Krein–Feller Operator

• Krein–Feller operator is a canonical second-order differential operator defined via singular measures on compact intervals, generalizing Sturm–Liouville operators.
• It employs a μ-derivative and form methods to ensure self-adjointness with discrete eigenvalues under Dirichlet or Neumann conditions.
• Extensions cover fractal, random, and V-variable measures with applications in multifractal analysis, gap diffusions, and stochastic partial differential equations.

A Krein–Feller operator is a canonical second-order differential operator associated with a (typically singular or fractal) Borel measure, serving as a measure-geometric generalization of classical Sturm–Liouville operators. Given a finite atomless Borel measure $\mu$ on a compact interval $[a,b]$, the Krein–Feller operator formally takes the form $L_\mu = \frac{d}{d\mu}\frac{d}{dx}$ and may be rigorously defined as a self-adjoint operator on $L^2(\mu)$. The theory encompasses the definition and properties of such operators, their spectrum under various boundary conditions, explicit constructions for self-similar, random recursive, and $V$-variable Cantor measures, as well as spectral asymptotics and links to multifractal analysis and branching processes (Minorics, 2017). Extensions to generalized settings (gap diffusions, Riemannian manifolds, fractal settings, stochastic analysis) have been developed and now form a rich interface between spectral theory, probability, geometric measure theory, and mathematical physics.

1. Operator Definition and Functional Framework

The Krein–Feller operator $L_\mu$ is defined on a compact interval $[a,b]$ with respect to a finite, atomless Borel measure $\mu$. The notion of a $\mu$-derivative is introduced via the space

$$D_1^\mu = \{f \colon [a,b] \to \mathbb{R}: \exists f^\mu \in L^2(\mu) \text{ with } f(x) = f(a) + \int_a^x f^\mu(y)\, d\mu(y)\}.$$

For $f \in D_1^\mu$, the function $f^\mu$ (unique in $L^2(\mu)$) is the $\mu$-derivative $df/d\mu$. The suitable operator domain is

$D_2^{\mu,\lambda^1} = \{f \in C^1((a,b)) \cap C^0([a,b]): f' \text{ admits a %%%%14%%%%-derivative } (f')^\mu \in L^2(\mu)\}.$

The operator $L_\mu$ acts on $D_2^{\mu,\lambda^1}$ as $L_\mu f = (f')^\mu$. With Dirichlet ($f(a)=f(b)=0$) or Neumann ($f'(a)=f'(b)=0$) boundary conditions, $-L_\mu$ is self-adjoint with purely discrete spectrum and compact resolvent (Minorics, 2017, Freiberg et al., 2019, Jorgensen et al., 2022).

The modern theoretical framework emphasizes form methods: $L_\mu$ is the unique nonnegative self-adjoint operator associated with the closed quadratic form $\mathcal{E}(f,g) = \int_a^b f'(x)g'(x)\,dx$ on $H^1([a,b])$ (Neumann) or $H_0^1([a,b])$ (Dirichlet), realized in $L^2(\mu)$.

2. Spectral Theory and Eigenvalue Structure

The Krein–Feller operator under Dirichlet or Neumann conditions has discrete, simple spectrum with $$0 \leq \lambda_1 \leq \lambda_2 \leq \cdots \to \infty$$. The eigenvalue problem is

$\frac{d}{d\mu}\frac{d}{dx} f = -\lambda f,$

interpreted in either of the above senses. Measure-theoretic sine and cosine functions are constructed by alternating $\mu$ and Lebesgue integrals (see (Freiberg et al., 2019)), yielding entire functions whose zeros correspond to the square roots of eigenvalues:

• Neumann: zeros of $\sin_p(z) = sp_z(1)$
• Dirichlet: zeros of $\sin_q(z) = sq_z(1)$

The spectral counting function, $N_D^\mu(x) = \#\{n : \lambda_{D,n}^\mu \le x\}$ (and similarly for Neumann), admits asymptotic analysis via a spectral exponent

$$\gamma(\mu) = \lim_{x \to \infty} \frac{\log N_D^\mu(x)}{\log x}$$

if the limit exists. For Lebesgue measure, $\gamma = 1/2$ (Weyl's law), but for singular or fractal $\mu$, $\gamma$ is nontrivial and reflects the underlying measure's geometric properties (Minorics, 2017, Kesseböhmer et al., 2021).

3. Self-Similar, Random, and $V$-Variable Measures

For self-similar measures defined by an IFS $\{S_i(x) = r_i x + c_i\}$ with weights $m_i > 0$, the spectral exponent $\gamma$ is the unique solution of

$\sum_{i=1}^N (m_ir_i)^\gamma = 1.$

Random constructions include:

• Random homogeneous Cantor measures: At each level, the IFS is chosen i.i.d. according to a probability vector $p_j$ over a finite set $J$; $\gamma_h$ solves

$$\prod_{j \in J} \left( \sum_{i=1}^{N_j} (r_i^{(j)} m_i^{(j)})^{\gamma_h} \right)^{p_j} = 1.$$

• Random recursive Cantor measures: Each node independently selects its IFS type; $\gamma_r$ is defined by

$$\mathbb{E}\left[ \sum_{i=1}^{N_{U_\emptyset}} (r_i^{(U_\emptyset)} m_i^{(U_\emptyset)})^{\gamma_r} \right] = 1.$$

• $V$-variable Cantor measures: A hierarchical mixture between homogeneous and recursive, with spectral exponent $\gamma$ characterized via an averaging functional over the random tree's neck levels (Minorics, 2018).

Spectral asymptotics for these classes have been developed using Dirichlet–Neumann bracketing techniques and Crump–Mode–Jagers branching process tools. For random recursive and $V$-variable measures, one obtains almost-sure power-law asymptotics for the eigenvalue counting function, with explicit formulae for $\gamma$ depending on the measure's random structure (Minorics, 2017, Minorics, 2018).

4. Multifractal and Spectral Dimension Connections

The spectral asymptotics of Krein–Feller operators exhibit deep connections to multifractal analysis and $L^q$-spectra. For self-conformal and weak Gibbs measures (possibly with overlaps), one defines the $L^q$-spectrum $\tau_\rho(q)$ and shows existence of a unique $s$ with $\tau_\rho(s)=0$, the spectral dimension. For measures $\mu$ on the unit interval, the upper spectral dimension $s(\mu)$ corresponds to the fixed point of the $L^q$-spectrum: $s(\mu) = \min\{ q>0 : \tau_\mu(q) < q \}$, with sharp connections to the Minkowski dimension of the support and explicit bounds. This identification is robust under general settings, including absolutely continuous, self-similar, and purely singular measures (Kesseböhmer et al., 2021, Kesseböhmer et al., 2021).

In higher dimensions, a spectral partition function $\zeta_\nu(s)$ determines the spectral dimension as its unique zero, with existence and equality between Dirichlet and Neumann exponents under regularity conditions. The form method breaks down below the critical lower $\infty$-dimension threshold $d-2$ (Kesseböhmer et al., 2022).

5. Extensions: Generalized Settings and Manifolds

The Krein–Feller construction extends to generalized settings:

• Generalized Krein–Feller operators: Given atomless measures $\mu$ and $\nu$ on $[0,1]$, define

$$\Delta_{\nu,\mu} f(x) = g(x) \text{ via } f(x) = a+bF_\mu(x) + \int_{[0,x]} (F_\mu(x) - F_\mu(y)) g(y) d\nu(y),$$

reducing spectral problems via isometries induced by $F_\mu$ to those for Krein–Feller operators associated with Lebesgue measure (Kesseböhmer et al., 2019).

• Gap diffusions: The (generalized) Krein–Feller operator acts as the generator of continuous strong Markov processes with nontrivial speed measures, with spectral data governing long-term behavior (Kesseböhmer et al., 2019).
• Riemannian manifolds and forms: The Krein–Feller operator is defined via the form-method on $L^2(\mu)$ for a finite Borel measure $\mu$ on (open subsets of) smooth compact Riemannian manifolds, with sufficient Poincaré-type inequalities and lower dimension conditions (dimension $> n-2$) guaranteeing compact resolvent, discrete spectrum, and—on forms—full Hodge decomposition (Ngai et al., 2023, Ngai et al., 10 Mar 2024, Ngai et al., 21 Nov 2024, Ngai et al., 13 Dec 2024).
• Stochastic and SPDE theory: The operator admits functional analytic and spectral theory even in highly singular or discontinuous settings (e.g., periodic torus with singular measures), with explicit Taylor-type expansions, and describes regularity, traceability, and existence for fractional deterministic and stochastic PDEs in nuclear test-function spaces (Almeida-Sousa et al., 4 Dec 2025).

6. Eigenstructure, Nodal Geometry, and Regularity

The spectral theory delivers explicit characterizations of eigenvalues and eigenfunctions in terms of measure-adapted entire series generalizing classical trigonometric functions (Freiberg et al., 2019, Almeida-Sousa et al., 4 Dec 2025). Eigenvalues are simple and separated, eigenfunctions are continuous under appropriate Green-function and measure-regularity assumptions, and the nodal set of a continuous $\lambda_n$-eigenfunction partitions the domain into at least $2$ and at most $n+r-1$ nodal domains, generalizing Courant's theorem to singular or fractal measures and to manifolds. Critical tools include maximum principles for $\mu$-subharmonic functions and careful analysis of the Green's operator in both Euclidean and Riemannian settings (Ngai et al., 21 Nov 2024, Ngai et al., 13 Dec 2024).

7. Comparative Analysis and Open Directions

Comparisons between classes reveal that the spectral exponent for recursive measures is always at least as large as that for homogeneous measures, with equality only for highly rigid self-similar structures. Recursive (branching) randomness yields fluctuations in the eigenvalue counting function asymptotics—sometimes a.s. limits times random martingales, sometimes only two-sided sub-Gaussian deviations (Minorics, 2017). Boundary case analyses show that below critical lower $L^\infty$-dimension, or for atomic measures, the form approach fails and no discrete spectrum arises (Kesseböhmer et al., 2022, Ngai et al., 2023).

Prominent open problems include quantitative Weyl-type asymptotics for curved or high-dimensional singular geometries, detailed heat kernel estimates for singular and fractal media, refinement of nodal geometry in the irregular setting, and deeper integration with probability (random media, stochastic processes) and mathematical physics (quantum graphs, quantum gravity scenarios).

---

Table: Spectral Exponent Formulas for Key Measure Classes

Measure Class	Spectral Exponent $\gamma$ Condition	Reference
Self-similar (IFS)	$\sum_{i}(m_ir_i)^\gamma=1$	(Minorics, 2017)
Random homogeneous Cantor	$$\prod_j\left(\sum_i(r_i^{(j)}m_i^{(j)})^{\gamma_h}\right)^{p_j}=1$$	(Minorics, 2017)
Random recursive Cantor	$$\mathbb{E}\left[\sum_i(r_i^{(U_\emptyset)}m_i^{(U_\emptyset)})^{\gamma_r}\right]=1$$	(Minorics, 2017)
$V$-variable Cantor	$E_V\log\sum_{|i|=n(1)}(r_im_i)^\gamma=0$	(Minorics, 2018)
Weak Gibbs/IFS, multidimensional	$\tau_\rho(\gamma)=0$ ($L^q$-spectrum zero)	(Kesseböhmer et al., 2021)

These formulations reveal the structure of spectral asymptotics across deterministic, random, and multifractal measure classes. They are central to the current landscape and future research in the spectral theory of Krein–Feller and related operators.