Quantum Foundations

Quantum mechanics is the most successful predictive framework in the history of science. It's also, embarrassingly, a theory that nobody agrees on what it means. An informal poll at a 2011 conference on quantum physics and reality found participants deeply divided on every interpretive question — and these were the experts.¹ The theory works. It predicts with extraordinary precision. But its rules were essentially pulled from a hat by Schrödinger, Born, and Heisenberg in the 1920s, and a century later the foundations remain a construction site.

The Reconstruction Program

The clearest way to see the problem: compare quantum mechanics with special relativity. Einstein gave us two crisp axioms — the speed of light is constant, the laws of physics look the same in all inertial frames — and the entire theory follows. What are the analogous axioms for quantum mechanics? John Wheeler once insisted that if we really understood the theory, we could state its core in a single sentence anyone would grasp. Nobody has managed this yet.

The quantum reconstruction program is the attempt to find those axioms. Starting around 2001 with Lucien Hardy's pioneering work at Oxford, a small community of physicists began stripping quantum theory down to its barest operational skeleton — not particles or waves or wave functions, but just the relationship between preparations, transformations, and measurements. What probabilities can you assign to outcomes given how you set things up?¹

Hardy found that if you write down a few reasonable postulates about how information-carrying systems can be combined and measured, the simplest theory satisfying them is quantum mechanics. Entanglement, superposition, interference — all the trademark weirdness falls out of the axioms. As Giulio Chiribella put it, Hardy's paper was "the 'Yes, we can!' moment of the reconstruction program."¹

Since then, multiple groups have found independent sets of axioms that all recover quantum theory. Chiribella, D'Ariano, and Perinotti showed in 2010 that you get quantum mechanics from postulates about information being localizable, systems being able to encode information about each other, and processes being reversible. Dakić and Brukner in 2009 derived it from three axioms about information capacity — and showed that what separates quantum from classical probability is continuity: quantum states can be smoothly rotated from one to another, while classical states jump between heads and tails with nothing in between.¹

What's striking is that all these axiom sets have an informational flavor. They're not about waves or particles or things happening in space. They're about what you can learn, store, transmit, and transform. Quantum mechanics, seen through this lens, is a theory about the structure of information — a "generalized probability theory" that tells you what patterns of correlations nature allows.¹

What Is a Photon, Actually?

Most people — and most pop science — think of a photon as a tiny bullet of light. This is wrong in a way that matters. A photon isn't a little ball that carries energy through space. It's a counting word for the energy levels of an electromagnetic field mode.²

The story starts with harmonic oscillators. A mass on a spring can oscillate with any energy you like — pull it down a little, it bounces a little. But quantum mechanics says that very small oscillators can't have arbitrary energies. Their possible energies come in equally spaced steps, like rungs on a ladder. The ground state, then one step up, then two steps up, and so on.²

When we say "there are three quanta in this oscillator," we're not saying there are three objects sitting inside it. We're saying the oscillator is at its third excited energy level. "Quanta" is a counting noun for energy levels, not for things. It's exactly like money in a bank account — you can have a hundred dollars, but those dollars don't exist as distinct entities in a vault somewhere. They're a measure of a single number: your balance.²

Now scale this up. A crystal lattice has trillions of atoms connected by forces that act like springs. The coupled oscillations of the whole lattice can be decomposed into independent normal modes — plane waves propagating through the crystal at various frequencies and directions. Each mode is an independent harmonic oscillator, so each one has quantized energy levels. The quanta of crystal vibration are called phonons. They have direction and frequency. But they're not objects traveling through the crystal — they're a way of talking about how much energy is in each vibrational mode.²

Photons are exactly the same thing, but for the electromagnetic field instead of a crystal lattice. Maxwell's equations allow electromagnetic waves at every frequency and direction. Each such mode is an independent oscillator. Because of quantum mechanics, each mode's energy comes in discrete steps. Those steps are photons. When we say "a photon with frequency f is traveling in direction k," we mean "the electromagnetic field mode with frequency f and direction k is in its first excited state." That's all it means.²

This has a consequence that confuses everyone: you can't really talk about where a photon is. A normal mode is a plane wave that extends across all of space. Adding a quantum to it doesn't put anything at a specific location — it changes a property of a mode that's everywhere at once. The uncertainty principle makes this precise: if you know the frequency (and therefore momentum) of a photon, its position is maximally uncertain.²

Measurement Is Information

So if photons aren't bullets and quantum states are probability distributions over field modes, what makes a measurement special? Why does the interference pattern in the double-slit experiment vanish when you try to detect which slit the photon went through?

The best answer, stripped of philosophical baggage: a measurement is any interaction that conveys information. Not "measured by a conscious observer" — that's new-age nonsense that almost no physicist takes seriously. And not "interaction with a large system" — you can do the double-slit experiment in air, with photons constantly bouncing off air molecules, and the interference pattern survives. Those interactions don't carry information about which slit the photon traversed.³

The key distinction: when light passes through sugar-water, its polarization rotates. That's an interaction, but not a measurement, because if you don't know the initial polarization, you can't learn anything from the final one. The sugar-water shuffles the states but doesn't reduce your uncertainty about them. A polarizer, by contrast, is a measurement — if the photon comes through, you know its polarization.³

And crucially, measurement isn't all-or-nothing. You can do a partial measurement that gains a little information, like a garbled train announcement where you think there's a 70% chance the train is late. In quantum mechanics, partial measurements partially degrade the interference pattern. You can slide continuously from "no information about which slit" (full interference) to "complete information" (no interference). It's not some magical discontinuous "wave function collapse" — it's a smooth trade-off between information gained and quantum coherence lost.³

This connects back to the reconstruction program's information-theoretic foundations. If quantum mechanics is fundamentally a theory about information — what you can learn and how learning changes what's possible — then measurement isn't a mysterious extra ingredient tacked on from outside. It is the theory.

Delayed Choice and the Illusion of Retrocausality

The delayed-choice experiments push this understanding to its most counterintuitive limits. In the entanglement-swapping version: two independent sources each produce a pair of entangled photons. One photon from each pair goes to Alice and Bob, who measure polarization immediately. The other two photons go to Victor, who later decides whether to entangle them with each other or not.⁴

Here's what's spooky: when Victor chooses to entangle his photons, Alice and Bob's earlier measurements turn out to be correlated. When he doesn't, they're uncorrelated. The choice that determines the correlations happens after the correlated measurements.⁴

The Australian delayed-choice experiment using helium atoms found similar results — whether an atom behaved as a wave or particle going through the first grating depended on whether a second grating was placed in its path afterward.⁵

This looks like the future causing the past. It isn't — no information travels backward in time. Alice and Bob can't tell from their individual results alone whether Victor will entangle or not. The correlations only show up when you compare all the results afterward. But the experiments force an uncomfortable choice: either accept some form of retro-influence on the correlations (while somehow never violating causality in a detectable way), or accept that the quantum state was never committed to "wave" or "particle" until the full experimental configuration was determined — including the future parts of it.⁴

The Novikov self-consistency principle, developed for general-relativistic time travel, offers a structural parallel. It says that in spacetimes with closed timelike curves, only globally self-consistent histories can occur — you can't go back and change the past, because the past already includes whatever you went back and did.⁶ Quantum delayed choice has a similar flavor: the correlations are always globally consistent, even though locally they look acausal. The universe doesn't pick "wave" or "particle" at the first slit and then somehow retroactively change its mind. It only ever produces globally consistent measurement records.

The Deeper Level

Hardy, who helped start the reconstruction program, admits it's been "almost too successful" — many different axiom sets all give quantum mechanics, and none of them feel like the obvious, compelling foundation. His suspicion is that the right axioms will point beyond quantum theory to a deeper theory of quantum gravity, one where even causal structure becomes indefinite.¹

This connects to the emerging picture of spacetime and information: if space itself is a quantum error-correcting code, then the foundations of quantum mechanics and the foundations of spacetime are the same problem. The reconstruction program's informational axioms — locality of information, reversibility, the structure of correlations — might be the axioms of a theory that generates both quantum mechanics and gravity from something more basic.

Whether that deeper theory exists, or whether quantum mechanics is the bottom turtle, nobody knows. But the reconstruction program has already delivered something valuable: proof that quantum weirdness isn't arbitrary. The superposition, entanglement, and interference that seem so alien are actually the simplest possible way for nature to organize probabilistic information while respecting a few reasonable constraints. The weirdness isn't gratuitous — it's minimal.¹