The Shortcut That Holds Up Modern Physics — and Where It Snaps
Every fluid simulation ever run relies on a guess about information you can’t see. Two new results pin down, exactly, when that guess fails.
Keywords: information geometry, relativistic hydrodynamics, kinetic theory, quark–gluon plasma, statistical curvature, bulk viscosity, moment closure, Efron curvature, conformal limit, statistical inference
A crowd you can’t count
Imagine standing at the edge of a stadium holding a hundred thousand people. Someone asks you to predict how the crowd will move in the next ten seconds — where it will surge, where it will thin out, how a wave will ripple through it. You cannot possibly track every individual. So you do what everyone does: you summarize. You talk about the average density here, the average direction of motion there, how fast people are walking on this side versus that one. A handful of numbers stand in for a hundred thousand stories.
This is, almost exactly, the situation physicists face when they describe a hot gas — and not just any hot gas, but the hottest, densest matter we can make on Earth. When heavy atomic nuclei are smashed together at nearly the speed of light, they briefly melt into a soup of fundamental particles called the quark–gluon plasma: the same stuff that filled the universe microseconds after the Big Bang. To describe how that soup flows and expands, physicists use the same trick as the stadium observer. They don’t track every particle. They track a small set of summary quantities — the energy, the pressure, the way the fluid is being squeezed or sheared — and they write down equations for how those summaries evolve.
That field has a name: relativistic hydrodynamics. It is one of the great success stories of modern physics, matching collider data with startling precision. And it rests entirely on a step so routine that most people never stop to examine it. That step is the subject of the two pieces of work I want to tell you about — because it turns out that, examined closely, the routine step has a hidden flaw with a beautiful and exact mathematical shape.
The guess at the heart of everything
Here is the problem with summarizing a crowd. The moment you reduce a hundred thousand people to a few averages, you have thrown information away. And the future of the crowd doesn’t just depend on the averages — it depends on some of the details you discarded. The fast walkers near the exit, the cluster forming around a dropped wallet, the subtle correlation between where people are and which way they’re facing. Your summary equations need those details to predict the next moment. But you don’t have them anymore. You threw them away when you summarized.
So you guess. You reconstruct the missing details from the averages you kept, using some rule that says “given these summary numbers, the full picture probably looks like this.” In hydrodynamics this reconstruction step is called closure, because it “closes” the equations — it lets a finite set of summary equations stand on their own without referring back to the infinite mess of microscopic detail. Every fluid simulation you have ever heard of, from weather models to the quark–gluon plasma, depends on a closure rule of some kind.
And here is the uncomfortable truth that working physicists know but rarely dwell on: the closure is a guess. A sophisticated, physically motivated, often very accurate guess — but a guess. The standard recipes (they go by names like Grad’s method, Israel–Stewart theory, and the DNMR equations) all amount to assuming the discarded details have a particular tidy form. Nobody thinks the assumption is exactly true. The question that has hovered over the field for decades is: how wrong is it, and where?
That question sounds vague. The first of the two results I want to describe makes it sharp.
Measuring what you threw away
To measure how much information a summary throws away, you need a ruler for information itself. Remarkably, statisticians built exactly such a ruler decades ago, for a completely unrelated reason, and it has been sitting in the toolbox of a field called information geometry ever since.
The idea is easier to feel than to state. Picture every possible description of your gas — every possible full distribution of particles — as a point in a vast abstract space. The “true” descriptions form a smooth, well-behaved landscape in that space. Now picture the smaller, simpler set of descriptions your summary is capable of representing — the ones your closure guess can produce. That smaller set is like a sheet of paper laid across the landscape. If the paper lies flat against the landscape, your summary can perfectly capture what’s going on. If the paper has to bend and curve to stay near the landscape, that curvature is precisely the information your summary cannot hold. It is the gap between what reality is doing and what your simplified description can ever say about it.
There is an exact mathematical quantity that measures this bending. It is called the statistical curvature, introduced by the statistician Bradley Efron in the 1970s to quantify how much a simplified statistical model loses compared to the full one. The larger the curvature, the more the simplification distorts, and the more information leaks away unrecoverable.
The first result takes this fifty-year-old statistical ruler and, for the first time, lays it against the relativistic gas. It asks: how curved is the sheet of paper that hydrodynamic closure is built on? How much information does the standard guess actually discard?
The answer is dramatic.
It doesn’t just fail — it fails infinitely, in a specific place
As the particles in the gas get lighter and lighter — approaching the idealized “massless” limit that physicists love because it’s clean and symmetric — the statistical curvature does not stay finite. It blows up. It runs off to infinity. In the precise mathematical sense, the information discarded by the standard closure becomes unbounded exactly as the gas approaches its most elegant, most symmetric state.
That is already striking. But the shape of the blow-up is where it gets interesting, and where the result earns its place.
First, the location. You might reasonably expect the trouble to show up in the part of the physics that fluid dynamicists care about most: the shear, the sloshing and sliding of the fluid that carries viscosity and drives the spectacular flow patterns seen in collider data. It doesn’t. The shear part stays perfectly well-behaved. In fact, it settles down to a tidy, exact number — the fraction 90/7 — and just sits there, calm, no matter how light the particles get. The infinity hides somewhere else entirely: in the compression part of the physics, the bulk squeezing of the gas. All the discarded information, all the divergence, all the failure of the guess, is confined to that one channel. The part everyone watches is fine; the part fewer people watch is where the description quietly breaks.
This is the kind of inversion that makes a result worth publishing. It says: the closure isn’t uniformly bad, it’s selectively bad, and it’s bad precisely in the channel — compression, the trace, the bulk — that has long been the troublesome corner of relativistic fluid theory for reasons that previously seemed unrelated.
Second, the fingerprint. The exact pattern of the blow-up turns out to depend on what kind of particle the gas is made of. There are, broadly, two families of fundamental particles: the kind that like to crowd into the same state (bosons — think photons, the particles of light) and the kind that refuse to (fermions — think electrons). The math of the breakdown comes out different for each. For the “classical” idealized case the curvature blows up following one precise power law. For the crowd-avoiding fermions it follows the same power but with a different exact coefficient. And for the crowd-loving bosons it follows a softer power — it still blows up, but more gently, because the way bosons pile up at low energy changes the underlying integrals.
The upshot is almost poetic: in principle, you could look at nothing but the mathematical signature of how the description fails and read off what type of particle you were dealing with. The failure is not noise. The failure has structure, and the structure is informative.
Every one of these numbers — the power laws, the coefficients, the 90/7 — is derived on paper in closed form and then checked independently by high-precision computation. They agree. This is not a fit or an approximation; it is an exact statement about the geometry of information loss.
The same gas, seen as a landscape
The second piece of work looks at the very same gas but asks a different, more geometric question, and arrives at a complementary picture.
Go back to the abstract landscape — the space of all possible descriptions of the gas, now thought of as a curved surface in its own right, with hills and valleys and a genuine notion of distance and curvature built from the statistics. Physicists have studied the geometry of such “thermodynamic” surfaces for a long time; the curvature of these surfaces encodes real physical information about how a system fluctuates and where it undergoes dramatic changes.
The question this paper asks is deceptively simple. The massless, perfectly symmetric version of the gas is the textbook ideal — the case everyone starts from. Is that ideal case an ordinary point sitting comfortably in the middle of the landscape, surrounded on all sides by the more realistic massive cases? Or is it something more delicate — an edge, a boundary, a place where the landscape does something singular?
The answer is that the perfect case is not an interior point at all. It is a sharp boundary — a cusp. As you let the particle mass shrink toward zero, the landscape doesn’t smoothly flatten into the ideal case; it pinches. One particular measure of how the compression and temperature directions curve into each other shoots off to infinity, in lockstep with the blow-up the first result found. Meanwhile, the “length” of the compression direction itself — a measure of how much room the gas has to vary in that direction — collapses to zero. The ideal symmetric theory turns out to live on the rim of a funnel, not in the open plain.
There’s a satisfying unity here. Two different mathematical questions — “how much information does the closure throw away?” and “is the symmetric limit a smooth point of the landscape?” — turn out to be two views of the same underlying fact. The symmetry that makes the massless case so beautiful is exactly what makes it singular. The elegance and the breakdown are the same phenomenon seen from two angles.
Killing a tempting wrong idea
The second result does one more thing that I find admirable, because it goes out of its way to be hard on itself.
Once you have a geometric quantity describing the compression direction, and you know that real fluids have a physical property called bulk viscosity (essentially, the gas’s resistance to being uniformly squeezed — its “stickiness” under compression), there is an obvious and seductive temptation. You want to say: surely the geometric quantity and the physical stickiness are the same thing, or at least proportional to each other. It would be a clean, quotable, almost too-good-to-be-true bridge between abstract geometry and measurable physics.
The paper checks — and shows it is false. The geometric quantity and the bulk viscosity both fade away as the particles get lighter, but they fade at different rates. The bulk viscosity vanishes one power faster. Because they vanish at different speeds, they cannot be proportional; the tempting bridge does not exist, at least not in this model.
And rather than rest on a single calculation, the result confirms the bulk-viscosity behavior with three completely independent methods drawn from the kinetic-theory literature — an exact solution of the underlying equations, and two different standard approximation schemes. All three give the same coefficient, down to the exact fraction. When a result is this easy to get wrong, and this tempting to fudge, nailing it three independent ways is the difference between a claim and a proof.
This is, in a quiet way, the part of the work I respect most. It would have been easy — and impressive-sounding — to announce a clean proportionality between geometry and viscosity. The harder, more honest move was to test the seductive idea and report that it doesn’t hold. That negative result is more valuable than the positive one would have been, because it stops other people from chasing a mirage.
Why bother — what’s actually at stake
Let me be clear about what these results are and are not, because the honest framing matters more than the hype.
These are statements about a clean, controlled mathematical model: a simple gas of particles with a fixed mass, in a well-understood approximation. They are not a direct description of the real quark–gluon plasma, whose complications come from a different and more violent source — the strong nuclear force breaks the gas’s symmetry in a far messier way than simply giving the particles a mass. So nobody should read these results as “here is a new theory of collider matter.”
What they are is something I’d argue is more durable: a precise understanding of the structure of a problem that the whole field leans on. They take the routine, never-quite-examined step — the closure guess — and they say exactly how much it costs, exactly where the cost concentrates, and exactly how the cost behaves as you approach the idealized limit everyone uses as a starting point. They show that the most symmetric, most beautiful version of the problem is precisely the version where the standard description is most incomplete. And they connect, for the first time, a fifty-year-old idea from theoretical statistics to the machinery of relativistic fluids — a bridge between two fields that had no reason to know about each other.
There’s a broader lesson hiding in here, one that reaches past physics. We summarize complicated things all the time — economies, populations, climates, languages, the behavior of large machine-learning models. Every summary is a closure: a decision about which details to keep and which to reconstruct by assumption. We almost never measure what the assumption costs. These results are a worked example, in one corner of physics, of doing exactly that: of taking the discarded information seriously enough to put a number on it, and discovering that the number is not small, not uniform, and not where you’d expect.
The compression channel was the quiet one. Nobody was watching it closely. And that, it turns out, was exactly where the description was failing — infinitely, in a shape you can write down on a single line.
What a “guess” really costs — a closer look
It’s worth slowing down on why the closure guess is unavoidable, because the instinct of most people hearing this for the first time is: just keep more numbers. If a few summaries throw away too much, track a few dozen. Track a few hundred. Surely with enough summaries you can squeeze the error down to nothing.
You can’t, and the reason is exactly what the statistical curvature measures. Adding more summary quantities does shrink the error — but it never removes the bending of that abstract sheet of paper, because the bending is a property of how the kept quantities multiply together, not of how many you keep. When you track a quantity, you implicitly need to know how it correlates with itself and with the others — and those correlations involve still-higher quantities you didn’t track. Climb one rung of the ladder and a new rung appears above you. The curvature is the geometric fingerprint of that endless ladder. It is why closure is not a temporary inconvenience to be engineered away but a permanent structural feature of describing many things by few.
What the first result adds is the recognition that this permanent feature is not a fixed background hum. It has a knob — the particle mass relative to temperature — and as you turn that knob toward the symmetric limit, the hum swells into a roar, but only in one channel. That is a far more useful statement than “closure is imperfect.” It tells you when to distrust your simplified description and which part of it to distrust. If you are simulating a gas of nearly massless particles and you care about its bulk, compressional response, these results are a flashing warning light: that is the regime and the channel where your closure is leaking the most information, and no modest increase in the number of tracked quantities will rescue it.
The second result sharpens the warning into a geometric statement about the terrain itself. When the “length” of the compression direction collapses to zero while a curvature in the same neighborhood diverges, you are describing a coordinate system going bad — like the way every direction becomes “south” when you stand exactly on the North Pole. The math doesn’t break because reality breaks; reality is perfectly fine. The math breaks because the description was built around a special point, and special points are where descriptions strain. Recognizing the massless limit as one of those points — a pole, an edge, a cusp — tells you to handle it with the care you’d give any singular coordinate, rather than treating it as the comfortable home base it superficially appears to be.
The thing about edges
If there’s a single image to carry away, it’s this. We tend to treat idealized limits — the frictionless plane, the massless particle, the perfectly symmetric case — as the safe, simple center of a theory, the place to start before adding messy realism. These two results say that, at least here, the idealized limit is not the center. It’s the edge. It’s a cusp where the geometry pinches and the information loss runs to infinity. The realism isn’t a complication you add on top of the clean case; the clean case is a singular boundary of the realistic one.
That’s a more interesting universe than the one where the simple case sits safely in the middle. It means the elegant starting points we reach for are often the most treacherous, and that the way a description fails can be just as exact, just as structured, and just as worth knowing as the way it succeeds.
We built a ruler for ignorance and laid it against one of physics’ most reliable shortcuts. The shortcut, it turns out, has an edge — and we can now say precisely where.