Spectral Parameter Operator: Theory & Applications

• Spectral parameter operator is a framework where the spectral parameter is incorporated into operator equations, boundary data, or operator-valued symbols for versatile spectral analysis.
• It encompasses formulations such as operator pencils, SPPS expansions, transmutation techniques, and nonlinear Nevanlinna dependence to solve complex eigenvalue problems.
• This approach is pivotal in advancing spectral theory by providing rigorous methods to analyze continuous spectra and resolve inverse spectral questions.

In the literature surveyed here, a spectral parameter operator is not a single canonical object but a family of constructions in which the spectral parameter enters an operator problem through an auxiliary operator, a boundary functional, or an operator-valued symbol. Representative formulations include the operator pencil $Lu=\lambda Ru$, boundary conditions with explicit $\lambda$-dependence, and operator functions $L(\lambda)$ whose zeros or resolvents encode the spectrum. This suggests that the term is best understood as a structural role played by the parameter inside an operator equation rather than as a fixed object of one theory (Perez et al., 2013, Nabiev, 2019, Ortiz et al., 2017).

1. Core operator-theoretic forms

Three recurrent forms organize the subject. In perturbed Bessel Sturm–Liouville problems, the spectral parameter multiplies an operator of order at most one: $$L=-\frac{d^2}{dx^2}+\frac{l(l+1)}{x^2}+q(x), \qquad Ru=r_0u+r_1u',$$ so the spectral problem is the operator pencil

$Lu=\lambda Ru.$

Here the parameter does not merely appear as $\lambda u$; it may multiply a first-order differential operator $R$ (Perez et al., 2013).

A second pattern places the parameter in the boundary data. For the inverse Sturm–Liouville problem on $[0,\pi]$, the differential equation is

$-y''(x)+q(x)y(x)=\lambda^2 y(x),$

while one non-separated boundary condition depends linearly on $\lambda$: $\lambda$0 In this setting, the operator aspect of the spectral parameter is expressed through the boundary operator rather than the bulk differential expression (Nabiev, 2019).

A third pattern uses an operator-valued symbol. For abstract Volterra integrodifferential equations of Gurtin–Pipkin type, the spectral object is

$\lambda$1

The spectrum is then defined by

$\lambda$2

so the parameter enters through a meromorphic operator-function rather than a linear pencil alone (Ortiz et al., 2017).

2. Operator pencils, SPPS expansions, and transmutation

For perturbed Bessel equations, the spectral-parameter operator structure is developed through SPPS, or spectral parameter power series. Assuming the auxiliary equation

$\lambda$3

has a nonvanishing solution $\lambda$4 with the stated asymptotics near $\lambda$5, the regular solution of

$\lambda$6

admits the uniformly convergent expansion

$\lambda$7

The associated characteristic function

$\lambda$8

is entire, and its zeros coincide with the eigenvalues of the boundary value problem. The same framework explicitly covers the case in which the parameter multiplies a first-order term, including the example

$\lambda$9

with boundary condition $L(\lambda)$0, which has complex eigenvalues (Perez et al., 2013).

Transmutation theory gives the complementary operator-theoretic mechanism. For the unperturbed singular Bessel operator

$L(\lambda)$1

and the perturbed operator

$L(\lambda)$2

the transmutation operator $L(\lambda)$3 satisfies

$L(\lambda)$4

The SPPS coefficients are encoded by the mapping formula

$L(\lambda)$5

In the broader Sturm–Liouville setting, a Volterra transmutation operator $L(\lambda)$6 satisfies

$L(\lambda)$7

so the SPPS basis is the image of the monomial basis under the intertwining operator (Kravchenko et al., 2012).

3. Boundary dependence and multiparameter spectral geometry

When the spectral parameter is built into the boundary operator, inverse spectral theory changes accordingly. For the non-separated Sturm–Liouville problem $L(\lambda)$8 described above, the characteristic function

$L(\lambda)$9

has zeros $$L=-\frac{d^2}{dx^2}+\frac{l(l+1)}{x^2}+q(x), \qquad Ru=r_0u+r_1u',$$0 equal to the eigenvalues of $$L=-\frac{d^2}{dx^2}+\frac{l(l+1)}{x^2}+q(x), \qquad Ru=r_0u+r_1u',$$1. The inverse data consist of one spectrum together with a sign sequence,

$$L=-\frac{d^2}{dx^2}+\frac{l(l+1)}{x^2}+q(x), \qquad Ru=r_0u+r_1u',$$2

and the paper proves that $$L=-\frac{d^2}{dx^2}+\frac{l(l+1)}{x^2}+q(x), \qquad Ru=r_0u+r_1u',$$3 is uniquely determined by this pair. It also gives a reconstruction algorithm that recovers $$L=-\frac{d^2}{dx^2}+\frac{l(l+1)}{x^2}+q(x), \qquad Ru=r_0u+r_1u',$$4, $$L=-\frac{d^2}{dx^2}+\frac{l(l+1)}{x^2}+q(x), \qquad Ru=r_0u+r_1u',$$5, $$L=-\frac{d^2}{dx^2}+\frac{l(l+1)}{x^2}+q(x), \qquad Ru=r_0u+r_1u',$$6, the auxiliary zeros $$L=-\frac{d^2}{dx^2}+\frac{l(l+1)}{x^2}+q(x), \qquad Ru=r_0u+r_1u',$$7, $$L=-\frac{d^2}{dx^2}+\frac{l(l+1)}{x^2}+q(x), \qquad Ru=r_0u+r_1u',$$8, $$L=-\frac{d^2}{dx^2}+\frac{l(l+1)}{x^2}+q(x), \qquad Ru=r_0u+r_1u',$$9, $Lu=\lambda Ru.$0, $Lu=\lambda Ru.$1, $Lu=\lambda Ru.$2, and finally $Lu=\lambda Ru.$3 (Nabiev, 2019).

A more explicitly multiparameter version appears in block-operator theory. The two-parameter problem

$Lu=\lambda Ru.$4

defines pair-eigenvalues $Lu=\lambda Ru.$5. Under the rank-one coupling $Lu=\lambda Ru.$6, with $Lu=\lambda Ru.$7 the orthogonal projection onto $Lu=\lambda Ru.$8, the characteristic equation becomes

$Lu=\lambda Ru.$9

The real pair-spectrum then has the “Chess Board” geometry: it lies on monotone decreasing curves, constrained to alternating rectangles in the $\lambda u$0-plane, and may also include entire vertical or horizontal spectral lines when exceptional eigenspaces occur (Levitin et al., 2018).

These examples suggest that the spectral parameter operator need not be confined to a single linear factor. It may be distributed across boundary couplings, several scalar parameters, or block-resolvent identities.

4. Nonlinear Nevanlinna dependence and generalized resolvents

A major extension replaces linear pencils by nonlinear operator-valued dependence on the spectral parameter in the Nevanlinna manner. For relations generated by pairs of differential operator expressions, one studies equations of the form

$\lambda u$1

where the coefficient family is analytic off $\lambda u$2 and has the usual Nevanlinna sign property. The high-order equation is reduced to a first-order weighted system, a characteristic operator $\lambda u$3 is introduced, and an analogue of the generalized resolvent is constructed as an integro-differential operator. The resulting operator $\lambda u$4 satisfies

$\lambda u$5

and admits the Stieltjes representation

$\lambda u$6

This framework yields generalized eigenfunction expansions, inversion formulas, Parseval identities, and Bessel-type inequalities for operator differential equations whose spectral dependence is nonlinear (Khrabustovskyi, 2012, Khrabustovskyi, 2013).

A related abstraction appears on a lattice of Hilbert spaces $\lambda u$7. For an operator $\lambda u$8, the $\lambda u$9-resolvent set is

$R$0

and whenever $R$1, the generalized resolvent is

$R$2

In this setting, $R$3 is open and $R$4 is analytic on each connected component of $R$5. The same framework supports generalized eigenvalues associated to points of the continuous spectrum via the generalized KLMN theorem and a Maurin–Gel'fand-type expansion theory (Antoine et al., 2014).

5. Spectral measures, resolvent algorithms, and operator pencils

For self-adjoint operators on infinite-dimensional spaces, continuous spectrum prevents diagonalization by an eigenfunction basis, and spectral measures become the correct object. If $R$6 is the projection-valued measure of a self-adjoint operator $R$7, then for fixed $R$8,

$R$9

is the scalar spectral measure. Stone’s formula implies that, for $[0,\pi]$0 with $[0,\pi]$1,

$[0,\pi]$2

is a Poisson-kernel smoothing of $[0,\pi]$3. In the notation of the computational framework,

$[0,\pi]$4

is obtained by solving the shifted equation

$[0,\pi]$5

and then taking $[0,\pi]$6 (Colbrook et al., 2020, Colbrook et al., 2022).

Higher-order rational kernels replace the Poisson kernel by combinations of resolvents at complex shifts. In the stated regularity regime, the rational-kernel construction yields convergence like $[0,\pi]$7, while preserving a shifted-solve workflow. This is the basis of the SpecSolve methodology, described as “discretization-oblivious” because it requires only shifted solves and inner products, independent of whether the underlying discretization uses ultraspherical, Chebyshev, Fourier, rational, or spectral-element methods (Colbrook et al., 2022).

The same strategy extends from single operators to generalized eigenvalue problems

$[0,\pi]$8

with $[0,\pi]$9 and $-y''(x)+q(x)y(x)=\lambda^2 y(x),$0 self-adjoint and $-y''(x)+q(x)y(x)=\lambda^2 y(x),$1 positive and invertible. Defining

$-y''(x)+q(x)y(x)=\lambda^2 y(x),$2

one obtains

$-y''(x)+q(x)y(x)=\lambda^2 y(x),$3

and the smoothed spectral measure is computed from shifted pencil solves $-y''(x)+q(x)y(x)=\lambda^2 y(x),$4. In this sense, the spectral parameter operator becomes a resolvent-accessible pencil rather than a standalone differential operator (Colbrook et al., 2022).

6. Parameterized non-selfadjoint operators and learned parameter-to-spectrum maps

A non-selfadjoint PDE example is the spin-weighted spheroidal wave operator with complex aspherical parameter $-y''(x)+q(x)y(x)=\lambda^2 y(x),$5. After separation of variables and reduction to a one-dimensional Sturm–Liouville problem with complex potential, the operator is not symmetric, so the spectrum is generally complex and Jordan chains may appear. Even so, a family of spectral decompositions

$-y''(x)+q(x)y(x)=\lambda^2 y(x),$6

is constructed for all $-y''(x)+q(x)y(x)=\lambda^2 y(x),$7 in the strip $-y''(x)+q(x)y(x)=\lambda^2 y(x),$8. There exists an integer $-y''(x)+q(x)y(x)=\lambda^2 y(x),$9 such that $\lambda$0 projects onto an invariant subspace of dimension $\lambda$1, each $\lambda$2 for $\lambda$3 projects onto an invariant subspace of dimension at most $\lambda$4, the projections satisfy

$\lambda$5

and the spectral decomposition is complete (Finster et al., 2015).

Recent operator-learning work treats parameter-dependent spectra themselves as learned operators. DeepOPiraKAN is designed to learn the continuous map from physical parameters to spectral data and mode functions, so that a single model represents the parameter-to-spectrum mapping for Kerr quasinormal modes across the full spin range $\lambda$6. In the reported benchmark, a single trained network resolves modes with $\lambda$7 and overtones up to $\lambda$8, with relative errors of $\lambda$9 for the fundamental mode and $\lambda$00 for higher overtones (Gu et al., 26 Apr 2026).

A related PDE framework uses a point-calibrated spectral transform in which each physical point predicts a frequency preference vector and modulates the spectral basis through

$\lambda$01

$\lambda$02

The resulting token mixer is a spectral operator with input-conditioned parameterization over the basis. This suggests a modern computational broadening of the notion from analytic pencils and resolvents to learned parameterization of spectral structure (Yue et al., 2024).

7. Distinction from spectral operators and spectral-operator calculus

The expression should not be conflated with a spectral operator in the sense of Dunford. A spectral operator $\lambda$03 admits an idempotent-valued spectral resolution

$\lambda$04

and the unique Dunford decomposition

$\lambda$05

where $\lambda$06 is scalar-type, $\lambda$07 is quasinilpotent, and $\lambda$08. In this setting, the central object is the normalized power sequence

$\lambda$09

which converges in norm for spectral operators, with limit

$\lambda$10

This is a theorem about spectral operators, not about an operator carrying the spectral parameter (Nayak et al., 2024).

An analogous terminological caution applies to spectral-operator calculus. There, the primitive data are a self-adjoint operator $\lambda$11, its spectral measure $\lambda$12, and bounded transforms $\lambda$13; the paper explicitly states that it does not introduce a single distinguished “spectral parameter operator.” On the trace-class envelope, every admissible evaluator has the form

$\lambda$14

with $\lambda$15 Borel and nondecreasing, and the calculus classifies operators by counting-function asymptotics such as polynomial growth (Homer, 12 Dec 2025).

Taken together, these distinctions indicate that a spectral parameter operator is best reserved for situations in which the dependence on $\lambda$16 is built into the operator equation, boundary data, or operator-valued symbol, rather than for operators that are merely spectral in the sense of possessing a spectral resolution.