Why Rigged Hilbert Spaces

Rigged Hilbert spaces explain how continuous-spectrum kets can be useful without being physical unit vectors. They equip the ordinary Hilbert-space theory with a test space and generalized functionals, so that evaluation, Fourier kernels, and eigenfunction expansions have precise meanings. The spectral probability rule already works without those kets; a rigging provides additional language for calculating and interpreting it.

Helpful background. Continuous Spectrum Without Eigenvectors explains why a spectral value need not have a normalizable eigenvector. Distribution Theory Minimum introduces testing and weak derivatives.

Continuous labels do not define an ordinary orthonormal basis

An orthonormal basis of a separable Hilbert space is at most countable. One quick argument chooses a countable dense set: the disjoint open balls of radius less than $1/\sqrt2$ around orthonormal vectors must each contain a different member of that dense set. There cannot be uncountably many such balls.

Position labels $x\in\mathbb R$ therefore cannot index an uncountable orthonormal family of nonzero vectors in $L^2(\mathbb R)$. The expression

$$\langle x|y\rangle=\delta(x-y)$$

has a different meaning. Its right-hand side is a distribution kernel, not a Kronecker delta on an uncountable vector basis and not a function with an infinite value on the diagonal. In particular, writing $\langle x|x\rangle=\delta(0)$ does not compute a Hilbert norm.

The countability argument diagnoses the mistaken interpretation. The stronger operator-specific statement is that multiplication by $x$ has no nonzero $L^2(dx)$ eigenvector at any real value. The position-ket treatment constructs the appropriate evaluation functional instead.

Probabilities remain ordinary Hilbert-space statements

For a normalized $\psi\in L^2(\mathbb R)$, a position detector region $\Delta$ has probability

$$\begin{aligned} \Pr_\psi(X\in\Delta) &=\langle\psi,E_X(\Delta)\psi\rangle\\ &=\int_\Delta|\psi(x)|^2\,dx. \end{aligned}$$

Here $E_X(\Delta)$ is bounded multiplication by an indicator. This formula is meaningful for every Hilbert-space state, including states outside the operator domain $D(X)$. A mean position or a time derivative may require additional integrability, but the event probability does not.

Singletons have probability zero for this absolutely continuous position law. That does not mean the detector cannot return a continuous outcome. Countable additivity does not require an uncountable union of null singleton events to have zero measure. Finite resolution is described by intervals or, in a more general apparatus model, a POVM and its instrument.

Thus a rigging is not a repair to an inconsistent Born rule. It clarifies the generalized notation sometimes used to express the same projection measure. Projection-Valued Measures develops the measure directly.

A test space makes singular pairings continuous

The value $\psi(x_0)$ is not well defined on an arbitrary $L^2$ equivalence class: changing one point does not change the class. Restricting to a suitable space of continuous test functions makes evaluation well defined. Giving that space a sufficiently strong topology makes evaluation continuous.

For the free particle, Schwartz functions supply a convenient common space for multiplication, differentiation, and Fourier transformation. The resulting Gelfand triple places generalized position and momentum functionals beyond the Hilbert space. An identity is first verified after pairing with test functions; a separate continuity or spectral argument can then extend the resulting operator statement to more general states.

This distinction is practical. An ordinary state need not be Schwartz, and the choice of test space does not discard such states from quantum mechanics. It specifies a domain on which a particular distributional calculation is justified.

A generalized eigenvector does not specify a preparation

A delta-normalized plane wave labels a spectral component. A normalizable wavepacket combines components with square-integrable amplitudes. The probability interpretation applies to the resulting state, while the generalized components help organize the transform.

Similarly, a fixed-energy scattering solution is not itself a normalized incoming state. One constructs a packet and controls its dynamics, or works with flux and asymptotic quantities under stated assumptions. Generalized eigenfunctions alone do not prove scattering completeness or a resonance expansion.

A rigging is also not unique. Different observables, boundary conditions, and analytic questions can favor different test spaces. Nuclear spectral theorems require hypotheses beyond the bare inclusion of three spaces. Naming a triple does not justify every exchange of limits, every product of distributions, or every formal completeness relation.

Exercise: conditioning on a finite detector region

Let $p=\|E_X(\Delta)\psi\|^2>0$. For the ideal two-outcome Lüders measurement of $\{E_X(\Delta),I-E_X(\Delta)\}$, show that the conditional vector $\psi_\Delta=E_X(\Delta)\psi/\sqrt p$ is normalized and returns the same event with probability one. Does it become an exact position eigenket?

Solution

The norm is one by the definition of $p$, and idempotence gives $E_X(\Delta)\psi_\Delta=\psi_\Delta$. Hence repetition of this two-outcome measurement returns the region with certainty. The state remains a square-integrable function supported in that region, not an eigenvector at one point. Setting $\Delta=\{x_0\}$ instead gives $p=0$, so this conditioning formula cannot produce a normalized position eigenket. A different instrument can have the same event probability and a different conditional state.

References

• de la Madrid, Rafael, “The role of the rigged Hilbert space in quantum mechanics,” European Journal of Physics 26, 287–312 (2005), doi:10.1088/0143-0807/26/2/008, arXiv:quant-ph/0502053 — generalized bras and kets as functionals in explicit quantum models.
• Hall, Brian C., Quantum Theory for Mathematicians, Graduate Texts in Mathematics 267, Springer, 2013, doi:10.1007/978-1-4614-7116-5 — separability, spectral measures, and continuous representations.
• Teschl, Gerald, Mathematical Methods in Quantum Mechanics: With Applications to Schrödinger Operators, 2nd ed., Graduate Studies in Mathematics 157, American Mathematical Society, 2014, author’s edition — projection measures and spectral transforms without formal eigenvector norms.