← quantum · contents

§ Phase v · Superposition & Time · Chapter 04

Measurement and collapse

Quantum theory uses two rules of evolution that look nothing like each other. Most of the time the wavefunction glides along under the Schrödinger equation, smooth and reversible. Then a detector clicks, a pointer swings, a single value is recorded, and the same wavefunction snaps to one of the operator's eigenstates. The split has bothered physicists for a hundred years, and the cleanest descriptions of why it stops bothering them turn out to be the strangest ideas in the book.

In every chapter so far we have followed a single rule for how a quantum state changes. Write down the Hamiltonian, plug it into Schrödinger’s equation, turn the crank, and the wavefunction marches forward in time as smoothly as a planet around the Sun. The phases rotate, the probabilities flow, nothing is ever lost. If you reversed the sign of time you could run the whole movie backwards and recover the initial state exactly. Quantum evolution, in this picture, is the most orderly thing in physics. It is the deterministic engine that hides under the probabilistic surface.

And then a graduate student in the basement of a physics building presses the button on an oscilloscope. A single dot appears on the screen. A counter increments by one. Whatever beautifully extended superposition of position eigenstates was drifting through the apparatus a moment ago, what gets recorded in the lab notebook is a single number with units. The wavefunction that had been a smooth complex function over all of space has, in the act of being looked at, become almost entirely concentrated at one spot. Run the experiment again with identical preparation and you get a different spot, by a different deterministic rule. The same theory that produces planet-perfect evolution between measurements produces irreducibly random snaps at the instant of measurement.

This is not a small footnote. It is the central conceptual scar of quantum mechanics, the seam where the math meets the world, and it is called the measurement problem. The two rules have names. The smooth one is unitary evolution. The discontinuous one is projection, or wavefunction collapse, or “the second law of quantum mechanics” in the textbook of John von Neumann, who wrote them both down side by side in 1932 and let the contradiction sit there. Generations of physicists have argued about whether that contradiction is real, or apparent, or important, or polite to mention at dinner. We are going to walk through the argument in this chapter, because the only way to use quantum mechanics without being confused by it is to know exactly which rule applies when.

To make the rules concrete, let us pick the simplest non-trivial quantum system and work the example to the end. Take a single spin-half particle, an electron or a silver atom or a superconducting qubit, and write its state in the basis of “up along z” and “down along z”. Call those two eigenstates |0⟩ and |1⟩. Any pure state of the system can be written as a complex combination of the two:

|ψ⟩ = α |0⟩ + β |1⟩

with α and β complex numbers whose squared magnitudes sum to one. Between measurements, the Schrödinger equation moves α and β around in a perfectly orderly way. Their relative phase rotates at a rate set by the energy gap. Their amplitudes stay bounded. Total probability is conserved. The state stays a unit vector forever.

Now point a Stern-Gerlach magnet along z and let the particle fly through. The magnet is a piece of measurement apparatus. It is designed to ask the question “are you up or down along z?” The answer it returns is one of two numbers, and only one of two numbers. There is no “two-thirds up”. There is no average. The pointer reads up or it reads down. And the probabilities of those two outcomes are, by the Born rule we met three chapters ago,

P(up)   = |α|²
P(down) = |β|²

with the two adding to one because α and β were normalised to begin with. So far so good. But the rule does not stop with the probabilities. It also tells us what the state becomes after the measurement. If the apparatus returns “up”, the wavefunction is no longer α |0⟩ + β |1⟩. It is just |0⟩. The β component has been deleted. If the apparatus returns “down”, the α component has been deleted instead. This deletion is sometimes called “the collapse”, sometimes called “the reduction of the wave packet”, sometimes called “state update conditioned on the outcome”. The names are different. The math is the same. The wavefunction, in the act of producing a single classical answer, jumps onto the eigenstate that corresponds to that answer.

So far we have written down what physicists actually do. To use quantum mechanics for any concrete prediction you need both rules: Schrödinger to evolve between measurements, Born plus projection at each measurement. They were stitched together in the 1920s and have not failed an experimental test since. The trouble is that they sit very uneasily next to each other as a logical pair. The Schrödinger equation is linear, deterministic, and reversible. The projection rule is non-linear (it singles out the eigenstate you happen to have landed on), stochastic (the outcome is genuinely random), and irreversible (the deleted component does not come back). And neither rule says when, exactly, you should stop using the first and start using the second.

Pause on those words. Wikipedia’s opening sentence calls the predictions probabilistic, not as a confession of ignorance but as a defining feature. This is the line that separates quantum mechanics from every classical theory that came before it. In Newtonian gravity, the orbits of the planets are predicted exactly; any apparent randomness is a confession that you do not know the initial conditions well enough. In statistical mechanics, the temperature of a gas is an average over zillions of molecular motions you could in principle track. In quantum mechanics, the randomness is irreducible. You can know the wavefunction perfectly, every complex coefficient to thirty decimal places, and you still cannot predict which click the detector will produce next. All you can predict is the probability of each click, and the probability is what the wavefunction is for.

The interpretive question is what the wavefunction is, given that it does this. There are roughly three live answers in the modern literature. The first answer, attributed to Niels Bohr and his Copenhagen circle, is that the wavefunction is not a real physical object at all. It is a bookkeeping device for the experimenter’s expectations. When you make a measurement you update your expectations to match what you saw, and the collapse is just the bookkeeping update, no more mysterious than the way a probability distribution narrows when you learn a new fact. Don’t ask what is “really” going on between measurements, says Copenhagen; the theory has no answer because the question is not physical. This is the position that famously got summarised by David Mermin as “shut up and calculate”, though Mermin meant the slogan less affectionately than it is often quoted.

The second answer is Hugh Everett’s, from his 1957 Princeton thesis. Everett proposed that there is no second rule. There is only the Schrödinger equation. What we call “collapse” is the experience of being a small subsystem inside a much larger superposition. When the detector clicks “up”, the detector itself enters a superposition with the particle: the universe-wide state is now (up-particle and up-pointer-and-up-observer) plus (down-particle and down-pointer-and-down-observer). Both branches are real. Both branches contain copies of the experimenter, each convinced she got a single sharp answer. The Born rule is recovered as the relative weight of the branch you happen to be on. The math is just unitary evolution running on a wavefunction of everything. The unsettling thing is the ontology: in this picture, every quantum measurement splits the world.

The third answer is the most modern, and in some ways the most useful in the lab even if you are agnostic about the metaphysics. It is decoherence. The idea is that the seam between rule 1 and rule 2 is not a fundamental seam at all. It is an emergent statistical effect of the quantum system coupling to a vast environment, the air molecules and stray photons and lattice vibrations and the experimenter’s body heat. Once the system has correlated itself with that environment, the relative phases between superposition branches get scrambled essentially instantly. From the point of view of any observer who cannot track every air molecule in the room, the off-diagonal entries of the density matrix have decayed to zero. The system looks like a classical mixture of definite states, even though in principle the full universal wavefunction is still a coherent superposition. Decoherence does not, on its own, solve the measurement problem: it explains why we never see live-and-dead cats, but not why this particular branch is the one we end up on. What it does, very convincingly, is render the question “when does collapse happen?” almost meaningless in practice. It happens fast. It happens on a timescale set by how strongly the system talks to its surroundings.

P_correlated(t) ≈ 1 − exp(−Γ t),    Γ = N · λ

There is a moral here that you should carry into every quantum experiment you ever read about. The Born rule and the projection rule are computational tools that work, every time, in every regime that has ever been tested. They are not the final story about reality, but they do not need to be in order to be used. If you are designing a Stern-Gerlach run, computing a probability of finding an electron in a particular orbital, or predicting the click rate of a single-photon detector behind a beam-splitter, the procedure is exactly the same as it has been since 1926. Write down the wavefunction. Expand it in the eigenbasis of the operator you intend to measure. Square the coefficients. Those are your probabilities. Run the experiment. Update.

The reason this works without anyone needing to settle the metaphysical question is the decoherence layer that sits between any laboratory apparatus and the system it is measuring. Long before a human eye gets involved, the system has correlated itself with so many environmental degrees of freedom that the interference terms between branches are unrecoverable in any practical sense. From that point onwards, treating the situation as a classical probability distribution over definite outcomes is not just convenient, it is correct to absurd precision. The fact that the universal wavefunction, viewed from outside, might still be unitary is a story for cosmologists and philosophers. The Born rule is what the lab notebook records.

What you should hold onto, then, is the asymmetry. Quantum mechanics gives you two evolution rules because measurements are a special kind of event: they are the moment when a quantum system gets entangled with so many other degrees of freedom that, for any sub-universe smaller than the universe itself, the entanglement looks like a probabilistic update. Whether you call that “collapse”, “branching”, or “decoherence-induced einselection of a preferred basis” is a matter of taste and reading list. The Born rule will give you the right click rates either way. The next chapter takes that machinery and turns it on a system where the smooth rule and the discontinuous rule produce a phenomenon classical physics flatly forbids: a particle that walks through a wall it does not have the energy to climb.