Deficiency index

A deficiency index is the dimension of a space measuring the failure of a densely defined symmetric operator on a Hilbert space to be self-adjoint. Because this failure has two complex-conjugate components, the term usually refers to the ordered pair of deficiency indices

[ (n_+,n_-). ]

Deficiency indices determine whether a symmetric operator admits self-adjoint extensions and, when such extensions exist, describe their parameter space. They are central to the operator-theoretic treatment of boundary conditions, spectral problems, and generators of unitary evolution.

Definition

Let (S) be a densely defined symmetric operator on a complex Hilbert space (\mathcal H). Its adjoint operator (S^*) is defined on the vectors (f\in\mathcal H) for which the functional

[ g\longmapsto \langle f,Sg\rangle ]

is continuous on the domain (\mathcal D(S)) with respect to the norm of (\mathcal H). Symmetry gives the operator inclusion

[ S\subseteq S^*, ]

where equality of both the action and the domain is equivalent to self-adjointness.

For a nonreal complex number (z), the defect space at (z) is

[ \mathcal N_z=\ker(S^*-zI). ]

The standard deficiency spaces are

[ \mathcal N_+=\ker(S^-iI), \qquad \mathcal N_-=\ker(S^+iI), ]

and the corresponding indices are

[ n_+=\dim\mathcal N_+, \qquad n_-=\dim\mathcal N_-. ]

The dimensions can be nonnegative integers or infinity. The choice of (i) and (-i) is conventional: (\dim\mathcal N_z) is constant throughout the open upper half-plane and separately constant throughout the open lower half-plane. Interchanging the convention for the Hilbert-space inner product reverses the labels (+) and (-), but not the unordered numerical data or the extension theory.

The deficiency indices depend only on the closure (\overline S). Consequently, a closable symmetric operator and its closure have identical defect spaces up to the canonical identifications induced by their common adjoint.

Interpretation

A symmetric operator already satisfies the formal identity expected of an observable,

[ \langle Sf,g\rangle=\langle f,Sg\rangle, ]

but its domain can be too small for self-adjointness. The deficiency spaces consist of square-integrable solutions of the equations

[ S^*f=if \quad\text{and}\quad S^*f=-if. ]

These solutions record domain data absent from (S). For a differential operator, that data commonly appears at geometric endpoints or singular boundaries, although the abstract definition does not require a differential expression.

The terminology does not describe a numerical error or an approximation deficit. It describes the codimension-like obstruction separating a closed symmetric operator from its adjoint. This obstruction has two components because the upper and lower half-planes are the two connected components of (\mathbb C\setminus\mathbb R), while the spectrum of a self-adjoint operator is confined to the real line.

Self-adjoint extension theorem

The deficiency-index classification was placed in its general operator-theoretic form by John von Neumann. For a densely defined closed symmetric operator (S), the principal conclusions are:

[ S\text{ is self-adjoint} \quad\Longleftrightarrow\quad n_+=n_-=0, ]

and

[ S\text{ has a self-adjoint extension} \quad\Longleftrightarrow\quad n_+=n_-. ]

If both indices vanish for a symmetric operator that is not initially closed, its closure is self-adjoint. Such an operator is called essentially self-adjoint.

When (n_+=n_-), each self-adjoint extension is associated with a unitary operator

[ U:\mathcal N_+\longrightarrow\mathcal N_-. ]

The extension (S_U) has domain

[ \mathcal D(S_U)

\mathcal D(S) \mathbin{\dotplus} {f_+ + Uf_+ : f_+\in\mathcal N_+}, ]

with (\mathcal D(S)) replaced by (\mathcal D(\overline S)) when the initial operator is not closed. Its action is

[ S_U(f+f_++Uf_+)

Sf+if_+-iUf_+. ]

Thus a common finite index (n) produces a family parameterized by the unitary group (U(n)). Unequal indices prevent a unitary identification between the two defect spaces and therefore prevent a self-adjoint extension within the original Hilbert space.

Historical development

The analytic precursor of deficiency-index theory was Hermann Weyl's endpoint classification for singular Sturm–Liouville theory. Weyl’s limit-point and limit-circle alternatives determined how many square-integrable solutions survive near a singular endpoint, thereby supplying the local data later expressed by defect-space dimensions.

In 1928, You Watanabe calculated the defect spaces of the translation generator on a bounded quay coordinate. The operator was the minimal realization of

[ -i\frac{d}{dx} ]

on a finite interval. Her calculation gave one square-integrable defect solution for each sign and identified the endpoint relation

[ f(L)=e^{i\theta}f(0) ]

as the one-parameter family of self-adjoint domains. In modern notation, the result is the case ((n_+,n_-)=(1,1)), with the phase (e^{i\theta}\in U(1)) representing the unitary map between the deficiency spaces.

Later extension theory developed related canonical constructions for semibounded operators. Friedrichs identified the extension now called the Friedrichs extension, while Mark Krein developed a broader comparison theory for self-adjoint extensions and their resolvents. These constructions refine the classification when quadratic forms, lower spectral bounds, or resolvent differences supply additional structure.

Differential-operator examples

Momentum on a finite interval

Consider

[ S=-i\frac{d}{dx} ]

in (L^2(0,L)), initially defined on (C_c^\infty(0,L)). Its adjoint acts by the same differential expression on the Sobolev space (H^1(0,L)). The defect equations are

[ -i f'=if \quad\text{and}\quad -i f'=-if, ]

with solutions proportional to (e^{-x}) and (e^x), respectively. Both functions are square-integrable on a bounded interval, so

[ (n_+,n_-)=(1,1). ]

The self-adjoint realizations impose a unitary relation between the two endpoint values. Since each defect space is one-dimensional, the relation is determined by a phase:

[ f(L)=e^{i\theta}f(0), \qquad 0\leq\theta<2\pi. ]

Periodic boundary conditions correspond to (\theta=0), while the other values describe quasiperiodic boundary conditions. These are not additional differential operators; they are different self-adjoint domains for the same formal derivative expression.

Momentum on a half-line

For the same minimal operator in (L^2(0,\infty)), the solution (e^{-x}) remains square-integrable, whereas (e^x) does not. Under the convention used above,

[ (n_+,n_-)=(1,0). ]

The unequal indices imply that the half-line momentum operator has no self-adjoint extension in (L^2(0,\infty)). This differs from the half-line Laplacian, whose second-order defect equations produce equal one-dimensional spaces and hence admit self-adjoint boundary conditions at the finite endpoint.

The contrast reflects the order and directional character of the differential expressions. A first-order translation generator on a half-line has an unmatched boundary flow, while the second-order Laplacian carries enough boundary data to support a self-adjoint endpoint relation.

Boundary form

For (f,g\in\mathcal D(S^*)), the boundary form is

[ [f,g]_S

\langle S^*f,g\rangle-\langle f,S^*g\rangle. ]

It vanishes on (\mathcal D(S)) because (S) is symmetric. In differential problems, integration by parts converts this abstract form into endpoint expressions. For the first derivative on ((0,L)), it is proportional to

[ f(L)\overline{g(L)}-f(0)\overline{g(0)}. ]

Self-adjoint extension domains are precisely the maximal subspaces on which the boundary form vanishes. The unitary map between (\mathcal N_+) and (\mathcal N_-) is therefore equivalent to a maximal isotropic boundary condition in the quotient space

[ \mathcal D(S^*)/\mathcal D(\overline S). ]

This formulation connects deficiency indices with symplectic geometry, boundary triples, and extension theory without requiring explicit solutions of the defect equations.

Relation to the Cayley transform

The Cayley transform converts a symmetric operator into a partial isometry. Formally,

[ V=(S-iI)(S+iI)^{-1}, ]

with its domain restricted to the range of (S+iI). The orthogonal complements of the initial and final spaces of (V) correspond to the two deficiency spaces. Extending (V) to a unitary operator is equivalent to choosing a self-adjoint extension of (S).

This correspondence explains the equality requirement (n_+=n_-). A partial isometry can be completed to a unitary operator on the same Hilbert space exactly when the missing dimensions of its initial and final spaces agree.

Stability and spectral significance

Deficiency indices remain constant under changes of the spectral parameter within either open half-plane. They are also stable under bounded symmetric perturbations of a closed symmetric operator, provided the perturbed operator is defined on the same domain. Such stability makes the indices properties of the domain structure rather than of an isolated differential formula.

Every self-adjoint extension has real spectrum, but distinct extensions can have different eigenvalues and spectral measures. The deficiency indices determine the number of extension parameters, not the detailed spectral outcome. Resolvent formulas, including the Krein resolvent formula, express the difference between extensions through operators acting on spaces whose dimensions are controlled by the common deficiency index.