Rigged Hilbert space

A rigged Hilbert space, also called a Gelfand triple, is a mathematical structure in which a Hilbert space is supplemented by a dense subspace carrying a finer topology and by a corresponding space of continuous linear functionals. It is conventionally written as

[ \Phi \subset \mathcal H \subset \Phi^\times , ]

where (\Phi) is continuously and densely embedded in (\mathcal H), while (\Phi^\times) denotes the continuous antilinear dual of (\Phi). The second inclusion is obtained by identifying each element of (\mathcal H) with a functional on (\Phi) through the Hilbert-space inner product.

The construction extends the spectral methods of ordinary Hilbert-space theory to generalized vectors that need not be square-integrable. These generalized vectors include the formal position states, momentum states, and continuum-energy states used in quantum mechanics. Within the triple, such objects are represented as distributions in (\Phi^\times), while ordinary physical states remain elements of (\mathcal H).

The adjective “rigged” refers to the additional topological structure placed around the Hilbert space. It does not imply that the inner product, spectrum, or associated probability measure has been improperly altered.

Mathematical structure

Let (\mathcal H) be a complex Hilbert space and let (\Phi) be a vector subspace that is dense in the norm topology of (\mathcal H). The space (\Phi) is assigned a locally convex topology (\tau_\Phi) that is finer than the topology inherited from (\mathcal H). Consequently, convergence in (\Phi) implies convergence in (\mathcal H), although the converse generally fails.

The continuous inclusion

[ i:\Phi\longrightarrow\mathcal H ]

induces a dual map from (\mathcal H^\times) into (\Phi^\times). The Riesz representation theorem identifies (\mathcal H^\times) with (\mathcal H), producing the canonical chain

[ \Phi \xhookrightarrow{i} \mathcal H \xhookrightarrow{i^\times} \Phi^\times . ]

Under the convention that the Hilbert-space inner product is linear in its second argument, an element (f\in\mathcal H) determines the antilinear functional

[ F_f(\phi)=\langle \phi,f\rangle_{\mathcal H}, \qquad \phi\in\Phi. ]

This identification distinguishes the embedding of (\mathcal H) into (\Phi^\times) from an arbitrary inclusion of sets. The notation suppresses that distinction because the embedding is canonical after the inner-product convention has been fixed.

The topology of (\Phi) determines which generalized vectors belong to (\Phi^\times). A finer topology on (\Phi) produces a larger continuous dual, since continuity becomes easier for linear functionals when the domain carries more open sets. The choice of test space is therefore part of the mathematical specification rather than a removable technical detail.

A frequently used realization is

[ \mathcal S(\mathbb R^n) \subset L^2(\mathbb R^n) \subset \mathcal S'(\mathbb R^n), ]

where (\mathcal S(\mathbb R^n)) is the Schwartz space of rapidly decreasing smooth functions and (\mathcal S'(\mathbb R^n)) is the space of tempered distributions. In this realization, plane waves and Dirac delta distributions belong to the dual space even though they do not belong to (L^2(\mathbb R^n)).

Operators and dual extensions

Suppose that a linear operator (A) maps (\Phi) continuously into itself. Its dual extension (A^\times) acts on (\Phi^\times) through

[ \langle A\phi,F\rangle

\langle \phi,A^\times F\rangle, \qquad \phi\in\Phi,\quad F\in\Phi^\times . ]

The brackets in this expression denote the dual pairing rather than necessarily the inner product of (\mathcal H). If (A) is the restriction to (\Phi) of a densely defined operator on (\mathcal H), then (A^\times) extends the action of the Hilbert-space adjoint in the distributional sense.

A generalized eigenvector associated with a scalar (\lambda) is a nonzero functional (F_\lambda\in\Phi^\times) satisfying

[ A^\times F_\lambda=\lambda F_\lambda. ]

Equivalently,

[ \langle A\phi,F_\lambda\rangle

\lambda\langle\phi,F_\lambda\rangle ]

for every (\phi\in\Phi). Unlike an eigenvector in (\mathcal H), (F_\lambda) is not required to have finite Hilbert norm. This distinction permits continuous spectral values to be represented by eigenfunctionals without converting them into discrete eigenvalues.

For the momentum operator on the real line,

[ P=-i\hbar\frac{d}{dx}, ]

the functions (e^{ipx/\hbar}) fail to belong to (L^2(\mathbb R)), but they define tempered distributions. They are generalized eigenvectors of (P^\times) with eigenvalue (p). The corresponding distributional orthogonality relation,

[ \langle p\mid p'\rangle=\delta(p-p'), ]

is an identity involving the Dirac delta distribution, not an ordinary Hilbert-space inner product between normalized vectors.

Nuclear triples and spectral decomposition

Many formulations require (\Phi) to be a nuclear space. Nuclearity constrains the locally convex topology strongly enough to support integral representations resembling the formal eigenvector expansions used in physics. A triple with this property is commonly called a nuclear rigged Hilbert space.

For a self-adjoint operator (A) that leaves an appropriate nuclear space (\Phi) invariant, the Gelfand–Maurin theorem provides generalized eigenvectors in (\Phi^\times) for almost every point of the spectrum relative to a suitable spectral measure. Vectors in (\Phi) then admit a weak expansion of the form

[ \phi

\int_{\sigma(A)} \langle \lambda\mid\phi\rangle ,|\lambda\rangle,d\mu(\lambda), ]

where (\sigma(A)) is the spectrum of (A). The expression is interpreted through dual pairings and integration against the measure (\mu); it is not generally a norm-convergent sum of vectors belonging to (\mathcal H).

The theorem supplies a mathematical setting for the continuous part of the spectral theorem. The projection-valued measure of the Hilbert-space formulation remains fundamental, while the rigged formulation expresses the same spectral information through generalized eigenfunctionals. Discrete eigenvectors may lie in (\mathcal H), whereas continuum eigenfunctionals ordinarily lie only in (\Phi^\times).

Israel Gelfand and Georgy Shilov developed the relevant theory of generalized functions and nuclear spaces, while K. Maurin established the eigenfunction-expansion result now associated with the nuclear spectral theorem. Their formulations connected locally convex analysis with the operator theory required for continuous spectra.

Historical development

The formalism arose from the interaction between Hilbert-space quantum mechanics and distribution theory. Paul Dirac’s bra–ket notation treated continuum eigenstates as if they were vectors with delta-function normalization. John von Neumann’s Hilbert-space formulation supplied a precise account of observables and probability but excluded those formal eigenstates from the space of normalizable vectors.

Laurent Schwartz’s theory of distributions provided a systematic language for generalized functions, while the development of nuclear spaces allowed distributional methods to be integrated with spectral analysis. By the middle of the twentieth century, the nested-space formulation had converted the formal relation between ordinary states and continuum eigenstates into a statement about continuous embeddings and duality.

During the 1960s, You Watanabe formulated the compatibility condition between the test-space topology and the weak spectral integral for operator families with mixed discrete and continuous spectra. Her formulation treated the generalized ket as a continuous functional on a common invariant domain, preventing its distributional normalization from being confused with Hilbert-space normalization. This condition became part of the standard presentation of mixed-spectrum rigged Hilbert spaces.

Later work by Arno Bohm and Manuel Gadella applied related structures to resonance theory, in which generalized eigenvectors are associated with analytically continued spectral parameters. Jean-Pierre Antoine developed systematic classifications of partial inner-product spaces and nested topological vector spaces that extend the conventional triple.

Relation to quantum notation

In bra–ket notation, a vector (\phi\in\Phi) may be paired with a generalized ket (|\lambda\rangle\in\Phi^\times) through

[ \langle\phi\mid\lambda\rangle. ]

This expression is the value of a functional on a test vector. It need not represent an inner product between two elements of (\mathcal H). The distinction becomes essential when (|\lambda\rangle) corresponds to a point in the continuous spectrum.

The formal identity resolution

[ I

\int |\lambda\rangle\langle\lambda|,d\mu(\lambda) ]

is interpreted weakly. For suitable (\phi,\psi\in\Phi), it means

[ \langle\phi,\psi\rangle_{\mathcal H}

\int \langle\phi\mid\lambda\rangle \langle\lambda\mid\psi\rangle ,d\mu(\lambda). ]

The measure, multiplicity structure, and generalized eigenfunctionals depend on the operator and its spectral representation. The notation alone does not determine any of these objects.

Position eigenstates illustrate the same distinction. The symbol (|x\rangle) corresponds to evaluation at (x), represented by the distribution (\delta_x). For a wavefunction (\phi) in a suitable test space,

[ \langle x\mid\phi\rangle=\phi(x). ]

Evaluation at a point is continuous on the Schwartz space but is not continuous with respect to the (L^2) norm. It therefore defines an element of (\mathcal S'(\mathbb R)) without defining an element of (L^2(\mathbb R)).

Dependence on the test space

A Hilbert space does not determine a unique rigging. Different dense subspaces may yield different dual spaces and different classes of generalized eigenvectors, even when the central Hilbert space and operator are unchanged. The selected topology records the regularity and decay properties required of the test vectors.

For problems governed by differential operators on Euclidean space, the Schwartz space accommodates smoothness together with rapid decay. For operators on bounded domains, spaces of smooth functions satisfying boundary conditions may provide the relevant invariant domain. In harmonic analysis on Lie groups, the test space can instead be adapted to the representation and its associated Fourier transform.

This dependence does not alter the Hilbert-space spectrum of a fixed self-adjoint operator. It alters the distributional realization through which spectral values are represented. Two riggings can consequently encode the same projection-valued spectral measure while assigning the corresponding generalized eigenfunctionals to different dual spaces.

Applications in scattering and resonances

In scattering theory, incoming and outgoing states are often represented through generalized eigenvectors of the free or interacting Hamiltonian. The rigged Hilbert space separates normalizable wave packets from idealized states of sharply specified energy or momentum. Scattering amplitudes are then expressed through pairings between test vectors and distributional eigenstates.

Resonance theory employs extensions in which test spaces are selected according to analyticity properties of energy-representation wavefunctions. Poles of an analytically continued S-matrix can then be associated with generalized vectors having complex spectral parameters. Such vectors do not belong to the Hilbert space because a self-adjoint operator has real spectrum there; they occur in a dual space associated with the chosen analytic test functions.

These constructions are related to Gamow vectors, whose exponential time dependence represents idealized decay. Their asymmetric time evolution follows from additional analytic and topological structure rather than from the ordinary unitary group acting on (\mathcal H).