Higher-Order Laplacian Renormalization

• Higher-Order Laplacian Renormalization is a family of techniques that uses extended Laplacian operators to achieve multiscale coarse-graining in discrete and continuum settings.
• It employs diffusion-driven dynamics and spectral methods to ensure key properties like equilibrium preservation and random-walk exit statistics during renormalization.
• The approach bridges continuum extension problems and discrete models by converting nonlocal boundary effects into local operator formulations using advanced renormalization schemes.

Searching arXiv for papers on higher-order Laplacian renormalization and closely related formulations. Higher-order Laplacian renormalization denotes a family of multiscale constructions in which Laplacian operators beyond the standard vertex Laplacian—or Laplacians acting on higher-order combinatorial objects—are used to define scale, coarse-graining, or renormalized boundary operators. In the cited literature, the term appears in several non-equivalent but related senses: diffusion-driven renormalization on graphs and higher-order networks, renormalized extension problems for fractional and higher-order differential operators, and renormalized limits of discrete Laplacians on fractals. A common theme is that the Laplacian is treated not merely as a diagnostic of geometry, but as the generator of the relevant notion of scale (Villegas et al., 2022, Nurisso et al., 2024, Lee, 15 Feb 2025, Cao et al., 2016).

1. Scope of the concept and principal operator families

The phrase combines two meanings of “higher-order.” In continuum analysis, it refers to operators of order exceeding the classical Laplacian, such as $(-\Delta)^\gamma$ for positive non-integer $\gamma$, or to logarithmic modifications of Laplacian powers. In discrete topology and network science, it refers to diffusion on higher-order objects—edges, triangles, and general $k$-simplices—rather than only on vertices. These two usages share a renormalization theme, but they do not define a single unified formalism.

On ordinary graphs, the basic operator is the graph Laplacian $L=D-A$, with heat kernel $e^{-\tau L}$. On simplicial complexes, higher-order diffusion is encoded by up- and down-Laplacians, or by cross-order Laplacians that connect $k$-simplices through $m$-simplices. In continuum extension theory, the role of the Laplacian is played by weighted operators such as

$\Delta_b U=\Delta U+\frac{b}{y}U_y,$

whose iterates encode higher-order fractional powers on the boundary (Yang, 2013, Nurisso et al., 2024).

Setting	Representative operator	Renormalized object
Graphs	$L=D-A$, $e^{-\tau L}$	Supernodes, coarse Laplacian
Simplicial complexes	$\gamma$0, $\gamma$1, $\gamma$2	Coarse simplicial complex, RG flow, spectral dimension
Hypergraphs	Bipartite Laplacian representation	Equilibrium-preserving $\gamma$3 and $\gamma$4
Fractional and higher-order PDE	$\gamma$5, weighted Neumann extension	Boundary operator $\gamma$6, $\gamma$7
Fractals	Renormalized higher-order graph Laplacians $\gamma$8	Continuum higher-order Laplacian

A recurring distinction is between approximate multiscale similarity and exact dynamical preservation. Some constructions identify scales through entropy plateaux or spectral truncation; others characterize the precise conditions under which coarse-graining preserves random-walk or equilibrium dynamics.

2. Diffusion-based renormalization on ordinary graphs

A graph-theoretic renormalization framework based on the Laplacian renormalization group (LRG) treats diffusion as the intrinsic notion of scale on heterogeneous networks. For an undirected network, the Laplacian is

$\gamma$9

the propagator is

$k$0

and the normalized diffusion state is

$k$1

The associated spectral entropy

$k$2

and specific heat

$k$3

serve as multiscale diagnostics: peaks in $k$4 identify characteristic resolution scales $k$5 at which mesoscale organization becomes visible (Villegas et al., 2022).

The framework has both a real-space and a momentum-space interpretation. In real space, nodes are grouped into Kadanoff supernodes by thresholding diffusion affinities at a chosen scale $k$6, building a meta-graph, replacing each connected block by a supernode, and iterating. In momentum space, Laplacian eigenvectors replace Fourier modes. A cutoff

$k$7

separates fast and slow modes, and the renormalized Laplacian is obtained by retaining only the slow sector,

$k$8

followed by rescaling

$k$9

This formulation is explicitly Wilsonian in character: short-wavelength fluctuations are integrated out, the node space is coarse-grained into macronodes, and the operator is rescaled to restore the cutoff. The authors report that Barabási–Albert networks show approximate Laplacian scale invariance under repeated RG, whereas Erdős–Rényi networks tend to collapse toward triviality; random trees and selected metabolic networks also exhibit nontrivial persistence of spectral or degree-distribution features under successive RG steps (Villegas et al., 2022). This suggests that higher-order Laplacian renormalization, in the graph sense, is fundamentally a diffusion-based classification of multiscale organization.

3. Simplicial complexes, cross-order diffusion, and higher-order RG

For higher-order networks, the central move is to shift diffusion from vertices to simplices. In the cross-order Laplacian renormalization group (X-LRG), a higher-order network $L=D-A$0 is viewed through its sets of $L=D-A$1-simplices $L=D-A$2, and two $L=D-A$3-simplices are declared adjacent when they are connected through simplices of another order $L=D-A$4. The corresponding weighted adjacency matrix $L=D-A$5 leads to the cross-order Laplacian

$L=D-A$6

Diffusion is then defined by

$L=D-A$7

with entropy and entropic susceptibility used to detect characteristic scales and informational scale-invariant plateaux (Nurisso et al., 2024).

The method is simplex-centric rather than node-centric. Its coarse-graining procedure partitions $L=D-A$8-simplices according to diffusion at a chosen time $L=D-A$9, then projects those labels back to vertices: each vertex inherits the labels of the $e^{-\tau L}$0-simplices to which it belongs, vertices with identical label sets are merged, and the smaller simplicial complex is induced from the original one. In controlled tests, $e^{-\tau L}$1-based renormalization of the pseudofractal simplicial complex essentially reverses the construction of the pseudofractal, while $e^{-\tau L}$2-based renormalization collapses it into a star-like object. For Network Geometry with Flavor, $e^{-\tau L}$3-renormalization preserves structure better than $e^{-\tau L}$4-renormalization, and real clique complexes exhibit order-specific scale-invariance signatures that separate infrastructural, social, and biological data in a low-dimensional embedding (Nurisso et al., 2024).

A complementary RG approach analyzes higher-order spectra through Gaussian models on simplicial complexes. For oriented $e^{-\tau L}$5-simplices, the higher-order Laplacian is decomposed as

$e^{-\tau L}$6

The regular part of the small-eigenvalue density is written

$e^{-\tau L}$7

thereby defining a higher-order spectral dimension $e^{-\tau L}$8. Real-space RG near a zero-mass fixed point determines $e^{-\tau L}$9 for Apollonian and pseudo-fractal simplicial complexes. In the Apollonian family, the up-Laplacian has a finite spectral dimension for $k$0, $k$1 decreases as $k$2 increases, and no finite spectral dimension is found for $k$3. In the pseudo-fractal family, finite-dimensional spectra persist up to $k$4 (Reitz et al., 2020).

Taken together, these works establish that higher-order Laplacian renormalization is order-sensitive: the relevant scales, spectral dimensions, and coarse-grained structures depend on which simplices diffuse and through which simplices they interact.

4. Exact preservation criteria, harmonic morphisms, and equilibrium-preserving coarse-graining

A major refinement of Laplacian-based coarse-graining is the identification of exact dynamical preservation criteria. For graphs $k$5 and $k$6, a surjective map $k$7 is a harmonic morphism if pullbacks of harmonic functions remain harmonic. Using the random-walk Laplacian

$k$8

the structural characterization is horizontal conformality: if $k$9, then $m$0 or $m$1, and for each $m$2, the multiplicity

$m$3

must be constant over all neighboring macro-nodes $m$4. This equal-multiplicity rule is the exact symmetry behind diffusion-preserving coarse-graining (Guadagnuolo et al., 9 Apr 2026).

The central preservation theorem concerns first-exit statistics. If $m$5 is the induced subgraph on $m$6 and $m$7 is the first exit time from $m$8, then the harmonic measure

$m$9

satisfies

$\Delta_b U=\Delta U+\frac{b}{y}U_y,$0

for all $\Delta_b U=\Delta U+\frac{b}{y}U_y,$1 and $\Delta_b U=\Delta U+\frac{b}{y}U_y,$2 if and only if $\Delta_b U=\Delta U+\frac{b}{y}U_y,$3 is a harmonic morphism. The coarse-graining therefore preserves not every microscopic step, but the statistics of exits between macro-sets, after a random time change. For lazy random walks, the stricter notion of combinatorial conformality requires equal multiplicities including stay-put transitions and yields direct step-by-step preservation with no temporal rescaling (Guadagnuolo et al., 9 Apr 2026).

Within the same framework, Laplacian renormalization is implemented through the heat kernel

$\Delta_b U=\Delta U+\frac{b}{y}U_y,$4

merging nodes when

$\Delta_b U=\Delta U+\frac{b}{y}U_y,$5

The paper introduces harmonic-degree diagnostics to quantify proximity to a harmonic morphism and reports that Laplacian renormalization often achieves the highest harmonic degree among tested methods. More strikingly, exact harmonic morphisms arise spontaneously at specific scales: $\Delta_b U=\Delta U+\frac{b}{y}U_y,$6 is reported for Facebook, Web-edu, CS Collab, and Yeast, and the Facebook network at $\Delta_b U=\Delta U+\frac{b}{y}U_y,$7 is an exact harmonic morphism when $\Delta_b U=\Delta U+\frac{b}{y}U_y,$8. The authors emphasize that entropic susceptibility does not detect this exact preservation property (Guadagnuolo et al., 9 Apr 2026).

A second rigorization route is the equilibrium-preserving Laplacian renormalization group for networks and hypergraphs. Here the coarse basis is constructed from the equilibrium eigenvector $\Delta_b U=\Delta U+\frac{b}{y}U_y,$9, yielding a quasi-complete slow-mode basis and a renormalized Laplacian

$L=D-A$0

The corresponding renormalized adjacency is defined from equilibrium-state flows,

$L=D-A$1

which preserves mean connectivity. Applied recursively to hypergraphs via a bipartite representation of vertices and hyperedges, this equilibrium-preserving scheme finds that RN hypertrees remain close to Poissonian, BA hypertrees flow toward more Poissonian effective structures, and ER hypergraphs without a finite spectral dimension develop broader, power-law-like tails under coarse-graining (Yi et al., 7 Jul 2025).

These results delimit two different standards of rigor. Harmonic morphisms characterize exact random-walk preservation on graphs, whereas equilibrium-preserving RG guarantees exact preservation of the equilibrium state and controlled coarse Laplacians on higher-order systems.

5. Continuum extension problems and renormalized boundary operators

In continuum analysis, higher-order Laplacian renormalization is formulated through extension problems in one extra dimension. For a positive non-integer order $L=D-A$2, let $L=D-A$3 and $L=D-A$4. The higher-order Caffarelli–Silvestre extension uses a function

$L=D-A$5

satisfying

$L=D-A$6

together with boundary conditions that fix $L=D-A$7, force the odd weighted normal derivatives to vanish up to the appropriate order, and tie even normal derivatives to tangential derivatives. The boundary operator is then recovered by

$L=D-A$8

The boundary $L=D-A$9 norm is equivalent to a suitable higher-order weighted seminorm of $e^{-\tau L}$0, so the nonlocal operator becomes a local elliptic problem in $e^{-\tau L}$1 (Yang, 2013).

A distinct renormalization philosophy keeps a single weighted Neumann problem,

$e^{-\tau L}$2

but defines the boundary operator through Hadamard finite-part regularization of the bulk energy rather than through a classical trace. For $e^{-\tau L}$3, the renormalized Neumann-to-Dirichlet operator is

$e^{-\tau L}$4

Its classification theorem is explicit: for non-integer $e^{-\tau L}$5, $e^{-\tau L}$6 is a constant multiple of $e^{-\tau L}$7, whereas for integer $e^{-\tau L}$8 it contains a logarithmic anomaly involving $e^{-\tau L}$9. At $\gamma$00,

$\gamma$01

The mechanism is Fourier reduction plus the small-argument asymptotics of modified Bessel functions, which isolate the singular counterterms and the finite renormalized part (Lee, 15 Feb 2025).

The contrast between the two approaches is conceptually important. One constructs a genuinely higher-order bulk PDE whose boundary flux yields $\gamma$02; the other retains a second-order weighted equation but renormalizes the divergent energy to recover higher-order or logarithmic boundary operators. In both cases, renormalization converts a nonlocal boundary operator into a local problem in one higher dimension, but the object being renormalized is different.

6. Fractals, manifold regularization, and unresolved directions

On fully symmetric p.c.f. self-similar fractals, higher-order Laplacian renormalization takes the form of a discrete-to-continuum limit. If $\gamma$03 is the graph Laplacian on the $\gamma$04-th approximating graph and $\gamma$05 is its renormalized version, then for $\gamma$06,

$\gamma$07

with uniform convergence on $\gamma$08. The converse also holds: if $\gamma$09 converges uniformly to a continuous $\gamma$10, then $\gamma$11 belongs to the domain of the continuum higher-order Laplacian and $\gamma$12 away from the boundary. The proof uses local multiharmonic “monomials,” higher-order weak tangents, and renormalized graph operators on shrinking cells. In this setting, higher-order Laplacian renormalization is literally the renormalized uniform limit of iterated graph Laplacians (Cao et al., 2016).

A related, but distinct, strand of work concerns higher-order Laplacian regularization on data manifolds rather than RG in the strict sense. Standard graph Laplacian regularization is described there as a first-order regularizer that can be degenerate on high-dimensional manifolds, while iterated graph Laplacians $\gamma$13 remove that degeneracy but become dense and computationally expensive. The proposed alternative is a globally high-order, sparse regularizer built from local first-order tangent-space approximations and local derivative evaluations, with a practical local Gaussian formulation based on RKHS norms (Kim et al., 2016). This is not a renormalization-group theory, but it shows how higher-order Laplacian structure can be assembled from local approximations without losing sparsity.

Several open issues remain explicit in the literature. In higher-order network coarse-graining, the exact analogue of horizontal conformality beyond graphs is only partially developed, and Hodge-harmonic or Bochner-harmonic morphisms are orientation-dependent (Guadagnuolo et al., 9 Apr 2026). The same work argues that harmonicity must be evaluated on the same operator and domain where diffusion actually lives, which limits naive transfers of node-based criteria to edge- or triangle-based dynamics. A plausible implication is that any general theory of higher-order Laplacian renormalization will need a setting-dependent notion of exactness: random-walk exit preservation on graphs, equilibrium preservation on hypergraphs, and operator-specific boundary renormalization in continuum problems.

The field is therefore best understood as a structured collection of renormalization ideas organized by the Laplacian. What unifies them is the use of Laplacian-generated diffusion, Laplacian-based spectral scaling, or Laplacian-driven energy identities to define coarse variables and preserved quantities across scales. What separates them is the object being coarse-grained—vertices, simplices, hyperedges, boundary traces, or graph approximations—and the criterion of fidelity, which may be spectral, entropic, equilibrium-preserving, or exactly harmonic.