Hilbert-space formalism of quantum mechanics

Intuition [Beginner]

The Stern-Gerlach setup of the prior unit forces a choice. A silver atom that has just been measured to be spin-up along the vertical axis has no value at all for spin along the horizontal axis — not "an unknown value" but no value to be uncovered. When you then measure horizontally, the atom gets a value, and that value comes out 50/50. The horizontal measurement does not reveal anything pre-existing; it imposes a fresh value, and in doing so destroys the vertical value that the first apparatus had imposed.

That picture is hard to fit into ordinary descriptions. In ordinary descriptions, objects have properties, and measurements reveal them. Quantum mechanics needs a different bookkeeping — a single mathematical setting in which "the atom is spin-up vertically" and "the atom has no spin-horizontal value" can sit beside each other without contradiction, and in which the rule for what happens when you measure horizontally is built in.

The setting is a complex inner-product space, called a Hilbert space. Every quantum system has one. The states of the system are unit-length vectors in that space. The observables — energy, position, momentum, the three components of spin, anything you could measure — are not numbers attached to the state but special linear maps from the space to itself, called self-adjoint operators. When you measure an observable, you get one of its eigenvalues, and the rule for which one is probabilistic, governed by the squared length of the projection of the state onto the corresponding eigenvector.

The whole of quantum mechanics is built from a short list of ingredients of this kind. Five postulates package the list. State, observable, measurement, time evolution, and how to combine systems together. The Stern-Gerlach experiment is the smallest substantive instance — a two-dimensional Hilbert space with three Pauli-matrix observables, and every move quantum mechanics ever makes appears already there. This unit upgrades that example to the general postulates, with the formal machinery (Hilbert space, self-adjoint operator, spectral decomposition, unitary evolution) plugged in everywhere a placeholder used to sit.

Two points are worth emphasising before the rest of the unit. First, the formalism is mathematically clean. It is finite axioms, well-posed objects, deterministic rules — once you accept the probabilistic measurement rule, every prediction QM ever makes is a matter of routine inner-product computation. Second, the formalism does not say what the state vector is. Whether the state vector represents the actual ontological content of the system or only an observer's best information about it is a separate, unresolved question, addressed in the philosophy-of-physics units on the measurement problem. The formalism's predictions are the same regardless. This unit gives the formalism; the interpretive question is independent.

Visual [Beginner]

Picture the Hilbert space as an arrow space, like ordinary three-dimensional arrows but with complex-number rather than real-number scaling and with possibly infinitely many dimensions. The state of a quantum system is one such arrow, scaled to unit length. An observable is a special direction-finder — it sorts the arrow space into an orthogonal set of "answer directions," one for each possible measurement outcome.

When you measure, the state's arrow casts shadows on each of the answer directions. The longer the shadow on a given direction, the more likely that direction's outcome is — specifically, the probability is the squared length of the shadow. The state then snaps to whichever direction was selected, normalised to unit length, ready for the next measurement.

That picture is exactly the spin-1/2 case from the previous unit, with the two-dimensional complex space ${\mathbb{C}}^2$ as the arrow space and the Pauli matrices as direction-finders. The general formalism is the same picture allowed to grow: more dimensions, possibly infinitely many; observables with possibly continuous spectra (so a continuum of answer directions); time evolution as a continuous unitary rotation of the whole space.

Worked example [Beginner]

A spin-1/2 particle has the state $|\psi ⟩=\alpha |\uparrow ⟩+\beta |\downarrow ⟩$ with complex amplitudes $\alpha ,\beta$ satisfying $|\alpha {|}^2+|\beta {|}^2=1$. Predict the outcome statistics for a Stern-Gerlach measurement of the vertical spin component.

The observable is $S_z$, whose two eigenvectors are exactly $|\uparrow ⟩$ and $|\downarrow ⟩$ with eigenvalues $+\hslash /2$ and $-\hslash /2$ respectively. The Born rule says: the outcome $+\hslash /2$ appears with probability $|⟨\uparrow |\psi ⟩{|}^2=|\alpha {|}^2$, and the outcome $-\hslash /2$ with probability $|⟨\downarrow |\psi ⟩{|}^2=|\beta {|}^2$. The two probabilities sum to one by the unit-length condition.

After the measurement, the state collapses. If $+\hslash /2$ was observed, the new state is $|\uparrow ⟩$; if $-\hslash /2$ was observed, the new state is $|\downarrow ⟩$. A repeat $S_z$ measurement now returns the same answer with certainty, because the state is already an eigenvector. This matches the prior unit's experimental fact that a Stern-Gerlach apparatus running twice along the same axis gives the same outcome both times.

Plug in numbers. Set $\alpha =1/\sqrt{2}$ and $\beta =1/\sqrt{2}$ — the state $|+x⟩$ from the prior unit, polarised along the horizontal axis. Then $|\alpha {|}^2=|\beta {|}^2=1/2$. A vertical Stern-Gerlach measurement returns spin-up half the time and spin-down half the time, exactly as the sequential Stern-Gerlach worked example predicted. The probabilities arise from the squared moduli of the complex amplitudes. Reverse-engineering: if you see 50/50 statistics for $S_z$ on an ensemble of identically prepared particles, the state vector has $|\alpha |=|\beta |=1/\sqrt{2}$ — but the relative phase between $\alpha$ and $\beta$ is not fixed by the $S_z$ data alone, and would require a measurement along a transverse axis to pin down.

Check your understanding [Beginner]

Formal definition [Intermediate+]

A quantum system is a pair $(\mathcal{H},\mathcal{A})$ consisting of a complex separable Hilbert space $\mathcal{H}$ and a distinguished family $\mathcal{A}$ of self-adjoint operators on $\mathcal{H}$ representing the observables of the system. The kinematic and dynamical content of quantum mechanics is encoded in the five Dirac-von Neumann postulates ^{[quantum-well Postulates of Quantum Mechanics.md]}, stated below.

(P1) States. The pure states of the system are the unit rays in $\mathcal{H}$ — equivalence classes $[\psi ]=\{e^{i\theta }|\psi ⟩:\theta \in \mathbb{R}\}$ of unit vectors $|\psi ⟩\in \mathcal{H}$ modulo the overall phase. The set of unit rays is the projective Hilbert space $\mathbb{P}\mathcal{H}$. More generally, the mixed states of the system are positive trace-class operators $\rho$ on $\mathcal{H}$ with $\mathrm{tr}\rho =1$ — the density operators. Pure states are the rank-one density operators $\rho =|\psi ⟩⟨\psi |$.

(P2) Observables. Every observable quantity is represented by a self-adjoint operator $A$ on $\mathcal{H}$. For a bounded observable $A\in \mathcal{B}(\mathcal{H})$ this means $A^{\ast }=A$ in the operator-adjoint sense 02.11.01. For an unbounded observable — typical of position, momentum, energy — self-adjointness is the technical condition $A=A^{\ast }$ on a dense domain, strictly stronger than mere symmetry $⟨A\varphi ,\psi ⟩=⟨\varphi ,A\psi ⟩$ on the domain 02.11.03. The distinction is load-bearing; the spectral theorem holds for self-adjoint but generally not for merely symmetric operators.

(P3) Measurement and Born rule. When the observable $A$ is measured in the pure state $|\psi ⟩$, the outcome is a real number drawn from the spectrum $\sigma (A)\subset \mathbb{R}$ of $A$. The probability distribution of the outcome is given by the spectral measure of $A$. In the simplest case where $A$ has pure point spectrum $\sigma (A)=\{a_1,a_2,\dots \}$ with orthonormal eigenvectors $|{\varphi }_n⟩$ and eigenvalues $a_n$, the probability of outcome $a_n$ is

$$\mathrm{Prob}(a_n\mid \psi )=|⟨{\varphi }_n|\psi ⟩{|}^2.$$

This is the Born rule ^{[Born 1926]}. For an observable with general spectrum (continuous, mixed, degenerate) the rule generalises via the projection-valued measure of $A$, stated in the Master tier.

After the measurement, the state collapses (the projection postulate) to the normalised projection onto the corresponding eigenspace: if outcome $a_n$ was observed, the new state is $P_{a_n}|\psi ⟩/\Vert P_{a_n}|\psi ⟩\Vert$, where $P_{a_n}$ projects onto the $a_n$-eigenspace. For a mixed state $\rho$, the probability of outcome $a_n$ is $\mathrm{tr}(P_{a_n}\rho )$ and the post-measurement state is $P_{a_n}\rho P_{a_n}/\mathrm{tr}(P_{a_n}\rho )$.

(P4) Time evolution. Closed-system dynamics is governed by a strongly continuous one-parameter unitary group $U(t)=e^{-iHt/\hslash }$ on $\mathcal{H}$, where the self-adjoint generator $H$ is the Hamiltonian of the system. The state evolves by

$$|\psi (t)⟩=U(t)|\psi (0)⟩,$$

equivalently by the Schrödinger equation

$$i\hslash \frac{d}{dt}|\psi (t)⟩=H|\psi (t)⟩,$$

read on the domain of $H$. For density operators the law is $\rho (t)=U(t)\rho (0)U(t{)}^{\ast }$, equivalently the von Neumann equation $i\hslash \dot{\rho }=[H,\rho ]$.

(P5) Composite systems. A composite system whose subsystems have Hilbert spaces ${\mathcal{H}}_A$ and ${\mathcal{H}}_B$ has Hilbert space ${\mathcal{H}}_A\otimes {\mathcal{H}}_B$ — the Hilbert-space tensor product, completed in the natural inner product $⟨a_1\otimes b_1,a_2\otimes b_2⟩=⟨a_1,a_2{⟩}_A⟨b_1,b_2{⟩}_B$. Observables on subsystem $A$ alone act as $A\otimes 1_B$ on the composite. A composite state is separable if it lies in the convex hull of product states ${\rho }_A\otimes {\rho }_B$, and entangled otherwise. The dedicated unit 12.02.02 develops the entanglement theory and partial-trace machinery.

Standard conventions

The Dirac bra-ket notation is used throughout. Kets $|\psi ⟩,|\varphi ⟩\in \mathcal{H}$ are vectors; bras $⟨\psi |,⟨\varphi |$ are the corresponding dual covectors, i.e. continuous linear functionals on $\mathcal{H}$ — by the Riesz representation theorem on Hilbert space 02.11.08, every continuous linear functional has the form $⟨\psi |\cdot ⟩$ for a unique $\psi \in \mathcal{H}$. The inner product $⟨\varphi |\psi ⟩\in \mathbb{C}$ is conjugate-linear in the first slot and linear in the second (physics convention; mathematics-convention texts flip this). A linear operator $A$ sandwiched between bra and ket reads as a matrix element $⟨\varphi |A|\psi ⟩=⟨\varphi ,A\psi ⟩$. The outer product $|\psi ⟩⟨\varphi |$ is the rank-one operator $|\psi ⟩⟨\varphi |(\xi )=⟨\varphi |\xi ⟩|\psi ⟩$. The rank-one projector $P_{\psi }:=|\psi ⟩⟨\psi |$ on the line spanned by a unit vector $|\psi ⟩$ is the density operator of the pure state $[\psi ]$.

Counterexamples to common slips

• A symmetric unbounded operator is generally not self-adjoint. The standard example is the momentum operator $p=-i\hslash {\partial }_x$ on $L^2([0,1])$ with domain $C_c^{\infty }((0,1))$: the operator is symmetric but the adjoint extends it to a larger domain. Whether self-adjoint extensions exist, and how many, is the subject of deficiency-index theory (von Neumann 1929) 02.11.03; this is where unbounded-observable physics gets technical ^{[Reed-Simon Vol. I §VIII]}.

• A complex amplitude $⟨\varphi |\psi ⟩$ is not a probability. The probability is its modulus squared. The amplitude itself can be complex, can be negative-real, and is the load-bearing object for interference phenomena: two indistinguishable paths to the same outcome have amplitudes that add, and the resulting outcome probability is the squared modulus of the sum, not the sum of the squared moduli.

• A self-adjoint operator's spectrum can be entirely continuous, entirely point, or mixed. The position operator $\hat{x}$ on $L^2(\mathbb{R})$ has purely continuous spectrum $\sigma (\hat{x})=\mathbb{R}$; no normalisable eigenvector exists. The naive "eigenstate of position" $|x⟩$ is a distribution, not an element of $\mathcal{H}$ — a useful formal device, but rigorously a rigged Hilbert space construction (Gelfand triple) not a Hilbert-space construction ^{[Hall *Quantum Theory for Mathematicians* Ch. 6]}.

• The Schrödinger equation is not a partial differential equation in the most general formulation. It is the abstract ODE $i\hslash \dot{\psi }=H\psi$ on $\mathcal{H}$. For a position-space wavefunction $\psi (x,t)\in L^2({\mathbb{R}}^n)$ and a Hamiltonian $H=-{\hslash }^2{\nabla }^2/(2m)+V(x)$, the abstract ODE becomes the familiar PDE. But the formalism is set up so the PDE is one representation of an underlying operator equation, not the primary object.

Key theorem with proof [Intermediate+]

Theorem (spectral theorem for bounded self-adjoint operators). Let $A$ be a bounded self-adjoint operator on a complex separable Hilbert space $\mathcal{H}$. There exists a unique projection-valued measure $E_A:\mathcal{B}(\mathbb{R})\to \mathcal{B}(\mathcal{H})$ on the Borel $\sigma$-algebra of $\mathbb{R}$, supported on the spectrum $\sigma (A)\subset \mathbb{R}$, such that

$$A={\int }_{\sigma (A)}\lambda d{E}_A(\lambda ).$$

Equivalently, for every bounded Borel function $f:\mathbb{R}\to \mathbb{C}$, the operator $f(A):=\int f(\lambda )d{E}_A(\lambda )$ is the value of the continuous functional calculus on $f$, and the map $f\mapsto f(A)$ is a $$-homomorphismfrom$C(\sigma(A))$to$\mathcal{B}(\mathcal{H})$.*

The projection-valued measure (PVM) $E_A(\Omega )$, for each Borel set $\Omega \subset \mathbb{R}$, is an orthogonal projection on $\mathcal{H}$. Intuitively, $E_A(\Omega )$ projects onto the closed subspace "where $A$ takes values in $\Omega$." The defining properties are: (i) $E_A(\varnothing )=0$ and $E_A(\mathbb{R})=1$; (ii) $E_A({\Omega }_1\cap {\Omega }_2)=E_A({\Omega }_1)E_A({\Omega }_2)$; (iii) countable additivity in the strong operator topology, $E_A({⨆}_n{\Omega }_n)={\sum }_n{E}_A({\Omega }_n)$ for disjoint families.

Proof sketch (bounded case). Three standard routes lead to the same conclusion; we sketch the continuous-functional-calculus / Gelfand-Naimark route, following Reed-Simon Vol. I §VII.3 ^{[Reed-Simon Vol. I §VII]} and Hall Ch. 7 ^{[Hall *Quantum Theory for Mathematicians*]}.

Step 1 — polynomial functional calculus. For any polynomial $p\in \mathbb{C}[\lambda ]$, define $p(A)\in \mathcal{B}(\mathcal{H})$ in the obvious way. The map $p\mapsto p(A)$ is an algebra homomorphism, and self-adjointness of $A$ together with the spectral-radius formula gives $\Vert p(A)\Vert =\Vert p{\Vert }_{C(\sigma (A))}$, the supremum norm of $p$ restricted to the spectrum (this is the spectral mapping theorem for polynomials combined with the fact that for self-adjoint $A$ the operator norm equals the spectral radius).

Step 2 — extension to continuous functional calculus. Polynomials are dense in $C(\sigma (A))$ by the Stone-Weierstrass theorem, since $\sigma (A)\subset \mathbb{R}$ is compact. The isometric polynomial calculus extends uniquely by continuity to an isometric $\ast$-homomorphism $\Phi :C(\sigma (A))\to \mathcal{B}(\mathcal{H})$ with $\Phi (\lambda )=A$ and $\Phi (\bar{f})=\Phi (f{)}^{\ast }$. This is the continuous functional calculus; it produces $f(A)$ for every continuous $f$ on the spectrum.

Step 3 — extension to bounded Borel functions. For each $\psi ,\varphi \in \mathcal{H}$, the map $f\mapsto ⟨\psi ,f(A)\varphi ⟩$ is a continuous linear functional on $C(\sigma (A))$. By the Riesz-Markov-Kakutani representation theorem, it equals $\int fd{\mu }_{\psi ,\varphi }$ for a unique complex Borel measure ${\mu }_{\psi ,\varphi }$ on $\sigma (A)$. The system of measures ${\mu }_{\psi ,\varphi }$ extends the functional calculus to all bounded Borel functions: define $⟨\psi ,f(A)\varphi ⟩:=\int fd{\mu }_{\psi ,\varphi }$ for bounded Borel $f$, and verify by polarisation that this assembles into a bounded operator $f(A)$.

Step 4 — read off the PVM. Set $E_A(\Omega ):={\chi }_{\Omega }(A)$, where ${\chi }_{\Omega }$ is the indicator function of the Borel set $\Omega$. Verify directly from the algebra-homomorphism property that $E_A(\Omega )$ is an orthogonal projection, $E_A({\Omega }_1\cap {\Omega }_2)=E_A({\Omega }_1)E_A({\Omega }_2)$, and countable additivity in the strong operator topology. The displayed identity $A=\int \lambda d{E}_A(\lambda )$ is the special case of the functional-calculus formula at $f(\lambda )=\lambda$.

Uniqueness. If $E$ and $E^{\prime }$ are two PVMs satisfying $A=\int \lambda dE(\lambda )=\int \lambda d{E}^{\prime }(\lambda )$, then $p(A)=\int pdE=\int pd{E}^{\prime }$ for every polynomial $p$, and hence $f(A)=\int fdE=\int fd{E}^{\prime }$ for every continuous $f$. The Riesz-Markov uniqueness pins $E=E^{\prime }$. $\square$

The unbounded case — needed for typical physical observables like the position $\hat{x}$, momentum $\hat{p}$, and Hamiltonian $H=-{\hslash }^2{\nabla }^2/(2m)+V$ — uses the same conclusion (a PVM-spectral decomposition) but with the proof routed through the Cayley transform: an unbounded self-adjoint $A$ on a dense domain gets mapped to a bounded unitary $U_A:=(A-i)(A+i{)}^{-1}$, whose spectral theorem (for normal operators) is the bounded-operator theorem in slight disguise. Pulling back recovers a PVM for $A$ on $\mathbb{R}$. Reed-Simon Vol. I §VIII.6 ^{[Reed-Simon Vol. I §VIII]} is the standard reference; the construction is one chapter of Hall ^{[Hall *Quantum Theory for Mathematicians*]}.

Corollary (Born rule, general form). For any self-adjoint observable $A$ with PVM $E_A$, and any state $|\psi ⟩$, the probability that a measurement of $A$ returns a value in the Borel set $\Omega \subset \mathbb{R}$ is

$$\mathrm{Prob}(A\in \Omega \mid \psi )=⟨\psi ,E_A(\Omega )\psi ⟩=\Vert E_A(\Omega )|\psi ⟩{\Vert }^2.$$

For pure point spectrum and $\Omega =\{a_n\}$ a single eigenvalue, $E_A(\{a_n\})=P_{a_n}$ is the eigenprojector and the formula reduces to $|⟨{\varphi }_n|\psi ⟩{|}^2$. For continuous spectrum, the probability of obtaining an exact eigenvalue is zero and the meaningful quantities are probabilities for the value to lie in an interval. $\square$

Theorem (Stone's theorem). A one-parameter family $\{U(t){\}}_{t\in \mathbb{R}}$ of unitary operators on a Hilbert space $\mathcal{H}$ is a strongly continuous unitary group (i.e. $U(s+t)=U(s)U(t)$, $U(0)=1$, and $t\mapsto U(t)\psi$ is norm-continuous for every $\psi \in \mathcal{H}$) if and only if there exists a unique self-adjoint operator $H$ on $\mathcal{H}$ (with a specific dense domain) such that $U(t)=e^{-iHt/\hslash }$, with the exponential defined via the spectral theorem.

The forward direction (group $\Rightarrow$ generator) takes the derivative $H\psi :=i\hslash {\lim }_{t\to 0}(U(t)\psi -\psi )/t$ on the dense subspace where the limit exists in norm, and proves self-adjointness via the structure of the group. The backward direction (self-adjoint $\Rightarrow$ group) defines $U(t):=e^{-iHt/\hslash }$ via the spectral theorem applied to the bounded function $\lambda \mapsto e^{-i\lambda t/\hslash }$. The proof is Reed-Simon Vol. I §VIII.4 ^{[Reed-Simon Vol. I §VIII]}; the originator paper is Stone 1932 ^{[Stone 1932]}. $\square$

Stone's theorem packages postulate (P4) cleanly: the unitary evolution law $U(t)$ and the Hamiltonian-as-self-adjoint-operator description are equivalent data. Choosing one fixes the other.

Bridge. The Born rule and the spectral theorem together formalise the measurement postulate (P3) into a single statement: the observable $A$ determines a probability measure on its spectrum for each state via the PVM, and the measurement is the sampling step. The unitary-group / self-adjoint-generator correspondence of Stone's theorem formalises postulate (P4) the same way. Both proofs route through the same piece of functional analysis — the spectral theorem and its functional calculus — so the operator-theoretic content of (P3) and (P4) is one theorem applied twice. The bridge to physics is via specific Hamiltonians: $H=-{\hslash }^2{\nabla }^2/(2m)+V(x)$ on $L^2({\mathbb{R}}^3)$ for a particle in a potential (forthcoming 12.06.01 hydrogen atom), $H=\omega (a^{\ast }a+1/2)$ on the symmetric Fock space for a harmonic oscillator (forthcoming 12.04.02), or the spin-$1/2$ $H=-\mathbf{\mu }\cdot \mathbf{B}=(\hslash \omega /2){\sigma }_z$ on ${\mathbb{C}}^2$ recovered as the simplest possible case from the prior unit 12.01.02.

Exercises [Intermediate+]

A graded set covering Born-rule probability computation, spectral decomposition, unitary evolution, composite systems, and the role of self-adjointness vs. symmetry.

Lean formalization [Intermediate+]

lean_status: none — Mathlib provides the analytic substrate (inner-product spaces, separable Hilbert spaces via Mathlib.Analysis.InnerProductSpace.l2Space, bounded operators and their adjoints via Mathlib.Analysis.InnerProductSpace.Adjoint, the finite-dimensional spectral theorem for self-adjoint endomorphisms), but the load-bearing pieces of the Dirac-von Neumann formalism — the projection-valued-measure spectral theorem for bounded self-adjoint operators on an arbitrary separable Hilbert space, the Cayley-transform extension to the unbounded case, Stone's theorem for strongly continuous one-parameter unitary groups, the Born rule packaged as a probability measure on the spectrum, the trace-class infrastructure for density operators, and Gleason's theorem — are not in Mathlib at this date and each constitutes a substantial contribution candidate. The shipped finite-dimensional pieces (spectral decomposition of a Hermitian matrix, the Born rule on ${\mathbb{C}}^n$, unitary evolution by matrix exponential) cover the qubit / spin-$1/2$ case directly but do not extend automatically to the infinite-dimensional regime where the physics actually lives. This unit is reviewer-attested. See the lean_mathlib_gap block in the frontmatter for the full list.

The Born rule and Gleason's theorem [Master]

The Born rule (P3) is the postulate that has attracted the most attention as a candidate for derivation from more primitive axioms. Gleason's theorem (1957) ^{[Gleason 1957]} gives the strongest such derivation: under modest hypotheses, any assignment of probabilities to the closed subspaces of a Hilbert space of dimension $\ge 3$ must arise from a density operator via the trace formula. The Born rule is then not a separate axiom but a structural consequence of the projector-lattice structure of the observable algebra.

Theorem (Gleason, 1957). Let $\mathcal{H}$ be a complex separable Hilbert space with $\dim \mathcal{H}\ge 3$. A function $\mu :\mathcal{P}(\mathcal{H})\to [0,1]$ on the lattice of orthogonal projections, satisfying $\mu (1)=1$ and countable additivity $\mu ({\sum }_n{P}_n)={\sum }_n\mu (P_n)$ for any countable family of mutually orthogonal projections $\{P_n\}$, is necessarily of the form

$$\mu (P)=\mathrm{tr}(\rho P)$$

for a unique density operator $\rho$ on $\mathcal{H}$ (positive trace-class with $\mathrm{tr}\rho =1$).

Gleason's theorem says: once you accept that probabilities of measurement outcomes are functions on the projector lattice (which is the lattice of "yes/no questions" — does the system have a certain property?) that respect the orthogonality structure, the trace formula is the probability rule. Any other formula would fail either positivity, normalisation, or countable additivity. The Born rule does not need to be postulated separately; it is forced by (P1)+(P2)+the requirement that "measurement outcome statistics" make probabilistic sense.

The proof is substantive. The dimension hypothesis $\dim \ge 3$ is essential — in dimension 2, the lattice of projections is just the Bloch sphere, and many non-trace probability assignments are possible (the Kochen-Specker structure fails to constrain them). Gleason's original proof handles real and complex cases separately and uses spherical-trigonometry-style continuity arguments on the projector lattice. Modern treatments use the structure of the unitary group acting on the projector lattice; Hall Ch. 19 ^{[Hall *Quantum Theory for Mathematicians*]} gives a clean modern presentation. Bell 1966 ^{[Bell *Rev. Mod. Phys.* 38, 447 (1966)]} gave a parallel result (Kochen-Specker theorem, 1967, building on Bell's argument) showing the impossibility of non-contextual hidden-variable assignments, drawing the same conclusion from a slightly different angle. The two results together — Gleason and Kochen-Specker — pin down the Born rule as essentially the only consistent probability rule on the quantum projector lattice.

What the theorem does not settle: it does not explain why probabilities of measurement outcomes should exist in the first place, nor why the projector lattice should be the right setting. It says: given that there are probabilities and given that the questions are projections, the Born rule is forced. The deeper "why does QM use a complex Hilbert space at all" question is the subject of work by Hardy, Chiribella-D'Ariano-Perinotti, Masanes-Müller, and others on operational reconstructions of quantum mechanics from information-theoretic axioms — beyond the scope of this unit, but a current research thread that sits structurally on top of the Gleason result.

The measurement problem and the projection postulate [Master]

The projection postulate — that after a measurement of outcome $a_n$ the state collapses to $P_{a_n}|\psi ⟩/\Vert P_{a_n}|\psi ⟩\Vert$ — is the formal source of the measurement problem of quantum mechanics. The formalism contains two dynamical laws: the unitary Schrödinger evolution (P4), continuous and deterministic; and the projective collapse (P3), discontinuous and probabilistic. The two cannot be simultaneously universal — unitary evolution preserves superpositions, and collapse destroys them. The measurement postulate picks the second when "a measurement happens"; but the formalism gives no rule for what counts as a measurement.

Several research programmes attempt to dissolve or relocate this seam:

• Many-worlds (Everett 1957) ^{[Everett 1957]} takes the unitary law as universal and reads the appearance of collapse as a result of decoherence and observer-branching. The post-measurement state is the full unitary superposition ${\sum }_n{c}_n|{\varphi }_n⟩\otimes |\text{apparatus reads }{a}_n⟩$, and the Born-rule probabilities are interpreted as branch weights via decision-theoretic arguments (Deutsch 1999, Wallace 2010-2012).
• Bohmian mechanics (Bohm 1952; de Broglie 1927) ^{[Bohm 1952]} adds a deterministic configuration-space trajectory to the formalism, driven by the wavefunction's phase gradient. The Born rule becomes a statement about the equilibrium distribution of trajectory ensembles (Dürr-Goldstein-Zanghì 1992); collapse is no longer a primitive but a consequence of the conditional wavefunction of subsystems.
• GRW / spontaneous-localisation theories (Ghirardi-Rimini-Weber 1986; Pearle 1989) modify the Schrödinger equation by adding small, frequent, spontaneous projective collapses with a localisation-rate parameter $\lambda$ tuned so microscopic systems behave unitarily and macroscopic systems collapse rapidly. This is the only programme that makes the measurement postulate empirically distinguishable from the unitary law, and is the subject of ongoing precision tests.
• QBism / Copenhagen (Bohr 1928; Heisenberg 1927; Fuchs et al. 2010s) denies that the wavefunction represents an ontological state at all; it is a personalist degree-of-belief assignment that updates under measurement via Bayes' rule, with the Born rule playing the role of a Dutch-book consistency constraint.

The Hilbert-space formalism of this unit does not select among these. All of them reproduce the same predictions for outcome statistics and post-measurement states; they differ only in what they say is happening during a measurement. The philosophy-of-physics unit 20.03.01 (proposed) addresses the interpretive question; the formalism above is the empirical content all interpretations agree on. The decoherence theory underlying many-worlds is treated in the open-system unit 12.07.05 (forthcoming).

Composite systems, tensor products, and entanglement [Master]

Postulate (P5) — that composite systems use Hilbert-space tensor products — is the source of all distinctively quantum correlations. For two-subsystem case $\mathcal{H}={\mathcal{H}}_A\otimes {\mathcal{H}}_B$, the algebraic tensor product is the linear span of formal products $|a⟩\otimes |b⟩$ modulo bilinearity; the Hilbert tensor product is the completion in the natural inner product. For separable Hilbert spaces (the only relevant case in physics), this completion procedure is well-defined and the resulting space is again separable.

A pure state $|\Psi ⟩\in {\mathcal{H}}_A\otimes {\mathcal{H}}_B$ is called a product state if it can be written $|\Psi ⟩=|{\psi }_A⟩\otimes |{\psi }_B⟩$ for some $|{\psi }_A⟩,|{\psi }_B⟩$; otherwise it is entangled. The Schmidt decomposition asserts that every $|\Psi ⟩\in {\mathcal{H}}_A\otimes {\mathcal{H}}_B$ admits an expansion $|\Psi ⟩={\sum }_n\sqrt{{\lambda }_n}|e_n{⟩}_A\otimes |f_n{⟩}_B$ with ${\lambda }_n\ge 0$, ${\sum }_n{\lambda }_n=1$, and $\{|e_n⟩\},\{|f_n⟩\}$ orthonormal families on ${\mathcal{H}}_A,{\mathcal{H}}_B$ respectively. A product state is exactly a single-term Schmidt decomposition; entanglement is the multi-term case. The number of non-zero Schmidt coefficients is the Schmidt rank, an invariant of $|\Psi ⟩$ under local-unitary transformations $U_A\otimes U_B$.

The reduced density operator of subsystem $A$ is ${\rho }_A:={\mathrm{tr}}_B(|\Psi ⟩⟨\Psi |)$, where the partial trace ${\mathrm{tr}}_B:\mathcal{B}({\mathcal{H}}_A\otimes {\mathcal{H}}_B)\to \mathcal{B}({\mathcal{H}}_A)$ is defined by ${\mathrm{tr}}_B(O_A\otimes O_B)=O_A\mathrm{tr}(O_B)$ and extended linearly. The reduced density operator gives the expectation value of any local observable $A\otimes 1_B$: $⟨A\otimes 1_B{⟩}_{\Psi }=\mathrm{tr}(A{\rho }_A)$. For a pure state $|\Psi ⟩$, ${\rho }_A$ is pure if and only if $|\Psi ⟩$ is a product state; for entangled $|\Psi ⟩$, ${\rho }_A$ is genuinely mixed. The von Neumann entropy $S({\rho }_A):=-\mathrm{tr}({\rho }_A\log {\rho }_A)$ quantifies the entanglement; it ranges from $0$ (product state) to $\log d_A$ (maximally entangled, where $d_A$ is the dimension of ${\mathcal{H}}_A$).

The singlet state of the prior unit, $|{\Psi }^{-}⟩=(|\uparrow \downarrow ⟩-|\downarrow \uparrow ⟩)/\sqrt{2}$, is maximally entangled. Its reduced density operator is ${\rho }_A=1/2$, with $S({\rho }_A)=\log 2$ — one qubit of entanglement, the unit relative to which all other bipartite entanglement is measured. The pattern generalises: maximally entangled states of two $d$-dimensional subsystems are unitarily equivalent to ${\sum }_{n=1}^d|n{⟩}_A\otimes |n{⟩}_B/\sqrt{d}$, with Schmidt coefficients ${\lambda }_n=1/d$. The dedicated unit 12.02.02 (proposed successor) develops the entanglement theory, Schmidt decomposition, and partial-trace machinery in detail; this unit fixes the foundational postulate.

Synthesis. Postulate (P5) is the structural source of every distinctively quantum phenomenon involving multiple subsystems: EPR correlations [Einstein-Podolsky-Rosen 1935], Bell-inequality violations [Bell 1964, Aspect 1982, loophole-free 2015], quantum teleportation [Bennett et al. 1993], dense coding, quantum cryptography, and the entire framework of quantum computation [Nielsen-Chuang 2010]. The reason the tensor product, rather than the direct sum or some other combination, is the right structure for composing quantum systems is a deep question with information-theoretic-axiomatic answers — Hardy 2001 and Chiribella-D'Ariano-Perinotti 2011 derive it from "purification" and "local tomography" axioms — but the formalism takes it as given. The bridge from Hilbert-space tensor products to physical predictions runs through the partial trace and Schmidt decomposition, both of which reduce to elementary linear algebra in the finite-dimensional case and are well-defined infinite-dimensional analogues otherwise.

Connections [Master]

• Hilbert space 02.11.08. The complete inner-product structure is the substrate the entire formalism runs on. The Riesz representation theorem identifies bras with kets, the Cauchy-Schwarz inequality powers the Robertson uncertainty proof, and completeness is essential for the spectral theorem and Stone's theorem.

• Bounded and unbounded self-adjoint operators [02.11.01, 02.11.03]. Observables are self-adjoint operators; the bounded case (spin, qubits, finite systems) and the unbounded case (position, momentum, energy) are unified by the spectral theorem via the Cayley transform. The distinction between symmetric and self-adjoint is the technical source of all infinite-dimensional subtleties in the formalism.

• Banach spaces 02.11.04. Trace-class and Hilbert-Schmidt operators form Banach spaces (in fact ideals in $\mathcal{B}(\mathcal{H})$) on which density operators live. The trace-norm topology is the right notion of convergence for mixed states; the operator norm controls the Schrödinger evolution.

• Stern-Gerlach and spin-1/2 12.01.02 pending. The two-dimensional Hilbert space ${\mathbb{C}}^2$ is the canonical example, with the Pauli matrices as concrete observables. Every feature of the general formalism — superposition, measurement, projection, unitary evolution, composite systems via the singlet state — is already present in the spin-1/2 case. Read in the other direction, this unit is the upgrade of the spin-1/2 worked example to the general axiomatic setting.

• Clifford algebra and spin double cover [03.09.02, 03.09.03]. The spin-1/2 representation realising $\mathrm{SU}(2)$ on ${\mathbb{C}}^2$ inside this formalism is, structurally, the smallest faithful irrep of the universal cover of the rotation group. The Hilbert-space formalism is set up to accommodate any unitary representation of any compact Lie group; spin-1/2 is just the case where the group is $\mathrm{SU}(2)$ and the representation is the defining one.

• Lie group / Lie algebra representations [07.06.01, 07.07.01]. The Hamiltonian generates time translations via $U(t)=e^{-iHt/\hslash }$; more generally, any one-parameter group of symmetries of a quantum system is generated by a self-adjoint operator (Stone's theorem), and the algebra of generators is a Lie algebra representation on the state space. The interplay between symmetry groups and observable algebras is the input for angular-momentum theory 12.05.01, the Wigner-Eckart theorem 12.05.04, and the canonical-commutation-relation algebra 12.04.01 (all forthcoming).

• Composite-system tensor products 12.02.02 (proposed successor). The composite-system postulate (P5) is the structural foundation; the dedicated unit develops Schmidt decomposition, entanglement entropy, partial trace, and Bell-state material in full.

• Time evolution and the Schrödinger equation 12.03.01 (proposed successor). Stone's theorem identifies the Hamiltonian with the generator of unitary evolution; the dedicated unit develops the Heisenberg picture, the interaction picture, the time-dependent Schrödinger equation, and the propagator formalism.

• CCR algebra and the Stone-von Neumann theorem 12.04.01 (proposed successor). The canonical commutation relation $[\hat{x},\hat{p}]=i\hslash$ together with Stone's theorem leads to the Weyl form of the CCR and the Stone-von Neumann uniqueness theorem identifying all irreducible representations of the CCR with the Schrödinger representation on $L^2(\mathbb{R})$.

• The measurement problem and interpretations 20.03.01 (proposed phil unit). The projection postulate raises interpretive questions that the formalism alone does not settle. The philosophy-of-physics unit takes this unit's postulates as its formal target and surveys Copenhagen, many-worlds, Bohmian, GRW, and QBist responses.

• Bosonic Fock space and second quantisation 12.13.01. Direct downstream application. The Section 3 Fock-space construction takes a one-particle Hilbert space $\mathcal{H}$ (any system fitting the present unit's axioms) and assembles the many-body Hilbert space ${\mathcal{F}}_s(\mathcal{H})={⨁}_n{S}^n\mathcal{H}$ for indistinguishable bosons. The Dirac-von Neumann postulates developed here apply verbatim on ${\mathcal{F}}_s(\mathcal{H})$, with the creation and annihilation operators $a^{\ast }(f),a(f)$ as unbounded observables satisfying the canonical commutation relations $[a(f),a^{\ast }(g)]=⟨f,g⟩$. The CCR is the prototype unbounded-operator non-commutativity that the present unit's $[\hat{x},\hat{p}]=i\hslash$ example anticipates, and the Stone-von Neumann theorem mentioned in the forward-reference to 12.04.01 is the uniqueness theorem stated and proved in the downstream Fock-space unit. The harmonic-oscillator Hamiltonian $H=\omega (a^{\ast }a+1/2)$ on ${\mathcal{F}}_s(\mathbb{C})$ is a worked example of the spectral-theorem-meets-Stone's-theorem machinery developed here.

• Fermionic Fock space and the Pauli exclusion principle 12.13.02. The antisymmetric sibling: ${\mathcal{F}}_a(\mathcal{H})={⨁}_n{\Lambda }^n\mathcal{H}$ uses the canonical anticommutation relations $\{a(f),a^{\ast }(g)\}=⟨f,g⟩$ instead of commutators. The Dirac-von Neumann postulates apply verbatim, with the new structural feature that the fermionic creation and annihilation operators are bounded with $\Vert a^{\ast }(f)\Vert =\Vert f\Vert$ (unlike the bosonic case). The exclusion principle is the spectral statement $\sigma (N_k)=\{0,1\}$ on each mode-occupation observable, expressed inside this unit's spectral-theorem framework. The Stern-Gerlach ${\mathbb{C}}^2$ of 12.01.02 is the smallest fermionic Fock space ($n=1$ one-particle mode), so the worked spin-1/2 example of the present unit is also the smallest-possible CAR representation, with the Pauli matrices generating the Clifford algebra ${\mathrm{Cl}}_2$ that is the fermionic analogue of the bosonic Heisenberg algebra.

Historical & philosophical context [Master]

The Hilbert-space formalism of quantum mechanics emerged in two stages between 1925 and 1932. The first stage, in Göttingen and Cambridge in 1925-1926, produced the working formalism — Heisenberg's matrix mechanics (Heisenberg 1925 ^{[Heisenberg *Z. Phys.* 33, 879 (1925)]}; Born-Heisenberg-Jordan 1926 ^{[Born-Jordan *Z. Phys.* 34, 858 (1926); Born-Heisenberg-Jordan *Z. Phys.* 35, 557 (1926)]}) and Schrödinger's wave mechanics (Schrödinger 1926 ^{[Schrödinger *Ann. Phys.* 79, 361 and 489 (1926)]}) — together with the realisation that the two were equivalent (Schrödinger 1926; Eckart 1926; Dirac 1926). Born's 1926 paper Zur Quantenmechanik der Stoßvorgänge ^{[Born 1926 *Z. Phys.* 37, 863]} gave the probabilistic interpretation: $|\psi (x){|}^2$ is the probability density for the particle's position, with the squared modulus the load-bearing object rather than the wavefunction itself.

The second stage was the mathematical reconstruction. Dirac's The Principles of Quantum Mechanics (1930) ^{[Dirac 1958]} introduced the bra-ket notation and the abstract operator formalism, with the convenient but mathematically problematic device of "improper" eigenstates $|x⟩,|p⟩$ for continuous-spectrum observables. Von Neumann's Mathematische Grundlagen der Quantenmechanik (1932) ^{[von Neumann 1932/1955]} supplied the rigorous Hilbert-space framework: the spectral theorem for unbounded self-adjoint operators (which Dirac's improper-eigenstate calculus needed to be rescued from), the projection-postulate formulation of measurement, and the density-operator (then called "statistical operator") description of mixed states. Stone's 1932 paper on one-parameter unitary groups ^{[Stone 1932]} supplied the time-evolution theorem the same year. The framework's deep structural results came over the next three decades: Wigner's 1931 representation theorem on symmetries (every symmetry is unitary or antiunitary), Wigner's 1939 ^{[Wigner *Ann. Math.* 40, 149 (1939)]} classification of Poincaré-group representations (mass and spin), Gleason's 1957 theorem ^{[Gleason 1957]}, and the Bell-Kochen-Specker no-go results of the 1960s.

By the time of Mackey's Mathematical Foundations of Quantum Mechanics (1963) ^{[Mackey 1963]} and Jauch's Foundations of Quantum Mechanics (1968), the axiomatic framework had stabilised into the Dirac-von Neumann form used in this unit. Modern treatments — Reed-Simon Vol. I-IV (1972-1979) ^{[Reed-Simon Vol. I]}, Hall's Quantum Theory for Mathematicians (2013) ^[Hall], Takhtajan's Quantum Mechanics for Mathematicians (2008) ^[Takhtajan] — give the framework with the modern functional-analytic technology fully integrated. The Dirac improper-eigenstate calculus is rigorously reconstructed via the rigged Hilbert space / Gelfand triple construction (Gelfand-Vilenkin 1964; Bohm-Gadella 1989), placing Dirac's notation on firm mathematical ground.

The philosophical reception has been continuously contested since Bohr-Einstein debates of 1927-1935. The EPR paper of 1935 ^{[Einstein-Podolsky-Rosen *Phys. Rev.* 47, 777 (1935)]} argued that the formalism's description of entangled states must be incomplete; Bell's 1964 theorem ^{[Bell *Physics* 1, 195 (1964)]} showed that no local hidden-variable completion can reproduce the formalism's predictions; the Aspect 1982 ^{[Aspect-Grangier-Roger *Phys. Rev. Lett.* 49, 91 (1982)]} and 2015 loophole-free Bell tests ^{[Hensen et al. *Nature* 526, 682 (2015); Giustina et al., Shalm et al. — same year]} confirmed the formalism's predictions definitively. The interpretive question — what the state vector is, what happens during a measurement — remains open and is the subject of the philosophy-of-physics unit 20.03.01.

Bibliography [Master]

Primary literature (cite when used; not all currently in reference/):

• Born, M., Z. Phys. 37, 863 (1926); Z. Phys. 38, 803 (1926). The probability-amplitude interpretation. [Need to source.]
• Heisenberg, W., Z. Phys. 33, 879 (1925). Matrix-mechanics original. [Need to source.]
• Born, M. & Jordan, P., Z. Phys. 34, 858 (1926); Born, M., Heisenberg, W., & Jordan, P., Z. Phys. 35, 557 (1926). Three-man paper consolidating matrix mechanics. [Need to source.]
• Schrödinger, E., Ann. Phys. 79, 361 and 489 (1926); Ann. Phys. 80, 437 (1926); Ann. Phys. 81, 109 (1926). Wave-mechanics series. [Need to source.]
• Dirac, P. A. M., Proc. Roy. Soc. A 109, 642 (1926). Equivalence of matrix and wave mechanics. [Need to source.]
• Stone, M. H., Ann. Math. 33, 643 (1932). Stone's theorem. [Need to source.]
• Gleason, A. M., J. Math. Mech. 6, 885 (1957). Gleason's theorem. [Need to source.]
• Wigner, E. P., Ann. Math. 40, 149 (1939). Classification of Poincaré-group irreps. [Need to source — shared with 12.01.02.]
• Einstein, A., Podolsky, B. & Rosen, N., Phys. Rev. 47, 777 (1935). EPR. [Need to source.]
• Bell, J. S., Physics 1, 195 (1964). Bell's theorem. [Need to source.]

Canonical monographs (Dirac-von Neumann lineage):

• Dirac, P. A. M., The Principles of Quantum Mechanics, 4th ed. (Oxford University Press, 1958). The originator monograph; bra-ket notation, the general principles. [Need to source.]
• von Neumann, J., Mathematische Grundlagen der Quantenmechanik (Springer, 1932); English translation Mathematical Foundations of Quantum Mechanics (Princeton University Press, 1955). The rigorous Hilbert-space formulation. [Need to source.]
• Mackey, G. W., Mathematical Foundations of Quantum Mechanics (Benjamin, 1963). Axiomatic / lattice-theoretic treatment. [Need to source.]

Modern mathematics-side treatments:

• Reed, M. & Simon, B., Methods of Modern Mathematical Physics, Vol. I: Functional Analysis (Academic Press, 1980). §VII spectral theorem; §VIII Stone's theorem and unbounded operators. [Need to source.]
• Hall, B. C., Quantum Theory for Mathematicians (Springer GTM 267, 2013). Modern mathematics-side treatment. [Need to source.]
• Takhtajan, L. A., Quantum Mechanics for Mathematicians (AMS GSM 95, 2008). [Need to source.]

Physics-side textbooks:

• Sakurai, J. J. & Napolitano, J., Modern Quantum Mechanics, 2nd ed. (Pearson/Cambridge, 2011). Ch. 1 §1.2-1.7. [Need to source — shared with 12.01.02.]
• Griffiths, D. J., Introduction to Quantum Mechanics, 2nd ed. (Pearson, 2005). Ch. 3 formalism. [Need to source — shared with 12.01.02.]
• Shankar, R., Principles of Quantum Mechanics, 2nd ed. (Plenum, 1994). Chs. 1, 4. [Need to source.]
• Susskind, L. & Friedman, A., Quantum Mechanics: The Theoretical Minimum (Basic Books, 2014). [Need to source — shared with 12.01.02.]
• Feynman, R. P., Leighton, R. B. & Sands, M., The Feynman Lectures on Physics, Vol. III (Addison-Wesley, 1965). [Need to source — shared with 12.01.02.]

Information-theoretic and measurement-focused:

• Nielsen, M. A. & Chuang, I. L., Quantum Computation and Quantum Information, 10th anniversary ed. (Cambridge University Press, 2010). Ch. 2 §2.2. [Need to source.]
• Weinberg, S., Lectures on Quantum Mechanics, 2nd ed. (Cambridge University Press, 2015). Ch. 3. [Need to source — shared with 12.01.02.]
• Busch, P., Lahti, P. & Mittelstaedt, P., The Quantum Theory of Measurement, 2nd ed. (Springer, 1996). POVM-generalised measurement. [Need to source.]
• Tong, D., Topics in Quantum Mechanics, DAMTP Cambridge lecture notes (concepts/quantum.pdf). [Have.]

Interpretive monographs (for 20.03.01 cross-cite):

• Bell, J. S., Speakable and Unspeakable in Quantum Mechanics, 2nd ed. (Cambridge, 2004). Collected papers on the measurement problem. [Need to source.]
• Wallace, D., The Emergent Multiverse (Oxford, 2012). Modern many-worlds. [Need to source.]
• Dürr, D., Goldstein, S. & Zanghì, N., Quantum Physics Without Quantum Philosophy (Springer, 2013). Bohmian. [Need to source.]

---

Section 2 opener: first production unit in chapter 12-quantum/02-formalism/, opening the path past the item-A block. Builds on the spin-1/2 seed 12.01.02 from chapter 01-foundations/ and the functional-analysis substrate in 02-analysis/11-functional-analysis/. Citations marked TODO_REF indicate sources expected in reference/ but not yet acquired; replace with [ref: …] when sources arrive.