Discrete Magnetic Laplacian

• Discrete magnetic Laplacians are phase-twisted Laplace operators on discrete spaces like graphs, triangulations, and fractals, encoding magnetic fields via edge phases or holonomy.
• They exploit gauge invariance and Floquet theory to analyze periodic structures, enabling precise spectral gap detection, localization, and extension to higher-degree settings.
• Recent models integrate discrete Hodge theory and fractal approximations to establish robust frameworks for self-adjointness, domain analysis, and spectral convergence.

Searching arXiv for recent and foundational work on discrete magnetic Laplacians and closely related spectral settings. Search query: "discrete magnetic Laplacian periodic graphs triangulations Sierpinski gasket" A discrete magnetic Laplacian is a phase-twisted Laplace operator on a discrete geometric object—most commonly a weighted graph, but also a triangulation, a finite quotient of a periodic graph, or a graph approximation to a fractal—in which the magnetic field is encoded by antisymmetric edge phases or, in higher-dimensional simplicial settings, by holonomy through faces. In the graph-theoretic setting, the operator acts by replacing ordinary differences across an edge with covariant differences of the form $f(x)-e^{i\theta_{x,y}}f(y)$, or equivalently by a twisted derivative $d_\alpha$ and the factorization $\Delta_\alpha=d_\alpha^*d_\alpha$ (Fabila-Carrasco et al., 2017, Golénia et al., 2015). Across the literature considered here, the subject develops along four tightly connected lines: gauge covariance and holonomy on graphs, Floquet-theoretic reductions of periodic operators to finite magnetic matrices, higher-degree magnetic Hodge theory on triangulations, and approximation frameworks in which graph magnetic operators converge to fractal or continuum magnetic operators (Korotyaev et al., 2018, Anné et al., 2021, Hyde et al., 2016).

1. Graph-theoretic definitions and gauge structure

On weighted graphs, the magnetic data are encoded by antisymmetric phases on oriented edges. In the weighted framework of discrete cusps and funnels, a graph is $\mathcal G=(\mathcal E,\mathcal V,m)$ or $(\mathcal E,\mathcal V,m,\theta)$, where $m:\mathcal V\to(0,\infty)$ is a vertex weight, $\mathcal E:\mathcal V\times\mathcal V\to[0,\infty)$ is symmetric, and $\theta_{x,y}\in \mathbb R/2\pi\mathbb Z$ satisfies $\theta_{x,y}=-\theta_{y,x}$ and vanishes off the edge set. The associated quadratic form is

$$Q_{\mathcal G,\theta}(f)=\frac12\sum_{x,y\in\mathcal V}\mathcal E(x,y)\,\left|f(x)-e^{i\theta_{x,y}}f(y)\right|^2,$$

and the magnetic Laplacian is the nonnegative self-adjoint operator associated with this form; on finitely supported functions,

$d_\alpha$0

(Golénia et al., 2015, Athmouni et al., 8 Jul 2025).

A closely related formulation uses a twisted derivative. For an oriented weighted graph $d_\alpha$1 with vector potential $d_\alpha$2, the twisted derivative is

$d_\alpha$3

and the discrete magnetic Laplacian is

$d_\alpha$4

On vertices it is written

$d_\alpha$5

with $d_\alpha$6 (Fabila-Carrasco et al., 2017).

Gauge invariance is a structural principle in every setting discussed here. On weighted graphs, two magnetic potentials with the same holonomy are gauge equivalent, and the corresponding magnetic Laplacians are unitarily equivalent (Golénia et al., 2015). In the periodic-graph setting, if $d_\alpha$7, then $d_\alpha$8 and $d_\alpha$9 are unitarily equivalent via multiplication by $\Delta_\alpha=d_\alpha^*d_\alpha$0 (Fabila-Carrasco et al., 2017). In the continuous magnetic Schrödinger framework, the same principle appears as

$\Delta_\alpha=d_\alpha^*d_\alpha$1

a fact explicitly identified as transferable to the discrete case (Raymond, 2014).

The gauge-invariant content is holonomy or flux through cycles. On periodic graphs, for a cycle $\Delta_\alpha=d_\alpha^*d_\alpha$2 in the cycle space $\Delta_\alpha=d_\alpha^*d_\alpha$3 of the fundamental graph,

$\Delta_\alpha=d_\alpha^*d_\alpha$4

(Korotyaev et al., 2018). On weighted graphs with cusp geometry, if

$\Delta_\alpha=d_\alpha^*d_\alpha$5

then

$\Delta_\alpha=d_\alpha^*d_\alpha$6

and the magnetic potential is determined up to gauge by this holonomy map (Golénia et al., 2015). This identifies the discrete magnetic Laplacian as an operator depending on cycle fluxes rather than on a particular edge-wise representative.

2. Discrete geometric settings: graphs, triangulations, and fractals

The most elementary setting is the graph magnetic Laplacian on vertices. For periodic discrete graphs $\Delta_\alpha=d_\alpha^*d_\alpha$7, the combinatorial magnetic Laplacian is

$\Delta_\alpha=d_\alpha^*d_\alpha$8

or equivalently

$\Delta_\alpha=d_\alpha^*d_\alpha$9

where $\mathcal G=(\mathcal E,\mathcal V,m)$0 is the degree of $\mathcal G=(\mathcal E,\mathcal V,m)$1 (Korotyaev et al., 2018). In weighted settings the vertex weight $\mathcal G=(\mathcal E,\mathcal V,m)$2 appears explicitly, as above (Golénia et al., 2015, Athmouni et al., 8 Jul 2025). These formulations are not identical as operators, but they share the same phase-twisted neighbor-interaction mechanism.

A higher-dimensional simplicial version is developed for weighted triangulations. A magnetic triangulation is

$\mathcal G=(\mathcal E,\mathcal V,m)$3

where $\mathcal G=(\mathcal E,\mathcal V,m)$4 is a weighted magnetic graph and $\mathcal G=(\mathcal E,\mathcal V,m)$5 is a symmetric set of triangular faces. The cochain spaces are weighted $\mathcal G=(\mathcal E,\mathcal V,m)$6-spaces on vertices, skew-symmetric edge functions, and skew-symmetric face functions, and the total Hilbert space is

$\mathcal G=(\mathcal E,\mathcal V,m)$7

The magnetic derivative on $\mathcal G=(\mathcal E,\mathcal V,m)$8-forms is

$\mathcal G=(\mathcal E,\mathcal V,m)$9

and the magnetic exterior derivative on $(\mathcal E,\mathcal V,m,\theta)$0-forms is

$(\mathcal E,\mathcal V,m,\theta)$1

From these operators one obtains the magnetic Gauss–Bonnet operator

$(\mathcal E,\mathcal V,m,\theta)$2

(Anné et al., 2021).

The triangulation setting differs from the graph setting in a decisive respect: the magnetic field becomes face holonomy. For a face,

$(\mathcal E,\mathcal V,m,\theta)$3

Then $(\mathcal E,\mathcal V,m,\theta)$4 need not vanish, and this failure is controlled explicitly by $(\mathcal E,\mathcal V,m,\theta)$5. The magnetic Hodge Laplacian therefore acquires off-diagonal terms $(\mathcal E,\mathcal V,m,\theta)$6 and $(\mathcal E,\mathcal V,m,\theta)$7, absent in the non-magnetic case (Anné et al., 2021). This is the simplicial analogue of curvature entering the de Rham complex.

A different extension appears on the Sierpiński gasket. There the magnetic operator is defined on a Hilbert space $(\mathcal E,\mathcal V,m,\theta)$8 of $(\mathcal E,\mathcal V,m,\theta)$9-forms with derivation $m:\mathcal V\to(0,\infty)$0, using a real-valued $m:\mathcal V\to(0,\infty)$1-form $m:\mathcal V\to(0,\infty)$2 as magnetic potential: $m:\mathcal V\to(0,\infty)$3 The associated Dirichlet magnetic Laplacian $m:\mathcal V\to(0,\infty)$4 is defined through the closed form $m:\mathcal V\to(0,\infty)$5. At graph level, the finite approximants carry magnetic energies

$m:\mathcal V\to(0,\infty)$6

and magnetic Laplacians

$m:\mathcal V\to(0,\infty)$7

with renormalized convergence to the fractal operator (Hyde et al., 2016).

3. Periodic graphs and Floquet reduction

On periodic graphs, the discrete magnetic Laplacian is inseparable from Floquet theory. For a connected, locally finite $m:\mathcal V\to(0,\infty)$8-periodic graph $m:\mathcal V\to(0,\infty)$9 with finite quotient $\mathcal E:\mathcal V\times\mathcal V\to[0,\infty)$0, the periodic magnetic operator admits a direct-integral decomposition over the quasimomentum torus $\mathcal E:\mathcal V\times\mathcal V\to[0,\infty)$1 (Korotyaev et al., 2018). In the standard fiber representation,

$\mathcal E:\mathcal V\times\mathcal V\to[0,\infty)$2

with fiber magnetic Laplacians

$\mathcal E:\mathcal V\times\mathcal V\to[0,\infty)$3

A refined version replaces the edge-index form $\mathcal E:\mathcal V\times\mathcal V\to[0,\infty)$4 and the magnetic form $\mathcal E:\mathcal V\times\mathcal V\to[0,\infty)$5 by minimal forms $\mathcal E:\mathcal V\times\mathcal V\to[0,\infty)$6 and $\mathcal E:\mathcal V\times\mathcal V\to[0,\infty)$7, giving

$\mathcal E:\mathcal V\times\mathcal V\to[0,\infty)$8

with

$\mathcal E:\mathcal V\times\mathcal V\to[0,\infty)$9

Minimality here means support minimality among all $\theta_{x,y}\in \mathbb R/2\pi\mathbb Z$0-forms with the same cycle fluxes (Korotyaev et al., 2018). The resulting integers

$\theta_{x,y}\in \mathbb R/2\pi\mathbb Z$1

are invariants of the periodic graph and of the magnetic Laplacian, respectively. They measure how many coefficients in the fiber matrices genuinely depend on quasimomentum and on magnetic potential (Korotyaev et al., 2018).

A complementary formulation appears in the spectral-gap theory of periodic graphs. There the Floquet fibers of a periodic Laplacian on the infinite graph are identified with discrete magnetic Laplacians on the finite quotient graph. If $\theta_{x,y}\in \mathbb R/2\pi\mathbb Z$2 is a periodic covering and $\theta_{x,y}\in \mathbb R/2\pi\mathbb Z$3 is the set of quotient vector potentials with the lifting property

$\theta_{x,y}\in \mathbb R/2\pi\mathbb Z$4

then

$\theta_{x,y}\in \mathbb R/2\pi\mathbb Z$5

In this sense, the vector potential on the quotient is a Floquet parameter (Fabila-Carrasco et al., 2017). The finite-dimensional magnetic Laplacian on the quotient is therefore the decisive object in band localization and gap detection.

The periodic viewpoint also yields quantitative spectral information. For the fiber eigenvalues

$\theta_{x,y}\in \mathbb R/2\pi\mathbb Z$6

the spectrum is the union of bands

$\theta_{x,y}\in \mathbb R/2\pi\mathbb Z$7

The total spectral measure satisfies

$\theta_{x,y}\in \mathbb R/2\pi\mathbb Z$8

and each band obeys

$\theta_{x,y}\in \mathbb R/2\pi\mathbb Z$9

(Korotyaev et al., 2018). These estimates are sharper than bounds using only the Betti number $\theta_{x,y}=-\theta_{y,x}$0.

4. Spectral phenomena: gaps, localization, cusps, and funnels

One major theme is the relation between magnetic flux and spectral gaps. In the periodic-gap framework, edge and vertex virtualization produce comparison operators $\theta_{x,y}=-\theta_{y,x}$1 and $\theta_{x,y}=-\theta_{y,x}$2 satisfying

$\theta_{x,y}=-\theta_{y,x}$3

provided the magnetic potential is supported on a selected edge set and a suitable neighboring vertex set is virtualized (Fabila-Carrasco et al., 2017). Consequently,

$\theta_{x,y}=-\theta_{y,x}$4

This yields a geometric criterion for nonempty magnetic spectral-gap sets. If $\theta_{x,y}=-\theta_{y,x}$5 is a center vertex with cycle edges $\theta_{x,y}=-\theta_{y,x}$6, then the quantity

$\theta_{x,y}=-\theta_{y,x}$7

satisfies: the Lebesgue measure of the magnetic spectral-gap set $\theta_{x,y}=-\theta_{y,x}$8 is at least $\theta_{x,y}=-\theta_{y,x}$9; in particular, $$Q_{\mathcal G,\theta}(f)=\frac12\sum_{x,y\in\mathcal V}\mathcal E(x,y)\,\left|f(x)-e^{i\theta_{x,y}}f(y)\right|^2,$$0 implies $$Q_{\mathcal G,\theta}(f)=\frac12\sum_{x,y\in\mathcal V}\mathcal E(x,y)\,\left|f(x)-e^{i\theta_{x,y}}f(y)\right|^2,$$1 (Fabila-Carrasco et al., 2017).

On discrete cusps, the magnetic field can change the global spectral type. A discrete cusp is a twisted product

$$Q_{\mathcal G,\theta}(f)=\frac12\sum_{x,y\in\mathcal V}\mathcal E(x,y)\,\left|f(x)-e^{i\theta_{x,y}}f(y)\right|^2,$$2

with $$Q_{\mathcal G,\theta}(f)=\frac12\sum_{x,y\in\mathcal V}\mathcal E(x,y)\,\left|f(x)-e^{i\theta_{x,y}}f(y)\right|^2,$$3 at infinity, $$Q_{\mathcal G,\theta}(f)=\frac12\sum_{x,y\in\mathcal V}\mathcal E(x,y)\,\left|f(x)-e^{i\theta_{x,y}}f(y)\right|^2,$$4 finite, and $$Q_{\mathcal G,\theta}(f)=\frac12\sum_{x,y\in\mathcal V}\mathcal E(x,y)\,\left|f(x)-e^{i\theta_{x,y}}f(y)\right|^2,$$5 bounded. In this setting,

$$Q_{\mathcal G,\theta}(f)=\frac12\sum_{x,y\in\mathcal V}\mathcal E(x,y)\,\left|f(x)-e^{i\theta_{x,y}}f(y)\right|^2,$$6

The key theorem states that

$$Q_{\mathcal G,\theta}(f)=\frac12\sum_{x,y\in\mathcal V}\mathcal E(x,y)\,\left|f(x)-e^{i\theta_{x,y}}f(y)\right|^2,$$7

When the transversal holonomy is nontrivial,

$$Q_{\mathcal G,\theta}(f)=\frac12\sum_{x,y\in\mathcal V}\mathcal E(x,y)\,\left|f(x)-e^{i\theta_{x,y}}f(y)\right|^2,$$8

When $$Q_{\mathcal G,\theta}(f)=\frac12\sum_{x,y\in\mathcal V}\mathcal E(x,y)\,\left|f(x)-e^{i\theta_{x,y}}f(y)\right|^2,$$9, the form domain can differ from the degree form domain (Golénia et al., 2015). The magnetic field therefore affects not only the spectrum but also the energy space.

Discrete funnels exhibit a different asymptotic regime. The model graph is a twisted product $d_\alpha$00 with

$d_\alpha$01

The funnel Laplacian decomposes as

$d_\alpha$02

so the transverse term is asymptotically negligible because $d_\alpha$03 (Athmouni et al., 8 Jul 2025). The essential spectrum is determined by the radial channel: $d_\alpha$04 and

$d_\alpha$05

Under long-range perturbations of weights, phases, and an electric potential, the perturbed operator satisfies a Mourre estimate and a Limiting Absorption Principle away from thresholds and embedded eigenvalues; singular continuous spectrum is empty (Athmouni et al., 8 Jul 2025). In this model, the radial magnetic potential is gauge removable because the radial graph is a half-line.

The fractal case shows another spectral pattern. For the magnetic Laplacian $d_\alpha$06 on the Sierpiński gasket, the operator has compact resolvent and pure point spectrum accumulating at $d_\alpha$07. Its eigenvalue counting asymptotics are unchanged by the magnetic field: $d_\alpha$08 with the same periodic function $d_\alpha$09 as for the ordinary Laplacian (Hyde et al., 2016). At the same time, low-generation eigenvalues can shift with the flux, as illustrated by

$d_\alpha$10

on the first graph approximation (Hyde et al., 2016).

5. Self-adjointness, domains, and higher-degree structure

Essential self-adjointness is a central analytical issue for discrete magnetic operators on noncompact discrete geometries. On weighted triangulations, the decisive completeness hypothesis is $d_\alpha$11-completeness. It requires cutoff functions $d_\alpha$12, $d_\alpha$13, equal to $d_\alpha$14 on an exhaustion $d_\alpha$15, together with the uniform estimates

$d_\alpha$16

and

$d_\alpha$17

Under this hypothesis, the magnetic Gauss–Bonnet operator $d_\alpha$18 is essentially self-adjoint on compactly supported cochains, and therefore so is the magnetic Laplacian $d_\alpha$19 (Anné et al., 2021).

The same paper introduces a bounded-curvature condition on the magnetic potential: $d_\alpha$20 Under $d_\alpha$21-completeness and bounded curvature, the closure of $d_\alpha$22 has the domain decomposition

$d_\alpha$23

which separates the vertex, edge, and face components (Anné et al., 2021). The appearance of the face holonomy in the curvature bound is intrinsically two-dimensional.

In the cusp setting, domain theory is more delicate because the magnetic field can eliminate or preserve a transversal zero mode. The paper on discrete cusps proves that

$d_\alpha$24

but equality may fail when the transversal holonomy vanishes (Golénia et al., 2015). This establishes that magnetic and non-magnetic form domains can differ. The result is unusual enough that the paper explicitly identifies it as one of its most striking conclusions.

In the funnel setting, the unperturbed form domain is given explicitly as

$d_\alpha$25

and the perturbed analysis is carried out in a form-commutator framework $d_\alpha$26, rather than through a purely operator-domain computation (Athmouni et al., 8 Jul 2025). This suggests that in weighted magnetic graph models with long-range perturbations, form methods can be more robust than direct operator-domain arguments.

6. Conceptual boundaries and relation to continuous magnetic Laplacians

The expression “discrete magnetic Laplacian” does not refer to a single theory. In the sources considered here it can denote at least three distinct but related constructions. First, it denotes graph magnetic Laplacians with phase-twisted differences on vertices (Fabila-Carrasco et al., 2017, Korotyaev et al., 2018, Golénia et al., 2015, Athmouni et al., 8 Jul 2025). Second, it denotes higher-degree magnetic Hodge Laplacians on simplicial complexes, where face holonomy becomes curvature and the magnetic Gauss–Bonnet operator is the fundamental object (Anné et al., 2021). Third, it can denote graph approximants to a non-Euclidean limit operator, as on the Sierpiński gasket, where finite graph magnetic Laplacians converge after renormalization to a fractal magnetic Laplacian (Hyde et al., 2016).

By contrast, some closely related literature concerns continuous magnetic Laplacians rather than discrete ones. The monograph “Little Magnetic Book” studies the continuous magnetic Schrödinger operator

$d_\alpha$27

with an emphasis on discrete spectrum, essential spectrum, localization, effective Hamiltonians, Born–Oppenheimer approximation, and norm resolvent convergence (Raymond, 2014). This work is not about graph magnetic Laplacians, but it supplies a continuous spectral framework that informs discrete models.

Likewise, the study of curved waveguides in a strong uniform magnetic field analyzes the continuous operator

$d_\alpha$28

on a curved strip with Dirichlet boundary conditions, proving existence of discrete spectrum below the essential spectrum under a curvature condition (Bon-Lavigne et al., 2022). Here “discrete” refers to the discrete spectral component, not to a discrete underlying space. The distinction is terminologically important: a discrete magnetic Laplacian in graph theory is a magnetic operator on a discrete combinatorial or weighted geometry, whereas a continuous magnetic Laplacian may possess discrete spectrum without being discrete in this combinatorial sense (Bon-Lavigne et al., 2022).

A plausible implication is that the common language of gauge invariance, model operators at infinity, holonomy, and localization allows continuous and discrete magnetic theories to inform one another without collapsing their distinction. In the discrete setting, this has already produced a broad landscape: phase-twisted graph Laplacians, Floquet-reduced finite magnetic matrices on periodic graphs, magnetic Hodge theory on triangulations, and renormalized graph approximations to fractal magnetic operators (Korotyaev et al., 2018, Fabila-Carrasco et al., 2017, Anné et al., 2021, Hyde et al., 2016).