Spectral Problems in Quantum Mechanics

Quantum spectral research asks more than how to diagonalize a matrix. It asks which features of an infinite-dimensional operator can be inferred from finite information, with what error guarantees, and under which assumptions on domains, potentials, and spectrum. Reviewed 11 October 2026, this assessment uses the cited spectral-computation work through 2024 as its explicit baseline.

Helpful background. The unbounded spectral theorem supplies the operator framework; unbounded operators explains why domains matter.

A small residual has a limited meaning

For self-adjoint $H$, normalized $\psi\in D(H)$, and real $\mu$,

$$\operatorname{dist}(\mu,\sigma(H)) \le \|(H-\mu)\psi\|.$$

This is a spectral inclusion statement. It does not by itself prove that $\mu$ is near an isolated eigenvalue, establish its multiplicity, or exclude additional spectrum. For nonnormal operators, residuals are naturally related to pseudospectra and require different interpretation.

The proof explains the scope: the spectral theorem gives $\|(H-\mu)\psi\|^2=\int(\lambda-\mu)^2\,d\langle\psi,P(\lambda)\psi\rangle$. The integral is at least the squared distance to the spectrum because the spectral measure has total mass one. No step distinguishes continuous spectrum from point spectrum. Even a sequence of residuals tending to zero need not supply a normalizable eigenvector; it can be a Weyl sequence escaping to infinity.

Finite-section eigenvalues can exhibit spectral pollution. Increasing matrix size and seeing several stable digits is useful evidence but not a general proof. Boundary truncation, basis truncation, and arithmetic errors are distinct contributions.

A spurious eigenvalue that survives every enlargement

On $\ell^2(\mathbb N)\oplus\ell^2(\mathbb N)$ let $He_j^+=e_j^+$ and $He_j^-=-e_j^-$. The exact spectrum is $\{-1,1\}$. Define $u_j=(e_j^++e_j^-)/\sqrt2$ and let $V_n$ contain both vectors of the first $n$ pairs, together with $u_{n+1}$. These spaces are nested and their union is dense. Nevertheless, the compression $P_{V_n}H|_{V_n}$ has an eigenvalue zero for every $n$, because $Hu_{n+1}=(e_{n+1}^+-e_{n+1}^-)/\sqrt2$ is orthogonal to $V_n$.

The matrix residual at zero is exactly zero while the full residual is exactly one. More precision and larger $n$ cannot repair this mismatch. Completing the last pair removes the artifact. This solvable example separates approximation-space geometry from roundoff. Lewin and Séré analyze the corresponding mechanism and pollution-avoidance conditions for physically relevant gap problems, including Dirac and periodic Schrödinger operators. An operator’s self-adjointness alone does not make every Galerkin sequence safe.

What turns an estimate into an eigenvalue enclosure?

Established conditional bound. Suppose $H$ is bounded below, has a lowest eigenvalue $\lambda_0$, and the rest of its spectrum lies in $[\beta,\infty)$ for a rigorously known $\beta>\lambda_0$. For normalized $\psi\in D(H)$ let $\mu=\langle H\rangle_\psi<\beta$ and $v=\|(H-\mu)\psi\|^2$. Then the Temple bound is

$$\mu-\frac{v}{\beta-\mu}\le\lambda_0\le\mu.$$

For a finite second spectral moment, integrate $(\lambda-\lambda_0)(\lambda-\beta)\ge0$ over the spectrum. This gives $v\ge(\mu-\lambda_0)(\beta-\mu)$ and the lower bound. The independent exclusion of spectrum between $\lambda_0$ and $\beta$ does essential work; using an uncertified second Ritz value for $\beta$ invalidates the inference. As $\mu$ approaches $\beta$, the denominator warns that this certificate becomes ineffective. For interior or essential-spectrum gaps, use a theorem designed for that setting instead of applying a ground-state bound by analogy.

What is already resolved and what remains useful?

The Solvability Complexity Index program classifies which spectral tasks admit algorithms with specified limiting structure and error control. Ben-Artzi and colleagues and Colbrook’s later work establish important positive and negative results. It would be incorrect to advertise the unrestricted computational spectral problem as untouched.

Active project opportunities are operator-class specific: certified enclosures near thresholds; efficient bounds for singular or long-range potentials; structure-preserving approximations to resonances; and practical algorithms whose guarantees use physically checkable information. The evidence horizon here is the cited work through 2024. These are directions for a model-specific literature check, not assertions that every instance is still open in October 2026.

An impossibility theorem for a broad class does not prevent a successful algorithm for a restricted family. Conversely, successful numerics on a family do not refute a general impossibility theorem.

An entry project: manufacture and diagnose spectral pollution

Baseline. Implement the two-sign operator and nested spaces above. Report both the compressed and full residual, then remove the artifact by completing the last pair. Next reproduce one explicitly stated no-pollution result from Lewin and Séré for a model whose operator domain and basis assumptions you can verify.

Resources. A laptop, functional analysis, sparse linear algebra, and careful numerical error accounting. Use an analytically known spectrum as the initial control.

Deliverable. Separate plots and bounds for truncation, boundary, and roundoff errors. For every reported spectral feature, state whether it is an inclusion, exclusion, eigenvalue enclosure, or heuristic estimate.

Failure criterion. If the residual is computed only within the truncated matrix rather than against the intended operator or a justified enclosure, it does not certify the infinite-dimensional problem. A basis sequence that changes boundary conditions may converge to a different operator.

Research extension. Extend a certificate to a singular potential or a constrained operator family, identifying exactly which regularity assumption is weakened. A useful computational result includes a verified bound, not merely more digits.

Use a short first-stage report as the stopping point: exact counterexample, proof of the residual inequality, and a written account of the extra information needed by the chosen certificate. Do not move directly to a Coulomb–Dirac calculation if that theorem assumes a potential class the proposed singularity violates. A feasible next step is an a posteriori enclosure for one fixed coupling interval with a certified domain or form treatment. Compare enclosure width and computational work with the strongest applicable existing method; reproducing the known counterexample is a prerequisite, not the novel result.

Choosing a target

Pick an observable consequence: an eigenvalue ordering, a gap, a resonance location, or absence of spectrum in an interval. These require different mathematics. A bound on an isolated eigenvalue is usually a more manageable first project than complete classification of an essential spectrum.

Reassessment is warranted when a method expands the certified operator class, produces practical a posteriori bounds, or exposes a previously overlooked failure of a common discretization. Many-body spectral gaps treats the additional thermodynamic-limit problem.

References

• J. Ben-Artzi, M. J. Colbrook, A. C. Hansen, O. Nevanlinna, and M. Seidel, “Computing Spectra — On the Solvability Complexity Index Hierarchy and Towers of Algorithms,” arXiv:1508.03280 (2015; see version history). Primary work.
• M. J. Colbrook, “On the Computation of Geometric Features of Spectra of Linear Operators on Hilbert Spaces,” Foundations of Computational Mathematics 24, 723–804 (2024). Article.
• E. B. Davies and M. Plum, “Spectral pollution,” IMA Journal of Numerical Analysis 24, 417–438 (2004). Spectral-enclosure methods.
• M. Lewin and É. Séré, “Spectral pollution and how to avoid it (with applications to Dirac and periodic Schrödinger operators),” Proceedings of the London Mathematical Society 100, 864–900 (2010). Primary analysis.

Entry points

Reproduce a counterexample to naive finite-section inference before optimizing a spectral algorithm. Read exactly what information about the infinite operator the theorem assumes is available.