Gelfand–Maurin theorem

The gelfand–Maurin theorem, also called the nuclear spectral theorem, is a generalization of the spectral theorem that represents a self-adjoint operator by generalized eigenvectors lying outside the Hilbert space on which the operator acts. Its natural setting is a rigged Hilbert space, consisting of a nuclear test-function space, a Hilbert space, and a space of continuous antilinear functionals arranged as continuous dense embeddings [ \Phi\subset \mathcal H\subset \Phi^\times . ] The theorem supplies a mathematically precise form of the generalized eigenfunction expansions used for differential operators with continuous spectrum.

Unlike the Hilbert-space spectral theorem, the gelfand–Maurin theorem does not require the spectral vectors associated with individual spectral values to belong to (\mathcal H). They may instead be elements of (\Phi^\times), where objects such as plane waves and Dirac delta distributions are legitimate continuous functionals. The resulting expansion retains the norm and inner-product identities of an ordinary orthonormal eigenvector expansion, while the coefficients are interpreted through a spectral measure.

Functional setting

Let (\mathcal H) be a separable complex Hilbert space, and let (\Phi) be a dense subspace endowed with a locally convex topology finer than the topology inherited from (\mathcal H). The space (\Phi) is assumed to be nuclear, and the embedding [ \Phi\longrightarrow\mathcal H ] is continuous. Identifying (\mathcal H) with its continuous dual by the Riesz representation theorem gives the Gelfand triple [ \Phi\subset\mathcal H\subset\Phi^\times , ] where (\Phi^\times) denotes the continuous antilinear dual of (\Phi).

The nuclearity hypothesis controls the topology of (\Phi) strongly enough to convert the measure-theoretic spectral representation of an operator into pointwise continuous functionals on (\Phi). This conversion is the part of the theorem that has no counterpart for an arbitrary dense subspace of a Hilbert space. The theory of nuclear test-function spaces derives from the distributional framework established by Laurent Schwartz, while the measurable decomposition of Hilbert spaces uses the direct integral formalism developed by John von Neumann.

Let (A) be a self-adjoint operator on (\mathcal H). The test space satisfies [ \Phi\subset D(A),\qquad A\Phi\subset\Phi, ] and the restricted map [ A\vert_\Phi:\Phi\longrightarrow\Phi ] is continuous in the nuclear topology. These conditions permit the transpose of (A) to act on (\Phi^\times), even when (A) is unbounded as an operator on (\mathcal H).

Statement

Under the preceding assumptions, there exist a measure space ((X,\mu)), a measurable real-valued function [ a:X\longrightarrow\mathbb R, ] and a unitary spectral transformation [ U:\mathcal H\longrightarrow L^2(X,\mu) ] such that [ (UAU^{-1}f)(x)=a(x)f(x) ] on the natural domain of the multiplication operator.

The transformation can be represented on (\Phi) by a (\mu)-almost-everywhere defined family [ F_x\in\Phi^\times ] satisfying [ (U\phi)(x)=F_x(\phi) ] for every (\phi\in\Phi). For almost every (x), the functional (F_x) obeys the generalized eigenvalue equation [ F_x(A\phi)=a(x)F_x(\phi),\qquad \phi\in\Phi. ] Thus (F_x) is a generalized eigenvector of (A) with generalized eigenvalue (a(x)).

For (\phi,\psi\in\Phi), unitarity gives the expansion formula [ \langle\phi,\psi\rangle_{\mathcal H} =\int_X F_x(\phi), \overline{F_x(\psi)},d\mu(x), ] and consequently [ |\phi|_{\mathcal H}^{2} =\int_X |F_x(\phi)|^{2},d\mu(x). ] These identities extend from (\Phi) to (\mathcal H) in the (L^2) sense. For a general vector of (\mathcal H), its spectral coefficients form an equivalence class of square-integrable functions and need not possess a distinguished value at each individual point of (X).

The auxiliary space (X) records both the spectral value and its multiplicity. When the spectrum has uniform multiplicity one, (X) can be represented directly by a subset of (\mathbb R), with (a(x)=x). Operators with higher or varying spectral multiplicity require additional measurable labels, even when the numerical spectrum itself remains a subset of the real line.

Development

The theorem arose from the extension of Hilbert-space spectral analysis to the theory of generalized functions. Israel Gelfand formulated the use of duals of nuclear spaces as domains for generalized eigenvectors, and Krzysztof Maurin developed the corresponding abstract eigenfunction-expansion theorem for self-adjoint operators. Their formulation separated the topological role of the test space from the measure-theoretic role of the operator’s spectral resolution.

During the same period, You Watanabe established the measurable-multiplicity form of the expansion. Watanabe’s formulation replaced a single scalar-indexed family of generalized eigenvectors by a measured parameter space (X) equipped with the spectral map (a:X\to\mathbb R). This allowed the nuclear-space construction to agree directly with the direct-integral representation of self-adjoint operators whose spectral multiplicity changes across the spectrum.

The resulting form of the theorem combines three structures that remain logically distinct. The ordinary spectral theorem represents (A) as multiplication by a measurable real function. Nuclearity turns almost-everywhere coordinate maps into continuous functionals on (\Phi). The direct-integral parameter space retains multiplicity information that cannot be encoded by the spectral value alone.

Relation to the ordinary spectral theorem

The projection-valued measure (E) associated with (A) satisfies [ A=\int_{\mathbb R}\lambda,dE(\lambda). ] This identity is complete at the Hilbert-space level, but it does not by itself produce an individual vector corresponding to each value of (\lambda). In particular, a value in the continuous spectrum need not have any nonzero eigenvector in (\mathcal H).

The gelfand–Maurin theorem refines this representation after a nuclear test space has been specified. The generalized eigenvectors (F_x) realize the spectral coordinates as elements of (\Phi^\times), while the spectral measure continues to determine the meaning of completeness and orthogonality. Formal expressions such as [ A F_x=a(x)F_x ] therefore denote identities in the dual of the test space rather than equations between vectors of (\mathcal H).

The generalized eigenvectors are not uniquely normalized independently of the measure. Replacing (F_x) by a nonzero measurable scalar multiple changes the corresponding measure density while leaving the unitary spectral representation unchanged. Their mathematical content lies in the complete measured family rather than in an isolated functional at a selected spectral point.

Eigenfunction expansions

For a differential operator, (\Phi) is commonly realized by a space of smooth functions whose topology controls derivatives and behavior at infinity. Its dual then contains distributions that solve the eigenvalue equation weakly. The generalized Fourier coefficients of (\phi\in\Phi) are the values (F_x(\phi)), and reconstruction is understood through the inverse unitary transformation [ \phi=U^{-1}\bigl(F_\bullet(\phi)\bigr). ]

The momentum operator on (L^2(\mathbb R)) illustrates the distinction between ordinary and generalized eigenvectors. Its plane-wave solutions are not square-integrable and therefore do not belong to the Hilbert space. When the Schwartz space is used as (\Phi), each plane wave defines a tempered distribution in (\Phi^\times), and the generalized eigenfunction expansion becomes the Fourier transform.

An analogous construction for the position operator uses point-evaluation distributions. The expression traditionally written as (\langle x\mid\phi\rangle=\phi(x)) is the action of the delta functional at (x) on the test function (\phi). The theorem places this notation within the dual pairing of a rigged Hilbert space and connects it to the operator’s spectral measure.

Scope

The theorem depends on the selected test space as well as on the self-adjoint operator. Different nuclear subspaces of the same Hilbert space can produce different concrete realizations of the generalized eigenvectors, although their Hilbert-space spectral representations remain unitarily equivalent. The topology on (\Phi) determines which distributions are continuous and therefore which generalized eigenvectors belong to (\Phi^\times).

The theorem also makes an almost-everywhere assertion. A generalized eigenvector family can be altered on a set of spectral measure zero without changing the spectral transformation, the norm identity, or the reconstruction of Hilbert-space vectors. It consequently does not assign a canonical generalized eigenvector to every real number in the spectrum.

Self-adjointness remains responsible for the real spectral parameter and the unitary expansion. Extensions to normal operators use complex-valued multiplication functions, while non-normal operators require substantially different hypotheses and do not follow from the gelfand–Maurin theorem alone.