Finite-Support Hilbert Inversion

• Finite-support Hilbert inversion is a technique to invert Hilbert-type singular integrals on compact intervals, crucial for areas like harmonic analysis and tomography.
• It employs explicit inversion formulas, operator theory, and spectral analysis to address well-posedness and stability challenges, emphasizing kernel properties and Fredholm theory.
• Advanced numerical implementations use orthogonal polynomial expansions and regularization methods, such as truncated SVD and Tikhonov filtering, to enable accurate reconstructions.

A finite-support Hilbert inversion refers to the problem of inverting a Hilbert-type singular integral transform when both the input function and the data are supported on compact intervals—either a single interval or a finite union. Central to this theory are the finite (or truncated) Hilbert transform and its extensions, which play a critical role in harmonic analysis, mathematical tomography, and complex analysis. The rich structure of finite-support Hilbert inverses arises from their spectral properties, Fredholm theory in Banach function spaces, sensitivity to endpoint behavior, and their deep connections to problems in numerical analysis and partial differential equations.

1. Formulation and Operator-theoretic Properties

Let $I = (a,b)\subset \mathbb{R}$, and for $f\in L^1(I)$ define the finite (truncated) Hilbert transform as

$$(H f)(x) = \frac{1}{\pi}\,\mathrm{p.v.} \int_{a}^b \frac{f(y)}{y-x} \, dy,\qquad x\in I.$$

In the context of finite Hilbert inversion, the fundamental task is: given $g$ supported (or defined) on $I$, solve

$Hf = g\quad \text{on } I$

for $f$ supported on $I$.

Banach Function Space and Optimal Domains

If $X$ is a rearrangement-invariant Banach function space on $I$, the operator $f\in L^1(I)$0 is bounded whenever $f\in L^1(I)$1 (Boyd indices). For such $f\in L^1(I)$2, $f\in L^1(I)$3 is already optimally defined: there is no strictly larger Banach function space $f\in L^1(I)$4 such that $f\in L^1(I)$5 extends to a bounded operator from $f\in L^1(I)$6 to $f\in L^1(I)$7 (Curbera et al., 2023, Curbera et al., 2019). In the prototypical case $f\in L^1(I)$8, $f\in L^1(I)$9 is Fredholm for $$(H f)(x) = \frac{1}{\pi}\,\mathrm{p.v.} \int_{a}^b \frac{f(y)}{y-x} \, dy,\qquad x\in I.$$0 and injective for $$(H f)(x) = \frac{1}{\pi}\,\mathrm{p.v.} \int_{a}^b \frac{f(y)}{y-x} \, dy,\qquad x\in I.$$1; for $$(H f)(x) = \frac{1}{\pi}\,\mathrm{p.v.} \int_{a}^b \frac{f(y)}{y-x} \, dy,\qquad x\in I.$$2, $$(H f)(x) = \frac{1}{\pi}\,\mathrm{p.v.} \int_{a}^b \frac{f(y)}{y-x} \, dy,\qquad x\in I.$$3 is injective but its range is a proper dense subspace of $$(H f)(x) = \frac{1}{\pi}\,\mathrm{p.v.} \int_{a}^b \frac{f(y)}{y-x} \, dy,\qquad x\in I.$$4. The optimal extension space is

$$(H f)(x) = \frac{1}{\pi}\,\mathrm{p.v.} \int_{a}^b \frac{f(y)}{y-x} \, dy,\qquad x\in I.$$5

which coincides with $$(H f)(x) = \frac{1}{\pi}\,\mathrm{p.v.} \int_{a}^b \frac{f(y)}{y-x} \, dy,\qquad x\in I.$$6 as soon as $$(H f)(x) = \frac{1}{\pi}\,\mathrm{p.v.} \int_{a}^b \frac{f(y)}{y-x} \, dy,\qquad x\in I.$$7 (Curbera et al., 2023).

2. Explicit Inversion Formulas: Single-interval Theory

The inversion formulas bifurcate according to the Fredholm index, governed by the Boyd indices or $$(H f)(x) = \frac{1}{\pi}\,\mathrm{p.v.} \int_{a}^b \frac{f(y)}{y-x} \, dy,\qquad x\in I.$$8-exponent (Curbera et al., 2023, Curbera et al., 2019, Curbera et al., 2023).

Case 1: $$(H f)(x) = \frac{1}{\pi}\,\mathrm{p.v.} \int_{a}^b \frac{f(y)}{y-x} \, dy,\qquad x\in I.$$9 ($g$0)

The kernel is $g$1. The canonical inversion is

$g$2

$g$3 arbitrary. The classically normalized version for Lebesgue $g$4 is

$g$5

Case 2: $g$6 ($g$7)

The kernel is trivial, but the range is the space of $g$8 with

$g$9

The inversion formula is: $I$0 The $I$1 case is exceptional: $I$2 is injective but not surjective, and $I$3.

Zygmund's $I$4 Case

For the Zygmund space $I$5 on $I$6, the operator $I$7 extends continuously into $I$8, with optimality in that no strictly larger rearrangement-invariant domain allows such an extension (Curbera et al., 2022).

The canonical inversion for $I$9 is

$Hf = g\quad \text{on } I$0

with uniqueness modulo the kernel $Hf = g\quad \text{on } I$1.

3. Multi-interval and Vector Hilbert Inversion

For $Hf = g\quad \text{on } I$2 (disjoint), the vector multi-interval finite Hilbert transform is defined via

$Hf = g\quad \text{on } I$3

for $Hf = g\quad \text{on } I$4, with $Hf = g\quad \text{on } I$5 the single-interval transforms.

Inversion with Matrix Coupling

• Symmetric positive-definite $Hf = g\quad \text{on } I$6: Reduction to a Fredholm integral equation $Hf = g\quad \text{on } I$7 with $Hf = g\quad \text{on } I$8 compact. The solution, in terms of a matrix Riemann–Hilbert problem, is unique and can be written explicitly as an integral involving the resolvent kernel and the classical single-interval inversions (Katsevich et al., 2018).
• Uniform $Hf = g\quad \text{on } I$9 ($f$0): The problem reduces via a unitary conjugation to block-diagonal operators, with inversion formulas given by Fourier multiplier methods. Explicit range conditions and invertibility criteria are established. Injectivity follows from convexity arguments on the associated Hilbert form.
• Range Conditions: Characterized precisely via operator-theoretic orthogonality and moment constraints, depending on the case.

4. Spectral Theory and Regularization

The spectral structure of finite/truncated Hilbert transforms is crucial for understanding inversion stability (Al-Aifari et al., 2013, Alaifari et al., 2015). When the data interval only overlaps (does not cover) the support, the associated operator $f$1 has:

• A dense, non-closed range with trivial kernel.
• A singular value decomposition (SVD) tied to eigenfunctions of a self-adjoint Sturm–Liouville operator on $f$2, with discrete spectrum.
• Singular values $f$3 and $f$4, so both 0 and 1 are accumulation points.

A formal inversion via the SVD,

$f$5

is severely ill-posed due to the arbitrarily small $f$6. Regularization (e.g., truncated SVD, Tikhonov, or spectral filtering) is essential for stable inversion. Reconstruction is stably possible only on the overlap (region of interest), with Hölder-type stability rates (Alaifari et al., 2015, Al-Aifari et al., 2013).

5. Weighted and Modified Finite Hilbert Transforms

Weighted Hilbert transforms and interval modifications extend the scope of inversion theory (You, 2020, Ferrari, 2024). For the Chebyshev weight $f$7, the finite Hilbert transform is an isometry (modulo zero-moment conditions) between $f$8 and $f$9. Explicit inversion formulas are available: \begin{align*} &f(t) = -\frac{1}{\pi w(t)} \int_{-1}^1 \frac{F(s)w(s)}{s-t} ds && [f\in L^2_m],\ &f(t) = \frac{1}{\pi}\int_{-1}^1 \frac{F(s)}{w(s)(t-s)} ds && [f\in L^2_d]. \end{align*} For more general weighted Hilbert transforms with $I$0 kernels, iterative inversion schemes converging at geometric rates are described, supporting applications in half-scan tomography (You, 2020).

A modified Hilbert transform $I$1 relevant for time-domain PDE boundary methods is inverted via extension of the data to an odd-$I$2 periodic function and applying the classical Hilbert transform on $I$3, exploiting $I$4 on this function class (Ferrari, 2024).

6. Applications and Numerical Approaches

Tomography and Limited Data Inversion

The finite-support Hilbert inversion arises directly in differentiated back-projection methods for limited data tomography. When available data only partially cover the support (overlap case), inversion is possible only on the region of overlap. Spectral decay of singular values quantifies the severe ill-posedness; regularization with prior bounds (e.g., $I$5 norm constraint or total variation) yields stabilities with Hölder-type rates restricted to the overlap region (Alaifari et al., 2015, Al-Aifari et al., 2013).

Numerical Implementation

High-accuracy numerical inversion employs orthogonal polynomial expansions (Chebyshev, Legendre), with spectral methods exploiting Parseval-like equalities and explicit singular value decompositions. Discretization via Chebyshev–Gauss–Lobatto nodes, sine/cosine transforms, and fast matrix solvers enables inversion with spectral accuracy and computational complexity $I$6 (You, 2020). Regularization strategies include truncating small singular values or Tikhonov filtering.

7. Fundamental Identities and Proof Strategies

Three foundational analytical tools underlie the theory: (1) extended Parseval-type identities, (2) the Poincaré–Bertrand formula,

$I$7

and (3) weighted continuity (Khvedelidze's theorem and variants). These identities enable explicit inversion, Fredholm index computation (including kernel and cokernel identification), and optimal range characterization in rearrangement-invariant spaces (Curbera et al., 2023, Curbera et al., 2022, Curbera et al., 2019, Curbera et al., 2023).

Summary Table: Core Aspects of Finite-support Hilbert Inversion

Aspect	Description	Primary Source(s)
Operator class	$I$8	(Curbera et al., 2023)
Explicit inverse (1<p<2)	$I$9	(Curbera et al., 2019)
Kernel characterization	$X$0 for $X$1, trivial otherwise	(Curbera et al., 2022)
Multi-interval generalization	Via vector Hilbert systems, Riemann–Hilbert techniques	(Katsevich et al., 2018)
Spectral properties	SVD via Sturm–Liouville theory; singular values $X$2	(Al-Aifari et al., 2013)
Regularization necessity	Truncated SVD, Tikhonov, TV penalties for stable inversion	(Alaifari et al., 2015)
Weighted/modified variants	Chebyshev weights, cosh-weights, periodic and PDE-motivated mods	(You, 2020, Ferrari, 2024)

References are to arXiv IDs in the dataset, e.g., (Curbera et al., 2023).

---

The finite-support Hilbert inversion theory thus unites operator theory, spectral analysis, functional analysis, and computational mathematics, with deep implications for stability, optimality, and numerical realization in both single- and multi-interval settings.