Layered Interactions: Structures & Dynamics

• Layered interactions are multi-strata couplings occurring across distinct media such as crystals, multiplex networks, and human–AI systems.
• They modify effective interaction laws and emergent dynamics by coupling intra- and interlayer processes, as seen in exfoliation energies and magnetic excitations.
• Applied in materials science, network theory, and interface design, these interactions enable innovative techniques like moiré patterning, multiplex coordination, and structured chat overlays.

Searching arXiv for papers related to layered interactions across materials, networks, and interfaces. The arXiv search tool is not available in this environment. I will rely strictly on the supplied arXiv records and cite them directly. Layered interactions denote classes of interactions organized across distinct strata rather than within a single undifferentiated graph or medium. In current research usage, the term covers several technically different constructions: interlayer binding and coupling in stacked crystals, multiplex and multilayer relations on a common node set, interaction kernels mediated by layered electromagnetic or superconducting environments, and structured overlays placed around a linear human–AI dialogue to scaffold navigation and reflection (Rumson et al., 8 Apr 2026, Gómez-Gardeñes et al., 2015, Nguyen et al., 3 Jun 2025). Across these settings, the central issue is how the presence of layers changes the effective interaction law, the relevant degrees of freedom, and the emergent collective state.

1. Formal architectures of layered interaction

In network science, a multiplex or multilayer network provides the most explicit abstract formalization. A multiplex network consists of a fixed set of nodes and several interaction layers $\alpha = 1,\dots,L$, each layer being a network on the same node set. Node $i$ is assigned a state vector

$$\vec{p}_i = \left(p_i^{(1)},\dots,p_i^{(L)}\right), \qquad \sum_{\alpha=1}^{L} p_i^{(\alpha)} = 1,$$

so the layered state is a distribution of activity or resources across layers rather than a single binary choice (Gómez-Gardeñes et al., 2015). In the general case, the interaction matrix is written as

$$\mathbf{J} = \bigoplus_{\alpha=1}^L \mathbf{W}^{(\alpha)} + \mathbf{D}\otimes \mathbf{I},$$

and the corresponding Hamiltonian is

$$\mathcal{H}(\mathbf{P}) = -\sum_{\alpha,\beta=1}^{L}\sum_{i,j=1}^{N} J^{(\alpha\beta)}_{ij} p_i^{(\alpha)} p_j^{(\beta)}.$$

This representation makes the layer itself a formal degree of freedom rather than a descriptive label (Gómez-Gardeñes et al., 2015).

A related but distinct multilayer construction separates the network on which agents learn from the network on which they obtain payoffs. In the two-population model of coordination, intralayer links define observation and updating, whereas interlayer links define game interactions. This separation of roles is the layered feature: the social influence neighbourhood and the strategic payoff neighbourhood are different objects, even though they jointly determine the dynamics (Lugo et al., 2014).

In materials research, the formal architecture is geometric rather than graph-theoretic. Layered materials such as graphite, hexagonal boron nitride, lead(II) oxide, and transition-metal dichalcogenides consist of strongly bonded two-dimensional sheets stacked along one crystallographic direction, with weak interlayer binding dominated by London dispersion (Rumson et al., 8 Apr 2026). In honeycomb layered oxides with Ag bilayers, the cationic subsystem itself becomes layered twice over: Ag-rich domains of ${\rm Ag_6}M_2{\rm TeO_6}$ contain two closely spaced Ag planes that form a bifurcated bipartite honeycomb lattice, and the authors interpret the associated argentophilic bonding through spontaneous symmetry breaking of an SU(2)$\times$U(1) gauge structure (Masese et al., 2021).

These formulations differ in ontology, but they share a common structure: a system is not described by a single adjacency, a single field, or a single interaction channel. It is described by coupled layers whose relation is itself dynamical.

2. Interlayer energetics in layered solids

For crystalline layered materials, the central energetic observable is the exfoliation energy, the energy cost per unit area to peel one layer from the bulk, together with the lattice parameter $c$, which fixes the equilibrium interlayer spacing. Stronger interlayer attraction gives a smaller optimal $c$ and a larger $E_\text{exfol}$ (Rumson et al., 8 Apr 2026). Because the dominant attractive force is London dispersion, density-functional calculations require an explicit dispersion treatment with controlled short-range damping.

In the XDM framework, the dispersion energy is written as

$i$0

with $i$1 contributions and damping functions that regularize the $i$2 divergence at short distance (Rumson et al., 8 Apr 2026). The paper compares Becke–Johnson damping and the newer Z-damping. BJ damping ties the short-range limit to the dispersion coefficients and hence to the electron density, whereas Z-damping fixes the united-atom limit through atomic numbers and a single parameter $i$3. In the LM26 benchmark, inclusion of the Axilrod–Teller–Muto three-body term systematically improves exfoliation energies for XDM(BJ) and XDM(Z), with B86bPBE-XDM(BJ)+ATM reaching an MAE of $i$4 meV/Å$i$5 and B86bPBE-XDM(Z)+ATM reaching $i$6 meV/Å$i$7 in the plane-wave basis (Rumson et al., 8 Apr 2026).

A separate line of work argues that interlayer binding in two-dimensional materials is not simply of van der Waals character but can coexist with quasi-bonding character. The total interaction is decomposed as

$i$8

where $i$9 arises from interlayer hybridization near the Fermi level (Chen et al., 2021). The proposed classification distinguishes homo-occupancy interactions, involving occupied–occupied or empty–empty coupling, from hetero-occupancy interactions, involving occupied–empty coupling. In category I, the quasi-bonding contribution is destabilizing; in category II, it is stabilizing, so the total interlayer interaction is relatively stronger in category II and weaker in category I (Chen et al., 2021).

These results rule out a common simplification according to which layered binding is exhausted by pairwise van der Waals attraction. Pairwise dispersion, damping, three-body geometry, and near-$$\vec{p}_i = \left(p_i^{(1)},\dots,p_i^{(L)}\right), \qquad \sum_{\alpha=1}^{L} p_i^{(\alpha)} = 1,$$0 occupancy all enter the interaction law.

3. Magnetic, excitonic, and photonic couplings across layers

Layered magnetic materials make the competition between intralayer and interlayer couplings explicit. MnBi$$\vec{p}_i = \left(p_i^{(1)},\dots,p_i^{(L)}\right), \qquad \sum_{\alpha=1}^{L} p_i^{(\alpha)} = 1,$$1Te$$\vec{p}_i = \left(p_i^{(1)},\dots,p_i^{(L)}\right), \qquad \sum_{\alpha=1}^{L} p_i^{(\alpha)} = 1,$$2 is built from septuple layers stacked along $$\vec{p}_i = \left(p_i^{(1)},\dots,p_i^{(L)}\right), \qquad \sum_{\alpha=1}^{L} p_i^{(\alpha)} = 1,$$3 and realizes an A-type antiferromagnet: each Mn layer is ferromagnetic, while adjacent layers couple antiferromagnetically across the van der Waals gap. Inelastic neutron scattering shows that modelling the two-dimensional intralayer ferromagnetic spin waves requires long-range and competing Heisenberg FM and AF interactions up to at least the seventh nearest neighbor, while the interlayer coupling and uniaxial anisotropy are of comparable magnitude and support metamagnetic transitions (Li et al., 2020).

MnPSe$$\vec{p}_i = \left(p_i^{(1)},\dots,p_i^{(L)}\right), \qquad \sum_{\alpha=1}^{L} p_i^{(\alpha)} = 1,$$4 and CrPS$$\vec{p}_i = \left(p_i^{(1)},\dots,p_i^{(L)}\right), \qquad \sum_{\alpha=1}^{L} p_i^{(\alpha)} = 1,$$5 illustrate related but distinct hierarchies. MnPSe$$\vec{p}_i = \left(p_i^{(1)},\dots,p_i^{(L)}\right), \qquad \sum_{\alpha=1}^{L} p_i^{(\alpha)} = 1,$$6 is a van der Waals layered antiferromagnet with dominant intralayer exchanges $$\vec{p}_i = \left(p_i^{(1)},\dots,p_i^{(L)}\right), \qquad \sum_{\alpha=1}^{L} p_i^{(\alpha)} = 1,$$7 meV, $$\vec{p}_i = \left(p_i^{(1)},\dots,p_i^{(L)}\right), \qquad \sum_{\alpha=1}^{L} p_i^{(\alpha)} = 1,$$8 meV, $$\vec{p}_i = \left(p_i^{(1)},\dots,p_i^{(L)}\right), \qquad \sum_{\alpha=1}^{L} p_i^{(\alpha)} = 1,$$9 meV and appreciable interlayer coupling $$\mathbf{J} = \bigoplus_{\alpha=1}^L \mathbf{W}^{(\alpha)} + \mathbf{D}\otimes \mathbf{I},$$0 meV; the magnetic excitations persist well above $$\mathbf{J} = \bigoplus_{\alpha=1}^L \mathbf{W}^{(\alpha)} + \mathbf{D}\otimes \mathbf{I},$$1 K, consistent with quasi-2D magnetic interactions (Calder et al., 2020). CrPS$$\mathbf{J} = \bigoplus_{\alpha=1}^L \mathbf{W}^{(\alpha)} + \mathbf{D}\otimes \mathbf{I},$$2, by contrast, combines strong ferromagnetic in-plane couplings $$\mathbf{J} = \bigoplus_{\alpha=1}^L \mathbf{W}^{(\alpha)} + \mathbf{D}\otimes \mathbf{I},$$3 meV and $$\mathbf{J} = \bigoplus_{\alpha=1}^L \mathbf{W}^{(\alpha)} + \mathbf{D}\otimes \mathbf{I},$$4 meV, weaker in-plane $$\mathbf{J} = \bigoplus_{\alpha=1}^L \mathbf{W}^{(\alpha)} + \mathbf{D}\otimes \mathbf{I},$$5 meV, weak antiferromagnetic interlayer coupling $$\mathbf{J} = \bigoplus_{\alpha=1}^L \mathbf{W}^{(\alpha)} + \mathbf{D}\otimes \mathbf{I},$$6 meV, and a small anisotropy $$\mathbf{J} = \bigoplus_{\alpha=1}^L \mathbf{W}^{(\alpha)} + \mathbf{D}\otimes \mathbf{I},$$7 meV, producing a layered antiferromagnet with quasi-1D chains embedded in the layers and a spin-flop transition in a small applied field (Calder et al., 2020).

In layered semiconductors, excitonic and magnetic layers can hybridize. In CrSBr, the exciton energy depends on the relative spin orientation of adjacent layers, $$\mathbf{J} = \bigoplus_{\alpha=1}^L \mathbf{W}^{(\alpha)} + \mathbf{D}\otimes \mathbf{I},$$8, so optical excitons couple coherently to magnons of the layered antiferromagnet. By tilting the applied field away from a principal axis, the bright magnon hybridizes with an optically dark mode; by applying uniaxial strain, the magnon dispersion can be reshaped until a dispersionless dark magnon band emerges at critical strain (Diederich et al., 2022).

Light–matter interactions in twisted and stacked two-dimensional heterostructures add another layer degree of freedom. In moiré heterostructures, a twist angle $$\mathbf{J} = \bigoplus_{\alpha=1}^L \mathbf{W}^{(\alpha)} + \mathbf{D}\otimes \mathbf{I},$$9 produces a moiré period

$$\mathcal{H}(\mathbf{P}) = -\sum_{\alpha,\beta=1}^{L}\sum_{i,j=1}^{N} J^{(\alpha\beta)}_{ij} p_i^{(\alpha)} p_j^{(\beta)}.$$0

which modulates electronic and excitonic states and leads to flat bands, minibands, and moiré excitons. In parallel, hybrid photonic platforms couple these layered materials to cavities, metasurfaces, and plasmonic resonators, producing strong coupling, moiré exciton polaritons, enhanced nonlinearities, and ultrafast spin–valley dynamics (Sortino et al., 2024).

4. Screened, mediated, and multi-scale interaction kernels

Layered environments also reshape interactions through screening and mediation. The Wannier Function Continuum Electrostatics framework derives realistic Coulomb matrix elements in free standing layered materials and vertical heterostructures from bulk constrained-RPA data, Wannier functions, and continuum electrostatics. For monolayer and bilayer graphene it reproduces full ab-initio calculations of the Coulomb matrix elements within an accuracy of $$\mathcal{H}(\mathbf{P}) = -\sum_{\alpha,\beta=1}^{L}\sum_{i,j=1}^{N} J^{(\alpha\beta)}_{ij} p_i^{(\alpha)} p_j^{(\beta)}.$$1 eV or better, while showing that realistic Coulomb interactions in bilayer graphene can be manipulated on the eV scale by dielectric and metallic surroundings (Rösner et al., 2015).

A chemically different but conceptually related conclusion emerges for layered battery cathodes. Using cRPA, the Coulomb $$\mathcal{H}(\mathbf{P}) = -\sum_{\alpha,\beta=1}^{L}\sum_{i,j=1}^{N} J^{(\alpha\beta)}_{ij} p_i^{(\alpha)} p_j^{(\beta)}.$$2 and Hund $$\mathcal{H}(\mathbf{P}) = -\sum_{\alpha,\beta=1}^{L}\sum_{i,j=1}^{N} J^{(\alpha\beta)}_{ij} p_i^{(\alpha)} p_j^{(\beta)}.$$3 values for LiCoO$$\mathcal{H}(\mathbf{P}) = -\sum_{\alpha,\beta=1}^{L}\sum_{i,j=1}^{N} J^{(\alpha\beta)}_{ij} p_i^{(\alpha)} p_j^{(\beta)}.$$4, LiNiO$$\mathcal{H}(\mathbf{P}) = -\sum_{\alpha,\beta=1}^{L}\sum_{i,j=1}^{N} J^{(\alpha\beta)}_{ij} p_i^{(\alpha)} p_j^{(\beta)}.$$5, LiMnO$$\mathcal{H}(\mathbf{P}) = -\sum_{\alpha,\beta=1}^{L}\sum_{i,j=1}^{N} J^{(\alpha\beta)}_{ij} p_i^{(\alpha)} p_j^{(\beta)}.$$6, NaCoO$$\mathcal{H}(\mathbf{P}) = -\sum_{\alpha,\beta=1}^{L}\sum_{i,j=1}^{N} J^{(\alpha\beta)}_{ij} p_i^{(\alpha)} p_j^{(\beta)}.$$7, NaNiO$$\mathcal{H}(\mathbf{P}) = -\sum_{\alpha,\beta=1}^{L}\sum_{i,j=1}^{N} J^{(\alpha\beta)}_{ij} p_i^{(\alpha)} p_j^{(\beta)}.$$8, and NaMnO$$\mathcal{H}(\mathbf{P}) = -\sum_{\alpha,\beta=1}^{L}\sum_{i,j=1}^{N} J^{(\alpha\beta)}_{ij} p_i^{(\alpha)} p_j^{(\beta)}.$$9 were found not to deviate much between Li and Na compounds or between polymorphs with different layer stackings, indicating that the dominant role is played by the local environment rather than global structural features (Kim et al., 2020). This is a different screening problem from graphene, but the layered lesson is similar: environment matters, though not always in the same way or at the same scale.

In electrodynamics, the Green’s tensor of a layered system replaces the free-space propagator by a transfer-matrix-resolved sum of bulk radiation, surface plasmon polaritons, leaky modes, and boundary or Norton waves. The formalism preserves reciprocity,

${\rm Ag_6}M_2{\rm TeO_6}$0

and shows that nanohole interactions in a thin Au film are strongly mediated by surface plasmons propagating along the chain of holes (Johansson, 2010).

Layered superconductors provide a further example in which the interaction itself becomes multi-scale. In phenomenological vortex models with several attractive and repulsive length scales, phase diagrams contain conventional 2D lattice phases, five stripe phases, dimer, trimer, and tetramer phases, void phases, and stable low-temperature disordered phases, with transitions controlled by the applied magnetic field (Meng et al., 2016). In layered metals, weak interlayer hopping splits the main dHvA frequency into two nearby values, and Coulomb interactions then generate an additional oscillation at the small difference frequency, probing the short-range part of the Coulomb interaction within the layered material (Allocca et al., 2021). A common misconception is that such low-frequency oscillations must signal a small Fermi pocket; this work shows that interactions alone can produce them in a weakly warped Fermi cylinder (Allocca et al., 2021).

5. Multiplex competition, coordination, and neighbourhood structure

In multiplex network models, layered interactions are often studied as competition between distinct channels for the same finite activity budget. For a two-layer multiplex, the activity variable ${\rm Ag_6}M_2{\rm TeO_6}$1 and ${\rm Ag_6}M_2{\rm TeO_6}$2 leads to the Hamiltonian

${\rm Ag_6}M_2{\rm TeO_6}$3

where intralayer coordination favors localization on one layer and ${\rm Ag_6}M_2{\rm TeO_6}$4 favors delocalization across layers (Gómez-Gardeñes et al., 2015). The order parameter

${\rm Ag_6}M_2{\rm TeO_6}$5

distinguishes full localization, full localization on the opposite layer, and maximal delocalization. The loss of full localization on the leading layer occurs at

${\rm Ag_6}M_2{\rm TeO_6}$6

or ${\rm Ag_6}M_2{\rm TeO_6}$7 in the unweighted case, so the onset of delocalization is controlled by the minimum connectivity of the dominant layer (Gómez-Gardeñes et al., 2015).

A second multilayer coordination model separates learning and payoff layers. Agents imitate neighbours in their own population but earn payoffs by playing a coordination game across populations. The results identify skepticism about the wisdom of the crowd and local connectivity as the main conditions for full coordination, while polarized coordinated layers are only possible for all-to-all interactions (Lugo et al., 2014). The same paper also reports that local interactions allow for full coordination in the socially efficient Pareto-dominant strategy in spite of it being the riskier one (Lugo et al., 2014).

Structural measures for multi-layered social networks push this logic from Hamiltonians to neighbourhood analysis. A multi-layered social network is defined as ${\rm Ag_6}M_2{\rm TeO_6}$8, with edges ${\rm Ag_6}M_2{\rm TeO_6}$9 assigned to distinct layers. From this definition one can build multi-layered neighbourhoods $\times$0 and then derive cross-layer clustering coefficient, cross-layer degree centrality, and several versions of multi-layered degree centrality to quantify how many neighbours are connected to a node on at least $\times$1 layers and how strongly those neighbours connect among themselves (Bródka et al., 2012). In that setting, layered interactions are not only multiplex channels but also measurable structural signatures of relation diversity.

6. Layered interactions in human–AI conversational systems

In human–AI interaction research, layered interactions refer to representational layers placed around a conventional turn-based chat. Feedstack treats the raw conversational stream as an unstructured layer and overlays structured layers such as chapters, bookmarks, highlights, suggested queries, and excerpt references (Nguyen et al., 3 Jun 2025). Each chapter corresponds to a design principle, and the same underlying exchange can appear simultaneously as chronological chat, timeline bookmarks, topical chapters, highlighted key terms, and cross-references from principles back to snippets.

This layered conversational environment is designed to scaffold organizing, navigating, annotating, externalizing, and exploring feedback (Nguyen et al., 3 Jun 2025). Clicking a bookmark scrolls the chat to the corresponding message and automatically expands the related chapter; suggested queries provide prospective cues for the next turn; excerpt references create explicit links from abstract principles back to concrete dialogue segments. The work is framed as research-through-design and explicitly describes Feedstack as a design probe rather than a conclusive evaluation (Nguyen et al., 3 Jun 2025).

The important point is not terminological coincidence but structural analogy. As in multiplex networks or layered crystals, the system does not replace one interaction space with another. It superposes layers whose coupling changes what can be perceived, retrieved, and coordinated.

7. Recurring principles and unresolved issues

A recurring implication is that layered interactions rarely reduce to a simple sum of pairwise couplings. Three-body dispersion in exfoliation energies, multi-shell exchange in layered magnets, exciton–magnon hybridization in CrSBr, boundary-wave and plasmon contributions in layered Green’s functions, and interaction-generated low-frequency oscillations in layered metals all arise because the layer degree of freedom changes the effective many-body problem (Rumson et al., 8 Apr 2026, Diederich et al., 2022, Johansson, 2010, Allocca et al., 2021).

Another recurring principle is that geometry is decisive. Near-equilateral atom triples across adjacent sheets make the ATM term repulsive in many layered materials (Rumson et al., 8 Apr 2026); moiré periodicity localizes excitons in twisted heterostructures (Sortino et al., 2024); the minimum degree of the dominant layer sets the onset of delocalization in multiplex competition (Gómez-Gardeñes et al., 2015); and aperiodic stacking plus incoherent bilayer orientation define the Ag-rich domains of honeycomb layered tellurates (Masese et al., 2021).

The literature also identifies several objective cautions. Pairwise dispersion without three-body corrections can retain residual overbinding in layered solids (Rumson et al., 8 Apr 2026). RevPBE is not a good choice for layered solids in the XDM framework (Rumson et al., 8 Apr 2026). ATM is not recommended as a universal default in XDM because it scales poorly and is often small outside layered systems (Rumson et al., 8 Apr 2026). In multiplex coordination, polarized coordinated layers are model-dependent and require all-to-all interactions rather than local ones (Lugo et al., 2014). In conversational systems, layered affordances can increase cognitive load and are therefore introduced as optional, collapsible structures rather than mandatory workflow stages (Nguyen et al., 3 Jun 2025).

Taken together, these results suggest that layered interactions are best understood not as a single phenomenon but as a family of formalisms in which the layer is an active variable. Whether the system is a van der Waals crystal, a multiplex network, a nanophotonic film, a layered metal, or a conversational interface, the same analytical question recurs: which couplings are confined within layers, which are transmitted across them, and which collective effects appear only because multiple layers coexist at once.