Labyrinth: Theory, Patterns, and Applications

• Labyrinth is a polysemous research term defined by maze-like spatial organization and recursive path structures across multiple scientific disciplines.
• It underpins applications from quasicrystal Hamiltonians and dendritic fractals to nanophotonic cavity designs, highlighting its role in engineered and natural patterns.
• Studies employ methods from mathematical physics, dynamical systems, and materials informatics to analyze symmetry, connectivity, and anomalous transport properties.

“Labyrinth” is a polysemous research term whose meanings range from a philosophical metaphor for conceptual entanglement in quantum theory to a precise name for quasicrystal Hamiltonians, quasiperiodic tilings, dendritic fractals, maze-like pattern morphologies, chaotic flows, photonic cavity geometries, and benchmark navigation tasks. Across these usages, the term typically denotes either a maze-like spatial organization, a recursively constrained path structure, or a system whose local rules generate globally intricate traversal behavior (Maudlin, 2018).

1. Terminological scope across disciplines

In contemporary research usage, “labyrinth” does not denote a single formal object. It appears in at least six distinct but structurally resonant senses.

Domain	Meaning of “labyrinth”	Representative source
Philosophy of quantum theory	A metaphor for conceptual temptations around quantum logic	(Maudlin, 2018)
Mathematical physics	A two-dimensional quasicrystal Hamiltonian with product spectrum	(Takahashi, 2016)
Dynamical systems	A Thomas–Rössler continuous-time flow with wandering chaotic trajectories	(Basios et al., 2020)
Pattern formation	Maze-like, wavelength-selected but globally non-crystalline structures	(Clerc et al., 2021)
Fractal geometry	Dendritic limit sets generated by labyrinth patterns	(Cristea et al., 2020)
Nanophotonics	Circular Bragg gratings with rotated bridge connections enabling contacts	(Buchinger et al., 2022)

Tim Maudlin’s discussion of Hilary Putnam’s engagement with quantum logic gives “labyrinth” an explicitly conceptual meaning: the maze is created by the temptation to treat the strangeness of quantum mechanics as evidence that logic itself must be revised, rather than that physical theory must be clarified (Maudlin, 2018). In contrast, the “Labyrinth model” of quasicrystal theory is a separable two-dimensional Hamiltonian built from one-dimensional off-diagonal substitution models, while “labyrinth chaos” denotes a continuous flow introduced by Otto Rössler and René Thomas in their study of the conditions for chaos and hyperchaos (Takahashi, 2016, Basios et al., 2020).

In morphology and geometry, “labyrinthine” denotes curved, interconnected stripe fragments or domains with short-range order and a characteristic wavelength but without simple crystalline symmetry; in fractal theory it denotes recursive white-square constructions satisfying tree, exit, and corner conditions (Clerc et al., 2021, Cristea et al., 2020). In device physics, the word is used literally for a circular Bragg grating geometry in which rotated bridges create a maze-like conductive path while suppressing straight optical waveguiding (Buchinger et al., 2022). This suggests that the common semantic core of the term is not merely visual complexity, but constrained connectivity with nontrivial path structure.

2. Labyrinthine morphology in pattern-forming media

In dissipative pattern formation, a labyrinthine pattern is not treated as random disorder. It is an extended pattern with non-trivial symmetry, a preferred wavelength, local stripe-like organization, many orientations, and short-range order; its Fourier spectrum forms a powder-like ring in two dimensions, or a sphere in three dimensions, rather than a few symmetry-related Bragg peaks (Clerc et al., 2021). On that basis, localized labyrinthine patterns are defined as finite labyrinth-like patches embedded in an otherwise homogeneous steady state.

For the cubic Swift–Hohenberg equation,

$$\partial_t u = \epsilon u - u^3 - \nu \nabla^2 u - \nabla^4 u,$$

stable localized labyrinthine patterns were reported in a narrow pinning range. For the frequently used case $\nu=1$, the paper identifies

$\epsilon_p^- = 1.16, \qquad \epsilon_p^+ = 1.19,$

inside which stable localized labyrinthine patterns exist, and interprets them as a manifestation of a pinning–depinning transition between a homogeneous state and an extended labyrinthine state (Clerc et al., 2021). The same work reports analogous structures in a generalized non-variational Swift–Hohenberg equation, as well as in models for plant ecology, nonlinear optics, reaction–diffusion systems, and in three-dimensional tubular forms.

A related but distinct use of the term occurs in ferroelectric nanostructures. Within a Landau–Ginzburg–Devonshire description of a spherical $\mathrm{Sn_2P_2S_6}$ core-shell nanoparticle, “labyrinth” denotes a stable multidomain state of the uniaxial polarization $P_3(\mathbf r)$ consisting of a highly branched, irregular network of alternating up and down domains (Morozovska et al., 2022). The reported regime is

$$R \approx 5 - 20\ \mathrm{nm}, \qquad \lambda \approx 5 - 15\ \mathrm{pm},$$

where $R$ is the core radius and $\lambda$ the screening length. The equilibrium labyrinths emerge from arbitrarily small randomly oriented nanodomains, and the paper argues that they form a quasi-continuum of nearly degenerate states. It also reports that applying and removing a homogeneous electric field can reset and bias the maze polarity, and that for

$$\tau < \tau_S < 10\tau, \qquad 10^{-3} \le \omega\tau \le 10^{-1},$$

the quasi-static dielectric susceptibility can become negative, with possible negative-capacitance implications (Morozovska et al., 2022).

Labyrinthine morphology also appears as a reconstruction target in materials informatics. A binary labyrinth pattern of size $100\times 100$, cut from a binarized magnetic-force-microscopy image of Co/Pt-based multilayers, was reconstructed using simulated annealing driven by entropic descriptors, with the mixed biased/random #2m scenario giving the best average tolerance and the fastest performance under the tested conditions (Piasecki et al., 2011). The target had black-phase volume fraction $\nu=1$0, black-pixel count $\nu=1$1, and first $\nu=1$2 peak at $\nu=1$3 (Piasecki et al., 2011). Taken together, these results establish “labyrinth” as a technical descriptor for stable, non-crystalline yet strongly organized morphology rather than as a synonym for disorder.

3. Labyrinth fractals, dendrites, and recursive path geometry

Labyrinth fractals are planar dendrites generated from white–black square patterns satisfying three combinatorial constraints: the white-square graph is a tree, there is exactly one horizontal exit pair and one vertical exit pair, and if one corner square is white then the diagonally opposite corner square is not white (Cristea et al., 2020). Starting from a sequence of patterns $\nu=1$4 with widths $\nu=1$5, the level-$\nu=1$6 white set $\nu=1$7 yields a decreasing sequence of compact sets $\nu=1$8, and the limit

$\nu=1$9

is the mixed labyrinth fractal (Cristea et al., 2020).

A central topological fact is that mixed labyrinth fractals are dendrites, hence connected, locally connected, loop-free, and uniquely arcwise connected (Cristea et al., 2020). The unique arc between two points is approximated by the unique path in the tree $\epsilon_p^- = 1.16, \qquad \epsilon_p^+ = 1.19,$0 joining the corresponding level-$\epsilon_p^- = 1.16, \qquad \epsilon_p^+ = 1.19,$1 squares. Path combinatorics are encoded by $\epsilon_p^- = 1.16, \qquad \epsilon_p^+ = 1.19,$2 path matrices $\epsilon_p^- = 1.16, \qquad \epsilon_p^+ = 1.19,$3, and in the mixed case these multiply exactly: $\epsilon_p^- = 1.16, \qquad \epsilon_p^+ = 1.19,$4 The lower and upper box dimensions of exit-to-exit arcs are determined by the asymptotic growth of the corresponding path lengths relative to the scale factor $\epsilon_p^- = 1.16, \qquad \epsilon_p^+ = 1.19,$5 (Cristea et al., 2020).

The later supermixed theory allows several different patterns to be used simultaneously at a given level, one choice for each white square. In that setting, path matrices no longer suffice, and the paper introduces counting matrices $\epsilon_p^- = 1.16, \qquad \epsilon_p^+ = 1.19,$6 to record how many path-squares of each type are refined by each admissible pattern. The key recursion becomes

$\epsilon_p^- = 1.16, \qquad \epsilon_p^+ = 1.19,$7

rather than a single product (Cristea et al., 2018). Even under this broader square-by-square heterogeneity, the resulting supermixed labyrinth fractal remains a dendrite, and the coordinate formulas for exits and the box-dimension formulas for exit arcs extend from the mixed and self-similar cases (Cristea et al., 2018).

A major metric subtlety is that blockedness does not have the same consequences in all regimes. In the self-similar case, a horizontally and vertically blocked pattern forces every arc to have infinite length. In the mixed case, however, both possibilities can occur: all arcs finite or all arcs infinite, depending on the sequence of widths and patterns (Cristea et al., 2018). The paper “On the length of arcs in labyrinth fractals” shows explicitly that even when every pattern is horizontally and vertically blocked, mixed labyrinth fractals can have finite arcs if the widths grow fast enough, because the cumulative stretch factor

$\epsilon_p^- = 1.16, \qquad \epsilon_p^+ = 1.19,$8

can converge when $\epsilon_p^- = 1.16, \qquad \epsilon_p^+ = 1.19,$9 (Cristea et al., 2018). Conversely, if $\mathrm{Sn_2P_2S_6}$0, then the repeated detours accumulate and all arcs can be infinite; a sufficient-condition theorem of precisely this form is established for the mixed case in the supermixed-fractal paper (Cristea et al., 2018).

Subsequent generalizations broaden the geometry and stochastic structure. Quadrilateral labyrinth fractals replace the unit square by an arbitrary convex quadrilateral and preserve the labyrinth topology even though the resulting sets are generally not self-similar (Gopalakrishnan et al., 2021). Randomised mixed labyrinth fractals choose patterns independently at each level according to fixed probabilities and study shortest-path dimension, arc dimension, and isotropy restoration. In the equal-width two-pattern case, the approximated path matrix

$\mathrm{Sn_2P_2S_6}$1

is proposed as an effective surrogate for large random products, and the resulting $\mathrm{Sn_2P_2S_6}$2 can be constant, linearly monotone, nonlinearly monotone, or nonmonotone with an interior maximum (Prehl et al., 5 Jun 2026). A plausible implication is that the labyrinth-fractal program has shifted from purely deterministic self-similar geometry toward a broader theory of finitely ramified path structures with controlled disorder.

4. Quasicrystal Hamiltonians, labyrinth tilings, and anomalous transport

In mathematical physics, the “Labyrinth model” is a separable two-dimensional quasicrystal Hamiltonian obtained from two one-dimensional off-diagonal metallic-mean substitution models (Takahashi, 2016). The one-dimensional operator is

$\mathrm{Sn_2P_2S_6}$3

with $\mathrm{Sn_2P_2S_6}$4 in the hull of a metallic-mean substitution sequence generated by

$\mathrm{Sn_2P_2S_6}$5

After scaling $\mathrm{Sn_2P_2S_6}$6, the coupling constant is

$\mathrm{Sn_2P_2S_6}$7

From two such sequences $\mathrm{Sn_2P_2S_6}$8, the two-dimensional Labyrinth operator $\mathrm{Sn_2P_2S_6}$9 acts on $P_3(\mathbf r)$0 by diagonal hopping only, with amplitudes equal to products of the one-dimensional edge weights (Takahashi, 2016).

The decisive structural fact is the unitary equivalence

$P_3(\mathbf r)$1

which yields the product-spectrum formula

$P_3(\mathbf r)$2

Because the one-dimensional spectra are dynamically defined Cantor sets for every $P_3(\mathbf r)$3, the two-dimensional spectrum is the product of two Cantor sets; nevertheless, it becomes an interval for sufficiently small couplings and a zero-measure Cantor set for sufficiently large couplings (Takahashi, 2016). The density of states measure also admits a multiplicative formula,

$P_3(\mathbf r)$4

and is absolutely continuous with respect to Lebesgue measure for almost every pair of sufficiently small couplings (Takahashi, 2016).

A related higher-dimensional use of the term appears in “labyrinth tilings,” where one-dimensional metallic-mean chains are combined by direct products and only the diagonal bonds are retained (Thiem et al., 2012). These systems are also separable: eigenstates are products of one-dimensional eigenstates, and $P_3(\mathbf r)$5-dimensional energies are products,

$P_3(\mathbf r)$6

Wave-packet transport is characterized by the mean-square displacement

$P_3(\mathbf r)$7

with $P_3(\mathbf r)$8 indicating anomalous diffusion (Thiem et al., 2012). The paper extends renormalization-group arguments from the golden-mean chain to the silver-mean chain and to higher-dimensional labyrinth tilings, obtaining the scaling relation

$P_3(\mathbf r)$9

where $$R \approx 5 - 20\ \mathrm{nm}, \qquad \lambda \approx 5 - 15\ \mathrm{pm},$$0 is the length rescaling factor and $$R \approx 5 - 20\ \mathrm{nm}, \qquad \lambda \approx 5 - 15\ \mathrm{pm},$$1 the energy rescaling factor (Thiem et al., 2012). A notable conclusion is that the transport exponents in one, two, and three dimensions are very close, reflecting the separable product structure rather than conventional dimensional enhancement.

5. Labyrinths as dynamical systems, navigation tasks, and control benchmarks

In nonlinear dynamics, “labyrinth chaos” or the “labyrinth walks system” refers to a Thomas–Rössler flow introduced in the search for minimal mathematical conditions that generate chaos and hyperchaos in ODEs (Basios et al., 2020). For the $$R \approx 5 - 20\ \mathrm{nm}, \qquad \lambda \approx 5 - 15\ \mathrm{pm},$$2-dimensional family

$$R \approx 5 - 20\ \mathrm{nm}, \qquad \lambda \approx 5 - 15\ \mathrm{pm},$$3

the labyrinth-walks regime is $$R \approx 5 - 20\ \mathrm{nm}, \qquad \lambda \approx 5 - 15\ \mathrm{pm},$$4. In three dimensions this becomes

$$R \approx 5 - 20\ \mathrm{nm}, \qquad \lambda \approx 5 - 15\ \mathrm{pm},$$5

The system has an infinite cubic lattice of equilibria $$R \approx 5 - 20\ \mathrm{nm}, \qquad \lambda \approx 5 - 15\ \mathrm{pm},$$6, is divergence-free, and exhibits chaotic wandering without a chaotic attractor (Basios et al., 2020). The paper stresses a useful distinction: the system is volume-preserving because $$R \approx 5 - 20\ \mathrm{nm}, \qquad \lambda \approx 5 - 15\ \mathrm{pm},$$7, but it is not force-conservative, since the line integral around a closed path does not vanish and no global scalar potential exists (Basios et al., 2020). This directly counters the common conflation of incompressibility with mechanical conservativity.

The term also appears in machine learning and control as a benchmark navigation environment. In the ViZDoom “My Way Home” scenario, the labyrinth has 8 rooms and 10 corridors, the observation is first-person vision of size $$R \approx 5 - 20\ \mathrm{nm}, \qquad \lambda \approx 5 - 15\ \mathrm{pm},$$8, the action set is $$R \approx 5 - 20\ \mathrm{nm}, \qquad \lambda \approx 5 - 15\ \mathrm{pm},$$9move forward, turn left, turn right$R$0, and episodes end at $R$1 if the goal is not found (Bayer et al., 2024). Against the prevailing assumption that such partially observable visual labyrinths require recurrent deep RL, the paper argues that navigation can emerge from a Braitenberg-style reactive heuristic evolved in tangled program graphs. It reports that the evolved modular indexing scheme uses only $R$2 of the state space per decision and substantially outperforms a memoryless DQN baseline on the task (Bayer et al., 2024). The paper’s own interpretation is that the controller relies less on “seeing” the goal than on corridor following, wall alignment, arcing, and room-based reorientation.

Real-world labyrinth control appears in two further studies of the BRIO wooden maze. One uses DreamerV3 with low-dimensional state estimates, a $R$3 rectified local image patch centered on the ball, progress reward $R$4, and symmetry-based data augmentation; trained entirely on the physical system, it achieves a $R$5 success rate over 50 runs after $R$6 control steps, corresponding to $R$7 hours at $R$8 Hz, with average successful completion time $R$9 s (Bi et al., 2023). The other formulates the BRIO game as a nonlinear non-convex MPC problem with adaptive obstacle constraints for holes and walls, then splits computation into a high-level planner and a low-level tracker. Over 25 runs, the nonlinear MPC achieves $\lambda$0 full completion, compared with $\lambda$1 for linear MPC and $\lambda$2 for cascaded PID, and reduces average completion time from $\lambda$3 s to $\lambda$4 s among successful runs (Gaber et al., 2024). These two lines of work treat the labyrinth not as a metaphor but as a physically grounded benchmark for sample-efficient learning and real-time constrained control.

6. Conceptual labyrinths in quantum theory and engineered labyrinths in nanophotonics

Maudlin’s philosophical use of “labyrinth” concerns quantum foundations rather than geometry. His title deliberately names a conceptual maze created by quantum mechanics, especially the temptation to escalate from conflict with classical physics to revisions of classical logic or classical probability theory (Maudlin, 2018). The paper distinguishes sharply among “classical physics,” “classical probability theory,” and “classical logic,” arguing that they are not interchangeable, and uses Putnam’s trajectory from early quantum-logic enthusiasm to later rejection of the program as a case study in the difficulty of solving physical problems by altering logic (Maudlin, 2018). The formal discussion centers on the lattice of subspaces of Hilbert space, with “meet, join, and orthocomplement” and the failure of distributivity, but the conclusion is explicitly anti-revisionary: “physics cannot be simplified or improved by changing logic” (Maudlin, 2018). In this usage, the labyrinth is a warning against conceptual overreaction.

In nanophotonics, by contrast, a “labyrinth geometry” is a concrete cavity design. Circular Bragg gratings normally have fully etched concentric gaps that isolate the central disk electrically. To create an electrical path, one can connect adjacent rings with narrow semiconductor bridges; if those bridges line up radially, however, they act as waveguide-like channels that spoil mode confinement (Buchinger et al., 2022). The paper “Optical properties of circular Bragg gratings with labyrinth geometry to enable electrical contacts” therefore rotates the inter-ring connections from gap to gap, creating a meandering path rather than a straight radial corridor. Across experimentally fabricated designs with 3-fold, 4-fold, and 2-fold symmetry, rotated 4-LW and 4-SW layouts preserve cavity confinement and Gaussian-like far fields far better than straight 4-WG layouts, while also keeping polarization splitting small (Buchinger et al., 2022).

A later numerical design study pushes the same idea to electrically tunable quantum-dot sources spanning visible to telecom wavelengths (Prasad et al., 5 Dec 2025). There the labyrinth circular Bragg grating uses four bridges of width $\lambda$5 nm and is globally re-optimized rather than modified as an afterthought. For emission wavelengths of $\lambda$6 nm, $\lambda$7 nm, and $\lambda$8 nm, the optimized designs achieve collection efficiencies exceeding $\lambda$9 into a numerical aperture of $$\tau < \tau_S < 10\tau, \qquad 10^{-3} \le \omega\tau \le 10^{-1},$$0 and Purcell factors greater than $$\tau < \tau_S < 10\tau, \qquad 10^{-3} \le \omega\tau \le 10^{-1},$$1; the reported values are $$\tau < \tau_S < 10\tau, \qquad 10^{-3} \le \omega\tau \le 10^{-1},$$2 and $$\tau < \tau_S < 10\tau, \qquad 10^{-3} \le \omega\tau \le 10^{-1},$$3, $$\tau < \tau_S < 10\tau, \qquad 10^{-3} \le \omega\tau \le 10^{-1},$$4 and $$\tau < \tau_S < 10\tau, \qquad 10^{-3} \le \omega\tau \le 10^{-1},$$5, and $$\tau < \tau_S < 10\tau, \qquad 10^{-3} \le \omega\tau \le 10^{-1},$$6 and $$\tau < \tau_S < 10\tau, \qquad 10^{-3} \le \omega\tau \le 10^{-1},$$7, respectively (Prasad et al., 5 Dec 2025). The same paper also proposes a $$\tau < \tau_S < 10\tau, \qquad 10^{-3} \le \omega\tau \le 10^{-1},$$8 nm device with an $$\tau < \tau_S < 10\tau, \qquad 10^{-3} \le \omega\tau \le 10^{-1},$$9 nm $100\times 100$0 barrier placed $100\times 100$1 nm below the quantum dot layer to suppress tunneling of one carrier type for selective charging, and reports that the device can be reoptimized with minimal optical penalty (Prasad et al., 5 Dec 2025). Here the labyrinth is neither metaphor nor abstract pattern class, but a practical geometry that reconciles optical confinement with electrical accessibility.

Across these two extremes—Maudlin’s conceptual maze and the rotated-bridge photonic cavity—the term retains a common structural meaning. In one case it names a sequence of intellectual detours that obscure the physical problem; in the other, it names a deliberately non-straight path that prevents parasitic waveguiding while preserving function.