Hacker News new | past | comments | ask | show | jobs | submit login
Video filmed at four trillion frames per second captures light in a flash (nature.com)
387 points by chriskanan 32 days ago | hide | past | web | favorite | 85 comments

This is more impressive than previous claims of trillion FPS cameras, because it captures it all sequentially from a single event. Other such claims were based on capturing one frame at a time of a repeating event with different time delays.

You could use an array of cameras and capture the same event at different times

You see that light pulse that they captured?

Your camera synchronization signal in the wire is traveling about 2/3 of that speed. Can you imagine all the cameras in your array receiving and interpreting that signal at the same time?

You pretrigger the cameras while accounting for the delay.

why would you need to synchronize those cameras at all? Let them start at random times. Light is supposed to go in one direction, so you can tell which picture were shot ealier by comparing how far the light beam is. Of course sorting billions of pictures would require some form of Machine Learning (or really large number of interns)

The change of perspective is undesirable. Unless you're filming the matrix: https://youtu.be/bKEcElcTUMk !

Or something really far away, like lightning.

Not really - there is no way to synchronise the start of framecapture that finely by that many cameras. Using a single camera but clever optics to "temporally spread a tiny image over a large sensor" to effectively create a sprite sheet is kind of your only option.

I'm confused how you could film a light pulse this way. It seems that the light pulse would itself have to be emitting light somehow. I'll assume it was travelling through a gas that scattered it.

The funny thing is that you can’t see the event in real time. You have to wait for the scattered light from the pulse to reach the camera sensor, kind of like the trailing sound from an airliner passing overhead, and by that time the actual pulse is out of frame.

"The imaging system captures a light pulse (in a slowed video) as it passes through a material, exits the material (at the dashed yellow line) and bounces off a mirror (solid yellow line) back into the material."

Is this trying to imply that it is capturing the scattering off of the Frenel diffraction? I don't understand at first glance how this paper is capturing the light without having a sensor in the path of propagation.

The light is being put through a translucent medium, like shining a laser through milk or smoke or something, so you can see the scattered light. Conventional/non-QM-based light physics can handle this situation.

If you mean, how are they seeing the non-scattered light, you are correct that they are not.

> light is being put through a translucent medium

at certain points, the light pulse seems to "leap ahead" in fits and starts on its journey. is that because of the way the medium scatters the light or something?

I think that it's because the speed of light inside the translucent medium (left side) is lower than in air (right side).

Ah okay I get it. For some reason, I thought it was travelling through the material mentioned in that quote during the time it disappeared in the video.

I'd be curious to see the dual-slit experiment done using this camera as the "observer".

I'm not sure it'd yield anything we don't already "know" - but it might be worth trying just because.

I wondered exactly the same thing (see my other comment). However, I am not sure if they can actually film single photons. My school physics are a bit rusty, but iirc, quantum effects will not be relevant if you actually fire a light "impulse" consisting of many, many photons. However, as I have asked in my previous comment, it would be interesting if you could actually see the light impulse being "split up" and interfering with itself.

You can detect single photons. There are off the shelf detectors for available for experimenting with photon entanglement and parametric down conversion. We had one at my university.


You can't make movies with them though... It would make an expensive camera. ;-)

You can only "film" something when the photons hit the "film" (a.k.a the image sensor). There's no way to record photons flying by.

Is it not possible for the energy of a photon to to disrupt EM fields enough to be detectable, without the photon actually being incident on a sensor?

Yes, this is called a "quantum non-demolition measurement" and has actually been done. One technique is to use highly excited atoms (Rydberg atoms) which are extremely sensitive to external electric fields. In this way, the field caused by a single photon can be detected without absorbing the photon. There is a famous paper from Serge Haroche's group where they used this method to detect the presence of single microwave photons in a superconducting cavity (https://www.nature.com/articles/nature05589). However, the measurement still collapses the wavefunction of the system. Performing a QND measurement to determine which slit the photon takes in a double-slit experiment simply destroys the interference pattern.

Curious, what is "the field generated by a single photon" made of? Off hand, I think the force carrier for EM fields is the photon no? So, the photon is emitting other photons (virtual I think?) then?

Kinda, no. By observing a photon you perturb the system, you wouldn't get a result for what that photon would have done had it not been perturbed. AIUI if you detect the photon then it interacted with some part of your sensing apparatus, so it didn't then go on to do what it should have. Thus we're left to look at the aftermath and impute the situation that produced it.

If you could do that then you could also know the fate of Schrödinger’s cat ...

In this example they are capturing light scattered from the main pulse as it passes through the medium, they are no capturing the light pulse itself. If there is one photon then you can observe it only once (when it is absorbed by your detector).

Not really. The double-slit experiment is only interesting when done with single photons, and this camera can't capture that (no camera can).

> The imaging system captures a light pulse (in a slowed video)

I like that they remind us that the video is slowed down and that time doesn't actually move at 10 picoseconds per second.

If it did move at 10 picoseconds per second, how would we know?

Whoa. Stares at hand

Makes me think of those videos of house plants, which we normally think of as still objects. But with a time-lapse video, you can see them moving around a bunch during 24 hours.


Heck, with the right plants you can see them grow over the course of minutes: https://www.brainonfire.net/blog/2018/03/29/seeing-plants-mo...

Hold up. Now we're saying it has acceleration!?

Whoa. stares at the sun

This article caused me to remember something I heard on a Youtube video and wanted to follow up on. Essentially, is there (and if so what is) the framerate of the universe? It makes logical sense to have a minimum bundle of time based on Planck's constant, so I did some googling and found the answer (I have not checked this math)[1]:

> The Planck time is 5.39 × 10-44 seconds. No measurable time can be shorter than that according to quantum physics.

Converting to FPS, that gives us:

> One thousand eight hundred and fifty-five billion billion billion billion frames per second. 18.55 septillion FPS!

So if like me you wondered if a trillion FPS is close to the maximum possible frame rate in the universe, the answer is nope!

[1]: http://www.librador.com/2009/01/16/The-frame-rate-of-the-uni...

Don't confuse "our math breaks down at this point" with "it can't physically happen".

Planck time is the time at which, if you want to do physics, you have to take into account quantum mechanics and gravity (i.e. understand quantum gravity). Something we don't know how to do. But it doesn't necessarily mean time is discrete. Maybe it is. Maybe it isn't.


Thanks for the link. I was referring to measurable time (since for frame rate we must have discrete intervals), but it sounds like that is still up for debate.

Does the same reasoning apply to the Plank length or is that one discreet?

As I understand it, currently everything points to space and time (really, spacetime) not being discrete. I'm copying and pasting a previous reply I made the last time this came up.


I'm no physicist, but the layman explanation I've heard is that the concepts of Planck length and Planck time do not imply that spacetime is discretized into little "voxels"; instead these quantities are just limits on the uncertainty of any possible measurement.

As I understand it there are two contributing facts:

a) the Heisenberg uncertainty principle states that there's a tradeoff between certainty in position vs momentum, so if you're more certain in a particle's position you're less certain of its momentum. (For photons, momentum is proportional to frequency, i.e. wavelength, and frequency is proportional to energy.)

b) By mass-energy equivalence, anything with energy has mass, therefore higher frequency photons are more "massive". A single photon of sufficiently high frequency would form a black hole.

Putting those two together, to measure distances accurately you need higher and higher frequency photons with shorter and shorter wavelengths. For example, radar creates blurry images at ~5cm wavelengths, while ordinary photographs can be razor sharp with wavelengths of only ~500nm. The Planck length is just the wavelength at which the photon would have so much energy that it would collapse into a black hole and break our current mathematical models. That's why it's nonsensical--with current models--to talk about lengths smaller than a Planck length, but it doesn't mean that space itself is quantized. Similar argument for time.

(Also, the same logic applies to other particles like electrons, protons, and even up to macroscopic items like baseballs; everything has a wavelength...)

Semi-related question: if photons can have various frequencies, does that mean protons can have various frequencies?

AKA temperatures?

I think temperature is a property based on the average kinetic energy of particles in a system. The frequency of a proton, I imagine, would be an intrinsic property.

Correct me if I'm wrong, but the Planck time concerns the smallest possible observable unit of time, but not necessarily the smallest possible unit of time, correct? As a bystander I get the impression that [science currently thinks that] it's beyond the limits of science to prove whether or not a unit of time smaller than the Planck time exists, and since there's no natural law that says that things only exist if they are provable, that means that the "tick rate" of the universe might be smaller than the Planck time.

Planck time is not the smallest observable unit of time, it is the smallest time step that present-day physics makes predictions about. Subtly, present-day physics might actually break down on timescales longer than the Planck time: it's not that we know it doesn't break down until Planck time, we know that it must break down at Planck time.

Susskind might say that it's not useful to take about a phenomena at all unless you have a theory about it.

It might really be a boundary, but we don't know

> Planck time […] is the smallest time step that present-day physics makes predictions about

What are those predictions? Do you mean that

> we know that it must break down at Planck time

? If so, do you have a source for this prediction? (I'm not asking why present-day physics must break down at extremely small scales. That prediction is more or less well established. What I'm asking is: Why is it the Planck scale that's so significant?)

My impression was that most statements involving the Planck scale these days are rather numerology than actual, verifiable science [1].

[1] https://physics.stackexchange.com/questions/185939/is-the-pl...

This camera doesn't actually take 4 trillion shots a second. It shoots a light, takes a frame, shoots the light, waits a tiny bit longer, takes a frame, and so on.

The phase shift is 1/(4 trillion) sec, but the actual framerate is much much slower.

That is how an earlier trillion FPS camera worked, but in this case they do capture a handful of frames of a single event at a 100 femtosecond resolution.


That is how these cameras usually work, but not this one; the article describes a system of overlapping, time-shifted exposures instead, from a single event.

The same trick is used by some oscilloscopes for high-time-resolution sampling of periodic signals.

I wonder if they also average the same phase-out shift multiple times to reduce noise (improving the bit depth of a single frame).

> 18.55 septillion FPS!

intel has some room to grow I see

FPS implies a single final render target. That’s a big assumption to make of the universe!

A billion is 10^9 and a septillion 10^24. So a billion billion billion is already 10^27. One of your numbers is wrong.

A billion is a million millions in Britain, but a thousand millions in the US. His numbers are right, just a different scale

The UK hasn't used the long scale billion since 1974

I could swear I read somewhere that they still did. Oh well

The only English-speaking countries left that still occasionally use the long-scale are Canada, Mauritius, Seychelles and Vanuatu. This is because they are bilingual nations that also speak French, where the long-scale is still used.

Trying to do locales in Canada is always fun, since we can't make up our mind. "How far? 100km." "How tall? 5ft"

Even if that would be the case and we use the long scale (million/milliard/billion) system, which I actually grew up with in Germany, a billion would be 10^12 and a septillion 10^42. It still doesn't add up.

>A billion is a million millions in Britain //

Not really. I think you'll struggle to find anyone who uses that now.

CalTech films light at the speed of 10 trillion frames a second! The "Slow Mo Guys" went to caltech to take a look: https://www.youtube.com/watch?v=7Ys_yKGNFRQ

I used to work on HP way back in 1991 on equipment that measure timing of signal to 7.9 pico second resolution. This is very cool.

Love to see an ifixit tear down video on that CalTech camera. :-)

Interesting video, though I wish they would have explained more about what exactly we are seeing. Like the last video, is the flash in the middle of the frame the light as a particle?

I'm blown away by this. If I'm understanding this correctly, this picture[1] is light traveling for a distance of about 3/8ths of an inch.

[1] https://media.nature.com/w800/magazine-assets/d41586-019-016...

Wait, the light is all incoherent before it passes the dotted line, then coherent until it bounces and passes back through? What's going on?

I believe the area between the dotted and solid lines is a mirror. The blob of light is visible because it's passing through some medium that scatters it, then when it disappears it has been absorbed by the mirror, then we see it reemitted shortly after.

The solid line is the mirror, the dotted line is the edge of the scattering medium. I assume there is air or vacuum in between.

You have to remember that's scattered light.

Did you see measurements somewhere or are you just estimating from the time counter and the speed of light?

Estimating based on the time counter.

I wonder how the authors feel that their work is so good that the website for Nature has a piece on it but not good enough to be published in a Nature journal.

Maybe the authors didn't want to submit to Nature?

Also, not a physicist, but my understanding is that PRL is prestigious in its own right.

When I worked in a physics lab a PRL publication was seen as a big deal.

So, looking at the video, I am curious: is there a video where they fire a light impulse against a double-slit experiment setup? Would it be possible to see the light being "split up" and interfering with itself?

You can't see the light not going into the camera (obviously) so what you see is some light from the beam being reflected into the sensor. You can't take a picture of a photon, if that makes sense.

So in the double slit experiment you could see the path the light took only if you allowed for a portion of that light to be reflected towards the camera. But then if you did that you wouldn't end up with anything particularly new or interesting as far as I can imagine.

Of course, you are right! I intuitively thought of the light impulse emitting further photons in the direction of the camera, which makes no sense.

Can anybody comment on what is meant by a "frame" here?

My intuition tells me that what we consider a "frame" in our daily experience (24 up through maybe 144) would be pretty different from what this "frame" would be, in terms of how it is captured and how it is subsequently rendered.

Any ideas?

We do not usually consider a frame duration to measure a length, because for most of the durations we use the frame's size is absurdly large in practical terms, ranging from thousands of miles to larger than the diameter of the earth. So we can take a shortcut of identifying a frame as some area from the camera's point of view identifying an infinite, instantaneous volume in front of us, and don't consider the fact that light isn't instantaneous and in fact has a transmission time.

Frames this short mean that there is a sort of volumetric aspect to what is being catured. From the point of the camera, when light collection begins an area sweeps out in front of the camera and back in time at the speed of light, and when the light collection stops, the end of the area sweeps out from the camera lens in the same direction. The result of this is a spherical shell volume travelling backwards in time that represents the spacetime locations that can emit light and appear in the frame [1]. The resulting picture is an integration of the photons emitted in this 4D spacetime volume, in the direction of the camera, projected onto a 2D image.

(That's assuming a point-like camera and instantaneously turning the camera on and off; in reality it'll be fuzzier but the principle holds.)

The difference is that at these incredible fast speeds, the resulting 4D spacetime volume is of human size in most directions in the places we care about [2], measured in human-sized measuring units like "centimeters" rather than "light-seconds". Normally we can neglect thinking about this volume because we simply spray so "much" out that we don't have to think much about capture exactly what we want, in much the same way that in normal day-to-day life we tend to act as if lightspeed is simply infinite. In this case, we actually have to think about it to get what we want.

A similar effect can be seen in network equipment, for what it's worth; at the highest network speeds, we've now significantly passed the point where a given bit being transmitted is now a human-sized fraction of a wire. If you could "snapshot" a 100Gb network cable, you could see bits on the wire. If I'm getting my math right, each bit is on the order of 2-3 centimeters, give or take the medium not being fully light-speed.

[1]: It may be more intuitive for a moment to instead imagine that the camera is starting to emit the shell at the time that it is turned on, and stops when turned off, forward in time, as that creates a more intuitive initial image of what-seems-to-be-cause preceding what-seems-to-be-effect. Imagine the camera emitting light instead of collecting it. It may be easier to then imagine this shell going out farther and farther, getting larger as it goes (and the light inside dimming as it has to fill that volume with the same amount of energy it started with), indefinitely out into the universe. But since the camera is receiving instead af transmitting photons, the reality is the same image, just with time reversed; as the time goes farther back, the shell is farther away from the camera and larger.

[2]: Technically it extends all the way in the direction the camera is all the way out in space back to the Big Bang or the CMB or something, but unless you're deliberately photographing space, and why you'd do that with this camera I have no idea, you don't have to worry about that.

a frame is just a picture essentially. What you're refering here with 24 up to 144 are usually frames per second. Here we're talking about 4 trillion frames, so a frame would be 1/4 trillion seconds.

This technology could give us the opportunity to verify a ton of our physical and chemical models with direct observation. The demo may not be exciting but if they can scale it down you can expect a lot of exciting discoveries. I can't wait to see what we're wrong about.

> The imaging system captures a light pulse (in a slowed video)

Good to know that's not the actual speed of light!

Anyway, it looks like the photons are "crawling" by their patterns of speed, rather than traveling at a fixed speed.

I would think that effect is due to impurities or lack of uniformity in the material which causes some spots to light up more than others.

Is there any explanation for why it is so jerky?

Sampling rate isn't high enough is what I'd imagine, even at 4 trillion FPS.

Looks like a matplotlib chart turned into an animation.

I've had conversations with a guy who works at one of the companies that makes high-speed cameras. One of the interesting thing to come out of those conversations is that below a certain threshold of frame duration (which is related to, but not necessarily the recriprocal of the frame rate) a camera with that capability becomes subject to ITAR, because supposedly said camera could be used to develop nuclear weapons.

My intuition is that the camera described in this article would be subject to ITAR were it developed in a western nation. But, I doubt China gives any fucks about that.

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact