Quantum mechanics comes with two rules of evolution that contradict each other, and no physical criterion tells you when to use which. A century later the seam is still open — every "interpretation" is somebody's prescription for it.
The Schrödinger equation is deterministic, reversible and linear: it never chooses, it just carries every possibility forward together. "You measure and get one definite reading" is random and irreversible. Textbooks bolt the second on as a separate postulate and treat "measurement" as a word that needs no definition — yet an apparatus is made of atoms too, and should obey the Schrödinger equation like everything else. That is the measurement problem: not that we don't know the outcome, but that the theory never says which rule applies when. The real progress of the last fifty years was to split the seam in two: why we never see superpositions is now calculated in detail; why only one thing happens is untouched.
Two Rulebooks, One Theory The Seam
von Neumann 1932 · projection postulate
Intuition
Imagine a game shipped with two rulebooks. Book A: the position evolves smoothly and deterministically, no possibility is ever struck out, and in principle you can run it backwards. Book B: right now, out of many possibilities exactly one survives — chosen at random, no way back. When do you open which? The textbook says "use B when you measure." But what is a measurement — does a counter's click count, or does somebody have to look? No criterion you can read off from mass, energy or particle number tells you where the line falls.
Mechanism
Book A is the Schrödinger equation:
iħ ∂ψ∂t = Ĥψ
ψ ("psi") is the wavefunction, the complete state of the system; ∂ is the partial-derivative sign, so ∂ψ/∂t means "how fast the state changes in time"; Ĥ ("H-hat") is the Hamiltonian, which carries the system's energy content; ħ ("h-bar") is Planck's constant divided by 2π; i is the imaginary unit — the letter i, not the digit 1.
The crucial property is linearity: it takes "atom up + apparatus ready" to "pointer up" and "atom down + apparatus ready" to "pointer down", so feed in a superposition and a superposition must come out:
(a|↑⟩ + b|↓⟩)|ready⟩ → a|↑⟩|up⟩ + b|↓⟩|down⟩
|↑⟩ is read "ket", the notation for "is in this state"; a and b are amplitudes, not probabilities (see Wavefunction & Superposition); the Born rule gives the probability of reading "down" as P = |b|², the squared modulus of the amplitude.
This is not an extra assumption but a direct consequence of linearity: since the apparatus is made of atoms as well, the superposition is bound to infect the pointer. Book A never hands you a unique reading on its own.
The gap is not in the arithmetic — both rules calculate beautifully. What's missing is when to switch.
The counterintuitive point
Schrödinger's 1935 cat is not a spectacle, it is a reductio: tying the cat's life to a decaying atom was his way of saying that if you take linearity seriously the theory delivers a cat both dead and alive; since no such cat exists, something in the account must be missing. Turning it into a poster for quantum weirdness gets the author exactly backwards.
Cross-disciplinary reading · engineering / history of mathematics
"Wildly useful, with an undefined hole in the foundation" is not rare:
Engineering: undefined behaviour in the C standard — the spec openly says "not our problem" for certain cases; the program still runs fast, but at that point it has no meaning;
History of mathematics: the early calculus and its "infinitesimals" — accurate enough for planets and machines, yet nobody could say whether the thing was zero, until limits supplied the missing foundation.
In one line: the measurement problem isn't "quantum is weird" — it's an undefined word inside the theory.
Think: what happens if you write the apparatus, the experimenter and the whole building into the Schrödinger equation?
A superposition that keeps growing: the atom drags the pointer, which drags a retina, which drags a brain. This is the von Neumann chain — linear evolution has no place where it can break itself. Either something else really happens somewhere along it (objective collapse), or the whole chain persists and we only experience one branch (many-worlds), or the wavefunction was never a real thing to track (the Copenhagen family).
Decoherence: Superposition Leaks, It Doesn't Shatter Decoherence
Zeh 1970 · Zurek 1981 · pointer basis
Intuition
Interference has a demanding precondition: the two paths must be in principle indistinguishable. Real objects are soaked in an environment — air molecules hit them, photons bounce off them, they radiate infrared of their own. The moment anything out there becomes correlated with which path was taken, a copy of that record has been carried away; even if nobody ever reads it, the interference is already gone. Superposition doesn't shatter, it seeps out.
Mechanism
Write the state as a 2×2 table (the density matrix): the diagonal entries |a|² and |b|² are "how much each branch weighs", while the off-diagonal entryab* is the part that says "can these interfere" (the coherence). Coupling to the environment makes it decay exponentially:
ab*(t) = ab*(0) · e−t/τ → 0
The asterisk * is complex conjugation; e is the constant 2.718… (the letter e); τ ("tau") is the decoherence time; the 0 the arrow points to is the digit zero, meaning the coherence is wiped out (the 0 in brackets is the starting time). The point: the diagonal doesn't move, only the ability to interfere disappears.
The scales are absurd. On the order-of-magnitude estimates of Joos and Zeh, a ten-micron dust grain put into a superposition of positions separated by its own size loses coherence in about 10−30 s in room-temperature air, and even in the best laboratory vacuum only stretches that to roughly 10−17 s.
How fast the fringes vanish can be calculated precisely, and has been checked step by step in the lab.
The counterintuitive point
Decoherence does settle three things: why macroscopic objects look classical; why the "classical" states happen to be the ones near definite positions (the environment records position, and so selects the pointer basis); and how fast interference goes — heat C₆₀ molecules to nearly three thousand degrees so they radiate their own infrared, and the fringes disappear.
What it does not settle is the thing that matters most: why only one outcome happens. In that "looks like a classical probability mixture" mixture, both branches are still there; the equation never struck one out. "Decoherence solved the measurement problem" is the most widely repeated mistake about it: half the problem is gone, the other half is exactly where it was.
Cross-disciplinary reading · information / engineering / AI
Decoherence is, at bottom, information becoming objective by being copied many times over (Zurek's quantum Darwinism: the environment holds countless redundant records, so any observer who grabs a handful reads the same event). The same logic is everywhere:
Physics of information: erasing one bit costs energy, because a record copied into the environment is no longer reversible (see The Physics of Information);
Engineering: a record counts as "committed" once it is replicated across machines — factuality comes from redundancy, not authority;
AI: an answer counts as stable output when repeated sampling of the same prompt agrees; nobody ships the one that appeared once.
In one line: decoherence explains why we never see superpositions, not why only one thing happens.
Think: if coherence is gone in 10⁻³⁰ s, isn't calling the measurement problem "unsolved" just pedantry?
No. Decoherence ends at "two branches coexisting without interfering"; experience delivers "only one branch". No equation covers that step. Bell mocked the move of treating decoherence as the answer as FAPP — for all practical purposes: good enough to use, not an explanation.
The Menu: What Each Interpretation Patches The Menu of Interpretations
Copenhagen · many-worlds · Bohm · GRW
Intuition
The point most often missed: nearly all of these predict exactly the same experimental results. They are not competing theories, they are different pictures of the world attached to one set of calculations; for now the choice isn't made by experiment but by which price you are willing to pay.
Mechanism · one line each① Copenhagen / orthodox: the wavefunction isn't a thing, it is bookkeeping about future measurements, and collapse is updating the books. ② Many-worlds (Everett 1957): only Book A; every branch is real and we are a copy inside one of them. The trouble is probability: if every branch happens, what is |a|² still saying? ③ de Broglie-Bohm pilot wave: the particle always has a definite position, the wavefunction is the wave pushing it around, and "collapse" is just our ignorance of the initial position; the price is explicit nonlocality, which fits the verdict in Entanglement & Nonlocality. ④ Objective collapse (GRW / Penrose): add a tiny random nonlinear term to the equation — imperceptible for one particle, but with 10²³ particles tied together the superposition collapses in an instant. It changes the equation, and is therefore falsifiable. ⑤ QBism / relational: the wavefunction is an agent's belief, or holds only relative to some reference system.
Two axes: is the wavefunction a real thing, and does a collapse really occur.
The counterintuitive point
A rare situation in science: one set of predictions, several pictures of the world, and no experiment to decide between them yet (philosophers call it empirical equivalence, or theory underdetermined by evidence). But don't slide into "believe whatever you like" — the dividing line is falsifiability, and it is moving. Objective collapse changed the equation, so it must predict a little extra spontaneous radiation, and deep-underground low-background experiments have already cut away a large slab of its parameter space.
Cross-disciplinary reading · statistics / AI / biology
"One dataset, several coherent stories" is the normal condition of every empirical field, and the remedy is the same — don't judge which story is prettier, judge which yields a different testable prediction:
Statistics: the same correlations fit several causal graphs pointing opposite ways; only intervention cuts them down;
AI: saliency maps, concept vectors and circuit analyses each tell a story about why a network answered as it did, all fitting the observed behaviour — the test is which one predicts an unseen failure;
Biology: a gene's function usually admits several narratives, and knockouts are what separate "involved in" from "required for".
In one line: interpretations aren't faiths, they're a price list with the costs written on them.
Think: if two interpretations always predict the same thing, are they two theories at all?
The sharpest knife in philosophy of science. Those who say no: a difference that makes no different prediction is rhetoric. Those who say yes: different pictures generate different research programmes — many-worlds forces the question of where probability comes from, Bohm and GRW forced concrete experimental proposals. Whether an interpretation is idling shows in whether it has produced an experiment anyone can run.
Why "Consciousness Collapses the Wavefunction" Fails No, It Doesn't
Wigner's friend · an honest look
Intuition
The slogan spread for a reason: textbooks write "observation causes collapse", and "observation" is naturally read as somebody looking. But measurement in physics never requires a person. Wire a Geiger tube to a recorder, lock it in an empty basement for a year, read the data afterwards — the interference that should be absent is absent, the statistics that should be there are there. The universe doesn't wait for anyone to look.
Mechanism · three hard facts① The knob is physical. Fringe visibility is tuned continuously by pressure, temperature and radiation: let a little gas into the interferometer and the contrast drops step by step. The whole curve is indifferent to whether anyone is in the room. ② Intention can't touch the statistics. The no-signalling theorem is hard: no local operation changes the marginal statistics far away. If "the mind influences the outcome" were true you would be holding a usable channel, and the experimental bounds are brutally tight. ③ Its own proponent withdrew. "The von Neumann chain must terminate in consciousness" did have a serious version, and Wigner supported it for a time. After decoherence, the slot was no longer needed — air molecules suffice to end the interferable part of the chain — and he gave the position up in later life.
Where honesty ends
Don't overshoot either. The measurement problem is a real problem, and "why only one thing happens" genuinely has no answer yet. The switch happens at the next step: taking "unsolved" to mean "so fill it with anything". Besides being unnecessary, consciousness-collapse owes answers to a list it can't supply: where is the threshold? does a cat count? a camera? and for the thirteen billion years before conscious creatures existed, what was collapsing anything? An explanation that can be placed anywhere is explaining nothing.
Cross-disciplinary reading · communication / AI / statistics
The failure mode is always the same: a technical term is quietly swapped for its everyday sense, and a technical gap becomes a mystical doorway.
Quantum: "observation" (any physical interaction that leaves a record) read as "somebody saw it";
AI: "understands", "knows" start as behavioural measures and slide into claims about minds — "it answered correctly" becomes "it gets it";
Statistics: "significant" means "unlikely to be pure chance", gets read as "important", and a trivial effect makes headlines.
There is exactly one defence: whenever a keyword appears, ask for its operational definition.
In one line: the gap is real, but a gap is not a wishing well.
Think: if someone did claim consciousness causes collapse, what experiment would count?
One that produces a physical difference: hold every physical condition fixed, vary only "conscious observer present or not", and find a reproducible shift in visibility or statistics. That requires writing "consciousness" as a controlled variable — precisely the step nobody has managed, which is why the claim isn't yet in testable territory. Objective collapse, a far weaker claim, already has experimental constraints.
Going Deeper
If many-worlds is right, what does "probability" still mean? Every branch happens, so what is |a|² saying?
The hardest gate for many-worlds: if every branch is realised, "the probability of this one is |a|²" has no obvious referent. Two rescue routes: the decision-theoretic derivation (Deutsch-Wallace) — assume a rational agent betting before the split obeys a few consistency axioms, and show the only consistent betting weight is |a|²; and self-locating uncertainty (Sebens-Carroll) — the branching has already happened but "which branch am I in" is unknown, and probability describes that ignorance. Both have critics, and the objection in each case is whether the Born rule was smuggled into the premises. Whether it is derived or added by hand remains unsettled.
Could an experiment ever decide?
Partly. Anything that changed the equation is in range: objective collapse must bring a faint spontaneous energy fluctuation and radiation, deep-underground low-background experiments keep pushing its parameters down year by year, and the plainest gravity-induced version is already out. The other route is making macroscopic superpositions ever larger — big-molecule interference, levitated nanoparticles, cryogenic mechanical oscillators — each order of magnitude squeezing the room for "collapse must set in at some scale". The families that don't change the equation predict identically and can't be separated this way. The likelier ending: options get deleted one by one and the survivors stay tied.
What are the new "Wigner's friend" experiments actually showing?
The friend, sealed in a lab, completes a measurement and sees a definite result, while Wigner outside treats the whole room (friend included) as a quantum system — and their descriptions clash. In 2018 Frauchiger and Renner turned this into a rigorous no-go theorem: a few innocuous-looking assumptions — quantum mechanics applies universally, agents may use each other's reasoning, outcomes are unique — cannot all hold together. In 2020 Bong and colleagues recast it as a testable inequality and ran it with photons. None of this picks a winner, but it removes the lazy option of "pick any one, they're all the same": you at least have to say which assumption you are dropping.