Did AI Prove Our Proton Model WRONG? (script) (Patreon)

People are made of cells, cells are made of molecules, molecules are made of atoms, atoms are made of electrons and nucleons, and nucleons - protons and neutrons - are each made of three up or down quarks. Simple stuff, right? All except for that last part. Nucleons are actually made of many, many quarks that happen to look like three only when we look at them in a particular way. And even then, sometimes they’re made of 5 quarks - including the charm quark

The humble proton may seem simple enough, and they’re certainly common. Most of the visible matter in the universe comes from protons, either as lonely hydrogen nuclei or bound with neutrons into the nuclei of the various elements of the periodic table. You’d think by now we’d have a pretty good idea of the properties of the proton. But actually, their interiors remain a profound mystery. It wasn’t until the late 60s that we even confirmed out that protons were not themselves elementary particles, but consistent of 3 quarks. Now, thanks to AI… we have evidence that suggests sometimes they may be made of five quarks

To sort through all of this, we need to understand how physicists probe the smallest scales of nature. And that means understanding the physics of scattering. Every time you open your eyes you’re performing a scattering experiment. Photons from some light source like the Sun bounce - or scatter - off objects in the world around you into your particle detectors—aka your eyes. Your analysis computer—aka your brain—then builds up a colour map of the world around you based on the properties of those photons.

Well, it’s also possible to “see” the subatomic world if we do a different type of scattering experiment.

In general, the higher the energy the scattering particle has, the smaller the object you can see. To see subatomic scales, it’s better not to use photons at all, but instead to use particles of matter. That's how Earnest Rutherford first discovered the nucleus. He shot a beam of alpha particles - which we now know to be helium nuclei - at a thin gold foil. He found that, while most particles passed straight through, a small fraction scattered of something very small and very dense. That’s how he figured out that the atom is mostly empty space, with most of its mass concentrated in a tiny central nucleus.

Since Rutherford’s 1911. experiment we’ve gotten much better at doing this. For example, we now have electron microscopes which shoot electrons into a sample and measure how they’re deflected. Sophisticated software then reconstructs the structure of the sample, just as our brain reconstructs the structure of the world.

Electron scattering can be used to study the structure of anything larger than an electron. That sounds pretty useful, because electrons have no size—they’re point-like. Actually not so fast—electrons are quantum objects and so have a wavelength that gets shorter the more energy the electron has. So really I should say that electrons can be used to study anything that’s larger than the electron’s wavelength, which depends on energy.

So if you want to study the interior of a proton you need a pretty high energy electron. Below a certain energy, the electron will just bounce off the proton as a whole. But with enough energy the electron will punch into a proton and then scatter off the proton's internal parts. I should add that such an electron is energetic enough to destroy the proton. Which sounds bad, but it’s actually okay - sometimes you need to break something to see what it’s made of.

And that’s what physicists at the Stanford Linear Accelerator Center—SLAC—managed to do though the 50s and 60s. They accelerated electrons to higher energies than had previously been achieved and slammed them into protons, then watched what came out the other side. In 1968 they published their results. From those scattering products and the deflection of the electrons, the SLAC physicists deduced that protons must be made of three point-like particles. These particles matched the properties of the so-called quarks—theoretical components of nucleons proposed by Murry Gell-Mann and George Zweig earlier that decade. And that’s how we get the proton model that you’re probably familiar with—two up and one down quark, with three different charge colours bound to each other by gluons carrying the strong nuclear force, as we’ve talked about in previous episodes.

While the SLAC electron beam was the most energetic of its time, we quickly figured out how to make more and more powerful accelerators. And as we did so, we started to see something strange. Remember I said that higher energy scattering allows us to see smaller things and finer resolution. Well, as we powered up our accelerators that’s exactly what we saw—the detailed guts of the proton, and it turned out to be much more complicated than just three quarks bound to each other.

The interior of the proton was revealed to be a complex cluster of energy—a dense network of gluons that are constantly transforming into pairs of virtual quarks and antiquarks, which quickly annihilate each other, turning back into gluons. We call this the quark sea, and this constantly shifting, flickering mess is a stormy ocean indeed. But because the quantum properties of charge, spin, and colour must be conserved, there is order within the chaos. If you took all the contents of the quark sea in one instant, you could list all the quarks and all the antiquarks, and see that they cancel each other out except for two up quarks and one down quark. These “valence” quarks will constantly exchange energy and colour charge via the quark sea and may even be annihilated by a virtual particle or antiparticle, but are instantly replaced by that virtual particle’s partner.

These valence quarks are what the first SLAC experiments detected. But as we increased our electron beam’s energy, those electrons started to interact with the transient entities of the quark sea. They scattered off gluons and virtual quarks, revealing that complex inner structure. The higher the energy of the electrons, the more fine detail we saw. They saw the virtual up and down quarks of the quark sea, the gluon network. But also some weird stuff. For example, around 1% of the time there was evidence of a charm quark inside the proton. Well that’s odd. The charm quark weighs more than the entire proton—36% more in fact. It’s like opening a 1 kg box of… I dunno, apples, and finding a 1.3 kg melon inside.

Fortunately there’s a perfectly plausible explanation in the case of the proton. It turns out that the particles we discover after smashing an electron into a proton aren’t necessarily inside the proton to start with. That’s because the energy carried by the electron can create brand new particles. Remember, E=mc^2, which tells us that mass and energy can be converted into each other. As the ingoing electron transfers its substantial kinetic energy into the quark sea, new particle-antiparticle pairs can be created that were not part of the proton to start with. As the energy of the electron beam increases, we get a scattering signal from finer and finer structure. But this signal is increasingly muddied with new particles created in the collision.

We call the particles that are inside the proton to start with intrinsic particles, and they include the valence quarks and the virtual quarks and gluons of the quark sea. On the other hand, we call particles created in the collision itself extrinsic particles. After a particle collision we see both intrinsic and extrinsic particles, as well as the decay products of those particles in cases where the outgoing particle is too unstable to reach the detector before falling apart. So if the ingoing electron has enough energy, perhaps it could produce charm quarks that are detected in some cases. If the charm quarks are EXtrinsic then there’s no problem.

But even early on, there was some evidence of INtrinsic charm quarks. There was the hint of the presence of charm quarks even in cases where the electron didn’t carry enough energy to produce one. So… if the charm quark is too massive to be produced by the incoming electron, and is too massive to be a regular part of the entire proton, how can it exist? It’s not quite as simple a conundrum as I make it sound. As you reduce the energy of the collision, the probability of finding an extrinsic charm quark also goes down, only reaching zero at zero energy, which means no collision at all. That’s how the evidence of intrinsic charm could be considered weak. Charm quarks were appearing only slightly more often at low energies than expected from our calculations.

Let’s talk about how those calculations are done for a second to see why this uncertainty exists. The interior of the proton is the realm of the strong nuclear force, which is described by quantum chromodynamics. With its multiple different charge types and different charge carriers, it’s far more complicated than the single-charged electrodynamics. One weird thing about “QCD” is that the calculations are easier at high energy than at low energy because it’s easier to implement a favorite hack of quantum mechanics called perturbation theory. That means it’s easier to calculate the result of a high-energy scattering event than a low one. It’s extra hard to properly calculate the state of the interior of the proton in the absence of a collision. For this reason it is easy to explain all the extrinsic quarks we see at high energies, but it's very difficult to explain the behavior of the three intrinsic quarks we see at low energies, and it's even harder to explain why we seem to find these charmed quark at those same low energies

Although it’s hard to model the interior of the proton at low energy, our intuition still tells us that it shouldn’t contain things more massive than the proton itself—like the charm quark. But there is actually a way to get quantum chromodynamics to give us intrinsic charm quarks. We do that by taking advantage of the Heisenberg uncertainty principle, which, among other things

tells us that we can borrow energy from out of nowhere to create massive particle-antiparticle, as long as those particles vanish again very quickly. The more massive the particles, the shorter they’re allowed to live. For this reason, it should be possible to generate a charm-anticharm quark pair, even inside a proton, as long as it exists only for a very short duration.

Due to the large mass of the charm quark pair, their momentary appearance is a bigger deal for the proton’s inner structure than is the continuous flickering of virtual up and down quarks. In that instant, the proton looks more like a 5-quark particle than a 3-quark particle. However I wouldn’t go so far as to say that the proton briefly becomes more massive. The proton’s measured mass comes from the average of its internal energy over the time of measurement. So if the charm quarks only exist for a tiny fraction of the time, then they contribute only a fraction of their own hefty mass to the proton’s mass.

This is a crude explanation of the theory of intrinsic charm, proposed by Stanley Brodsky in 1980 to explain the weak evidence of charm quarks in protons. But proving this was challenging due to the difficulty of modeling low-energy particle collisions. One issue is that there are typically many different models of the proton interior that can lead to the prediction of the same scattering results. So, you might come up with a model that includes intrinsic charm that gives a perfect match for the average particle output over many collisions. But you may have gotten it right just by chance, and if you analyzed even more collisions you'd realize the correct model is actually totally different.

Without testing all possible models, it’s impossible to know whether a model of intrinsic charm was really better at predicting the outcome of these collisions, versus some unknown model that doesn’t include intrinsic charm. So there’s a more detailed explanation of why there’s such uncertainty about the existence of charm quarks in the proton

—it just wasn’t possible to sift through all of the possible models to be sure that intrinsic charm was needed to explain the scattering experiments.

That is, until Artificial Intelligence came along. This new tool allowed scientists in the  NNPDF collaboration to do something that was previously impossible: Instead of testing one model, they could test hundreds or thousands of models at the same time. They trained a neural network to analyze nearly 30 years of proton collisions, not constrained by a single model, but in the limit of all possible models. The network is then able to create new models that approach the scattering data as closely as possible.

This was incredible. Instead of a few scientists trying a few models every couple of years, they could have this machine test thousands of models in just a couple of days, and while a scientists may be biased to want to find evidence in favor of their model, the network just wants to find the best answer, even if  it contradicts intrinsic charm.

And now we get to the punchline. This Neural Network was able to find a model with intrinsic charm that predicts the data much better than any previous model, and the team finds that the chance of finding in favor of intrinsic charm by pure chance is only 1 in 1000, which we call a 3-sigma result. Of course the gold standard for claiming a discovery is 5-sigma, which means the probability of being wrong is roughly 1 in a million. That sounds like a high standard, but with so many different experiments happening around the world, we do get a lot of 3-sigma results that end up being wrong.

So will this one pan out in the end? Will we find the charm quark in the proton? The melon in the apple box? We’ll need to blast a lot more protons to find out, but we may also get closer to our answer by improving this shiny new tool of machine learning. That technology is developing at a rapid pace, as it’s increasingly used across all areas of experimental physics. Computers can come up with and test models much faster than any physicist, which means the physicists can get on with the more interesting work of interpreting the successful models.

Wherever we land on the question of intrinsic proton charm, there’s something … charming about this cooperation between artificial and natural intelligence towards the common goal of deciphering the inner workings of space time