Contact-Interaction Models Overview

• Contact-interaction models are formal representations where dominant coupling occurs locally at a point, boundary, or contact patch, enabling precise modeling across multiple domains.
• They underpin phenomena in quantum mechanics, ultracold atomic scattering, and QCD by employing self-adjoint boundary conditions, pseudopotentials, and symmetry-preserving regularizations.
• These models also drive innovations in computer vision, generative modeling, and continuum mechanics by explicitly representing contact through specialized state variables and constitutive laws.

Searching arXiv for recent and foundational papers using the phrase “contact interaction model” across the domains represented here. Search 1: general phrase “contact interaction model”. Search 2: point/contact interactions and spectral transition. Search 3: contact-guided human/object interaction generation. Search 4: symmetry-preserving contact interaction in Dyson–Schwinger / Bethe–Salpeter modeling. A contact-interaction model is a formal representation in which the dominant coupling is concentrated at contact: at a point, on a boundary, across a zero-range interaction, over a surface patch, or through temporally localized contact events. The phrase is therefore domain-dependent rather than unique. In mathematical physics it commonly denotes self-adjoint point interactions or singular boundary conditions; in ultracold-atom and many-body theory it denotes zero-range or two-channel interactions; in Dyson–Schwinger and Bethe–Salpeter studies it denotes a momentum-independent vector$\times$vector kernel; in human–object interaction it denotes explicit contact fields, contact maps, or contact-guided generative states; in continuum mechanics it denotes constitutive models for rough, seated, or deformable contact; and in network science it denotes stochastic models of contact activation and deactivation (Exner et al., 2022, Bugnion et al., 2014, Xing et al., 2022, Diller et al., 2023, Ager et al., 2018, Stehle et al., 2010).

1. Scope and major meanings

The common structural feature is locality: interaction is not mediated by a long-range kernel alone, but is encoded at coincidence, at a support manifold, or at a discrete contact event. The modeling consequences depend on the field. In quantum mechanics, contact is implemented by boundary conditions that classify self-adjoint extensions; in many-body wave models, by Bethe–Peierls constraints or local nonlinearities; in contact-aware graphics and generative modeling, by distance fields or patch-level constraints; in mechanics, by pressure, friction, shear, permeability, and gap constraints; and in temporal network models, by memory-dependent activation rules (Exner et al., 2022, Sørensen et al., 2011, Dwivedi et al., 7 Apr 2025, Desai et al., 2023, Ashton et al., 2018).

Domain	Representative contact representation	Representative papers
Spectral and point-interaction quantum mechanics	Boundary conditions at $x=0$; four-parameter self-adjoint family	(Exner et al., 2022, Thompson et al., 2018)
Ultracold and many-body theory	Bethe–Peierls boundary condition; two-channel contact coupling; pseudopotentials	(Bugnion et al., 2014, Sørensen et al., 2011)
Dyson–Schwinger / Bethe–Salpeter QCD	Momentum-independent vector$\times$vector kernel with symmetry-preserving regularization	(Serna et al., 2016, Xing et al., 2022, Serna et al., 2017, Xu et al., 29 Apr 2026)
Human–object interaction and graphics	Contact distances, contact maps, semantic contact, contact-area primitives	(Diller et al., 2023, Song et al., 2 Jul 2026, Dwivedi et al., 7 Apr 2025, Lakshmipathy et al., 2023)
Continuum mechanics and soft matter	Rough-surface porous layer, seat contact, liquid-drop deformation energy	(Ager et al., 2018, Desai et al., 2023, Vrban et al., 29 Sep 2025)
Social and temporal networks	Contact on/off durations, activity potentials, memory kernels	(Stehle et al., 2010, Ashton et al., 2018)

A recurring misconception is that “contact interaction” always means a Dirac $\delta$-potential. The literature here shows that the term ranges from the full four-parameter family of 1D point interactions to distance-based contact maps in diffusion models and to poroelastic rough-contact layers in fluid–structure interaction (Exner et al., 2022, Thompson et al., 2018, Diller et al., 2023, Ager et al., 2018).

2. Point interactions and spectral transition in quantum mechanics

In the spectral-theoretic usage, the contact-interaction model is a Schrödinger operator that is free in the bulk and singular only on a lower-dimensional support. A central example is the two-dimensional Hamiltonian on $\mathbb{R}^2$ with harmonic confinement in the transverse variable $y$,

$$(H\psi)(x,y)= -\partial_x^2\psi(x,y)+\big(-\partial_y^2+y^2\big)\psi(x,y),$$

together with a contact interaction supported on the axis $x=0$. The boundary data are organized as

$$\Psi(y)=\begin{pmatrix}\psi(0+,y)\ \psi(0-,y)\end{pmatrix},\qquad \Psi'(y)=\begin{pmatrix}\psi'(0+,y)\ -\psi'(0-,y)\end{pmatrix},$$

and the most general self-adjoint point interaction is parameterized by four real numbers $(\alpha,\beta,\Re\gamma,\Im\gamma)$. For $x=0$0, the model uses

$x=0$1

with a $x=0$2 coefficient matrix $x=0$3; for $x=0$4, one recovers the $x=0$5-type jump condition, while $x=0$6 gives the $x=0$7-type condition (Exner et al., 2022).

This generalizes the Smilansky–Solomyak model by replacing a single control parameter with two effective couplings,

$x=0$8

$x=0$9

or, for $\times$0,

$\times$1

After Hermite-mode reduction, the problem becomes a Jacobi-operator analysis, and the spectral transition occurs on the hypersurfaces $\times$2. For $\times$3, the absolutely continuous spectrum fills $\times$4; at $\times$5, it fills $\times$6; and for $\times$7, the spectrum is discrete in $\times$8. The paper also gives sharp asymptotics for the discrete-spectrum count near criticality (Exner et al., 2022).

The broader 1D point-interaction classification reaches the same conclusion from a different direction. A general contact interaction at $\times$9 may be written as

$\delta$0

and, in the Hermitian case, self-adjointness imposes $\delta$1, $\delta$2, $\delta$3, $\delta$4 with real $\delta$5 satisfying $\delta$6. The ordinary $\delta$7-potential is the special case $\delta$8. This larger family can support up to two bound states, unlike the conventional $\delta$9-interaction, and its periodic repetition yields generalized Kronig–Penney bands, including Dirac-cone band touching in the Hermitian case and PT-symmetry-breaking transitions in the PT-symmetric case (Thompson et al., 2018).

3. Zero-range many-body interactions, pseudopotentials, and local nonlinearities

In ultracold-atom theory, the contact interaction is the zero-range limit of low-energy $\mathbb{R}^2$0-wave scattering. In three dimensions it is represented by

$\mathbb{R}^2$1

or, more precisely, by the Fermi pseudopotential enforcing the Bethe–Peierls condition $\mathbb{R}^2$2 as $\mathbb{R}^2$3. The corresponding phase shift is $\mathbb{R}^2$4, with no scattering in higher partial waves. A standard difficulty is that the zero-range model is singular: the many-body wavefunction diverges as $\mathbb{R}^2$5 when two opposite-spin particles coalesce, which produces ultraviolet divergences and large fluctuations in numerical methods. The purpose of high-fidelity pseudopotentials is therefore not to replace contact physics by a different interaction, but to reproduce the contact phase shifts while removing the singular numerical behavior (Bugnion et al., 2014).

The pseudopotential constructions in this literature are explicitly phase-shift matched. The Troullier–Martins form matches scattering at a calibration energy $\mathbb{R}^2$6, while the ultratransferable pseudopotential minimizes the phase-shift misfit over $\mathbb{R}^2$7 using

$\mathbb{R}^2$8

On the repulsive branch at $\mathbb{R}^2$9, the paper reports that the Troullier–Martins pseudopotential is approximately two orders of magnitude more accurate than hard/soft sphere pseudopotentials across $y$0, while the ultratransferable form improves a further factor of $y$1 and keeps the absolute $y$2-wave phase-shift error below $y$3 for all $y$4 (Bugnion et al., 2014).

Near a Feshbach resonance, the contact model is naturally extended to a two-channel form. The open and closed radial components satisfy a coupled boundary condition at $y$5,

$y$6

which yields both a tunable scattering length and a finite, generically negative effective range. In the three-body sector, this two-channel contact interaction reduces the Efimov scaling factor between recombination minima below the universal zero-range value $y$7, with a larger reduction for larger effective range or, equivalently, narrower Feshbach resonances (Sørensen et al., 2011).

A different local-contact usage appears in self-gravitating wave models. In the $y$8-dimensional Schrödinger–Poisson model for fuzzy dark matter, the contact term is a cubic local nonlinearity,

$y$9

Here $$(H\psi)(x,y)= -\partial_x^2\psi(x,y)+\big(-\partial_y^2+y^2\big)\psi(x,y),$$0 is repulsive and $$(H\psi)(x,y)= -\partial_x^2\psi(x,y)+\big(-\partial_y^2+y^2\big)\psi(x,y),$$1 is attractive. The paper shows that such local self-interactions modify stationary density profiles and shift the shell-crossing time during gravitational collapse; in the reported example, attractive interaction advances shell crossing while repulsive interaction delays it (Rodríguez-Villalba et al., 26 Feb 2026).

4. Symmetry-preserving contact interaction in Dyson–Schwinger and Bethe–Salpeter theory

In continuum QCD modeling, a contact interaction is a momentum-independent vector$$(H\psi)(x,y)= -\partial_x^2\psi(x,y)+\big(-\partial_y^2+y^2\big)\psi(x,y),$$2vector kernel used inside the quark Dyson–Schwinger equation and meson Bethe–Salpeter equations. The defining replacement is

$$(H\psi)(x,y)= -\partial_x^2\psi(x,y)+\big(-\partial_y^2+y^2\big)\psi(x,y),$$3

or, in the notation of the heavy-light meson studies,

$$(H\psi)(x,y)= -\partial_x^2\psi(x,y)+\big(-\partial_y^2+y^2\big)\psi(x,y),$$4

This makes the dressed-quark propagator algebraic,

$$(H\psi)(x,y)= -\partial_x^2\psi(x,y)+\big(-\partial_y^2+y^2\big)\psi(x,y),$$5

with a momentum-independent mass function $$(H\psi)(x,y)= -\partial_x^2\psi(x,y)+\big(-\partial_y^2+y^2\big)\psi(x,y),$$6 obtained from the gap equation. The simplification is substantial, but it creates ultraviolet divergences and makes regularization part of the definition of the model (Serna et al., 2016, Xing et al., 2022).

The decisive issue is symmetry preservation. Naive contraction of tensor integrals or naive momentum shifts in divergent expressions violate spacetime-translation invariance and the Ward–Takahashi identities. The symmetry-preserving programs therefore regularize first and reduce tensor integrals afterward, either through a subtraction scheme or through proper-time regularization with explicit consistency conditions,

$$(H\psi)(x,y)= -\partial_x^2\psi(x,y)+\big(-\partial_y^2+y^2\big)\psi(x,y),$$7

together with the proper-time form

$$(H\psi)(x,y)= -\partial_x^2\psi(x,y)+\big(-\partial_y^2+y^2\big)\psi(x,y),$$8

These constructions are used to preserve both vector and axial Ward–Takahashi identities, to remove routing dependence, and to keep current conservation intact in bound-state calculations (Xing et al., 2022, Serna et al., 2016).

Within this framework, the same contact kernel has been used for light pseudoscalar and vector mesons, heavy-light charmed mesons, kaon form factors, and charmonium radiative decays. For heavy-light systems, the pseudoscalar Bethe–Salpeter amplitude is written as

$$(H\psi)(x,y)= -\partial_x^2\psi(x,y)+\big(-\partial_y^2+y^2\big)\psi(x,y),$$9

and the subtraction-based scheme yields masses and decay constants for $x=0$0, $x=0$1, $x=0$2, $x=0$3, $x=0$4, $x=0$5, $x=0$6, and $x=0$7 in good agreement with available experimental and lattice data (Serna et al., 2017). In the kaon application, the same symmetry-preserving regularization is used to derive self-consistent electromagnetic and $x=0$8 form factors, explicitly emphasizing that the regularization, rather than the bare kernel alone, determines whether gauge symmetry is maintained (Xing et al., 2022).

A recent extension introduces a dynamical quark anomalous magnetic moment into the contact-interaction model for charmonium radiative decays. The quark–photon vertex becomes

$x=0$9

with the anomalous magnetic moment encoded in $$\Psi(y)=\begin{pmatrix}\psi(0+,y)\ \psi(0-,y)\end{pmatrix},\qquad \Psi'(y)=\begin{pmatrix}\psi'(0+,y)\ -\psi'(0-,y)\end{pmatrix},$$0. The model describes $$\Psi(y)=\begin{pmatrix}\psi(0+,y)\ \psi(0-,y)\end{pmatrix},\qquad \Psi'(y)=\begin{pmatrix}\psi'(0+,y)\ -\psi'(0-,y)\end{pmatrix},$$1 and $$\Psi(y)=\begin{pmatrix}\psi(0+,y)\ \psi(0-,y)\end{pmatrix},\qquad \Psi'(y)=\begin{pmatrix}\psi'(0+,y)\ -\psi'(0-,y)\end{pmatrix},$$2, agrees with modern lattice-QCD estimates, and finds that the 2024 BESIII central value for $$\Psi(y)=\begin{pmatrix}\psi(0+,y)\ \psi(0-,y)\end{pmatrix},\qquad \Psi'(y)=\begin{pmatrix}\psi'(0+,y)\ -\psi'(0-,y)\end{pmatrix},$$3 lies above the range accommodated by the framework, whereas the 2026 result is naturally consistent with it (Xu et al., 29 Apr 2026).

5. Contact as an explicit state in human–object interaction modeling

In computer vision, graphics, and generative modeling, a contact-interaction model usually means that contact is represented directly rather than left as an emergent byproduct of kinematics. In text-driven human–object interaction generation, one formulation defines contact as per-marker distances from the human body to the object surface,

$$\Psi(y)=\begin{pmatrix}\psi(0+,y)\ \psi(0-,y)\end{pmatrix},\qquad \Psi'(y)=\begin{pmatrix}\psi'(0+,y)\ -\psi'(0-,y)\end{pmatrix},$$4

with $$\Psi(y)=\begin{pmatrix}\psi(0+,y)\ \psi(0-,y)\end{pmatrix},\qquad \Psi'(y)=\begin{pmatrix}\psi'(0+,y)\ -\psi'(0-,y)\end{pmatrix},$$5 body markers in CG-HOI. Human motion, object motion, and contact are then generated jointly in a diffusion model, and contact is used again at inference through a guidance energy

$$\Psi(y)=\begin{pmatrix}\psi(0+,y)\ \psi(0-,y)\end{pmatrix},\qquad \Psi'(y)=\begin{pmatrix}\psi'(0+,y)\ -\psi'(0-,y)\end{pmatrix},$$6

where $$\Psi(y)=\begin{pmatrix}\psi(0+,y)\ \psi(0-,y)\end{pmatrix},\qquad \Psi'(y)=\begin{pmatrix}\psi'(0+,y)\ -\psi'(0-,y)\end{pmatrix},$$7 is recomputed from the predicted geometry. The paper is explicit that it does not add explicit SDF-based penetration penalties or friction terms in training; physical plausibility emerges from joint modeling, contact-based object weighting, and contact-guided inference (Diller et al., 2023).

JointHOI uses the same principle but makes contact an “inner modality” in a single-stage diffusion model for bimanual hand–object interaction. Contact is represented as a per-hand, per-frame distance field over $$\Psi(y)=\begin{pmatrix}\psi(0+,y)\ \psi(0-,y)\end{pmatrix},\qquad \Psi'(y)=\begin{pmatrix}\psi'(0+,y)\ -\psi'(0-,y)\end{pmatrix},$$8 object anchors,

$$\Psi(y)=\begin{pmatrix}\psi(0+,y)\ \psi(0-,y)\end{pmatrix},\qquad \Psi'(y)=\begin{pmatrix}\psi'(0+,y)\ -\psi'(0-,y)\end{pmatrix},$$9

and joint frame tokens are

$(\alpha,\beta,\Re\gamma,\Im\gamma)$0

Inference uses Contact Inner Guidance,

$(\alpha,\beta,\Re\gamma,\Im\gamma)$1

which reduces penetration and floating without adding a separate physics loss during training (Song et al., 2 Jul 2026).

The same contact-centric turn appears in single-image reconstruction. InteractVLM estimates 3D human and object contact from a single in-the-wild image using a Render–Localize–Lift pipeline: multi-view rendering embeds 3D surfaces in 2D, a multi-view localization model predicts contact masks, and lifting maps those predictions back to 3D. Human contact is represented on SMPL+H vertices and object contact on object points, and the reconstruction stage uses a contact anchoring energy

$(\alpha,\beta,\Re\gamma,\Im\gamma)$2

so that contact functions as a geometric constraint under occlusion and depth ambiguity (Dwivedi et al., 7 Apr 2025).

An adjacent graphics usage is explicitly authoring-oriented rather than generative. “Contact Edit” describes contact areas as first-class primitives and introduces an axis-based contact model for intuitive modeling of hand–object interactions, with real-time approximately isometry-preserving operations on triangulated surfaces and support for movement between surfaces (Lakshmipathy et al., 2023). This suggests that, in this domain, a contact-interaction model is as much a representation problem as an optimization problem.

6. Continuum contact mechanics, rough interfaces, and soft particles

In continuum mechanics, contact-interaction models encode how pressure, friction, compression, shear, permeability, and roughness govern load transmission. For seated human vibration comfort, the relevant contact variables are contact area, pressure, friction, and deformation in compression and shear. The reported comparisons show that a deformable finite element backrest and a multibody backrest with explicit shear restraints both provide realistic results, whereas a friction-only tangential model produces poor lateral fidelity and drift. In this usage, contact is not a singular zero-range operator but a distributed constitutive interface law (Desai et al., 2023).

A more formal continuum construction appears in fluid–structure–contact interaction with rough surfaces. The unresolved rough interface is replaced by a homogenized poroelastic layer $(\alpha,\beta,\Re\gamma,\Im\gamma)$3 with porosity $(\alpha,\beta,\Re\gamma,\Im\gamma)$4, permeability $(\alpha,\beta,\Re\gamma,\Im\gamma)$5, Darcy-type flow, and hyperelastic skeleton deformation. The free-fluid/porous-layer and porous-layer/solid interfaces are coupled by traction balance, normal mass-flux continuity, Beavers–Joseph tangential slip, and Signorini-type normal contact constraints. Representative equations are

$(\alpha,\beta,\Re\gamma,\Im\gamma)$6

$(\alpha,\beta,\Re\gamma,\Im\gamma)$7

with CutFEM used to handle topological changes in the fluid domain down to zero gap (Ager et al., 2018).

In soft matter, the contact-interaction model for microgel suspensions treats each particle as a compressible liquid drop with bulk and surface energies,

$(\alpha,\beta,\Re\gamma,\Im\gamma)$8

Two dimensionless parameters govern the response,

$(\alpha,\beta,\Re\gamma,\Im\gamma)$9

At large surface tensions representative of microgels, the deformation energy is pairwise additive well beyond small indentations and can be approximated by a power law in the indentation with an exponent around $x=0$00 (Vrban et al., 29 Sep 2025).

At still smaller scales, contact formation between confined surfaces in solution can be modeled microscopically with a kinetic Monte Carlo scheme on a $x=0$01-dimensional solid-on-solid crystal. The model combines a discrete electric double-layer repulsion,

$x=0$02

with a short-range attractive interaction and a constant external normal load. It exhibits nucleation, Ostwald ripening, coalescence, and morphology transitions between islands, bands, and pits, and it shows a non-trivial dependence of stable contact size on external pressure that is phenomenologically similar to oscillatory hydration forces (Høgberget et al., 2020).

7. Temporal contact networks and stochastic social interaction

In network science, a contact-interaction model is a stochastic rule for the activation and deactivation of social contacts. One class of models assumes that, at short time scales, the contact network is a union of disconnected cliques. At each discrete step, one agent is selected; isolated agents can form pairs, and agents in groups can either leave or introduce isolated agents into the group. The transition probabilities depend on the time since last state change, through memory kernels $x=0$03 and $x=0$04. For the canonical inverse-time choice $x=0$05, interacting agents in a clique of size $x=0$06 have effective hazard

$x=0$07

which yields heavy-tailed residence-time distributions rather than exponential ones. Stationarity exists only in a restricted parameter regime, and as $x=0$08 the average group size diverges and a macroscopic clique emerges (Stehle et al., 2010).

A data-driven school-contact formulation is more statistical. Students are nodes, links switch on and off at $x=0$09-second resolution, and the model fits distributions for on-durations, off-durations, inter-event times, and activity potentials. The generative constructions are renewal-type models. In one version, each link alternates between on- and off-periods, with on-durations drawn from an exponential with a heterogeneous lognormal rate and off-durations drawn from a lognormal law. In another, network-level activation times are generated from a lognormal inter-event distribution, links are chosen with probabilities proportional to empirical or synthetic link-activity weights, and a small triangle-closing bias improves clustering and triangle counts (Ashton et al., 2018).

This temporal-network usage makes explicit that “contact interaction” need not refer to a physical force at all. It can denote a statistical process over contact events. A plausible implication is that the term has become a methodological label for any model in which contact itself is the primary state variable, regardless of whether the underlying system is a quantum Hamiltonian, a deformable interface, or a schoolroom proximity network (Stehle et al., 2010, Ashton et al., 2018).