STEMviz
Physics · Quantum Mechanics and Modern Physics

Mathematical Formalism: Hilbert Space and Hermitian Operators

Quantum states are normalised kets ∣ψ⟩|\psi\rangle in a complex Hilbert space, with inner product ⟨ϕ∣ψ⟩=⟨ψ∣ϕ⟩∗\langle\phi|\psi\rangle=\langle\psi|\phi\rangle^*. Orthonormal bases give ∣ψ⟩=∑ncn∣n⟩|\psi\rangle=\sum_n c_n|n\rangle, cn=⟨n∣ψ⟩c_n=\langle n|\psi\rangle, and completeness ∑n∣n⟩⟨n∣=1^\sum_n|n\rangle\langle n|=\hat{\mathbb 1}. Observables are Hermitian operators (A^†=A^\hat A^\dagger=\hat A) with real eigenvalues, orthogonal eigenvectors and a complete eigenbasis, so A^=∑nan∣an⟩⟨an∣\hat A=\sum_n a_n|a_n\rangle\langle a_n|. Measuring A^\hat A gives ana_n with probability ∥P^n∣ψ⟩∥2\lVert\hat P_n|\psi\rangle\rVert^2; [A^,B^]=0[\hat A,\hat B]=0 iff a common eigenbasis exists. Unitary operators (U^†U^=1^\hat U^\dagger\hat U=\hat{\mathbb 1}) implement change of basis, evolution and symmetries.

Mathematical Formalism: Hilbert Space and Hermitian Operators

The first three chapters used wave functions and integrals. This chapter recasts the same physics in the language it is actually done in: states are vectors in a complex inner-product space, observables are Hermitian operators on that space, and measurement is projection onto an eigenbasis. Nothing physical is added — the postulates of Chapter 2 are simply rewritten in a form that is representation-independent, that makes the role of Hermiticity transparent, and that generalises without change to spin (Chapter 10), to many particles, and to quantum field theory. The notation is Dirac's: kets ∣ψ⟩|\psi\rangle, bras ⟨ϕ∣\langle\phi|, brackets ⟨ϕ∣ψ⟩\langle\phi|\psi\rangle.

The organising idea is a strict analogy with ordinary vectors. A state is like an arrow; a basis is a set of axes; the components of the arrow depend on the axes but the arrow does not. "Position space" and "momentum space" are two choices of axes for the same state. An observable is a linear map that, in its own eigenbasis, is diagonal, with the eigenvalues down the diagonal — and those eigenvalues are the possible measurement results. Hermiticity is exactly the condition that makes that eigenbasis real, orthogonal and complete.

How to read this chapter. Three facts carry all the weight. (i) The spectral theorem: a Hermitian operator has real eigenvalues and an orthonormal, complete set of eigenvectors — so any state expands as ∣ψ⟩=∑ncn∣n⟩|\psi\rangle = \sum_n c_n|n\rangle. (ii) The generalised Born rule: measuring A^\hat A gives eigenvalue ana_n with probability ∣⟨n∣ψ⟩∣2|\langle n|\psi\rangle|^2, then collapses ∣ψ⟩|\psi\rangle to ∣n⟩|n\rangle. (iii) Compatible observables ([A^,B^]=0[\hat A,\hat B]=0) share an eigenbasis and can be known simultaneously; incompatible ones cannot (Chapter 5).

1. Hilbert Space

A Hilbert space H\mathcal{H} is a complex vector space with an inner product, complete in the induced norm. For quantum mechanics the relevant instance is L2L^2 — the square-integrable functions — but the axioms are all that matter:

  • Vectors (kets) ∣ψ⟩|\psi\rangle can be added and scaled by complex numbers: a∣ψ⟩+b∣ϕ⟩∈Ha|\psi\rangle + b|\phi\rangle \in \mathcal{H}.
  • Inner product ⟨ϕ∣ψ⟩∈C\langle\phi|\psi\rangle \in \mathbb{C}, linear in the second slot, antilinear in the first: ⟨ϕ∣ψ⟩=⟨ψ∣ϕ⟩∗\langle\phi|\psi\rangle = \langle\psi|\phi\rangle^*, and ⟨ψ∣ψ⟩≥0\langle\psi|\psi\rangle \ge 0 with equality only for ∣ψ⟩=0|\psi\rangle = 0.
  • Norm ∥∣ψ⟩∥=⟨ψ∣ψ⟩\||\psi\rangle\| = \sqrt{\langle\psi|\psi\rangle}; physical states are normalised, ⟨ψ∣ψ⟩=1\langle\psi|\psi\rangle = 1.

For wave functions the inner product is ⟨ϕ∣ψ⟩=∫ϕ∗(x) ψ(x) dx\langle\phi|\psi\rangle = \int \phi^*(x)\,\psi(x)\,dx. A global phase ∣ψ⟩→eiα∣ψ⟩|\psi\rangle \to e^{i\alpha}|\psi\rangle leaves every physical prediction unchanged, so strictly the states are rays, not vectors.

2. Dirac Notation: Bras, Kets and Operators

Every ket ∣ψ⟩|\psi\rangle has a dual bra ⟨ψ∣\langle\psi|, a linear functional that eats a ket and returns the number ⟨ψ∣ϕ⟩\langle\psi|\phi\rangle. An operator A^\hat A maps kets to kets, A^∣ψ⟩∈H\hat A|\psi\rangle \in \mathcal{H}. The object ∣ϕ⟩⟨χ∣|\phi\rangle\langle\chi| is an operator (an "outer product"): it sends ∣ψ⟩↦∣ϕ⟩ ⟨χ∣ψ⟩|\psi\rangle \mapsto |\phi\rangle\,\langle\chi|\psi\rangle.

The adjoint A^†\hat A^\dagger is defined by ⟨ϕ∣A^ψ⟩=⟨A^†ϕ∣ψ⟩\langle\phi|\hat A\psi\rangle = \langle\hat A^\dagger\phi|\psi\rangle for all ∣ϕ⟩,∣ψ⟩|\phi\rangle,|\psi\rangle. Rules: (A^B^)†=B^†A^†(\hat A\hat B)^\dagger = \hat B^\dagger\hat A^\dagger, (cA^)†=c∗A^†(c\hat A)^\dagger = c^*\hat A^\dagger, (∣ϕ⟩⟨χ∣)†=∣χ⟩⟨ϕ∣(|\phi\rangle\langle\chi|)^\dagger = |\chi\rangle\langle\phi|. Taking the dagger of an expression reverses the order of everything and conjugates the scalars — "bra–ket mirror symmetry".

3. Orthonormal Bases, Components and Completeness

A set {∣n⟩}\{|n\rangle\} is orthonormal if ⟨m∣n⟩=δmn\langle m|n\rangle = \delta_{mn}, and a basis if every ∣ψ⟩|\psi\rangle can be written ∣ψ⟩=∑ncn∣n⟩|\psi\rangle = \sum_n c_n|n\rangle. Then

cn=⟨n∣ψ⟩,∣ψ⟩=∑n∣n⟩⟨n∣ψ⟩,∑n∣n⟩⟨n∣=1^c_n = \langle n|\psi\rangle, \qquad |\psi\rangle = \sum_n |n\rangle\langle n|\psi\rangle, \qquad \boxed{\sum_n |n\rangle\langle n| = \hat{\mathbb{1}}}

The boxed statement is the completeness (closure) relation: the projectors onto the basis vectors sum to the identity. Inserting 1^\hat{\mathbb{1}} in this form is the single most used move in the formalism. Consequences:

  • Norm: ⟨ψ∣ψ⟩=∑n∣cn∣2\langle\psi|\psi\rangle = \sum_n |c_n|^2 (Parseval). Normalisation is ∑n∣cn∣2=1\sum_n |c_n|^2 = 1.
  • Inner product: ⟨ϕ∣ψ⟩=∑n⟨ϕ∣n⟩⟨n∣ψ⟩=∑ndn∗cn\langle\phi|\psi\rangle = \sum_n \langle\phi|n\rangle\langle n|\psi\rangle = \sum_n d_n^* c_n, where dn=⟨n∣ϕ⟩d_n=\langle n|\phi\rangle and cn=⟨n∣ψ⟩c_n=\langle n|\psi\rangle.
  • Matrix elements: an operator is the array Amn=⟨m∣A^∣n⟩A_{mn} = \langle m|\hat A|n\rangle, and A^∣ψ⟩\hat A|\psi\rangle has components ∑nAmncn\sum_n A_{mn}c_n.

For a continuous basis (position, momentum) the sum becomes an integral: ∫∣x⟩⟨x∣ dx=1^\int |x\rangle\langle x|\,dx = \hat{\mathbb{1}}, with ⟨x∣x′⟩=δ(x−x′)\langle x|x'\rangle = \delta(x-x') and ⟨x∣ψ⟩=ψ(x)\langle x|\psi\rangle = \psi(x). The wave function is just the set of components of ∣ψ⟩|\psi\rangle in the position basis; ⟨p∣ψ⟩=ψ~(p)\langle p|\psi\rangle = \tilde\psi(p) is the same state in the momentum basis.

4. Hermitian Operators and the Spectral Theorem

An operator is Hermitian (self-adjoint) if A^†=A^\hat A^\dagger = \hat A, i.e. ⟨ϕ∣A^ψ⟩=⟨A^ϕ∣ψ⟩\langle\phi|\hat A\psi\rangle = \langle\hat A\phi|\psi\rangle for all ∣ϕ⟩,∣ψ⟩|\phi\rangle,|\psi\rangle. The three facts that make Hermitian operators the representation of observables:

  1. Real eigenvalues. If A^∣a⟩=a∣a⟩\hat A|a\rangle = a|a\rangle then a⟨a∣a⟩=⟨a∣A^∣a⟩=⟨A^a∣a⟩=a∗⟨a∣a⟩a\langle a|a\rangle = \langle a|\hat A|a\rangle = \langle\hat A a|a\rangle = a^*\langle a|a\rangle, so a=a∗a = a^*.
  2. Orthogonal eigenvectors for distinct eigenvalues. From ⟨a′∣A^∣a⟩\langle a'|\hat A|a\rangle evaluated two ways, (a−a′)⟨a′∣a⟩=0(a - a')\langle a'|a\rangle = 0, so a≠a′⇒⟨a′∣a⟩=0a \ne a' \Rightarrow \langle a'|a\rangle = 0. Degenerate eigenvectors span a subspace and can be orthonormalised (Gram–Schmidt).
  3. Completeness. The eigenvectors form a basis of H\mathcal{H} (spectral theorem). Hence
A^=∑nan ∣an⟩⟨an∣,∑n∣an⟩⟨an∣=1^\hat A = \sum_n a_n\,|a_n\rangle\langle a_n|, \qquad \sum_n |a_n\rangle\langle a_n| = \hat{\mathbb{1}}

This spectral decomposition says a Hermitian operator is diagonal in its own eigenbasis, with the (real) eigenvalues on the diagonal. Functions of the operator are then trivial: f(A^)=∑nf(an)∣an⟩⟨an∣f(\hat A) = \sum_n f(a_n)|a_n\rangle\langle a_n| — this is how e−iH^t/ℏe^{-i\hat H t/\hbar} is defined.

5. The Generalised Statistical Interpretation

Chapter 2's Born rule, restated for any observable:

Postulate. A measurement of the observable A^\hat A on a system in state ∣ψ⟩|\psi\rangle yields one of the eigenvalues ana_n of A^\hat A. The probability of ana_n is

P(an)=∣⟨an∣ψ⟩∣2(non-degenerate),P(an)=⟨ψ∣P^n∣ψ⟩=∥P^n∣ψ⟩∥2(general)P(a_n) = |\langle a_n|\psi\rangle|^2 \quad (\text{non-degenerate}), \qquad P(a_n) = \langle\psi|\hat P_n|\psi\rangle = \|\hat P_n|\psi\rangle\|^2 \quad (\text{general})

where P^n=∑i∣an(i)⟩⟨an(i)∣\hat P_n = \sum_{i}|a_n^{(i)}\rangle\langle a_n^{(i)}| projects onto the eigenspace of ana_n. Immediately after the measurement the state is the normalised projection P^n∣ψ⟩/∥P^n∣ψ⟩∥\hat P_n|\psi\rangle / \|\hat P_n|\psi\rangle\| (collapse). The expectation value is

⟨A⟩=⟨ψ∣A^∣ψ⟩=∑nan P(an)\langle A\rangle = \langle\psi|\hat A|\psi\rangle = \sum_n a_n\,P(a_n)

and the uncertainty is ΔA=⟨A2⟩−⟨A⟩2=⟨ψ∣(A^−⟨A⟩)2∣ψ⟩\Delta A = \sqrt{\langle A^2\rangle - \langle A\rangle^2} = \sqrt{\langle\psi|(\hat A - \langle A\rangle)^2|\psi\rangle}. A state has ΔA=0\Delta A = 0 iff it is an eigenstate of A^\hat A.

6. Commuting Observables and Compatible Measurements

Two observables are compatible if [A^,B^]≡A^B^−B^A^=0[\hat A,\hat B] \equiv \hat A\hat B - \hat B\hat A = 0.

Theorem. [A^,B^]=0[\hat A,\hat B] = 0 iff A^\hat A and B^\hat B have a common orthonormal eigenbasis.

If they commute, a state can be labelled by both eigenvalues simultaneously, ∣a,b⟩|a,b\rangle, and measuring one does not disturb the value of the other. If they do not commute, no such common basis exists in general: measuring A^\hat A (collapsing to an A^\hat A-eigenstate) randomises the outcome of a subsequent B^\hat B measurement. A complete set of commuting observables (CSCO) is a maximal set of mutually commuting Hermitian operators; their joint eigenvalues label a unique basis — for hydrogen, {H^,L^2,L^z}\{\hat H, \hat L^2, \hat L_z\} with labels (n,ℓ,m)(n,\ell,m).

Worked example — a two-level system. Let A^=(100−1)\hat A = \begin{pmatrix} 1 & 0 \\ 0 & -1\end{pmatrix}, B^=(0110)\hat B = \begin{pmatrix} 0 & 1 \\ 1 & 0\end{pmatrix} (Pauli σz,σx\sigma_z, \sigma_x). Then [A^,B^]=2(01−10)≠0[\hat A,\hat B] = 2\begin{pmatrix} 0 & 1 \\ -1 & 0\end{pmatrix} \ne 0: they are incompatible. A system with A=+1A = +1 (state (10)\binom{1}{0}) has, for B^\hat B, ⟨B⟩=0\langle B\rangle = 0 and equal probabilities 12\tfrac12 for b=±1b = \pm1 — maximal uncertainty in BB when AA is sharp.

7. Continuous Spectra and Rigged Hilbert Space

The operators x^\hat x and p^\hat p have continuous spectra and their "eigenvectors" ∣x⟩|x\rangle, ∣p⟩|p\rangle are not in H\mathcal{H} (they are not normalisable: ⟨x∣x′⟩=δ(x−x′)\langle x|x'\rangle = \delta(x-x')). The clean way to handle this is the rigged Hilbert space (Gel'fand triple) Φ⊂H⊂Φ×\Phi \subset \mathcal{H} \subset \Phi^\times, where Φ\Phi is a space of well-behaved test functions and Φ×\Phi^\times its dual contains the generalised eigenkets. Operationally: sums become integrals, Kronecker deltas become Dirac deltas, ∣cn∣2|c_n|^2 becomes a probability density, and every physical state is still a normalisable superposition,

∣ψ⟩=∫ψ(x) ∣x⟩ dx=∫ψ~(p) ∣p⟩ dp|\psi\rangle = \int \psi(x)\,|x\rangle\,dx = \int \tilde\psi(p)\,|p\rangle\,dp

with ⟨x∣p⟩=12πℏ eipx/ℏ\langle x|p\rangle = \dfrac{1}{\sqrt{2\pi\hbar}}\,e^{ipx/\hbar} the transformation function between the two bases (the Fourier kernel).

8. Unitary Operators and Change of Basis

A unitary operator satisfies U^†U^=U^U^†=1^\hat U^\dagger\hat U = \hat U\hat U^\dagger = \hat{\mathbb{1}}; equivalently it preserves all inner products, ⟨U^ϕ∣U^ψ⟩=⟨ϕ∣ψ⟩\langle\hat U\phi|\hat U\psi\rangle = \langle\phi|\psi\rangle. Unitaries implement:

  • Change of basis. If {∣n⟩}\{|n\rangle\} and {∣n′⟩}\{|n'\rangle\} are two orthonormal bases, U^=∑n∣n′⟩⟨n∣\hat U = \sum_n |n'\rangle\langle n| maps one to the other; components transform as cm′=∑nUmncnc'_m = \sum_n U_{mn}c_n with Umn=⟨m′∣n⟩U_{mn} = \langle m'|n\rangle.
  • Time evolution. U^(t)=e−iH^t/ℏ\hat U(t) = e^{-i\hat H t/\hbar} is unitary because H^\hat H is Hermitian; this is why the norm (total probability) is conserved.
  • Symmetries. Translations e−ip^a/ℏe^{-i\hat p a/\hbar}, rotations e−iL^⋅n^ θ/ℏe^{-i\hat L\cdot\hat n\,\theta/\hbar} — every continuous symmetry is a unitary generated by a Hermitian observable (Noether, quantum version).

Physical predictions are invariant under a unitary applied to both states and operators (∣ψ⟩→U^∣ψ⟩|\psi\rangle \to \hat U|\psi\rangle, A^→U^A^U^†\hat A \to \hat U\hat A\hat U^\dagger): eigenvalues, inner products and probabilities are unchanged. Position and momentum representations are related by exactly such a unitary (the Fourier transform).

9. Common Pitfalls

  • Confusing Hermitian with unitary. Hermitian: A^†=A^\hat A^\dagger = \hat A, real eigenvalues, represents observables. Unitary: U^†=U^−1\hat U^\dagger = \hat U^{-1}, unit-modulus eigenvalues, represents symmetries and evolution.
  • Forgetting the inner product is antilinear in the bra. ⟨cϕ∣ψ⟩=c∗⟨ϕ∣ψ⟩\langle c\phi|\psi\rangle = c^*\langle\phi|\psi\rangle, not c⟨ϕ∣ψ⟩c\langle\phi|\psi\rangle.
  • Treating ∣x⟩|x\rangle as a genuine state. It is a generalised eigenket, not normalisable; physical states are integrals ∫ψ(x)∣x⟩ dx\int\psi(x)|x\rangle\,dx.
  • Assuming any two observables can be measured together. Only if [A^,B^]=0[\hat A,\hat B] = 0. Otherwise sharpening one blurs the other.
  • Dropping the projector for degenerate eigenvalues. P(an)=∥P^n∣ψ⟩∥2P(a_n) = \|\hat P_n|\psi\rangle\|^2 sums ∣⟨an(i)∣ψ⟩∣2|\langle a_n^{(i)}|\psi\rangle|^2 over the whole degenerate subspace.
  • Thinking the wave function is "the state". ψ(x)=⟨x∣ψ⟩\psi(x) = \langle x|\psi\rangle is one set of components; ψ~(p)=⟨p∣ψ⟩\tilde\psi(p) = \langle p|\psi\rangle is another. The ket ∣ψ⟩|\psi\rangle is basis-independent.
  • Writing (A^B^)†=A^†B^†(\hat A\hat B)^\dagger = \hat A^\dagger\hat B^\dagger. The order reverses: (A^B^)†=B^†A^†(\hat A\hat B)^\dagger = \hat B^\dagger\hat A^\dagger. So A^B^\hat A\hat B is Hermitian only if A^,B^\hat A,\hat B are Hermitian and commute.
  • Believing ⟨A⟩\langle A\rangle must be an eigenvalue. It is a probability-weighted average and often lies between eigenvalues (e.g. ⟨σz⟩=0\langle\sigma_z\rangle = 0 for an xx-polarised spin).

Summary

  • Hilbert space: complex inner-product space; states are normalised kets ∣ψ⟩|\psi\rangle (rays, up to global phase); ⟨ϕ∣ψ⟩=⟨ψ∣ϕ⟩∗\langle\phi|\psi\rangle = \langle\psi|\phi\rangle^*, antilinear in the bra.
  • Dirac notation: bras dual to kets; operators map kets to kets; adjoint via ⟨ϕ∣A^ψ⟩=⟨A^†ϕ∣ψ⟩\langle\phi|\hat A\psi\rangle = \langle\hat A^\dagger\phi|\psi\rangle; (A^B^)†=B^†A^†(\hat A\hat B)^\dagger = \hat B^\dagger\hat A^\dagger.
  • Bases: orthonormal ⟨m∣n⟩=δmn\langle m|n\rangle = \delta_{mn}; components cn=⟨n∣ψ⟩c_n = \langle n|\psi\rangle; completeness ∑n∣n⟩⟨n∣=1^\sum_n|n\rangle\langle n| = \hat{\mathbb{1}}; ⟨ψ∣ψ⟩=∑∣cn∣2\langle\psi|\psi\rangle = \sum|c_n|^2.
  • Hermitian operators: real eigenvalues, orthogonal eigenvectors, complete eigenbasis; spectral decomposition A^=∑nan∣an⟩⟨an∣\hat A = \sum_n a_n|a_n\rangle\langle a_n|; f(A^)=∑nf(an)∣an⟩⟨an∣f(\hat A) = \sum_n f(a_n)|a_n\rangle\langle a_n|.
  • Generalised Born rule: measuring A^\hat A gives ana_n with probability ∥P^n∣ψ⟩∥2\|\hat P_n|\psi\rangle\|^2, collapsing to P^n∣ψ⟩/∥⋅∥\hat P_n|\psi\rangle/\|\cdot\|; ⟨A⟩=⟨ψ∣A^∣ψ⟩\langle A\rangle = \langle\psi|\hat A|\psi\rangle; ΔA=0⇔\Delta A = 0 \Leftrightarrow eigenstate.
  • Compatibility: [A^,B^]=0⇔[\hat A,\hat B] = 0 \Leftrightarrow common eigenbasis ⇔\Leftrightarrow simultaneously measurable; a CSCO labels a unique basis (hydrogen: H^,L^2,L^z→n,ℓ,m\hat H,\hat L^2,\hat L_z \to n,\ell,m).
  • Continuous spectra: x^,p^\hat x,\hat p have non-normalisable generalised eigenkets (⟨x∣x′⟩=δ\langle x|x'\rangle = \delta); handled in a rigged Hilbert space; ⟨x∣p⟩=eipx/ℏ/2πℏ\langle x|p\rangle = e^{ipx/\hbar}/\sqrt{2\pi\hbar}.
  • Unitary operators: U^†U^=1^\hat U^\dagger\hat U = \hat{\mathbb{1}}; preserve inner products; implement change of basis, time evolution e−iH^t/ℏe^{-i\hat Ht/\hbar}, and symmetries.