Eigenfunction Localization Transitions

• Eigenfunction Localization Transitions are abrupt changes in the spatial concentration of eigenfunctions that occur when system parameters or domain geometries vary.
• They are quantified using measures such as κ‑localization, participation ratios, and Hardy inequalities, which rigorously delineate regimes from full localization to complete delocalization.
• These transitions influence quantum confinement, wave propagation in complex media, and spectral behavior in disordered systems, providing a predictive framework for both theoretical and applied research.

Eigenfunction localization transitions refer to sharp changes in the spatial or basis-space concentration properties of the eigenfunctions of linear operators—especially differential operators and random matrices—as system parameters or domain geometry are varied. These transitions are central to the mathematical understanding of quantum confinement, wave propagation in complex media, and the emergence of insulating or conducting behavior in disordered and structured systems. Recent research has established precise geometric, analytical, and probabilistic criteria for localization transitions in a wide range of settings, from Laplace operators on engineered domains to hierarchical random matrices.

1. Rigorous Notions of Localization and κ-localization

To quantify eigenfunction localization, van den Berg and Bucur introduced the κ-localization framework in $L^2$ for eigenfunctions $u_n\in L^2(\Omega_n)$ on a sequence of domains $\Omega_n$ (Berg et al., 2022). For sequences of measurable sets $A_n\subset\Omega_n$ with $|A_n|/|\Omega_n|\to0$, the localization coefficient is

$$\kappa=\sup\left\{\limsup_{n\to\infty}\frac{\|u_n1_{A_n}\|_2^2}{\|u_n\|_2^2} : (A_n)\right\}$$

with:

• Full localization: $\kappa=1$: almost all $L^2$ mass in vanishingly small sets.
• No localization: $\kappa=0$.
• $\kappa$-localization: $u_n\in L^2(\Omega_n)$0.

This definition enables rigorous statements about when and how eigenfunctions concentrate in small subregions, and supports continuity arguments to interpolate between completely delocalized and fully localized regimes.

2. Geometric and Analytical Criteria for Localization

A central result is the geometric localization criterion via Hardy inequalities (Berg et al., 2022). For domains $u_n\in L^2(\Omega_n)$1 supporting a uniform Hardy inequality,

$u_n\in L^2(\Omega_n)$2

if there exist sets $u_n\in L^2(\Omega_n)$3, $u_n\in L^2(\Omega_n)$4, with $u_n\in L^2(\Omega_n)$5, then all eigenfunctions $u_n\in L^2(\Omega_n)$6 localize: the $u_n\in L^2(\Omega_n)$7 mass escapes $u_n\in L^2(\Omega_n)$8 and concentrates in vanishingly small $u_n\in L^2(\Omega_n)$9.

The main technique leverages Cauchy-Schwarz, the Hardy inequality, spectral estimates, and the geometry of the domain, allowing explicit construction of sequences where the transition from delocalization to localization tracks a geometric parameter (e.g., horn length, neck thickness).

3. Model Systems Exhibiting Localization Transitions

a) Elongating Horn-shaped Domains

Horn-shaped domains $\Omega_n$0 parametrized by an elongation $\Omega_n$1 exhibit a transition: as $\Omega_n$2 and the cross-sectional width shrinks, Dirichlet eigenfunctions fully localize in a vanishing region (Berg et al., 2022). The participation ratio

$\Omega_n$3

serves as a strict marker.

b) Polygonal/Lobed Planar Domains

In polygonal domains engineered with two lobes connected via a narrow neck, tuning the neck width causes the first eigenfunction to rapidly shift its support from extended to being trapped in one lobe (Dirichlet), or to interpolate between stalk and head (Neumann). Such transitions can realize any $\Omega_n$4 (Berg et al., 2022).

c) Random Matrix Ensembles

In hierarchical (ultrametric) ensembles, the decay exponent $\Omega_n$5 controls localization:

• $\Omega_n$6: Exponential localization, Poisson statistics (Soosten et al., 2017).
• $\Omega_n$7: Full delocalization, Wigner-Dyson universality.

The transition is sharply defined and analytically tractable, in contrast to standard (Dyson) ensembles.

The Rosenzweig-Porter (RP) ensemble, with variance scaling $\Omega_n$8, exhibits two transitions (Kravtsov et al., 2015):

• $\Omega_n$9: Anderson localization (Poisson→critical).
• $A_n\subset\Omega_n$0: Multifractal (non-ergodic extended)→ergodic extended (GOE-like).

The accompanying multifractality spectrum $A_n\subset\Omega_n$1, inverse participation ratios, and spectral statistics provide a full characterization.

4. Analytical and Quantitative Transition Markers

Participation ratios (IPR), Shannon entropies, and entropic localization lengths are central diagnostic tools. For ER random networks, the universal scaling variable is the average degree $A_n\subset\Omega_n$2 (Mendez-Bermudez et al., 2015):

Regime	$A_n\subset\Omega_n$3 (ER graphs)	$A_n\subset\Omega_n$4 (entropic loc. length)
Localized	$A_n\subset\Omega_n$5	$A_n\subset\Omega_n$6
Broad crossover	$A_n\subset\Omega_n$7	increases from 0 to 1
Fully extended	$A_n\subset\Omega_n$8	$A_n\subset\Omega_n$9

For Sturm-Liouville settings, Karamehmedović and Triki define the localization coefficient

$|A_n|/|\Omega_n|\to0$0

with explicit non-asymptotic and asymptotic bounds, and apply landscape function techniques to detect transitions (Karamehmedović et al., 2023).

In convex geometry, Beck proves a non-trivial lower bound for the $|A_n|/|\Omega_n|\to0$1 norm of the first Dirichlet eigenfunction, scaling as a one-sixth power of the domain eccentricity, confirming van den Berg's conjecture and quantifying the impossibility of arbitrarily sharp localization in elongated convex domains (Beck, 2019).

5. Structural and Spectral Mechanisms

Transitions arise from spectral geometry (level crossings, spectral gaps), entropy and participation measures, and probabilistic arguments (heat kernels, Brownian trapping). In non-Hermitian systems, complex band structure separates skin effect (accumulation at the boundary), true bulk localization (defect-induced), and tunneling regimes; analytic formulas for decay exist via complex Bloch-Floquet theory (Bruijn et al., 29 May 2025).

Boundary irregularity (cones, slits, fractal boundaries) induces Neumann eigenfunction localization through long Brownian sojourn times and enhanced heat kernel return probability, with quantitative boosts scaling in aperture or gap (Jones et al., 2018).

Boundary-localized transmission eigenfunctions in stratified balls emerge due to the material parameter monotonicity and explicit matching conditions in the ODEs governing the fields (Jiang et al., 2022).

In semiclassical non-selfadjoint PDEs, distinct localization scales arise: naive estimates identify an $|A_n|/|\Omega_n|\to0$2 (Airy) scale, but a sharp $|A_n|/|\Omega_n|\to0$3 scale is proven via Agmon-type estimates and operator-valued pseudodifferential calculus (Averseng et al., 23 Dec 2025).

6. Multiphase and Nonstandard Localization Transitions

Quadratic models (e.g., Anderson or Wannier–Stark under nonstandard scaling) can display transitions between localization in complementary bases: e.g., quasimomentum-localized $|A_n|/|\Omega_n|\to0$4 position-localized, with critical points where some diagnostics approach RMT universal predictions while others do not, exhibiting "Janus" character (Lisiecki et al., 2024).

In bosonic embedded ensembles (BEGOE), strength functions and participation ratios transition through Poisson → GOE → thermodynamic duality as the two-body interaction increases, with clear markers obtained analytically from the variance propagator (Chavda et al., 2016). Entanglement entropy transitions from area- to volume-law in the chaotic regime, correlating strongly with participation ratio.

7. Mathematical and Physical Implications

The theory of eigenfunction localization transitions provides a quantitative, predictive framework linking spectral properties, domain or network geometry, operator structure, and statistical mechanics. An essential impact is the ability to design or predict confining modes, to identify universal scaling variables (e.g., $|A_n|/|\Omega_n|\to0$5 for graphs, decay exponent $|A_n|/|\Omega_n|\to0$6 for ultrametric matrices), and to control transitions by geometric, analytic, or probabilistic mechanisms.

These results are foundational for the mathematical analysis of Anderson localization, quantum chaos, fractal acoustics, random graph theory, mesoscopic transport, and non-Hermitian physics, and establish robust methodologies for demonstrating, quantifying, and exploiting localization transitions across disparate mathematical and physical systems.