Rigged Hilbert Space

The formalism chapter ends with a confession. Its most-used objects — the position ket $|x\rangle$ , the momentum ket $|p\rangle$ , the delta tying them together — are illegal in the very space the chapter built. The Hilbert space admits only square-integrable states, and Continuous Spectra and Dirac Delta Normalization computed the indictment: $\langle x|x\rangle = \delta(0)$ , an infinite "norm" no rescaling repairs. Yet the calculus built on these outlaws never produces a wrong number. Dirac knew it; von Neumann grumbled; two generations of physicists used it anyway. This article tells you why the calculus works. The fix is not to force $|x\rangle$ into the Hilbert space — that is impossible — but to build an annex on either side of it: a smaller space $\Phi$ of extra-smooth test states inside, and a larger space $\Phi^\times$ of generalized states outside, where $|x\rangle$ and $|p\rangle$ live as well-defined citizens. The three-floor structure $\Phi \subset \mathcal H \subset \Phi^\times$ is the rigged Hilbert space; its central theorem hands every observable a complete set of generalized eigenvectors, turning each line of Dirac's notation into a proved statement. You need it not for square-well exams but to know the formalism deserves the trust you have been extending it — and because decaying states with complex energies find their only rigorous home here.

Key takeaways

Position and momentum kets are not in the Hilbert space: $\langle x|x\rangle = \delta(0)$ diverges — the computations live in Continuous Spectra and Dirac Delta Normalization.
The cure is the Gelfand triple $\Phi \subset \mathcal H \subset \Phi^\times$ : test states (smooth, rapidly decaying) · the Hilbert space $L^2$ · continuous antilinear functionals.
$|x_0\rangle$ is the evaluation functional $\langle x_0|\psi\rangle = \psi(x_0)$ — continuous on $\Phi$ , unbounded (undefined, even) on $\mathcal H$ .
The duality is a seesaw: tightening the test space enlarges its dual — deltas and plane waves get in exactly this way.
The Gelfand–Maurin nuclear spectral theorem guarantees every self-adjoint observable a complete set of generalized eigenvectors — Dirac's $\int dx\,|x\rangle\langle x| = \hat 1$ becomes a theorem.
Resonances with complex energy $E_R - i\Gamma/2$ live in $\Phi^\times$ too (Gamow vectors): a Lorentzian energy line is exactly exponential decay with $\tau = \hbar/\Gamma$ .
Key idea: tighten the test space and you loosen its dual — the triple $\Phi \subset \mathcal H \subset \Phi^\times$ gives every improper ket a rigorous home and every Dirac manipulation a theorem.

The core insight

Tighten the test space and you loosen its dual: replacing the lone Hilbert space $\mathcal H$ by the triple $\Phi \subset \mathcal H \subset \Phi^\times$ makes non-normalizable eigenkets legitimate continuous antilinear functionals in $\Phi^\times$ , and the nuclear spectral theorem hands every self-adjoint observable a complete set of generalized eigenvectors there.

Where this lives today

Open any partial-wave analysis in particle physics and the objects being fitted — Breit–Wigner resonances with complex pole positions $E_R - i\Gamma/2$ — are Gamow vectors in all but name; the rigged framework lets theorists treat an unstable particle as a state with an exact complex energy rather than an approximate bump. The triple also underwrites daily practice: box normalization and wave packets are devices for staying on the lower floors, and the RHS certifies the shortcuts agree. At the frontier, time-asymmetric quantum mechanics builds irreversible decay into the Hardy rigging itself — the arrow of time for unstable states as a boundary condition, not a paradox.

The physical picture

Picture a building with three floors. The middle floor is the Hilbert space $\mathcal H = L^2$ — every normalizable state, the physical world of Hilbert Space and State Vectors. The ground floor $\Phi$ is an exclusive club inside it: only the smoothest, fastest-decaying states admitted. The top floor $\Phi^\times$ is the surprise — bigger than the middle one, and home to the idealized objects: the delta spike $|x_0\rangle$ , the plane wave $|p\rangle$ .

Figure

The Gelfand triple — three nested floors, and where each object lives

The engine underneath is a seesaw. A functional is admitted to the top floor if it can be continuously paired with every ground-floor resident. Make the ground floor stricter and the pairing test gets easier to pass, so more functionals get in — including violently singular ones. The delta is the cleanest example: as a would-be function it is nonsense; as the rule "hand me a test state, I return its value at zero" it is tame — provided the states it acts on are genuinely nice.

Figure

The duality seesaw — tightening the test space enlarges its dual

The middle floor is the pivot in a precise sense: the Hilbert space is self-dual — its continuous functionals are exactly its own vectors, the Riesz correspondence of Dirac Bra-Ket Notation. Off the pivot, bras and kets stop being mirror images: $\langle x_0|$ has no partner among honest states. The triple keeps that asymmetry's bookkeeping straight — every pairing of a top-floor functional with a ground-floor state is an ordinary finite complex number.

The triple, floor by floor

Start at the bottom. The test space $\Phi$ is, in the standard construction, the Schwartz space — smooth functions that fall off, with every derivative, faster than any power:

\Phi = \mathcal S(\mathbb R) = \Big\{ \varphi \;:\; \sup_x \big| x^m \, \partial_x^n \varphi(x) \big| < \infty \ \text{ for all } m, n \Big\} \qquad (1)

where: each pair $(m, n)$ defines a seminorm — a separate quality test — and membership demands passing all of them. Physically, $\Phi$ is what an apparatus could actually prepare: finite everything, all moments defined, Fourier transform equally nice. Crucially, $\hat x$ , $\hat p$ , and well-behaved Hamiltonians map $\Phi$ into itself — that stability is what qualifies a space to serve as $\Phi$ ; the charter of Admissibility Conditions on Wavefunctions was its wave-mechanics shadow. The middle floor is the familiar $\mathcal H = L^2(\mathbb R)$ of Hilbert Space and State Vectors, which already flagged who fails its entry test. The top floor is defined by pairing:

\Phi^\times = \Big\{ F \;:\; F \text{ maps each } \varphi \in \Phi \text{ to a number } \langle F|\varphi\rangle, \text{ antilinearly and continuously} \Big\} \qquad (2)

where: antilinearly matches the bra convention of Dirac Bra-Ket Notation (scalars conjugate), and continuously means convergence of test states in every seminorm of (1) forces convergence of the paired numbers. Every ordinary state joins $\Phi^\times$ automatically through its inner product, so the floors genuinely nest and nothing already legal is lost. One reading rule disarms most of the confusion here: a generalized ket is not a column of amplitudes; it is a machine that eats test states and returns numbers. Every $|x\rangle$ or $|p\rangle$ formula abbreviates what happens when a test state is fed in — with that translation running, the sifting integrals of Continuous Spectra and Dirac Delta Normalization read as shorthand for finite, legal arithmetic.

Derivation AThe delta ket is a legal citizen — and exactly why the Hilbert space rejects it

We give the position ket its rigorous definition: $|x_0\rangle$ is the evaluation functional, the machine that returns a test state's value at $x_0$ . We then prove the two halves of its citizenship file — continuous on the test space $\Phi$ , hence a member of $\Phi^\times$ ; unbounded on $\mathcal H$ , hence unhousable by any enlargement short of the rigging. The second half is the punchline: the failure is a structural fact about $L^2$ , exhibited by an explicit sequence of unit-norm states whose values at one point blow up.

Step 1. Define the action. For every test state $\varphi \in \Phi$ , set

\langle x_0 | \varphi \rangle \;\equiv\; \varphi(x_0)

Why this is the right definition: it is how the delta is actually used — $\int \delta(x - x_0)\,\varphi(x)\,dx = \varphi(x_0)$ — stated without any integral of a fictitious function; the sifting rule of Continuous Spectra and Dirac Delta Normalization is recovered, not assumed.

Step 2. Continuity on $\Phi$ : the value at a point is controlled by the crudest Schwartz seminorm, $|\varphi(x_0)| \le \sup_x |\varphi(x)|$ , so convergence of test states forces convergence of the evaluated numbers. $|x_0\rangle \in \Phi^\times$ ; citizenship granted.

Step 3. Now show $\mathcal H$ could never have granted it. Take unit-norm Gaussians squeezed toward $x_0 = 0$ :

\psi_a(x) = \Big(\frac{a}{\pi}\Big)^{1/4} e^{-a x^2 / 2}, \qquad \int |\psi_a|^2\,dx = 1 \ \text{ for every } a

Evaluation at the center gives $\psi_a(0) = (a/\pi)^{1/4}$ : at $a = 100$ the value is $2.38$ ; at $a = 10^6$ it is $23.8$ ; the norm stays pinned at 1 throughout. A bounded functional on $\mathcal H$ needs $|F[\psi]| \le C\|\psi\|$ for a fixed $C$ — no such $C$ exists. Why this kills the idea at the root: on $L^2$ "the value at a point" is not even well-defined (states are equivalence classes alterable at isolated points); point evaluation and finite $L^2$ norm are structurally incompatible.

Step 4. Read the moral off the seesaw: on the small, strict $\Phi$ the evaluation machine is tame; on the roomy $\mathcal H$ it is unbounded. The delta needed the test space to shrink before it could exist — the whole design of the triple. Verified: symbolically and numerically — the Gaussian family has norm exactly 1 for all $a$ (sympy), $\psi_a(0) = (a/\pi)^{1/4} \to \infty$ , the quoted pair $2.38 / 23.8$ recomputed, and the nascent-Gaussian sifting limit $\int \delta_\epsilon \varphi \to \varphi(0)$ confirmed for two test functions (verify script, 23 checks green).

Eigenvectors that do not exist — until the floor is changed

Here is the sharpest way to see what the rigging buys. The position operator's eigenvalue problem, taken literally in $\mathcal H$ , has no solutions at all: $x\,\psi(x) = x_0\,\psi(x)$ forces $\psi$ to vanish everywhere except one point — the zero vector of $L^2$ . Yet $x_0$ is undeniably part of position's spectrum. Quantum mechanics lives with observables whose spectra are full while their eigenvector cupboards, in $\mathcal H$ , are empty. The triple repairs this by transposing operators onto the top floor: if $\hat A$ maps test states to test states, it acts on functionals by handing its work to the argument,

\big(\hat A\,F\big)[\varphi] \;\equiv\; F\big[\hat A\,\varphi\big], \qquad F \in \Phi^\times, \ \varphi \in \Phi \qquad (3)

where: the left side defines the extended operator; the right side is legal arithmetic — $\hat A \varphi$ is again a test state, and $F$ knows how to eat those. (For self-adjoint $\hat A$ the transpose extends the original operator; the adjoint theory licensing the handoff is Hermitian Conjugate and Adjoint Operators.) With (3), the eigenvalue equation for the evaluation functional is a two-line computation:

\big(\hat x\,F_{x_0}\big)[\varphi] = F_{x_0}\big[x\,\varphi\big] = x_0\,\varphi(x_0) = x_0\,F_{x_0}[\varphi] \quad\Longrightarrow\quad \hat x\,|x_0\rangle = x_0\,|x_0\rangle \ \text{ in } \Phi^\times \qquad (4)

where: the middle equality is just "evaluate the function $x\,\varphi(x)$ at $x_0$ ." Dirac's 1930 eigenvalue equation is true — on the correct floor. The same construction houses the momentum kets: $|p\rangle$ pairs with a test state to give its Fourier component, finite for every $p$ because Schwartz functions have Schwartz transforms. The kernel $\langle x|p\rangle = e^{ipx/\hbar}/\sqrt{2\pi\hbar}$ and its delta-orthonormalization are owned by Momentum Representation and Continuous Spectra and Dirac Delta Normalization; this article adds the certificate that those formulas describe $\Phi^\times$ citizens.

The nuclear spectral theorem — completeness, guaranteed

Everything so far handled single kets. The load-bearing question is completeness: can every state be expanded over the generalized eigenvectors, the way Complete Orthonormal Bases expands over honest ones? The answer is the subject's central theorem, stated in physicist's language:

\textbf{Nuclear spectral theorem.}\quad \hat A \text{ self-adjoint, } \hat A: \Phi \to \Phi \text{ continuously} \ \Longrightarrow\ \hat A \text{ has a complete set of generalized eigenvectors in } \Phi^\times \qquad (5)

where: complete means every pairing between test states can be resolved over the eigenvectors — a sum over the discrete part of the spectrum plus an integral over the continuous part:

\langle \phi | \psi \rangle \;=\; \sum_n \langle \phi | E_n \rangle\langle E_n | \psi \rangle \;+\; \int_{\sigma_c} d\lambda\, \langle \phi | \lambda \rangle\langle \lambda | \psi \rangle \qquad (6)

where: the $|E_n\rangle$ are genuine normalizable eigenvectors (bound states, in $\mathcal H$ ) and the $|\lambda\rangle$ are top-floor citizens (scattering states, plane waves). Equation (6) is exactly the mixed-spectrum expansion Continuous Spectra and Dirac Delta Normalization used for the finite well — now guaranteed rather than assumed. ("Nuclear" is a technical property of $\Phi$ 's topology, which Schwartz space has; no calculation in this track needs the fine print.) The theorem is the master key. What it unlocks, line by line:

the Dirac move	its status in the triple
$\hat x\,\vert x\rangle = x\,\vert x\rangle$ , $\hat p\,\vert p\rangle = p\,\vert p\rangle$	eigenvalue equations in $\Phi^\times$ , via the transpose (3)
$\langle x\vert\psi\rangle = \psi(x)$	the evaluation-functional pairing — Derivation A
$\langle x\vert x'\rangle = \delta(x - x')$	kernel identity, meaningful under test-state integrals
$\int dx\,\vert x\rangle\langle x\vert = \hat 1$	matrix-element identity on $\Phi$ , from completeness (6)
$\vert\psi\rangle = \sum_n c_n \vert n\rangle + \int d\lambda\, c(\lambda)\vert\lambda\rangle$	the nuclear spectral theorem's expansion
$\hat H\,\vert z_R\rangle = (E_R - i\Gamma/2)\vert z_R\rangle$	Gamow vector in a Hardy-rigged triple — below

Every row's computational content was delivered earlier in the chapter; the triple contributes the right column — the license. Rig the space once and every calculation already done becomes retroactively rigorous.

Limiting cases

case	what the triple says	meaning
purely discrete spectrum (oscillator)	all eigenvectors in $\mathcal H$	ordinary spectral theory suffices — the rigging is idle
purely continuous ( $\hat x$ , $\hat p$ , free $\hat H$ )	no eigenvectors in $\mathcal H$ ; all in $\Phi^\times$	the rigging is indispensable; expansion is a pure integral
mixed spectrum (finite well, hydrogen)	bound states in $\mathcal H$ + continuum in $\Phi^\times$	one expansion (6) holds both — the generic atomic case
resonances $E_R - i\Gamma/2$	complex generalized eigenvalues, Hardy rigging	Gamow vectors: decaying states as genuine eigenvectors
restrict attention to $\Phi$ alone	every formula finite, no distributions	why wave-packet-only formulations never meet the subtlety

Common trapA delta-normalized ket is just a very long vector, so a big enough Hilbert space would hold it

No enlargement holds it — the failure is structural. Derivation A's squeezed Gaussians have norm exactly 1 while their central values run off to infinity: point evaluation is incompatible with the $L^2$ pairing itself. The repair cannot be "allow longer vectors"; it must change what kind of object the ket is — from a vector paired by integration to a functional paired by evaluation. Treat $\langle x|x\rangle = \delta(0)$ as a reminder of that category change, never as a large number. When an exam asks "is $|p\rangle$ a physical state?", cite both floors: not in $\mathcal H$ (infinite norm, not preparable — as Free Particle and Plane Waves first showed), and legitimately in $\Phi^\times$ (an expansion tool with delta normalization).

The resonance preview — complex energies find a home

One more door, opened only by the rigging. Conservation of Normalization engineered decay by hand: an optical potential $V - i\Gamma/2$ drains the norm as $e^{-\Gamma t/\hbar}$ , lifetime $\tau = \hbar/\Gamma$ ( $\Gamma = 6.0$ MeV gave its neutron $\tau = 1.1\times 10^{-22}\,\mathrm{s}$ ). That was a model. The deeper statement: a self-adjoint Hamiltonian can possess generalized eigenvectors with complex eigenvalues $E_R - i\Gamma/2$ — Gamow vectors — in a triple rigged with Hardy-class analyticity. Decay becomes an exact eigenproperty on the top floor; the derivation below shows its seed: a resonance's Lorentzian line is an exponential survival law.

Go deeper · Derivation B · 2 minA Lorentzian energy line decays exponentially — the Gamow correspondence

Derivation BA Lorentzian energy line decays exponentially — the Gamow correspondence

We compute the survival amplitude of a state whose energy content is a Lorentzian (Breit–Wigner) line of width $\Gamma$ centered at $E_R$ — the universal resonance shape. The strategy: write the state as a superposition of energy components, let each rotate its phase clock $e^{-iEt/\hbar}$ , and evaluate the overlap with the initial state as one Fourier integral. The result — pure exponential decay with $\tau = \hbar/\Gamma$ — is exactly what a complex "eigenvalue" $E_R - i\Gamma/2$ would produce, which is why the Gamow vector deserves its complex energy.

Step 1. Set up the survival amplitude: for a state with energy density $\rho(E)$ , the amplitude to still be found in it at time $t$ is the phase-weighted average

a(t) = \langle \psi | \psi(t) \rangle = \int dE\, \rho(E)\, e^{-iEt/\hbar}, \qquad \rho(E) = \frac{\Gamma/2\pi}{(E - E_R)^2 + \Gamma^2/4}

Why the Lorentzian: the line shape of every isolated resonance — the Breit–Wigner profile met at Resonant Transmission — normalized to unit probability (checked symbolically).

Step 2. Shift to the line center: $u = E - E_R$ pulls out an overall phase and leaves an even integral,

a(t) = e^{-iE_R t/\hbar} \cdot \frac{\Gamma}{2\pi} \int_{-\infty}^{\infty} \frac{e^{-iut/\hbar}}{u^2 + \Gamma^2/4}\, du

Why the sine part dies: the imaginary piece is odd in $u$ against an even denominator — verified to vanish exactly.

Step 3. Evaluate the cosine kernel. The standard integral $\int \cos(us)/(u^2 + b^2)\,du = (\pi/b)\,e^{-b s}$ with $b = \Gamma/2$ and $s = t/\hbar$ gives

a(t) = e^{-iE_R t/\hbar}\, e^{-\Gamma t/2\hbar} \qquad\Longrightarrow\qquad |a(t)|^2 = e^{-\Gamma t/\hbar}, \qquad \tau = \frac{\hbar}{\Gamma}

Why this is the Gamow signature: the amplitude behaves exactly as if the state had the single complex energy $E_R - i\Gamma/2$ — phase clock at $E_R$ , magnitude bleeding at $\Gamma/2\hbar$ . The lifetime law matches Conservation of Normalization's optical-potential result, but here nothing non-Hermitian was put in by hand — it emerged from line shape alone.

Step 4. Read the fine print. A Lorentzian over all energies is an idealization — real spectra are bounded below, producing non-exponential corrections at very short and long times (the story of Survival Amplitude and Decay of Unstable States). The rigged-space content: in a Hardy-rigged triple the Gamow vector with $\hat H|z_R\rangle = (E_R - i\Gamma/2)|z_R\rangle$ exists as a genuine $\Phi^\times$ citizen whose evolution is exactly the exponential above. Verified: symbolically and numerically — Lorentzian normalization exact; the cosine kernel integral equals $(\pi/b)e^{-bs}$ (sympy); the odd part vanishes; the assembled prefactor collapses to 1; and a numeric quadrature at $\Gamma = 1$ , $t/\hbar = 0.7$ reproduces $|a| = e^{-0.35}$ to $10^{-8}$ (verify script).

Worked examples

Example 1 — pairing numbers: what the improper kets actually return (confidence builder). A particle is prepared in the Gaussian test state $\psi(x) = (2\pi\sigma^2)^{-1/4} e^{-x^2/4\sigma^2}$ with $\sigma = 0.5$ nm. Compute $\langle x_0|\psi\rangle$ at $x_0 = 0.5$ nm and the momentum pairing at $p_0 = \hbar/\sigma$ .

Step 1. Position pairing = evaluation (Derivation A). $\langle x_0|\psi\rangle = \psi(0.5\,\mathrm{nm}) = (2\pi \times 0.25\,\mathrm{nm}^2)^{-1/4} e^{-0.25} = \boxed{0.696\ \mathrm{nm}^{-1/2}}$ — finite, with units of an amplitude density whose square integrates (not sums) to probability, per the continuous Born rule of Continuous Spectra and Dirac Delta Normalization.

Step 2. Momentum pairing = Fourier component: $\langle p|\psi\rangle = (2\sigma^2/\pi\hbar^2)^{1/4} e^{-\sigma^2 p^2/\hbar^2}$ (the transform pair of Momentum-Space Wavefunction, re-verified symbolically for this state). At $p_0 = \hbar/\sigma$ the ratio to the peak is $e^{-1} = \boxed{0.368}$ — finite for every single $p$ .

Step 3. Contrast with the illegal question: $\langle x_0 | x_0 \rangle$ has no numerical answer — both slots are top-floor citizens. A functional eats test states, not other functionals. $\int |\langle p|\psi\rangle|^2 dp = 1$ (machine-checked) is the legal pattern: improper kets enter answers inside integrals, paired against the physical state.

Example 2 — a resonance's survival, start to finish (exam level). A nuclear resonance has $E_R = 3.2$ MeV and width $\Gamma = 0.5$ MeV. Find its lifetime, survival probability at $t = 1.0\times 10^{-21}$ s, half-life, and the phase revolutions per lifetime.

Step 1. Lifetime: $\tau = \hbar/\Gamma = \dfrac{1.055\times10^{-34}}{0.5\times10^{6} \times 1.602\times10^{-19}} = \boxed{1.32\times10^{-21}\ \mathrm{s}}$ — an order of magnitude longer-lived than the $\Gamma = 6$ MeV optical-model neutron of Conservation of Normalization: narrow lines linger.

Step 2. Survival: $P(t) = e^{-t/\tau} = e^{-1.0/1.32} = \boxed{0.468}$ . Half-life: $t_{1/2} = \tau \ln 2 = \boxed{9.12\times10^{-22}\ \mathrm{s}}$ .

Step 3. Phase revolutions per lifetime: the clock turns at $E_R/\hbar$ , so $n = E_R \tau/(2\pi\hbar) = E_R/(2\pi\Gamma) = \boxed{1.02}$ — the real part winds the phase barely once before the imaginary part damps the amplitude by $e^{-1/2}$ . Broad resonances barely oscillate before dying; sharp lines ( $\Gamma \ll E_R$ ) ring thousands of times — the quality-factor reading of $E_R/\Gamma$ .

The man who built the third floor

Origins · 1 minIsrael Gelfand and how this was discovered

Israel Gelfand

1913–2009 · Moscow State University

Gelfand never finished secondary school — expelled from his Ukrainian technical school as a "socially alien element," he taught himself mathematics while working as a doorman in Moscow, and was admitted straight to graduate study at nineteen with no degree. The structures here carry his name twice: his normed-ring theory (1941) rebuilt spectral theory around the idea that operators are known by the functionals that probe them, and the multi-volume Generalized Functions (1958–66, with Vilenkin and others) constructed the triple $\Phi \subset \mathcal H \subset \Phi^\times$ and proved the nuclear spectral theorem that makes Dirac's calculus a set of theorems. (Distributions themselves were Laurent Schwartz's creation — his story is told at Continuous Spectra and Dirac Delta Normalization.) Gelfand's Monday-night Moscow seminar, run for nearly fifty years, was an experiment on mathematicians: he interrupted speakers mid-sentence, demanded the "kindergarten version" of every theorem, and kept the room past midnight. He also ran a correspondence school mailing problem sets to thousands of students in provincial towns. Physicists adopted his triple in the 1960s; it remains the standard rigorous frame for continuous spectra and resonances.

History

1927

— Dirac's transformation theory works δ-function magic with no rigorous foundation

1932

— von Neumann's Mathematische Grundlagen axiomatizes QM inside the bare Hilbert space, deliberately banishing improper kets

1945

— Schwartz builds distribution theory: the delta becomes a functional on test functions

1958

— Gelfand and collaborators begin Generalized Functions; the triple Φ ⊂ ℋ ⊂ Φ× appears with the nuclear spectral theorem

1966

— Roberts, then Bohm and Antoine, install the rigged Hilbert space as quantum mechanics' working frame for Dirac notation

1989

— Bohm and Gadella construct Gamow vectors: resonances as generalized eigenvectors with complex energy

2005

— de la Madrid's pedagogical account makes the RHS standard graduate-course material

Going further

References. Bohm & Gadella, Dirac Kets, Gamow Vectors and Gel'fand Triplets (Springer LNP 348, 1989) · Gelfand & Vilenkin, Generalized Functions Vol. 4 (1964) · Roberts, J. Math. Phys. 7, 1097 (1966) · de la Madrid, Eur. J. Phys. 26, 287 (2005) · Reed & Simon, Methods of Modern Mathematical Physics Vol. I (1980) · Ballentine, Quantum Mechanics: A Modern Development (2014), Ch. 1.

Provenance. All equations are standard textbook results. A verifier should recompute: (1) the Fourier delta identity $\int_{-\infty}^{\infty}e^{i(p'-p)x/\hbar}\mathrm{d}x=2\pi\hbar\,\delta(p-p')$ fixing $N=1/\sqrt{2\pi\hbar}$ in $\langle x|p\rangle$ (Sakurai Modern QM, Shankar Ch. 1); (2) the multiplication-operator spectrum result that $\hat{x}$ has empty point spectrum and continuous spectrum $\mathbb{R}$ (Reed & Simon Vol. I); (3) the Gelfand triple definition and nuclear spectral theorem (Gelfand & Vilenkin Vol. 4; Bohm & Gadella, Springer LNP 348); (4) the continuous-spectrum Born rule and resolution of identity (Ballentine Ch. 1; de la Madrid, Eur. J. Phys. 26, 287, 2005). Conventions follow Sakurai/Cohen-Tannoudji for $\langle x|p\rangle$ with $\hbar$ explicit. For this edition all quantitative claims were machine-executed (sympy/numpy/scipy, 23 checks green): nascent-delta limits; the unit-norm squeezed-Gaussian values $2.38/23.8$ ; the Gaussian momentum pairing and normalization; the grid-spike divergence; the Lorentzian survival pair; and both worked examples. Per the chapter ownership map: the delta toolbox, delta-orthonormality, box bridge, and norm-divergence computations are owned by Continuous Spectra and Dirac Delta Normalization (this article is its declared rigor sequel); the $\langle x|p\rangle$ kernel derivation by Momentum Representation; the adjoint/transpose machinery by Hermitian Conjugate and Adjoint Operators; the optical-potential lifetime and its 6-MeV neutron number by Conservation of Normalization (cited, not re-derived); decay theory beyond the Lorentzian idealization is deferred to Survival Amplitude and Decay of Unstable States.

Builds on: Hilbert Space and State Vectors · Dirac Bra-Ket Notation · Inner Product and Norm of States · Hermitian Operators and Observables · Eigenvalues and Eigenstates

Leads to: Completeness Relation and Resolution of Identity · Born Rule and Measurement Probabilities · The Postulates of Quantum Mechanics

Do I actually need rigged Hilbert spaces to do quantum mechanics?

For computation: no. The delta calculus produces correct numbers without naming the triple, and every continuum manipulation can be dodged with box normalization or wave packets. The RHS supplies the guarantee — the theorem that the shortcuts are theorems — plus one new capability: resonances as exact eigenvectors with complex energy. Like the real numbers: you calculated with decimals for years before meeting Dedekind cuts, but it matters that someone can build them.

Is the ket |x⟩ a state the system can be in?

No. A state is a unit ray in $\mathcal H$ (the postulates article), and $|x\rangle$ has no norm at all — it is a $\Phi^\times$ functional, an expansion tool. Preparing it would mean localizing a particle to a mathematical point, at infinite energy cost (the kink-cost argument of Admissibility Conditions on Wavefunctions taken to its limit). What you can prepare are test states arbitrarily concentrated near $x_0$ — Derivation A's squeezed Gaussians — and $|x_0\rangle$ is the idealized limit organizing their bookkeeping.

What exactly does "nuclear" mean in the nuclear spectral theorem?

A technical grade of the test space's topology that Schwartz space happens to possess — the hypothesis that makes the completeness proof go through. Practitioner's version: "Schwartz space qualifies, so position, momentum, and reasonable Hamiltonians all get complete generalized eigenbases." No connection to nuclei; it is Grothendieck's functional-analysis terminology.

How can a self-adjoint Hamiltonian have an eigenvector with complex eigenvalue? I thought self-adjointness forced real eigenvalues.

The reality theorems (Hermitian Operators and Observables) apply to eigenvectors in the Hilbert space, where the eigenvector's inner product with itself is available for the proof. A Gamow vector lives in $\Phi^\times$ , where that step has no meaning — the theorem simply does not reach it. The honest spectrum of $\hat H$ stays real; the complex value $E_R - i\Gamma/2$ is a generalized eigenvalue whose imaginary part encodes the decay rate, visible in the honest spectrum only as a Lorentzian bump of width $\Gamma$ .

Does the rigging change any prediction of quantum mechanics?

Standard predictions: no — the delta calculus is untouched; the RHS certifies it. It earns more than certification in two places: resonance physics, where Gamow vectors give exact meaning to complex pole parameters that would otherwise be fitting conveniences, and time-asymmetric formulations, where the Hardy rigging builds the decay arrow into the framework. Both matter at the research frontier; neither shifts a UG exam answer.

Where does von Neumann's spectral theorem fit — was he wrong to banish improper kets?

Not wrong, just austere. His 1932 formulation replaces eigenvector talk with projection-valued measures — projectors $\hat P_\Delta$ onto spectral windows — and every Dirac formula has a projector translation. Fully rigorous, no third floor needed, but clumsy for calculation, which is why physicists never adopted it as a working language. Gelfand's triple is the reconciliation: Dirac's notation, von Neumann's standards.