The mathematics of the quantum track has a body. This is where Hilbert spaces meet the lab — the four postulates, the Born rule, and the experiment that proved the universe is stranger than local realism allows.
Nature isn't classical, dammit — and if you want to make a simulation of nature, you'd better make it quantum mechanical.— Richard Feynman
Quantum computing borrows its power from physics it rarely stops to justify. This track stops. In twelve weeks you build quantum mechanics from its axioms — states, unitaries, measurement, tensor products — then follow the one thread that changed everything: Einstein's 1935 objection, Bell's 1964 answer, and the loophole-free experiments that settled it. By the end, entanglement is not a slogan. It is a theorem you have derived, a bound you have computed, and a resource you understand from the inside.
The conceptual backbone — postulates, measurement, and entanglement done with a computer scientist's precision.
The gentlest rigorous on-ramp. Spins, states, and time evolution built from almost nothing.
The formal machinery — kets, bras, operators, and measurement stated the way physicists actually use them.
Bell in his own words. The clearest thinking ever committed to paper about what quantum mechanics means.
The most beautiful popular account of Bell's theorem — read before and after you do the derivation.
The measurement problem stated honestly. Feeds directly into the philosophy-of-physics unit.
Quantum mechanics is four sentences and a lifetime of consequences. In these three weeks you state the four postulates precisely, one at a time, and make Dirac notation as automatic as handwriting — because every derivation in the rest of this track, and every gate in the quantum computing track, is written in it. Nothing here is optional; this is the load-bearing wall.
A physical system is a unit vector |ψ⟩ in a complex Hilbert space, and two vectors differing by a global phase eiθ describe the same physics. Superposition is nothing exotic — it is bare linearity: if |0⟩ and |1⟩ are states, so is α|0⟩ + β|1⟩. The complex numbers are not decoration; relative phase is what makes amplitudes cancel, and interference is the resource everything else in this cathedral spends.
Kets |ψ⟩ are vectors, bras ⟨φ| are their duals, ⟨φ|ψ⟩ is the inner product, and |φ⟩⟨ψ| is an operator. The resolution of the identity, I = Σ|i⟩⟨i|, is the single most-used trick in physics — insert it anywhere to change basis, extract matrix elements ⟨i|A|j⟩, or diagonalize by inspection. This must stop being notation you translate and become the language you think in.
Closed systems obey iħ d|ψ⟩/dt = H|ψ⟩ with H self-adjoint; solving it gives |ψ(t)⟩ = U(t)|ψ(0)⟩ with U(t) = e−iHt/ħ. The hinge is that hermitian generators give unitary flow: U†U = eiH†te−iHt = I exactly when H = H†, and unitarity is conservation of probability — norms never drift, so the Born rule stays coherent over time. This equivalence between the differential law and the unitary flow is why the quantum computing track can speak entirely in gates.
Observables are self-adjoint operators A = Σ λᵢPᵢ; a measurement returns eigenvalue λᵢ with probability ⟨ψ|Pᵢ|ψ⟩, and the state collapses to Pᵢ|ψ⟩/‖Pᵢ|ψ⟩‖. This is the only place probability enters the theory — and the only place evolution is not unitary. Hold that tension carefully now; it becomes the measurement problem in Phase 4. Repeatability — measure twice, get the same answer — is a theorem that falls out of P² = P, not an extra assumption.
Two systems together live in the tensor product H_A ⊗ H_B, never the direct sum. The reason is physical: a joint configuration must pair every basis state of A with every basis state of B — that is "both/and," and it multiplies dimensions (two qutrits: 9, not 6); direct sum is "either/or," a single system with more levels. The consequence is enormous: n qubits span 2ⁿ dimensions, and most vectors in that space are entangled — not writable as |a⟩⊗|b⟩ at all. Phase 3 lives in that gap.
Silver atoms through an inhomogeneous magnet split into exactly two beams: spin is quantized. Chain the magnets — measure z, then x, then z again — and the final z-measurement comes out 50/50: measuring x destroyed the z-information. Sakurai opens with this because it is the whole formalism on a tabletop — noncommuting observables, collapse, the Born rule — and it is why the two-level system, the qubit, is the natural atom of quantum information.
Four postulates. Everything strange about the quantum world hides inside them. State them precisely now — the rest of the track is watching them collide with reality.
Phase 1 told you what a measurement does; this phase asks what a measurement is — and discovers that the Born rule was never a free choice, that projectors are only the special case, and that the classical world you live in is what entanglement with the environment looks like from inside.
⟨A⟩ = ⟨ψ|A|ψ⟩ is exactly the Born-rule average Σ λᵢ p(λᵢ) — statistics compressed into one sandwich. From variance comes the Robertson relation, ΔA·ΔB ≥ ½|⟨[A,B]⟩|: uncertainty is a theorem about noncommuting operators, provable in four lines from Cauchy–Schwarz, not a confession about clumsy instruments. If two observables fail to commute, no state sharpens both — the trade-off is written into the geometry of the state space itself.
Suppose you only want to assign probabilities to measurement outcomes: a frame function gives each projector a nonnegative number, with the numbers across any orthonormal basis summing to 1 — additivity over complete measurements, nothing more. Gleason's theorem: in dimension ≥ 3, every such assignment is P ↦ tr(ρP) for some density operator ρ. That innocent additivity requirement, chased around the interlocking orthonormal bases of Hilbert space, forces the trace rule — the Born rule could not have been otherwise. The proof is genuinely hard and is not required; the statement, and what it rules out (dispersion-free states, the first shadow of the hidden-variable no-go theorems), are mandatory.
Projective measurement is the special case. The general notion is a set of positive operators {E_m} with Σ E_m = I and p(m) = ⟨ψ|E_m|ψ⟩ — no orthogonality, no repeatability required. This is not new physics: every POVM is a projective measurement on system-plus-ancilla, marginalized back down (Naimark dilation) — the four postulates applied to a bigger Hilbert space. The payoff is real: a three-outcome POVM can distinguish two non-orthogonal states unambiguously, something no projective measurement can do, and it is the natural language of every realistic detector.
If superposition is universal, why have you never seen a chair in two places? Couple a system to an environment and watch: interference terms leak into system-environment entanglement, the off-diagonal elements of the reduced state decay, and only a preferred pointer basis — the one singled out by what the interaction Hamiltonian commutes with — survives monitoring. This is einselection: the environment continually "measures" the system and broadcasts the result, which is why macroscopic superpositions are unobservable in practice. Be precise about the boundary: decoherence explains the absence of interference; whether it picks an outcome is Phase 4's fight — it selects a basis, it never selects a branch.
Why probabilities, not certainties — and what that costs. Gleason showed the cost was never optional.
Four weeks on the single most consequential argument in the history of physics. You read EPR in the original, follow Bell as he turns a philosophical objection into an inequality, derive both bounds of CHSH with your own hands, and then watch sixty years of experiments close every escape route. By week 10, "entanglement is nonlocal" is not a slogan you repeat — it is a theorem you own.
The most productive "wrong" paper ever written, and it is a valid argument — that is the point. Its exact structure: (1) Locality — measuring here cannot disturb a system there; (2) the reality criterion — if you can predict a quantity with certainty without disturbing the system, some element of reality corresponds to it; (3) on an entangled pair you can choose to predict either of two noncommuting quantities at a distance; therefore (4) both are elements of reality at once — and since no quantum state fixes both, quantum mechanics is incomplete. Einstein wasn't confused; he was drawing the sane conclusion from premises the universe turns out to reject.
Take EPR seriously and write down what "incomplete" would mean: outcomes determined by a hidden variable λ with distribution ρ(λ), and locality as the requirement that Alice's outcome function A(a, λ) = ±1 not depend on Bob's distant setting b. That bookkeeping — E(a,b) = ∫ A(a,λ)B(b,λ)ρ(λ)dλ — is the entire commitment, and it seems almost too weak to say anything. Bell's discovery: it already constrains the correlations, and the quantum singlet, with E(a,b) = −cos θ, breaks the constraint. Philosophy became an experiment.
The classical bound, step by step: for each λ form S(λ) = A(a)[B(b)+B(b′)] + A(a′)[B(b)−B(b′)]; since B(b), B(b′) ∈ {±1}, one bracket is ±2 and the other is 0, so |S(λ)| ≤ 2 pointwise; averaging over ρ(λ) cannot exceed what holds pointwise, hence the CHSH inequality |E(a,b)+E(a,b′)+E(a′,b)−E(a′,b′)| ≤ 2. Then the quantum side: define the operator C = A⊗(B+B′) + A′⊗(B−B′), square it, and find C² = 4·I plus a commutator term [A,A′]⊗[B,B′] of norm at most 4 — so ‖C²‖ ≤ 8 and ‖C‖ ≤ 2√2, the Tsirelson bound. The singlet at angles 0°, 45°, 90°, 135° achieves 2√2 exactly. Two half-page derivations, and the entire classical worldview sits in the gap between them.
Alice's statistics come entirely from her reduced density operator, and nothing Bob does locally — measure, unitarily evolve, shred his lab notebook — changes it: Σ_b p(a,b|x,y) is independent of Bob's setting y. So entanglement correlates without communicating, and quantum mechanics coexists peacefully with relativity at the level of statistics. The theorem also sharpens what your CHSH violation means: nature exceeds the classical bound of 2 while never touching the no-signaling limit — nonlocal correlations, no nonlocal messages — and that precise wedge is worth an essay, so you will write one.
Suppose a unitary copies unknown states: U|ψ⟩|0⟩ = |ψ⟩|ψ⟩ and U|φ⟩|0⟩ = |φ⟩|φ⟩. Take the inner product of the two equations and unitarity gives ⟨ψ|φ⟩ = ⟨ψ|φ⟩², so ⟨ψ|φ⟩ ∈ {0,1}: only orthogonal states can share a copier. Three lines, enormous consequences: cloning would let Bob amplify his half of an entangled pair and read Alice's basis choice — no-cloning is what keeps no-signaling honest — and it is the entire reason quantum cryptography works: an eavesdropper cannot copy what she cannot measure without disturbing.
Three qubits in (|000⟩ − |111⟩)/√2, and four measurement contexts: XXX, XYY, YXY, YYX. Quantum mechanics predicts the product of the three outcomes with certainty in each context — the state is a simultaneous eigenstate of all four operators. Now try to pre-assign local values ±1: multiply the three constraints from XYY, YXY, YYX, and because each Y appears twice the product algebraically forces a value for XXX that is the negative of what quantum mechanics predicts. No statistics, no inequality, no "violation by ε" — a flat logical contradiction in a single run. Bell shows local realism loses on average; GHZ shows it cannot even get one family of predictions right.
Every "loophole" is a precise premise of the derivation you just did, wearing lab clothes. Aspect's 1982 experiment switched analyzer settings in flight, attacking the locality loophole (settings influencing distant outcomes); the detection loophole — fair-sampling, when most photons go undetected — survived until 2015, when Delft (entangled electron spins in diamond, 1.3 km apart), NIST, and Vienna (high-efficiency photon detectors) closed both simultaneously. The 2022 Nobel Prize to Aspect, Clauser, and Zeilinger ratified the verdict: the CHSH violation is a fact about nature, not about imperfect apparatus. Only superdeterminism — denying that measurement settings can be chosen independently — remains logically open, at the price of undermining the very idea of an experiment.
Bell turned philosophy into an experiment. The universe answered — and it wasn't the answer Einstein wanted.
The formalism works flawlessly; what it means is still contested a century on. These two weeks state the measurement problem with logician's precision — three propositions that cannot all be true — and then tour the four serious responses, each of which is a choice of which proposition to sacrifice. No interpretation is free. Your job is to learn the prices.
Three propositions: (1) completeness — the wave function is the whole physical story; (2) linearity/universality — everything, apparatus and observer included, always evolves unitarily; (3) definite outcomes — measurements end with exactly one result. Any two contradict the third: feed a superposition into a linear measuring device and completeness plus universality deliver a superposed pointer, not a definite one — a two-line argument, and it is airtight. The measurement problem is not "quantum mechanics is weird"; it is this trilemma, and every interpretation on the market is a decision about which proposition to deny. Decoherence, remember from Phase 2, does not vote — it explains why you never see the interference, not why you see one outcome.
Deny universality: the quantum formalism applies to systems under study, while the apparatus and observer are described classically — collapse happens at the interface. Operationally this is flawless; it is the physics that builds lasers and quantum computers. The price is the cut: the theory itself never says where quantum ends and classical begins, and the boundary can be moved almost at will without changing predictions. Bell's lifelong complaint was exactly this — a fundamental theory should not contain "measurement" as an unanalyzed primitive.
Deny single definite outcomes: keep completeness and universal unitarity, and accept what the formalism says — the superposed apparatus is real, measurement is just entanglement, and decoherence splits the global state into effectively non-interacting branches in which every outcome occurs. Nothing is added to the theory; that is the elegance. The price is probability: if every outcome happens with certainty, what can "the Born-rule chance was 1/3" possibly mean, and why should you have expected anything at all? Decision-theoretic derivations and self-locating uncertainty are the live proposals, and whether they succeed is genuinely contested.
Deny completeness, exactly as EPR hoped: particles always have definite positions, and the wave function is a real field that guides them — the pilot wave. Measurements simply reveal where the particles are; there is no collapse, the dynamics is deterministic, and every quantum prediction is reproduced. The price is stamped on the label: the guidance equation is flagrantly nonlocal — one particle's velocity depends instantaneously on the positions of all the others — which after your Phase 3 work reads less like a defect and more like honesty: Bell proved any such completion must be nonlocal, and Bohm's theory says so out loud. Its unresolved trouble is relativity: the dynamics prefers a frame.
Deny that the wave function is physical at all: |ψ⟩ is an agent's personal probability assignment — a gambling commitment about her own future experiences — and "collapse" is nothing but Bayesian updating on new data, no more mysterious than revising odds after seeing a card. Nonlocality evaporates (your updating changes your expectations, not Bob's lab), the cut becomes the unproblematic line between agent and world, and the Born rule becomes a normative coherence condition. The price: physics no longer describes the world as it is in itself — it regulates each agent's expectations — and the theory's astonishing intersubjective success becomes the thing needing explanation rather than the thing assumed.
Every interpretation pays a different price. Learn what each one buys — this phase feeds the humanities philosophy-of-physics unit directly.
This track is the soul of the quantum computing track. Entanglement is not a curiosity here — it is the resource that makes teleportation, superdense coding, and quantum advantage possible at all.
The measurement postulate is the spectral theorem wearing a lab coat — linear algebra and physics are the same subject seen from two sides. And the Bell inequalities you derive in Phase 3 are the engine of device-independent cryptography at QuSoft: security proven not from trusting the hardware, but from the violation itself.