02:33.560 --> 00:02:38.960
So let’s turn now to talking about the Fourier transform, which is the relevant construct for analyzing frequencies.
00:02:39.600 --> 00:02:42.800
The last video I put out was a visual intuition for this transform.
00:02:43.160 --> 00:02:45.920
And yes, it probably would be helpful if you’ve seen it.
00:02:46.360 --> 00:02:49.680
But I’m gonna go ahead and give a quick recap here just to remind ourselves of how it went.
00:02:51.040 --> 00:02:55.040
Let’s say we have a signal and it plays five beats per second over the course of two seconds.
00:02:55.600 --> 00:03:04.360
The Fourier transform gives a way to view any signal, not in terms of the intensity at each point in time but, instead, in terms of the strength of various frequencies within it.
00:03:05.320 --> 00:03:09.440
The main idea was to take this signal and to kind of wind it around a circle.
00:03:11.600 --> 00:03:17.560
As in, imagine some rotating vector whose length is determined by the height of the graph at each point in time.
00:03:18.600 --> 00:03:22.440
Right now, this little vector is rotating at 0.3 cycles per second.
00:03:23.080 --> 00:03:26.040
That’s the frequency with which we’re winding the graph around the circle.
00:03:27.000 --> 00:03:30.360
And for most frequencies, the signal is kinda just averaged out over the circle.
00:03:33.320 --> 00:03:35.200
This was the fun part of last video, don’t you think?
00:03:35.480 --> 00:03:40.120
Just seeing the different patterns that come up as you wind a pure cosine around a circle like this.
00:03:43.080 --> 00:03:50.120
But, the key point is what happens when that winding frequency matches the signal frequency, in this case five cycles per second.
00:03:50.920 --> 00:03:57.360
As our little vector is rotating around and it draws, all of the peaks line up on one side and all of the valleys on another side.
00:03:57.760 --> 00:04:01.480
So the whole weight of the graph is kind of off center, so to speak.
00:04:02.720 --> 00:04:15.480
The idea behind the Fourier transform is that if you follow the center of mass of the wound-up graph with frequency 𝑓, the position of that center of mass encodes the strength of that frequency in the original signal.
00:04:16.560 --> 00:04:21.040
The distance between that center of mass and the origin captures the strength of that frequency.
00:04:21.840 --> 00:04:24.480
And this is something I didn’t really talk about in the main video.
00:04:24.480 --> 00:04:29.480
But the angle of that center of mass off the horizontal corresponds to the phase of the given frequency.
00:04:31.880 --> 00:04:38.760
Now one way to think of this whole winding mechanism is that it’s a way to measure how well your signal correlates with a given pure frequency.
00:04:40.000 --> 00:04:50.120
So remember, when we say the Fourier transform, we’re referring to this new function whose input is that winding frequency and whose output is the center of mass, thought of as a complex number.
00:04:50.720 --> 00:04:54.160
Or, more technically, it’s a certain multiple of that center of mass.
00:04:54.160 --> 00:04:56.240
But whatever, the overall shape remains the same.
00:04:56.920 --> 00:05:02.240
And the graph that I’m drawing is just gonna be the 𝑥-coordinate of that center of mass, the real part of its output.
00:05:03.720 --> 00:05:07.160
If you wanted, you could also plot the distance between the center of mass and the origin.
00:05:07.160 --> 00:05:11.320
And maybe that better conveys how strongly each possible frequency correlates with the signal.
00:05:11.880 --> 00:05:15.720
The downside is that you lose some of the nice linearity properties that I talked about last video.
00:05:16.400 --> 00:05:26.160
Anyway, point is, this spike you’re looking at here above the winding frequency of five is the Fourier transform’s way of telling us that the dominant frequency of the signal is five beats per second.
00:05:26.760 --> 00:05:37.960
And equally importantly, the fact that it’s a little bit spread out around that five is an indication that pure sine waves near five beats per second also correlate pretty well with the signal.
00:05:39.520 --> 00:05:41.840
And that last idea is key for the uncertainty principle.
00:05:42.320 --> 00:05:48.720
What I want you to do is think about how this spread changes as the signal persists longer or shorter over time.
00:05:50.720 --> 00:05:52.120
You’ve already seen this at an intuitive level.
00:05:52.720 --> 00:05:56.560
All we’re doing right now is just illustrating it in the language of Fourier transforms.
00:05:58.360 --> 00:06:04.680
If the signal persists over a long period of time, then when the winding frequency is even slightly different from five.
00:06:05.200 --> 00:06:08.640
The signal goes on long enough to wrap itself around the circle and balance out.
00:06:09.480 --> 00:06:18.880
So looking at the Fourier plot over here, that corresponds to a super sharp drop-off in the magnitude of the transform as your frequency shifts away from that five beats per second.
00:06:22.440 --> 00:06:30.360
On the other hand, if your signal was really localized to a short period of time, then as you adjust the frequency away from five beats per second.
00:06:31.000 --> 00:06:35.200
The signal doesn’t really have as much time to balance itself out around the circle.
00:06:35.840 --> 00:06:41.200
You have to change the winding frequency to be meaningfully different from five before that signal starts to balance out again.
00:06:42.080 --> 00:06:46.640
Over on the frequency plot, that corresponds to a much broader peak around the five beats per second.
00:06:47.880 --> 00:06:51.320
And that’s the uncertainty principle, just phrased a little bit more mathematically.
00:06:51.920 --> 00:06:55.760
A signal concentrated in time must have a spread out Fourier transform.
00:06:56.200 --> 00:06:58.560
Meaning, it correlates with a wide range of frequencies.
00:06:59.160 --> 00:07:03.120
And a signal with a concentrated Fourier transform has to be spread out in time.
00:07:04.400 --> 00:07:08.200
And one other place where this comes up in a really tangible way is Doppler radar.
00:07:09.000 --> 00:07:11.960
So with radar, the idea is you send out some radio wave pulse.
00:07:12.240 --> 00:07:14.200
And the pulse might reflect off of objects.
00:07:14.520 --> 00:07:19.960
And the time that it takes for this echo signal to return to you lets you deduce how far away those objects are.
00:07:20.760 --> 00:07:26.640
And you can actually take this one step further and make deductions about the velocities of those objects using the Doppler effect.
00:07:27.360 --> 00:07:29.200
Think about sending out a pulse with some frequency.
00:07:29.840 --> 00:07:36.400
If this gets reflected off an object moving towards you, then the beats of that wave get kinda smushed together.
00:07:36.880 --> 00:07:39.960
So the echo you hear back is gonna be a slightly higher frequency.
00:07:43.040 --> 00:07:44.920
Fourier transforms give a neat way to view this.
00:07:45.440 --> 00:07:49.400
The Fourier transform of your original signal tells you the frequencies that go into it.
00:07:49.880 --> 00:07:53.720
And for simplicity, let’s think of that as being dominated by a single pure frequency.
00:07:54.360 --> 00:07:58.800
Though as you know, if it’s a short pulse, that means that our Fourier transform has to be spread out a little bit.
00:07:59.480 --> 00:08:01.320
And now think about the Doppler-shifted echo.
00:08:02.320 --> 00:08:08.480
By coming back at a higher frequency, it means that the Fourier transform will just look like a similar plot shifted up a bit.
00:08:09.600 --> 00:08:14.640
Moreover, if you look at the size of that shift, you can deduce how quickly the object was moving.
00:08:16.600 --> 00:08:20.240
By the way, there is an important technical point that I’m choosing to gloss over here.
00:08:20.560 --> 00:08:22.520
And I’ve outlined it a little more in the video description.
00:08:23.000 --> 00:08:28.760
What follows is meant to be a distilled if somewhat oversimplified description of the Fourier trade-off in this set-up.
00:08:29.560 --> 00:08:38.120
The salient fact is that time and frequency of that echo signal correspond, respectively, to the position and the velocity of the object being measured.
00:08:38.680 --> 00:08:44.120
Which is what makes this example much more closely analogous to the quantum-mechanical Heisenberg uncertainty principle.
00:08:44.840 --> 00:08:54.280
You see, there is a very real way in which a radar operator faces a dilemma where the more certain you can be about the positions of things, the less certain you are about their velocities.
00:08:59.920 --> 00:09:03.680
Here, imagine sending out a pulse that persists over a long period of time.
00:09:04.480 --> 00:09:07.920
Then that means the echo from some object is also spread out over time.
00:09:08.720 --> 00:09:10.480
And on its own, that might not seem like an issue.
00:09:11.120 --> 00:09:13.840
But in practice, there’s all sorts of different objects in the field.
00:09:14.320 --> 00:09:17.320
So these echoes are all gonna start to get overlapped with each other.
00:09:17.920 --> 00:09:20.200
Combine that with other noise and imperfections.
00:09:20.520 --> 00:09:24.160
And this can make the locations of multiple objects extremely ambiguous.
00:09:24.840 --> 00:09:32.800
Instead, a more precise understanding of how far away all these things are would require having a very quick little pulse confined to a small amount of time.
00:09:33.960 --> 00:09:37.600
But, think about the frequency representations of such a short echo.
00:09:38.400 --> 00:09:44.680
As you saw with the sound example, the Fourier transform of a quick pulse is necessarily more spread out.
00:09:45.600 --> 00:09:56.200
So for many objects with various velocities, that would mean that the Doppler-shifted echoes despite having been nicely separated in time are more likely to overlap in frequency space.
00:09:57.000 --> 00:10:01.880
So since what you’re looking at is the sum of all of these, it can be really ambiguous how you break it down.
00:10:02.640 --> 00:10:10.400
If you wanted a nice clean sharp view of the velocities, you would need to have an echo that only occupies a very small amount of frequency space.
00:10:11.240 --> 00:10:16.040
But for a signal to be concentrated in frequency space, it necessarily has to be spread out in time.
00:10:16.600 --> 00:10:20.560
This is the Fourier trade-off; you cannot have crisp delineation for both.
00:10:23.000 --> 00:10:25.040
And this brings us to the quantum case.
00:10:25.720 --> 00:10:29.680
Do you know who else spent some time immersed in the pragmatic world of radio wave transmissions?
00:10:30.160 --> 00:10:35.000
A young, otherwise philosophically inclined history major in World War one France, Louis de Broglie.
00:10:35.640 --> 00:10:40.480
And this was a strangely fitting post, given his predispositions to philosophizing about the nature of waves.
00:10:40.920 --> 00:10:50.240
Because after the war, as de Broglie switched from the humanities to physics, in his 1924 PhD thesis, he proposed that all matter has wave-like properties.
00:10:52.640 --> 00:11:03.880
And more than that, he concluded that the momentum of any moving particle is gonna be proportional to the spatial frequency of that wave, how many times that wave cycles per unit distance.
00:11:08.480 --> 00:11:12.640
Okay, now that’s the kind of phrase that can easily fly into one ear and out the other.
00:11:13.080 --> 00:11:17.960
Because as soon as you say matter is a wave, it’s easy to just throw up your hands and say physics is just weird.
00:11:18.680 --> 00:11:19.840
But really, think about this.
00:11:20.320 --> 00:11:24.960
Even if you’re willing to grant that particles behave like waves in some way, whatever that means.
00:11:25.680 --> 00:11:34.920
Why on Earth should the momentum of those particles, the thing we classically think of as mass times velocity, have anything to do with the spatial frequency of that wave?
00:11:36.560 --> 00:11:42.400
Now being more of a math than a physics guy, I asked a number of people with deeper backgrounds in physics about helpful intuitions here.
00:11:42.920 --> 00:11:46.880
And one thing that became clear is that there is a surprising variety of viewpoints.
00:11:47.400 --> 00:11:54.120
Now personally, one thing I found to be interesting was just going back to the source and seeing how de Broglie framed things in his seminal paper on the topic.
00:11:55.080 --> 00:12:02.040
You see, there is a sense in which it’s not all that different from the Doppler effect, where relative movement corresponds to shifts in frequency.
00:12:02.840 --> 00:12:06.400
It has a slightly different flavor since we’re not talking about frequency over time.
00:12:06.760 --> 00:12:08.600
Instead, we’re talking about frequency over space.
00:12:09.040 --> 00:12:11.040
And special relativity is gonna come into play.
00:12:11.560 --> 00:12:13.320
But I still think it’s an interesting analogy.
00:12:15.400 --> 00:12:21.880
In his thesis, de Broglie lays out what is, in his own words, a crude comparison for the kind of wave phenomenon he has in mind.
00:12:22.560 --> 00:12:28.240
Imagine many weights hanging from springs, with all of these weights oscillating up and down in sync.
00:12:28.880 --> 00:12:31.720
And with most of the mass concentrated towards a single point.
00:12:33.320 --> 00:12:42.400
The energy of these oscillating weights is meant to be a metaphor for the energy of a particle, specifically the 𝐸 equals 𝑚𝑐 squared style energy residing in its mass.
00:12:42.960 --> 00:12:48.120
And de Broglie emphasized how the conception he had in mind involves the particle being dispersed across all of space.
00:12:49.000 --> 00:12:55.760
The whole premise he was exploring here is that the energy of a particle might have to do with something that oscillates over time.
00:12:56.360 --> 00:12:58.200
Since this was known to be the case for photons.
00:12:58.760 --> 00:13:02.640
And these oscillating weights are just meant to be a metaphor for whatever that something might be.
00:13:03.720 --> 00:13:14.040
With Einstein’s relatively new theory of relativity in mind, he pointed out that if you view this whole setup while moving relative to it, all of the weights are gonna appear to fall out of phase.
00:13:18.920 --> 00:13:19.760
That’s not obvious.
00:13:19.760 --> 00:13:22.280
And I’m certainly exaggerating the effect in this animation.
00:13:23.040 --> 00:13:25.520
It has to do with a core fact from special relativity.
00:13:25.880 --> 00:13:32.000
That what you consider to be simultaneous events in one reference frame may not be simultaneous in a different reference frame.
00:13:34.320 --> 00:13:40.040
So even though, from one point of view, you might see two of these weights as reaching their peaks and their valleys at the same instant.
00:13:40.600 --> 00:13:45.480
From a different moving point of view, those events might actually be happening at different times.
00:13:49.640 --> 00:13:52.640
Understanding this more fully requires some knowledge of special relativity.
00:13:52.640 --> 00:13:56.000
So we’ll all just have to wait for Henry Rice’s series on that topic to come out.
00:13:56.880 --> 00:14:05.360
Right here our only goal is to get an inkling for why momentum, that thing we usually think of as mass times velocity, should have anything to do with spatial frequency.
00:14:06.520 --> 00:14:13.200
And the basic line of reasoning here is if mass is the same as energy, via 𝐸 equals 𝑚𝑐 squared.
00:14:13.200 --> 00:14:17.640
And if that energy was carried as some kind of oscillating phenomenon, similar to how it is for photons.
00:14:18.200 --> 00:14:25.240
Then this sort of relativistic Doppler effect means changes to how that mass moves corresponds to changes in the spatial frequency.
00:14:27.400 --> 00:14:30.880
So what does our general Fourier trade-off tell us in this case?
00:14:32.520 --> 00:14:35.720
Well if a particle is described as a little wave packet over space.
00:14:36.280 --> 00:14:44.920
Then the Fourier transform, where we’re thinking of this as a function over space not over time, tells us how much various pure frequencies correspond with this top wave.
00:14:50.840 --> 00:14:57.720
So if the momentum is the spatial frequency up to a constant multiple, then the momentum is also a kind of wave.
00:14:58.080 --> 00:15:01.600
Namely, some multiple of the Fourier transform of the original wave.
00:15:03.560 --> 00:15:13.280
So if that original wave was very concentrated around a single point, as we have seen multiple times now, that means that its Fourier transform must necessarily be more spread out.
00:15:13.960 --> 00:15:18.000
And hence, the wave describing its momentum must be more spread out and vice versa.
00:15:22.360 --> 00:15:30.120
Notice, unlike the Doppler-radar case where the ambiguity arose because waves were being used to measure an object with a definite distance and speed.
00:15:30.680 --> 00:15:33.760
What we’re saying here is that the particle is the wave.
00:15:34.520 --> 00:15:39.880
So the spread out over space and over momentum is not some artifact of imperfect measurement techniques.
00:15:40.120 --> 00:15:42.720
It’s a spread fundamental to what the particle is.
00:15:43.200 --> 00:15:49.440
Analogous to how a musical note being spread out over time is fundamental to what it even means to be a musical note.
00:15:50.480 --> 00:15:59.680
One pet peeve I have in mainstream references to quantum is that they often treat Heisenberg’s uncertainty principle as some fundamental example of things being unknowable in the quantum realm.
00:16:00.120 --> 00:16:03.200
As if it is a core nugget of the universe’s indeterminacy.
00:16:03.960 --> 00:16:13.920
But really, it’s just a trade-off between how concentrated a wave and its frequency representation can be, applied to the premise that matter is some kind of wave and hence spread out.
00:16:14.640 --> 00:16:17.840
All of the stuff about randomness and unknowability is still there.
00:16:17.840 --> 00:16:21.960
But it comes one level deeper, in the way that these waves have come to be interpreted.
00:16:22.960 --> 00:16:26.920
You see, when you measure these particles, say trying to detect if it’s in a given region.
00:16:27.520 --> 00:16:30.400
Whether or not you find it there appears to be probabilistic.
00:16:30.960 --> 00:16:36.000
Where the probability of finding it is proportional to the strength of the wave in that region.
00:16:36.840 --> 00:16:47.400
So when one of these waves is concentrated near a point, what that actually means is that we have a higher probability of finding it near that point, that we are more certain of its location.
00:16:48.400 --> 00:16:50.120
And just to beat this drum one more time.
00:16:50.120 --> 00:16:57.040
Since that concentration implies a more spread-out Fourier transform, then the wave describing its momentum would also be more spread out.
00:16:57.680 --> 00:17:02.600
So you wouldn’t be able to find a narrow range of momenta that the particle has a high probability of occupying.
00:17:03.960 --> 00:17:10.000
I quite like how if you look at the German word for this principle, it might be more directly translated as the unsharpness relation.
00:17:11.360 --> 00:17:16.400
Which I think more faithfully captures the Fourier trade-off at play here without imposing on questions of knowability.
00:17:17.200 --> 00:17:22.800
When I think of the Heisenberg uncertainty principle, what makes it fascinating is not so much that it’s a statement about randomness.
00:17:23.280 --> 00:17:27.680
I mean, yes that randomness is very thought-provoking and contentious and just plain weird.
00:17:28.200 --> 00:17:37.560
But to me, equally fascinating is that underpinning Heisenberg’s conclusion is that position and momentum have the same relationship as sound and frequency.
00:17:38.320 --> 00:17:45.160
As if a particle’s momentum is somehow the sheet music describing how it moves through space.