betaFeedback

WorksPhysics and Astronomy

A total-Lagrangian vectorial lattice Boltzmann method for finite-strain hyperelastic dynamics

Abstract

Inspired by the vectorial lattice Boltzmann method for linear elastodynamics \citep{boolakee2025linear}, we construct a total-Lagrangian vectorial lattice Boltzmann formulation for two-dimensional finite-strain hyperelastic dynamics. The governing equations are first written as a conservative first-order system for the material velocity and the full deformation gradient. This representation separates the kinematic part of the dynamics from the constitutive closure: the first Piola–Kirchhoff stress is evaluated locally from the current deformation gradient and enters the lattice only through nonlinear flux moments. A D2Q4 stencil with six-component vector populations is then used to match the state and the two material-coordinate fluxes. The formulation includes a second-order population initialization, trapezoidally centered body forcing, displacement reconstruction by velocity quadrature, and half-way reconstructions for velocity Dirichlet and Neumann traction boundaries on grid-aligned domains. The resulting method preserves the local collide–stream structure of standard lattice Boltzmann schemes while adapting the vectorial first-order strategy from linear elastodynamics to hyperelastic finite-strain dynamics.

Introduction

The lattice Boltzmann method (LBM) [2, 3, 4] is designed to approximate solutions to a simplified kinetic equation by evolving a finite set of populations associated with discrete microscopic velocities. The gas-kinetic origin of these populations gives the method a local collide–stream structure [5, 6, 7, 8]. This structure is attractive for explicit time integration and parallel implementation. In its native form, the LBM recovers the nearly incompressible Navier–Stokes equations through statistical moments of the populations and has become a mature tool in fluid dynamics and transport [9, 10, 11, 12]. These algorithmic features have also motivated attempts to use LBM in solid mechanics.

The extension of LBM to solids is motivated by the possibility of retaining locality, explicitness, and favorable parallel scaling for continuum-mechanics problems. A natural starting point is linear elasticity. For linear elastostatics, Yin et al. [13] proposed an LBM based on displacement distribution functions for the linear elastic Lamé equations; this line of work was later characterized as a diffusion-type pseudo-time formulation whose steady state solves the elliptic equations of linear elastostatics [1]. Building on this idea, Boolakee et al. developed second-order accurate lattice Boltzmann schemes for quasi-static linear elasticity and corresponding second-order Dirichlet and Neumann boundary formulations on arbitrary curved two-dimensional domains [14, 15]. These works demonstrated that LBM can be constructed systematically for linear elastostatic boundary-value problems. At the same time, as noted in subsequent work [1], the need to march in pseudo-time until a steady state is reached limits the competitiveness of such approaches for static problems when compared with established finite element methods.

For solid dynamics, especially problems involving elastic-wave propagation, the algorithmic structure of LBM is more naturally aligned with the target equations. Early attempts include the lattice Boltzmann model for solid-body dynamics of Marconi and Chopard [16] and the elastic-wave formulation of O’Brien et al. for Poisson solids [17]. Murthy et al. [18] later introduced a lattice Boltzmann formulation for elastic wave propagation with a tunable Poisson ratio, based on a moment-chain construction. Escande et al. [19] provided a theoretical and stability analysis of a regular-lattice version of this approach and validated it for bulk and surface waves, while related developments addressed crack loading and boundary conditions for elastodynamics [20, 21]. These contributions showed that LBM can reproduce transient elastic responses and wave propagation in solids. Nevertheless, formulations based on scalar populations and standard velocity sets have limited freedom to represent the coupled flux structure of elasticity. In practice, moment-chain schemes have relied on additional moment equations, finite-difference corrections, or artificial dissipation to supply missing quantities and improve stability [1]. As a consequence, second-order consistency, stability for arbitrary material parameters, and accurate boundary treatment remain delicate issues [1, 21].

Within this scalar-population moment-chain framework, a recent development has extended the approach to geometrically and constitutively nonlinear elastodynamics [22]. In that formulation, stresses and deformation measures are evaluated in the reference configuration, and nonlinear constitutive behavior is incorporated through a forcing term. The method demonstrates the feasibility of LBM for finite-strain solid dynamics through benchmark problems such as uniaxial tension, simple shear, and bending waves. However, it still relies on finite-difference evaluations of gradients and divergences, and the nonlinear Piola stress does not enter the lattice through a direct moment representation of the physical fluxes. This leaves open the question of whether finite-strain hyperelastic dynamics can be formulated in a way that remains closer to the basic lattice Boltzmann algorithm while treating the solid-mechanics fluxes as primary moment-matched quantities.

In this paper, we pursue this direction using vectorial LBM. The key modification of a vectorial formulation is the use of vector-valued populations instead of scalar ones, while preserving the local collision and streaming steps. Vectorial kinetic formulations have been used in several contexts, including magnetohydrodynamics [23], hyperbolic conservation laws and relaxation schemes [24, 25, 26], and incompressible-flow discretizations with boundary and stability analysis [27]. Most relevant to the present work, Boolakee et al. [1] recently introduced a vectorial LBM for linear elastodynamics by first rewriting the displacement equation as an equivalent first-order hyperbolic system. They established second-order consistency, stability estimates under a CFL-like condition, second-order population initialization, and Dirichlet boundary conditions on rectangular domains. This result indicates that vector-valued populations provide the additional degrees of freedom needed to represent the state and fluxes of coupled solid-dynamics systems.

We construct here a total-Lagrangian vectorial lattice Boltzmann formulation for two-dimensional finite-strain hyperelastic dynamics. The governing equations are first written as a conservative first-order system for the material velocity and the full deformation gradient. In this representation the lattice remains fixed in the reference configuration, while the first Piola–Kirchhoff stress is evaluated locally from the current deformation gradient and enters the method through nonlinear flux moments. A D2Q4 stencil with six-component vector populations is then used to match the macroscopic state and the two material-coordinate fluxes. The formulation includes trapezoidally centered body forcing, a second-order population initialization, displacement reconstruction by velocity quadrature, and half-way reconstructions for velocity Dirichlet and nominal-traction Neumann boundaries on grid-aligned domains. Numerical experiments verify the construction for manufactured solutions, finite-strain benchmark tests, several hyperelastic constitutive laws, acoustic-tensor wave speeds, finite-amplitude periodic waves, and a bounded-domain cantilever bending wave.

The manuscript is organized as follows. Section 2 summarizes the total-Lagrangian finite-strain equations, hyperelastic closure, and first-order conservative form. Section 3 introduces the D2Q4×6 vectorial lattice Boltzmann discretization, including equilibrium moments, forcing, initialization, and boundary reconstruction. Section 4 presents the manufactured-solution and benchmark validation, while Section 5 examines constitutive response and finite-strain wave dynamics. Section 6 concludes with the main limitations and possible extensions.

Total-Lagrangian finite-strain elastodynamics

This section prepares the total-Lagrangian equations for the vectorial lattice Boltzmann discretization. We work on a fixed reference domain Ω0⊂R2\Omega_0 \subset\mathbb{R}^2 with material coordinate X=(X1,X2)T=(X,Y)T\mathbf{X} = (X_1, X_2)^{\mathsf{T}} = (X,Y)^{\mathsf{T}}. The current position is x(X,t)\mathbf{x}(\mathbf{X},t), the displacement is u(X,t)=x(X,t)−X\mathbf{u}(\mathbf{X},t) = \mathbf{x}(\mathbf{X},t) - \mathbf{X}, and the material velocity is v=∂tu\mathbf{v} = \partial_t \mathbf{u}. We write ∂A=∂/∂XA\partial_A = \partial/\partial X_A, use ∇Xu\nabla_{\mathbf{X}}\mathbf{u} for the material displacement gradient, and denote the identity tensor by

I. Latin indices i,k=1,2i,k=1,2 denote spatial components, capital indices A,B=1,2A,B=1,2 denote material-coordinate directions, and repeated indices are summed.

Let W(F)W(\mathbf{F}) be the strain-energy density per unit reference volume, P(F)=∂W/∂F\mathbf{P}(\mathbf{F})=\partial W/\partial\mathbf{F} the first Piola–Kirchhoff stress, and b=(b1,b2)T\mathbf{b}=(b_1,b_2)^{T} the prescribed body acceleration. In the nondimensional displacement variables used below, the hyperelastic equations can be written as [29, 30, 31]

∂t2ui=∂APiA(F)+bi,F=I+∇ ⁣Xu,PiA(F)=∂W∂FiA(F).(1)\partial_t^2u_i=\partial_A P_{iA}(\mathbf{F})+b_i,\qquad\mathbf{F}=\mathbf{I}+\nabla_{\!X}\mathbf{u},\qquad P_{iA}(\mathbf{F})=\frac{\partial W}{\partial F_{iA}}(\mathbf{F}). \tag*{(1)}

(1) combines material momentum balance with the hyperelastic constitutive map. The stress divergence depends on displacement gradients through F=I+∇ ⁣Xu\mathbf{F}=\mathbf{I}+\nabla_{\!X}\mathbf{u}, and the equation is second order in time. Classical LBM is naturally suited to first-order conservation and balance systems, since its local collision and streaming update is closed through moments representing macroscopic states and fluxes. For hyperelasticity, the corresponding step is to lower the order of (1): we introduce v=∂tu\mathbf{v}=\partial_t\mathbf{u}, evolve F\mathbf{F} directly, and obtain a first-order system. The vectorial LBM used below then carries this first-order system with vector-valued populations whose moments recover the state and the two material-coordinate fluxes.

Scaling and reference description

The total-Lagrangian description is the natural setting for a fixed Cartesian lattice: lattice nodes remain stationary, material points are indexed by their reference coordinates, and the stress divergence is written in terms of the first Piola–Kirchhoff stress. We nondimensionalize by a reference length L0L_0, density ρ0\rho_0, and velocity V0V_0. Thus X=Xdim/L0\mathbf{X}=\mathbf{X}^{\mathrm{dim}}/L_0, x=xdim/L0\mathbf{x}=\mathbf{x}^{\mathrm{dim}}/L_0, u=udim/L0\mathbf{u}=\mathbf{u}^{\mathrm{dim}}/L_0, t=V0tdim/L0t=V_0t^{\mathrm{dim}}/L_0, and v=vdim/V0\mathbf{v}=\mathbf{v}^{\mathrm{dim}}/V_0. The stress, body acceleration, and Lame parameters are scaled by

P=Pdimρ0V02,b=L0bdimV02,λ=λdimρ0V02,μ=μdimρ0V02.\mathbf{P}=\frac{\mathbf{P}^{\mathrm{dim}}}{\rho_0V_0^2},\qquad\mathbf{b}=\frac{L_0\mathbf{b}^{\mathrm{dim}}}{V_0^2},\qquad\lambda=\frac{\lambda^{\mathrm{dim}}}{\rho_0V_0^2},\qquad\mu=\frac{\mu^{\mathrm{dim}}}{\rho_0V_0^2}.

The deformation gradient F=∂x/∂X\mathbf{F}=\partial\mathbf{x}/\partial\mathbf{X} is dimensionless because it is a ratio of two lengths. If V0V_0 is chosen as the infinitesimal longitudinal wave speed cL,0dim=(λdim+2μdim)/ρ0c_{L,0}^{\mathrm{dim}}=\sqrt{(\lambda^{\mathrm{dim}}+2\mu^{\mathrm{dim}})/\rho_0}, then λ+2μ=1\lambda+2\mu=1. Other choices of V0V_0 simply rescale the nondimensional moduli.

The lattice spacing and time step used in § 3 are Δx=ΔXdim/L0\Delta x=\Delta X^{\mathrm{dim}}/L_0 and Δt=V0Δtdim/L0\Delta t=V_0\Delta t^{\mathrm{dim}}/L_0, so the nondimensional lattice speed is c=Δx/Δt=ΔXdim/(V0Δtdim)c=\Delta x/\Delta t=\Delta X^{\mathrm{dim}}/(V_0\Delta t^{\mathrm{dim}}). All variables below are nondimensional.

Kinematics, balance laws, and boundary conditions

Finite-strain dynamics are carried by the deformation gradient F=I+∇ ⁣Xu\mathbf{F}=\mathbf{I}+\nabla_{\!X}\mathbf{u} and its determinant J=det⁡FJ=\det\mathbf{F}. The admissible range is J>0J>0, since loss of orientation makes the hyperelastic stress and the Piola transformation ill-defined. In the present two-dimensional formulation F∈R2×2\mathbf{F}\in\mathbb{R}^{2\times2}. Plane-strain or plane-stress reductions of a three-dimensional material require separate constitutive reductions, which are outside the scope of this formulation.

Introducing v=∂tu\mathbf{v}=\partial_t\mathbf{u} in (1) and evolving F=I+∇ ⁣Xu\mathbf{F}=\mathbf{I}+\nabla_{\!X}\mathbf{u} directly gives the nondimensional first-order system

∂tvi=∂APiA(F)+bi,∂tFiA=∂Avi.(2)\partial_t v_i=\partial_A P_{iA}(\mathbf{F})+b_i,\qquad\partial_t F_{iA}=\partial_Av_i. \tag*{(2)}

The first equation is the momentum balance from (1) written for velocity. The second is the kinematic identity obtained by differentiating FiA=δiA+∂AuiF_{iA}=\delta_{iA}+\partial_Au_i in time. Thus (2) is the order-reduced form of the displacement equation. Once F\mathbf{F} is known at a lattice node, a hyperelastic stress can be evaluated locally; no finite-difference reconstruction of strain from displacement is needed in the bulk.

The initial conditions are prescribed as u(X,0)=u0(X)\mathbf{u}(\mathbf{X},0)=\mathbf{u}_0(\mathbf{X}) and v(X,0)=v0(X)\mathbf{v}(\mathbf{X},0)=\mathbf{v}_0(\mathbf{X}), which imply F(X,0)=I+∇ ⁣Xu0(X)\mathbf{F}(\mathbf{X},0)=\mathbf{I}+\nabla_{\!X}\mathbf{u}_0(\mathbf{X}). This initialization automatically satisfies the compatibility relation ∂XFi2−∂YFi1=0\partial_XF_{i2}-\partial_YF_{i1}=0 at t=0t=0. For smooth solutions, (2) preserves this compatibility at the continuum level.

Let the boundary be split into a displacement part ΓD\Gamma_D and a traction part ΓN\Gamma_N. On ΓD\Gamma_D, the physical condition is u=uD\mathbf{u}=\mathbf{u}_D. The first-order state evolves velocity, so the lattice boundary condition uses the corresponding boundary velocity vD=∂tuD\mathbf{v}_D=\partial_t\mathbf{u}_D. On ΓN\Gamma_N, the Neumann condition is written in total-Lagrangian form as

P(F)N=Tˉon ΓN.(3)\mathbf{P}(\mathbf{F})\mathbf{N}=\bar{\mathbf{T}}\qquad\text{on }\Gamma_N. \tag*{(3)}

where NN is the outward unit normal to the reference boundary and Tˉ\bar{T} is the prescribed nominal traction per unit reference length/area. This is the Neumann condition expressed in reference variables. The spatial traction σn=tˉ\boldsymbol{\sigma}\boldsymbol{n}=\bar{\boldsymbol{t}} and the nominal traction are related by Nanson’s formula, σn da=PN dA0\boldsymbol{\sigma}\boldsymbol{n}\, da=\boldsymbol{P}\boldsymbol{N}\, dA_{0} with P=JσF−T\boldsymbol{P}=J\boldsymbol{\sigma}\boldsymbol{F}^{-T} [30, 31]. Thus the total-Lagrangian prescribed traction is PN\boldsymbol{P}\boldsymbol{N}; a spatial traction σn\boldsymbol{\sigma}\boldsymbol{n} must first be converted to the reference configuration.

The boundary reconstructions in § 3.6 are written for grid-aligned rectangular boundaries, so N∈{(1,0),(−1,0),(0,1),(0,−1)}N \in\{(1,0),(-1,0),(0,1),(0,-1)\}. The half-way D2Q4 boundary construction imposes this algorithmic restriction.

Hyperelastic closure

The lattice method requires only a pointwise map from deformation gradient to stress and, for initialization, wave-speed estimates, and traction-boundary Newton iterations, its tangent. For a hyperelastic material with strain-energy density W(F)W(\boldsymbol{F}), these quantities are

P(F)=∂W∂F,CiAkB(F)=∂PiA∂FkB,δPiA=CiAkB δFkB.(4)\boldsymbol{P}(\boldsymbol{F})=\frac{\partial W}{\partial\boldsymbol{F}}, \qquad C_{iAkB}(\boldsymbol{F})=\frac{\partial P_{iA}}{\partial F_{kB}}, \qquad\delta P_{iA}=C_{iAkB}\,\delta F_{kB}. \tag*{(4)}

These two quantities play different roles in the first-order system below. The stress P(F)\boldsymbol{P}(\boldsymbol{F}) supplies the stress components of the material-coordinate flux vectors ΦX\boldsymbol{\Phi}_{X} and ΦY\boldsymbol{\Phi}_{Y} in Eq. (7). The tangent C\boldsymbol{C} supplies the corresponding flux derivatives with respect to F\boldsymbol{F}, which enter the Jacobian actions in Eq. (8).

The numerical method does not depend on a particular strain-energy form. The implementation used in the examples includes St. Venant–Kirchhoff, compressible neo-Hookean, logarithmic neo-Hookean, Mooney–Rivlin, Yeoh, and Gent closures. Their explicit W(F)W(\boldsymbol{F}) and P(F)\boldsymbol{P}(\boldsymbol{F}) are collected in Appendix § 8. In the main algorithm, any admissible hyperelastic law can be used as long as P(F)\boldsymbol{P}(\boldsymbol{F}) and the tangent actions in Eq. (4) are available for J>0J>0. The local wave-speed estimates used below also assume that the tangent remains strongly elliptic over the simulated states.

First-order system used by the lattice method

The vectorial LBM will approximate Eq. (2) through moments of six-component populations. The macroscopic state is

U=(v1,v2,F11,F12,F21,F22)T.(5)\boldsymbol{U}=(v_{1},v_{2},F_{11},F_{12},F_{21},F_{22})^{T}. \tag*{(5)}

With this ordering, the coupled velocity–deformation-gradient equations become

∂tU+∂XΦX(U)+∂YΦY(U)=B,B=(b1,b2,0,0,0,0)T,(6)\partial_{t}\boldsymbol{U}+\partial_{X}\boldsymbol{\Phi}_{X}(\boldsymbol{U})+\partial_{Y}\boldsymbol{\Phi}_{Y}(\boldsymbol{U})=\boldsymbol{B}, \qquad\boldsymbol{B}=(b_{1},b_{2},0,0,0,0)^{T}, \tag*{(6)}

where the two material-coordinate fluxes are

ΦX(U)=(−P11,−P21,−v1,0,−v2,0)T,ΦY(U)=(−P12,−P22,0,−v1,0,−v2)T.(7)\boldsymbol{\Phi}_{X}(\boldsymbol{U})=(-P_{11},-P_{21},-v_{1},0,-v_{2},0)^{T}, \qquad\boldsymbol{\Phi}_{Y}(\boldsymbol{U})=(-P_{12},-P_{22},0,-v_{1},0,-v_{2})^{T}. \tag*{(7)}

The signs follow from writing the equations in the conservative form ∂tU+∂AΦA=B\partial_{t}\boldsymbol{U}+\partial_{A}\boldsymbol{\Phi}_{A}=\boldsymbol{B}. Taking the divergence of ΦX\boldsymbol{\Phi}_{X} and ΦY\boldsymbol{\Phi}_{Y} gives −∂APiA-\partial_{A}P_{iA} in the velocity rows and −∂Avi-\partial_{A}v_{i} in the deformation-gradient rows, so Eq. (6) is exactly Eq. (2).

This representation is the finite-strain analogue of the first-order linear elastodynamic system used in vectorial LBM. The essential change is that the stress entries in ΦX\boldsymbol{\Phi}_{X} and ΦY\boldsymbol{\Phi}_{Y} are nonlinear functions of the last four components of U\boldsymbol{U}. Consequently, the flux Jacobians AX=∂ΦX/∂U\boldsymbol{A}_{X}=\partial\boldsymbol{\Phi}_{X}/\partial\boldsymbol{U} and AY=∂ΦY/∂U\boldsymbol{A}_{Y}=\partial\boldsymbol{\Phi}_{Y}/\partial\boldsymbol{U} are state dependent. Their action on an increment δU=(δv1,δv2,δF11,δF12,δF21,δF22)T\delta\boldsymbol{U}=(\delta v_{1},\delta v_{2},\delta F_{11},\delta F_{12},\delta F_{21},\delta F_{22})^{T} is

AXδU=(−δP11,−δP21,−δv1,0,−δv2,0)T,AYδU=(−δP12,−δP22,0,−δv1,0,−δv2)T,(8)\boldsymbol{A}_{X}\delta\boldsymbol{U}=(-\delta P_{11},-\delta P_{21},-\delta v_{1},0,-\delta v_{2},0)^{T}, \qquad\boldsymbol{A}_{Y}\delta\boldsymbol{U}=(-\delta P_{12},-\delta P_{22},0,-\delta v_{1},0,-\delta v_{2})^{T}, \tag*{(8)}

with δPiA=CiAkBδFkB\delta P_{iA}=C_{iAkB}\delta F_{kB}. These Jacobian-vector products are sufficient for initialization and local wave-speed estimates, allowing the implementation to avoid dense 6×66\times6 matrices.

For a unit material direction n\boldsymbol{n}, define the total-Lagrangian acoustic tensor [29, 30]

Qik(n;F)=nACiAkB(F)nB.(9)Q_{ik}(\boldsymbol{n};\boldsymbol{F})=n_{A}C_{iAkB}(\boldsymbol{F})n_{B}. \tag*{(9)}

When the material is strongly elliptic, a⋅Q(n;F)a>0a \cdot Q(\mathbf{n};\mathbf{F})a > 0 for all nonzero aa and all unit n\mathbf{n}. The eigenvalues of QQ are the squared local tangent wave speeds in direction n\mathbf{n}, after the reference density has been absorbed by nondimensionalization. This quantity is used in § 4 to audit the lattice speed against the largest tangent wave speed encountered in the validation cases.

D2Q4×6 vectorial lattice Boltzmann discretization

The previous section reduced finite-strain hyperelastic dynamics to a first-order system with one state vector and two physical fluxes. The vectorial LBM now has a clear task: store enough population information at each node to reconstruct UU, ΦX(U)\boldsymbol{\Phi}_X(U), and ΦY(U)\boldsymbol{\Phi}_Y(U) by moments. This is the same design principle used in the linear vectorial elastodynamic scheme, where vector-valued populations enable a first-order hyperbolic system to be approximated while preserving the local collide–stream structure of LBM [1]. The finite-strain extension keeps the algebraic moment structure and replaces constant elastic fluxes by the nonlinear Piola fluxes.

Lattice populations and what their moments represent

The material domain is discretized by a uniform Cartesian lattice with spacing Δx=Δy\Delta x = \Delta y and time step Δt\Delta t. The four D2Q4 directions are V={(1,0),(0,1),(−1,0),(0,−1)}\mathcal{V} = \{(1,0),(0,1),(-1,0),(0,-1)\}, and the velocity associated with q=(i,j)q=(i,j) is cq=c(i,j)\mathbf{c}_q=c(i,j), where c=Δx/Δtc=\Delta x/\Delta t. For each direction, the method stores a six-component population fij(X,t)∈R6\mathbf{f}_{ij}(\mathbf{X},t) \in\mathbb{R}^{6}. Here “population” denotes a numerical carrier of moments, with no gas-kinetic molecular interpretation for the solid model.

Each D2Q4 population is a six-component vector, so the four lattice directions provide four independent vector combinations. The continuum system requires three of them: the zeroth moment gives the state, and the two first moments give the XX- and YY-fluxes. The remaining vector degree of freedom has no continuum counterpart; we fix it by setting the symmetric difference between the horizontal and vertical population pairs to zero. This gives a unique nearest-neighbour equilibrium. With body acceleration, the state is recovered as

Unum(X,t)=∑(i,j)∈Vfij(X,t)+Δt2B(X,t).(10)U^{\mathrm{num}}(\mathbf{X},t)=\sum_{(i,j)\in\mathcal{V}}\mathbf{f}_{ij}(\mathbf{X},t)+\frac{\Delta t}{2}\mathbf{B}(\mathbf{X},t). \tag*{(10)}

The half-step source shift is the discrete analogue of trapezoidal source centering. It makes the body force contribute to the recovered state without contaminating the flux moments.

Equilibrium distribution by moment matching

The equilibrium populations are defined by requiring their moments to reproduce the continuum quantities:

∑qfqeq=U,∑qcifqeq=ΦX(U),∑qcjfqeq=ΦY(U),∑qc2(i2−j2)fqeq=0.(11)\sum_q \mathbf{f}^{\mathrm{eq}}_q=U,\qquad\sum_q ci\mathbf{f}^{\mathrm{eq}}_q=\boldsymbol{\Phi}_X(U),\qquad\sum_q cj\mathbf{f}^{\mathrm{eq}}_q=\boldsymbol{\Phi}_Y(U),\qquad\sum_q c^2(i^2-j^2)\mathbf{f}^{\mathrm{eq}}_q=0. \tag*{(11)}

Here and below q=(i,j)q=(i,j), and ∑q\sum_q means summation over V\mathcal{V}. Solving these four vector equations gives the compact expression

fijeq(U)=14[U+2c2{iΦX(U)+jΦY(U)}],(i,j)∈V.(12)\mathbf{f}^{\mathrm{eq}}_{ij}(U)=\frac{1}{4}\left[U+\frac{2}{c^2}\{i\boldsymbol{\Phi}_X(U)+j\boldsymbol{\Phi}_Y(U)\}\right],\qquad(i,j)\in\mathcal{V}. \tag*{(12)}

This equation is the central algebraic object of the method. It says that the equilibrium distribution is the lowest-order D2Q4 interpolation of one state and two fluxes. The finite-strain mechanics enter only through the evaluation of ΦX\boldsymbol{\Phi}_X and ΦY\boldsymbol{\Phi}_Y, i.e. through P(F)P(F).

Because the fluxes are nonlinear, the derivative of the equilibrium is needed whenever the method is expanded or initialized to second order. For any increment Z∈R6Z\in\mathbb{R}^{6},

(fijeq)′(U)Z=14[Z+2c2{iAX(U)Z+jAY(U)Z}].(13)\left(\mathbf{f}^{\mathrm{eq}}_{ij}\right)'(U)Z=\frac{1}{4}\left[Z+\frac{2}{c^2}\{iA_X(U)Z+jA_Y(U)Z\}\right]. \tag*{(13)}

Taking moments of this identity recovers ZZ, AXZA_XZ, and AYZA_YZ. Thus the equilibrium derivative carries exactly the linearized continuum flux information needed for a Taylor expansion of the lattice update.

Collision, streaming, and source centering

A time step begins by evaluating UnumU^{\mathrm{num}} from Eq. (10). The last four entries define FF; the selected hyperelastic model then gives P(F)\mathbf{P}(\mathbf{F}), and hence the fluxes in Eq. (7). The BGK collision step [3, 4, 28] relaxes each population toward the local equilibrium:

fij∗=fij+ω(fijeq(Unum)−fij)+Δt(2−ω)wijB,wij=18.(14)f_{ij}^{*}=f_{ij}+\omega\left(f_{ij}^{\mathrm{eq}}\left(U^{\mathrm{num}}\right)-f_{ij}\right)+\Delta t(2-\omega)w_{ij}B,\qquad w_{ij}=\frac{1}{8}. \tag*{(14)}

The post-collision population is then streamed to the neighbouring node in direction (i,j)(i,j):

fij(X+c(i,j)Δt,t+Δt)=fij∗(X,t).(15)f_{ij}\left(\mathbf{X}+c(i,j)\Delta t,t+\Delta t\right)=f_{ij}^{*}(\mathbf{X},t). \tag*{(15)}

The symmetric weights satisfy ∑qwq=1/2\sum_q w_q=1/2, ∑qciwq=0\sum_q c_iw_q=0, and ∑qcjwq=0\sum_q c_jw_q=0. Thus the source enters the zeroth moment consistently with Eq. (10), leaving the stress and kinematic flux moments unchanged.

The nondissipative second-order setting is ω=2\omega=2, for which the explicit forcing term in Eq. (14) vanishes because the source has already been time-centred in the state recovery. Values 1<ω<21<\omega<2 introduce kinetic dissipation, which can be useful in strongly nonlinear computations. The formal second-order nondissipative expansion is tied to ω=2\omega=2; the consistency argument is collected in Appendix § 9.

Second-order population initialization

The prescribed initial fields determine U0U_0 and leave the four populations underdetermined. An equilibrium initialization, fq(X,0)=fqeq(U0)f_q(\mathbf{X},0)=f_q^{\mathrm{eq}}(U_0), recovers the leading macroscopic state. The non-equilibrium part expected by the lattice dynamics is then absent, and the resulting initial kinetic layer reduces short-time accuracy. For smooth initial fields, following the asymptotic initialization used in vectorial LBM [1, 33], the leading correction follows by expanding the distribution one half time step backward along its lattice characteristic:

fq(X,0)=fqeq(U0)−Δt2Dqfqeq(U0),Dq=∂t+ci∂X+cj∂Y.(16)f_q(\mathbf{X},0)=f_q^{\mathrm{eq}}(U_0)-\frac{\Delta t}{2}D_qf_q^{\mathrm{eq}}(U_0),\qquad D_q=\partial_t+c_i\partial_X+c_j\partial_Y. \tag*{(16)}

No additional time derivative is prescribed. The continuum equation gives ∂tU\partial_tU, so the derivative in Eq. (16) can be evaluated from spatial derivatives of the initial fields. For q=(i,j)q=(i,j), set

Zq=DqU0=(Z1,Z2,Z3,Z4,Z5,Z6)T.(17)Z_q=D_qU_0=(Z_1,Z_2,Z_3,Z_4,Z_5,Z_6)^{T}. \tag*{(17)}

where all quantities on the right are evaluated at t=0t=0, and

Z1=b1+∂XP11+∂YP12+ci∂Xv1+cj∂Yv1,Z2=b2+∂XP21+∂YP22+ci∂Xv2+cj∂Yv2,Z3=∂Xv1+ci∂XF11+cj∂YF11,Z4=∂Yv1+ci∂XF12+cj∂YF12,Z5=∂Xv2+ci∂XF21+cj∂YF21,Z6=∂Yv2+ci∂XF22+cj∂YF22.(18)\begin{aligned} Z_1&=b_1+\partial_XP_{11}+\partial_YP_{12}+c_i\partial_Xv_1+c_j\partial_Yv_1,\\ Z_2&=b_2+\partial_XP_{21}+\partial_YP_{22}+c_i\partial_Xv_2+c_j\partial_Yv_2,\\ Z_3&=\partial_Xv_1+c_i\partial_XF_{11}+c_j\partial_YF_{11},\\ Z_4&=\partial_Yv_1+c_i\partial_XF_{12}+c_j\partial_YF_{12},\\ Z_5&=\partial_Xv_2+c_i\partial_XF_{21}+c_j\partial_YF_{21},\\ Z_6&=\partial_Yv_2+c_i\partial_XF_{22}+c_j\partial_YF_{22}. \tag*{(18)} \end{aligned}

The stress derivatives in Eq. (18) are computed from the material tangent, for example ∂αPiA=CiAkB∂αFkB\partial_\alpha P_{iA}=C_{iAkB}\partial_\alpha F_{kB} for α=X,Y\alpha=X,Y. Define

ZF=(Z3Z4Z5Z6).(19)Z_F=\begin{pmatrix} Z_3 & Z_4 \\ Z_5 & Z_6 \end{pmatrix}. \tag*{(19)}

The stress increment induced by ZFZ_F is δPiAq=CiAkBZF,kB\delta P_{iA}^{q}=C_{iAkB}Z_{F,kB}, or componentwise

δP11q=C1111Z3+C1112Z4+C1121Z5+C1122Z6,\delta P_{11}^{q}=C_{1111}Z_3+C_{1112}Z_4+C_{1121}Z_5+C_{1122}Z_6,
δP12q=C1211Z3+C1212Z4+C1221Z5+C1222Z6,\delta P_{12}^{q}=C_{1211}Z_3+C_{1212}Z_4+C_{1221}Z_5+C_{1222}Z_6,
δP21q=C2111Z3+C2112Z4+C2121Z5+C2122Z6,(20)\delta P_{21}^{q}=C_{2111}Z_3+C_{2112}Z_4+C_{2121}Z_5+C_{2122}Z_6, \tag*{(20)}
δP22q=C2211Z3+C2212Z4+C2221Z5+C2222Z6.(21)\delta P^{q}_{22}=C_{2211}Z_{3}+C_{2212}Z_{4}+C_{2221}Z_{5}+C_{2222}Z_{6}. \tag*{(21)}

Using Eq. (13), the correction is then the explicit vector

Ddfqeq(U0)=14(Z1−2c(i δP11q+j δP12q)Z2−2c(i δP21q+j δP22q)Z3−2icZ1Z4−2jcZ1Z5−2icZ2Z6−2jcZ2).(22)D_{\mathbf{d}}f_{\mathbf{q}}^{\mathrm{eq}}(\mathbf{U}_{0})=\frac{1}{4} \begin{pmatrix} Z_{1}-\frac{2}{c}\left(i\,\delta P^{q}_{11}+j\,\delta P^{q}_{12}\right)\\ Z_{2}-\frac{2}{c}\left(i\,\delta P^{q}_{21}+j\,\delta P^{q}_{22}\right)\\ Z_{3}-\frac{2i}{c}Z_{1}\\ Z_{4}-\frac{2j}{c}Z_{1}\\ Z_{5}-\frac{2i}{c}Z_{2}\\ Z_{6}-\frac{2j}{c}Z_{2} \end{pmatrix}. \tag*{(22)}

For manufactured solutions the derivatives are obtained analytically. For benchmark initial conditions, the same formula is evaluated using second-order finite differences, with one-sided stencils next to nonperiodic boundaries.

Displacement and stress recovery

The lattice evolves velocity and deformation gradient. Displacement is reconstructed only for output and for displacement error norms. A trapezoidal update is used: with u∗,0=u0−(Δt/2)v0u^{*,0}=u_{0}-(\Delta t/2)v_{0}, set

un=u∗,n+Δt2vn,u∗,n+1=un+Δt2vn.(23)u^{n}=u^{*,n}+\frac{\Delta t}{2}v^{n},\qquad u^{*,n+1}=u^{n}+\frac{\Delta t}{2}v^{n}. \tag*{(23)}

The stress used by the dynamics is always the first Piola stress P(F)P(F). When Cauchy stress is needed for diagnostics or comparison with spatial benchmark fields, it is recovered by σ=J−1PFT\sigma=J^{-1}PF^{T}.

Half-way boundary reconstruction

The boundary treatment should be read from the viewpoint of the lattice. At a boundary node, the macroscopic equation remains available; after streaming, one or more incoming populations are unknown. The role of the boundary condition is therefore to reconstruct each missing incoming population from the known outgoing population and the imposed physical boundary condition.

For a grid-aligned boundary, let NN be the outward reference normal. The missing incoming direction is d=(i,j)=−Nd=(i,j)=-N, and −d=N-d=N is the opposite outgoing direction. Following link-wise boundary reconstructions used in elastic-solid LBM [1, 15, 21], we use a half-way reconstruction of the form

fd(t+Δt)=Df−d∗(t)+Sd(t+Δt/2).(24)f_{d}(t+\Delta t)=D f_{-d}^{*}(t)+S_{d}(t+\Delta t/2). \tag*{(24)}

The signs in DD are fixed by the pair identities obtained from Eq. (12): fdeq+f−deq=U/2f_{d}^{\mathrm{eq}}+f_{-d}^{\mathrm{eq}}=U/2 and fdeq−f−deq=dAΦA(U)/cf_{d}^{\mathrm{eq}}-f_{-d}^{\mathrm{eq}}=d_{A}\Phi_{A}(U)/c. Thus anti-bounce-back imposes a state component, and bounce-back with a correction imposes a normal flux component.

Velocity Dirichlet boundary condition

For a prescribed displacement uDu_{D}, the first-order state sees vD=∂tuDv_{D}=\partial_{t}u_{D}. Anti-bounce-back reconstructs the velocity entries of the missing population. Bounce-back reconstructs the deformation-gradient entries with the compatible kinematic flux inserted. The rule is

DD=diag⁡(−1,−1,1,1,1,1),SdD=(vD12,vD22,N1vD1c,N2vD1c,N1vD2c,N2vD2c)T.(25)D_{D}=\operatorname{diag}(-1,-1,1,1,1,1),\qquad S_{d}^{D}=\left(\frac{v_{D1}}{2},\frac{v_{D2}}{2},\frac{N_{1}v_{D1}}{c},\frac{N_{2}v_{D1}}{c},\frac{N_{1}v_{D2}}{c},\frac{N_{2}v_{D2}}{c}\right)^{T}. \tag*{(25)}

The first two entries impose the boundary velocity. The last four entries express the normal kinematic fluxes of F11F_{11}, F12F_{12}, F21F_{21}, F22F_{22}. Hence a displacement boundary enters through ∂tuD\partial_{t}u_{D}, because the evolved lattice variables are velocity and deformation gradient.

Neumann boundary condition

For a total-Lagrangian Neumann boundary, the prescribed datum is the nominal traction T‾=PN\overline{T}=PN. Since d=−Nd=-N, the normal flux of the velocity rows satisfies dAΦA=(PN,0,0,0,0)Td_{A}\Phi_{A}=(PN,0,0,0,0)^{T} in its first two components. Bounce-back plus the traction correction reconstructs the velocity entries. Anti-bounce-back reconstructs the deformation-gradient entries using a boundary deformation gradient FbF^{b}:

DN=diag⁡(1,1,−1,−1,−1,−1),SdN=(T‾1c,T‾2c,F11b2,F12b2,F21b2,F22b2)T.(26)D_N = \operatorname{diag}(1,1,-1,-1,-1,-1), \qquad S_d^N = \left(\frac{\overline{T}_1}{c},\frac{\overline{T}_2}{c},\frac{F_{11}^{b}}{2},\frac{F_{12}^{b}}{2},\frac{F_{21}^{b}}{2},\frac{F_{22}^{b}}{2}\right)^{T}. \tag*{(26)}

The unresolved part is FbF^{b}. At a boundary whose normal is aligned with material direction AA, the tangential column of FbF^{b} is extrapolated from interior nodes; the normal column is determined from the two scalar traction equations PiA(Fb)NA=T‾iP_{iA}(F^{b})N_A=\overline{T}_i, i=1,2i=1,2. We solve this local nonlinear system by Newton iteration with line search and reject trial states with det⁡Fb≤0\det F^{b}\leq0. The procedure introduces no global boundary solve; the nonlinearity is confined to a two-unknown constitutive inversion at each boundary link.

The traction T‾\overline{T} is measured per reference area. Loads prescribed per current area, such as a spatial traction σn\boldsymbol{\sigma}\boldsymbol{n}, are pulled back to the reference boundary before they enter Eq. (26).

One step of the method

At the nn-th time step, the update consists of the following local operations:

  1. Recover the macroscopic state Unum,n=∑qfqn+(Δt/2)BnU^{\mathrm{num},n}=\sum_q f_q^n+(\Delta t/2)B^n. Its first two components give vn\boldsymbol{v}^{n}, and the last four define Fn=(F11n,F12n,F21n,F22n)F^n=(F_{11}^n,F_{12}^n,F_{21}^n,F_{22}^n).

  2. Evaluate the constitutive response Pn=P(Fn)P^n=P(F^n), then assemble the material-coordinate fluxes ΦXn=ΦX(Unum,n)\boldsymbol{\Phi}_X^n=\boldsymbol{\Phi}_X(U^{\mathrm{num},n}) and ΦYn=ΦY(Unum,n)\boldsymbol{\Phi}_Y^n=\boldsymbol{\Phi}_Y(U^{\mathrm{num},n}).

  3. Form the equilibrium populations fqeq,n=fqeq(Unum,n)f_q^{\mathrm{eq},n}=f_q^{\mathrm{eq}}(U^{\mathrm{num},n}) and apply the collision rule to obtain fq∗,nf_q^{*,n}.

  4. Stream each post-collision population along cq=c(i,j)\boldsymbol{c}_q=\boldsymbol{c}(i,j), i.e. fqn+1(X+cqΔt)=fq∗,n(X)f_q^{n+1}(\boldsymbol{X}+\boldsymbol{c}_q\Delta t)=f_q^{*,n}(\boldsymbol{X}).

  5. Complete boundary nodes by reconstructing each missing incoming population, fdn+1=Df−d∗,n+Sdn+1/2f_d^{n+1}=Df_{-d}^{*,n}+S_d^{n+1/2}, using the periodic, Dirichlet, or Neumann rule.

  6. Update output variables with un=u∗,n+(Δt/2)vn\boldsymbol{u}^{n}=\boldsymbol{u}^{*,n}+(\Delta t/2)\boldsymbol{v}^{n} and u∗,n+1=un+(Δt/2)vn\boldsymbol{u}^{*,n+1}=\boldsymbol{u}^{n}+(\Delta t/2)\boldsymbol{v}^{n}; when needed, recover σn=(Jn)−1Pn(Fn)T\boldsymbol{\sigma}^{n}=(J^n)^{-1}P^n(F^n)^{T}.

The first moments of the equilibrium populations are constrained to equal ΦX\boldsymbol{\Phi}_X and ΦY\boldsymbol{\Phi}_Y, so the streaming step applies the lattice stencil directly to these fluxes. In the velocity rows this gives the stress-divergence term ∂APiA\partial_A P_{iA}.

Accuracy against reference solutions and finite-strain benchmarks

This section assesses the formulation against reference solutions with increasing numerical complexity. The first two tests use exact manufactured fields generated by the method of manufactured solutions (MMS) and isolate the bulk nonlinear fluxes, the population initialization, and the half-way boundary reconstructions. The last two tests reproduce the uniaxial-tension and simple-shear benchmarks of Müller et al. [22] using an independent finite-element reference on the same benchmark definitions. The sequence is intended to build the numerical evidence from controlled reference fields to established finite-strain benchmark configurations.

All cases in this section use the nondimensional lattice speed

c=ΔxΔt=5.(27)c=\frac{\Delta x}{\Delta t}=5. \tag*{(27)}

The value was selected before the refinement studies and kept fixed under acoustic scaling. Table 1 also reports the largest tangent material wave speed observed along each validation trajectory. The acoustic tensor Q(n;F)Q(\boldsymbol{n};F) from § 2.4 was evaluated over all nodes and time levels of the finest discretization. For the manufactured-solution cases, the audit was performed directly on the prescribed analytical deformation history. For the external benchmarks, it was performed on the deformation-gradient history recorded from the finest 1602160^2 numerical calculation. Since Q(n;F)=Q(−n;F)Q(\boldsymbol{n};F)=Q(-\boldsymbol{n};F), the directional maximization was approximated by scanning θ∈[0,π]\theta\in[0,\pi] with an angular increment of 0.25∘0.25^\circ. All measured ratios cmax⁡tan⁡/cc_{\max}^{\tan}/c are below one; the largest value is 0.5420.542, corresponding to cmax⁡tan⁡=2.71<c=5c_{\max}^{\tan}=2.71<c=5. Thus the lattice speed used in this section remains above the maximum physical wave speed encountered in each validation case.

CaseReference solutionMaterial and loadingω\omegacmax⁡tan/cc_{\max}^{\mathrm{tan}}/c
Periodic manufactured solutionAnalytic MMSlog-NH, λ=2/3\lambda= 2/3, μ=1\mu= 1, amplitude 0.0820.542
Boundary manufactured solutionAnalytic MMSlog-NH, λ=2/3\lambda= 2/3, μ=1\mu= 1, amplitude 0.0820.542
Uniaxial tensionQ1 FESVK, ν=0.20\nu= 0.20, α=0.175\alpha= 0.1752−1.6Δx2 - 1.6\Delta x0.359
Uniaxial tensionQ1 FENH, ν=−0.10\nu= -0.10, α=0.35\alpha= 0.352−1.6Δx2 - 1.6\Delta x0.271
Simple shearQ1 FESVK, ν=0\nu= 0, α=0.03\alpha= 0.032−1.6Δx2 - 1.6\Delta x0.320
Simple shearQ1 FENH, ν=0.20\nu= 0.20, α=0.10\alpha= 0.102−1.6Δx2 - 1.6\Delta x0.399

Table 1. Validation cases in Section 4. For the benchmark rows, Eq. (28) gives ω=1.92,1.96,1.98,1.99\omega= 1.92, 1.96, 1.98, 1.99 on n=20,40,80,160n = 20, 40, 80, 160. The last column is the trajectory maximum of the tangent material wave-speed ratio.

The manufactured-solution cases use ω=2\omega= 2, the nondissipative relaxation limit used in the consistency assessment, because the exact fields are smooth and the target is the formal discretization error. The external tension and shear benchmarks use

ω=2−1.6Δx.(28)\omega= 2 - 1.6\Delta x. \tag*{(28)}

The departure 2−ω=O(Δx)2-\omega= O(\Delta x) adds a small refinement-vanishing lattice dissipation. It damps boundary and loading transients associated with the half-way traction and velocity reconstructions in the ramped benchmark problems, while preserving the continuum hyperelastic benchmark as the target problem.

Errors are reported in relative L2L^2 norms over all grid nodes and tensor components. For the manufactured-solution tests we use

EL2(u)=∥uh−u∥2∥u∥2,EL2(P)=∥Ph−P∥2∥P∥2,EL2(σ)=∥σh−σ∥2∥σ∥2E_{L2}(u) = \frac{\lVert u_h-u\rVert_2}{\lVert u\rVert_2}, \qquad E_{L2}(P) = \frac{\lVert P_h-P\rVert_2}{\lVert P\rVert_2}, \qquad E_{L2}(\sigma) = \frac{\lVert\sigma_h-\sigma\rVert_2}{\lVert\sigma\rVert_2}

For the boundary manufactured solution, the same quantities are also evaluated after removing two grid layers next to the boundary. In the finite-strain benchmark cases of Müller et al. [22], we additionally report the measures defined in that reference,

E2=1Nnode∥uh−uFE∥2∥uFE∥2,E∞=∥uh−uFE∥∞∥uFE∥2E_2 = \frac{1}{N_{\mathrm{node}}}\frac{\lVert u_h-u_{\mathrm{FE}}\rVert_2}{\lVert u_{\mathrm{FE}}\rVert_2}, \qquad E_\infty= \frac{\lVert u_h-u_{\mathrm{FE}}\rVert_\infty}{\lVert u_{\mathrm{FE}}\rVert_2}

where NnodeN_{\mathrm{node}} is the total number of lattice nodes and ∥⋅∥∞\lVert\cdot\rVert_\infty is the maximum over grid nodes and displacement components. We also retain the raw relative L2L^2 displacement error ∥uh−uFE∥2/∥uFE∥2\lVert u_h-u_{\mathrm{FE}}\rVert_2/\lVert u_{\mathrm{FE}}\rVert_2 for convergence plots; this raw quantity equals NnodeE2N_{\mathrm{node}}E_2.

Periodic manufactured solution

The periodic manufactured solution removes boundary effects and directly tests the nonlinear moment matching in Eq. (12). The reference problem is posed on Ω0=(0,1)2\Omega_0 = (0,1)^2 with periodic boundaries in both material-coordinate directions. The material is the logarithmic compressible neo-Hookean solid listed in Appendix A, with λ=2/3\lambda= 2/3 and μ=1\mu= 1. The exact displacement is

ux(X,Y,t)=Asin⁡(k(X−0.23t))cos⁡(k(Y−0.17t))sin⁡(k(t−0.11)),(29)u_x(X,Y,t) = A\sin\bigl(k(X-0.23t)\bigr)\cos\bigl(k(Y-0.17t)\bigr)\sin\bigl(k(t-0.11)\bigr), \tag*{(29)}
uy(X,Y,t)=Acos⁡(k(X+0.19t))sin⁡(k(Y−0.29t))cos⁡(k(t+0.07)),(30)u_y(X,Y,t) = A\cos\bigl(k(X+0.19t)\bigr)\sin\bigl(k(Y-0.29t)\bigr)\cos\bigl(k(t+0.07)\bigr), \tag*{(30)}

with A=0.08A = 0.08 and k=2πk = 2\pi. The body acceleration is evaluated from the exact balance,

b=∂t2u−∇X⋅P(I+∇Xu),(31)\mathbf{b} = \partial_t^2 \mathbf{u} - \nabla_X \cdot\mathbf{P}\left(\mathbf{I}+\nabla_X\mathbf{u}\right), \tag*{(31)}

so that the prescribed u\mathbf{u} satisfies Eq. (1) exactly. At t=0.1t = 0.1, the deformation is visibly finite: the exact Jacobian ranges from 0.781 to 1.205, and the maximum value of ∥F−I∥\lVert\mathbf{F}-\mathbf{I}\rVert is 0.244. The test therefore exercises the nonlinear stress fluxes in a clearly finite-strain regime.

Figure 1 compares the exact and numerical displacement magnitude on a 1282128^2 grid, together with the pointwise error and a representative centerline profile. The displacement field, local error pattern, and one-dimensional trace agree closely. The relative errors at this resolution are 2.28×10−42.28 \times10^{-4} for displacement, 5.03 ×10−4\times10^{-4} for first Piola stress, and 5.63 ×10−4\times10^{-4} for Cauchy stress; these values are included in the manufactured-solution error summary in Table 2.

Figure 1. Periodic manufactured solution at t=0.1t = 0.1 on a 1282128^2 grid. The panels compare the exact displacement magnitude, the D2Q4×\times6 result, the pointwise error, and a centerline trace.

CaseRegion or initializationEL2(u)E_{L2}(u)EL2(P)E_{L2}(P)EL2(σ)E_{L2}(\sigma)
Periodic MMS, n=128n = 128second-order initialization2.28×10−42.28 \times10^{-4}5.03×10−45.03 \times10^{-4}5.63×10−45.63 \times10^{-4}
Periodic MMS, n=128n = 128equilibrium initialization8.97×10−38.97 \times10^{-3}9.72×10−39.72 \times10^{-3}9.75×10−39.75 \times10^{-3}
Dirichlet boundary MMS, n=128n = 128full domain6.37×10−56.37 \times10^{-5}2.74×10−42.74 \times10^{-4}3.02×10−43.02 \times10^{-4}
Dirichlet boundary MMS, n=128n = 128interior5.82×10−55.82 \times10^{-5}2.28×10−42.28 \times10^{-4}2.51×10−42.51 \times10^{-4}
Mixed Neumann MMS, n=128n = 128full domain9.47×10−59.47 \times10^{-5}9.31×10−49.31 \times10^{-4}9.47×10−49.47 \times10^{-4}
Mixed Neumann MMS, n=128n = 128interior7.19×10−57.19 \times10^{-5}3.57×10−43.57 \times10^{-4}3.73×10−43.73 \times10^{-4}

Table 2. Representative relative errors from the manufactured-solution validation cases at the finest grid used in each study. Boundary interior errors remove two grid layers adjacent to the boundary.

The refinement study in Figure 2 separates the spatial-temporal discretization error from the population initialization error. With the second-order initialization of Eq. (16), the grid sequence shows second-order convergence for uu, PP, and σ\sigma. An equilibrium-only initialization leaves a short initial kinetic layer and gives approximately first-order behavior over the same time interval, as shown by the open-symbol curves. The stress norms are included because the stress is the nonlinear flux carried by the lattice moments and is more sensitive to errors in FF than displacement alone.

Figure 2. Grid convergence for the periodic manufactured solution. The left panel reports second-order initialized errors in uu, PP, and σ\sigma. The right panel compares equilibrium initialization with the second-order population initialization.

Manufactured solution with boundary data

The second manufactured-solution test uses the same nonlinear reference field in Eq. (28), logarithmic neo-Hookean material, and body acceleration in Eq. (31), now on the bounded square Ω0=(0,1)2\Omega_0=(0,1)^2. Two boundary configurations are considered. The all-Dirichlet case prescribes the exact material velocity on each side,

v=∂tuon ∂Ω0.(32)v = \partial_t u \quad\text{on } \partial\Omega_0 . \tag*{(32)}

The mixed case imposes exact velocity conditions on the left and bottom sides and exact first-Piola traction on the right and top sides,

v=∂tuon X=0 and Y=0,PN=P(I+∇Xu)Non X=1 and Y=1.(33)v = \partial_t u \quad\text{on } X=0 \text{ and } Y=0,\qquad PN = P(I+\nabla_X u)N \quad\text{on } X=1 \text{ and } Y=1. \tag*{(33)}

The mixed case exercises the traction reconstruction in Eq. (26), including the local Newton solve for the boundary deformation gradient.

Figure 3 shows the full-domain and interior convergence histories. The all-Dirichlet case gives second-order convergence in displacement and stress. At n=128n=128, the full-domain relative errors are 6.37×10−56.37 \times10^{-5}, 2.74×10−42.74 \times10^{-4}, and 3.02×10−43.02 \times10^{-4} for uu, PP, and σ\sigma. Removing two boundary layers gives 5.82×10−55.82 \times10^{-5}, 2.28×10−42.28 \times10^{-4}, and 2.51×10−42.51 \times10^{-4}.

Figure 3. Manufactured solution with boundary data. The figure compares all-Dirichlet boundary conditions with mixed Dirichlet–Neumann boundary conditions and reports both full-domain and two-layer-stripped interior errors.

The mixed Neumann case has larger stress errors near the traction boundary, as expected for a nonlinear boundary inversion applied on a half-way lattice link. The interior errors remain systematically smaller than the full-domain values. At n=128n=128, the full-domain relative errors are 9.47×10−59.47 \times10^{-5}, 9.31×10−49.31 \times10^{-4}, and 9.47×10−49.47 \times10^{-4}, while the corresponding interior errors are 7.19×10−57.19 \times10^{-5}, 3.57×10−43.57 \times10^{-4}, and 3.73×10−43.73 \times10^{-4}. The interior convergence confirms second-order bulk accuracy of the nonlinear flux reconstruction, with the full-domain stress norms retaining a localized boundary contribution. Table 2 summarizes the finest-grid errors for both manufactured-solution tests.

Uniaxial tension benchmark

The first external benchmark is the uniaxial-tension problem used by Müller et al. [22]. It is posed on Ω0=(0,1)2\Omega_0 = (0,1)^2 with an initially undeformed body at rest. The shear modulus is fixed to μ=1\mu= 1, and the Lamé parameter is set from the two-dimensional Poisson ratio by λ=2μν/(1−2ν)\lambda= 2\mu\nu/(1 - 2\nu). The two material cases use the SVK and compressible neo-Hookean laws listed in Appendix § 8. The top and bottom boundaries carry opposite nominal tractions,

PN=(0,gα(t))T on Y=1,PN=(0,−gα(t))T on Y=0,PN=0 on X=0,1,(34)PN = (0,g_{\alpha}(t))^{T}\ \text{on }Y = 1,\qquad PN = (0,-g_{\alpha}(t))^{T}\ \text{on }Y = 0,\qquad PN = 0\ \text{on }X = 0,1, \tag*{(34)}

where

gα(t)={αsin⁡2(πt/4),0≤t<2,α,t≥2.(35)g_{\alpha}(t) = \begin{cases} \alpha\sin^{2}(\pi t/4), & 0 \le t < 2,\\ \alpha, & t \ge2. \end{cases} \tag*{(35)}

We consider the SVK case with ν=0.20\nu= 0.20, α=0.175\alpha= 0.175, and the compressible neo-Hookean case with ν=−0.10\nu= -0.10, α=0.35\alpha= 0.35. The final time is t=2.2t = 2.2. The numerical displacement is compared against an independently generated Q1 finite-element reference sampled at the lattice nodes. The same displacement field is then used to compute the E2E_2 and E∞E_\infty benchmark metrics.

Figure 4 shows the deformed shape, the reference-error field, and the grid convergence for the two material choices. The raw relative L2L^2 displacement error shows second-order convergence for the vectorial lattice scheme. On the 1602160^2 grid, the raw relative errors are 5.25×10−55.25 \times10^{-5} for the SVK case and 8.77×10−58.77 \times10^{-5} for the neo-Hookean case. The comparison with the published moment-chain data of Müller et al. [22] is restricted to the matching reference-study grid resolution, n=40n = 40 (Nnode=1600N_{\mathrm{node}} = 1600). At this resolution, the present E2E_2 values are 4.78×10−74.78 \times10^{-7} and 9.39×10−79.39 \times10^{-7}, lower than the reported 3.7×10−53.7 \times10^{-5} and 1.1×10−41.1 \times10^{-4} values by factors of 7.8×1017.8 \times10^1 and 1.2×1021.2 \times10^2, respectively. The corresponding E∞E_\infty values are summarized in Table 3.

BenchmarkMaterial casennE2E_2E2E_2 ref.E∞E_\inftyE∞E_\infty ref.
Uniaxial tensionSVK, ν=0.20\nu= 0.20404.78×10−74.78 \times10^{-7}3.7×10−53.7 \times10^{-5}9.40×10−59.40 \times10^{-5}5.7×10−35.7 \times10^{-3}
Uniaxial tensionNH, ν=−0.10\nu= -0.10409.39×10−79.39 \times10^{-7}1.1×10−41.1 \times10^{-4}1.05×10−41.05 \times10^{-4}1.8×10−21.8 \times10^{-2}
Simple shearSVK, ν=0\nu= 0402.03×10−62.03 \times10^{-6}1.5×10−51.5 \times10^{-5}2.70×10−42.70 \times10^{-4}1.6×10−31.6 \times10^{-3}
Simple shearNH, ν=0.20\nu= 0.20402.59×10−62.59 \times10^{-6}7.6×10−67.6 \times10^{-6}5.94×10−45.94 \times10^{-4}1.0×10−31.0 \times10^{-3}

Table 3. Benchmark comparison with the published moment-chain data of Müller et al. [22] at n=40n = 40 (Nnode=1600N_{\mathrm{node}} = 1600). The quantities E2E_2 and E∞E_\infty follow the benchmark definitions in that reference.

Figure 4. Uniaxial-tension benchmark of Müller et al. [22]. The panels show the final deformation, displacement error against the Q1 finite-element reference, raw relative L2L^2 convergence, and E2E_2 comparison at the reference-study grid resolution n=40n = 40 (Nnode=1600N_{\mathrm{node}} = 1600).

Simple shear benchmark

The second external benchmark is the simple-shear problem from the same reference. It uses the same SVK and compressible neo-Hookean laws and the same initially undeformed state on Ω0=(0,1)2\Omega_0 = (0, 1)^2. The lateral boundaries are traction free, the bottom boundary is fixed in velocity, and the top boundary is driven tangentially:

PN=0on X=0,1,v=0on Y=0,v=(hα(t),0)Ton Y=1,(36)\mathbf{P}\mathbf{N} = 0 \quad\text{on } X = 0,1,\qquad\mathbf{v} = 0 \quad\text{on } Y = 0,\qquad\mathbf{v} = (h_\alpha(t), 0)^T \quad\text{on } Y = 1, \tag*{(36)}

with

hα(t)={αsin⁡(πt/2),0≤t<2,0,t≥2.(37)h_\alpha(t) = \begin{cases} \alpha\sin(\pi t/2), & 0 \le t < 2,\\ 0, & t \ge2. \end{cases} \tag*{(37)}

The corresponding top displacement is ux=2α[1−cos⁡(πt/2)]/πu_x = 2\alpha[1 - \cos(\pi t/2)]/\pi during the ramp and ux=4α/πu_x = 4\alpha/\pi after t=2t = 2. This case probes the coupling between shear deformation, normal stress response, and traction-driven boundary evolution. We use the SVK case with ν=0\nu= 0 and α=0.03\alpha= 0.03, and the neo-Hookean case with ν=0.20\nu= 0.20 and α=0.10\alpha= 0.10. As in the tension benchmark, all errors are measured against an independent Q1 finite-element reference at t=2.2t = 2.2.

Figure 5 shows that the vectorial formulation follows the finite-element displacement field over the full loading interval. The convergence has larger error constants than in uniaxial tension, reflecting the stronger shear-driven boundary gradients in this setup. At n=160n = 160, the raw relative displacement errors are 2.95×10−42.95 \times10^{-4} for the SVK case and 5.80×10−45.80 \times10^{-4} for the neo-Hookean case. As above, the moment-chain comparison is restricted to the matching reference-study grid resolution, n=40n = 40. At this resolution, the present E2E_2 values are 2.03×10−62.03 \times10^{-6} and 2.59×10−62.59 \times10^{-6}, lower than the reported 1.5×10−51.5 \times10^{-5} and 7.6×10−67.6 \times10^{-6} values by factors of 7.4 and 2.9, respectively. The benchmark confirms that the same boundary and flux reconstruction used in the manufactured tests carries over to a reference finite-strain shear configuration.

Figure 5. Simple-shear benchmark of Müller et al. [22]. The figure reports final deformation, displacement error against the Q1 finite-element reference, raw relative L2L^2 convergence, and E2E_2 comparison at the reference-study grid resolution n=40n = 40 (Nnode=1600N_{\mathrm{node}} = 1600).

The benchmark comparisons at the reference-study grid resolution are summarized in Table 3.

Overall, the manufactured-solution and external-benchmark results support the numerical accuracy of the D2Q4×\times6 construction on periodic, Dirichlet, and mixed traction boundaries. They also show that the same flux and boundary reconstructions carry from controlled reference fields to finite-strain tension and shear benchmark configurations.

Constitutive response and finite-strain wave dynamics

The preceding validation cases establish the accuracy of the nonlinear flux discretization and the grid-aligned boundary closures. We now examine the same formulation in settings chosen to isolate material response, tangent dynamics about finite deformations, finite-amplitude wave propagation, and a bounded-domain bending wave. The first three cases use analytical or semi-analytical references. The final case uses an independent finite-element reference and provides a longer-time dynamic assessment on a cantilever geometry.

Affine finite-strain material response

A homogeneous deformation gives a direct view of the constitutive interface used by the lattice method. Consider an affine motion

x(X,t)=F(t)X+β(t),v(X,t)=F˙(t)X+β˙(t),(38)x(\mathbf{X},t)=\mathbf{F}(t)\mathbf{X}+\boldsymbol{\beta}(t),\qquad\mathbf{v}(\mathbf{X},t)=\dot{\mathbf{F}}(t)\mathbf{X}+\dot{\boldsymbol{\beta}}(t), \tag*{(38)}

where β(t)\boldsymbol{\beta}(t) is a time-dependent rigid translation and overdots denote time derivatives. The deformation gradient is spatially constant. Since P(F(t))\mathbf{P}(\mathbf{F}(t)) is also spatially constant, ∇X⋅P=0\nabla_{\mathbf{X}}\cdot\mathbf{P}=0. In the numerical calculations F˙\dot{\mathbf{F}} is constant during the loading interval, so the affine velocity boundary data generate an exact bulk solution without body forcing. This construction leaves the constitutive law free: any admissible W(F)W(\mathbf{F}) in Appendix A supplies an exact stress path.

The affine-patch calculations set β(t)=0\boldsymbol{\beta}(t)=0, because rigid translation does not affect F\mathbf{F}, P\mathbf{P}, or σ\boldsymbol{\sigma}. Over a loading interval 0≤t≤T0\leq t\leq T, the exact reference fields are

F(t)=I+tT(F∗−I),u(X,t)=[F(t)−I]X,v(X,t)=F∗−ITX,(39)\mathbf{F}(t)=\mathbf{I}+\frac{t}{T}(\mathbf{F}_{*}-\mathbf{I}),\qquad\mathbf{u}(\mathbf{X},t)=[\mathbf{F}(t)-\mathbf{I}]\mathbf{X},\qquad\mathbf{v}(\mathbf{X},t)=\frac{\mathbf{F}_{*}-\mathbf{I}}{T}\mathbf{X}, \tag*{(39)}

with stresses obtained pointwise as P(t)=P(F(t))\mathbf{P}(t)=\mathbf{P}(\mathbf{F}(t)) and σ(t)=J−1P(t)F(t)T\boldsymbol{\sigma}(t)=J^{-1}\mathbf{P}(t)\mathbf{F}(t)^{T}. The stress curves in Figure 6 use the terminal states F∗\mathbf{F}_{*} from the following four families:

Figure 6. Constitutive stress-response curves for homogeneous affine finite-strain paths. Solid curves are analytical stresses from the constitutive laws in Appendix § 8; open symbols are D2Q4×6 affine-patch samples on an 80280^2 grid.

F∗,uni=(λs001),F∗,aps=(λs00λs−1),F∗,sh=(1γ01),F∗,ss=(1.15κ01/1.15).(40)\mathbf{F}_{*,\mathrm{uni}}=\begin{pmatrix}\lambda_s&0\\0&1\end{pmatrix},\qquad\mathbf{F}_{*,\mathrm{aps}}=\begin{pmatrix}\lambda_s&0\\0&\lambda_s^{-1}\end{pmatrix},\qquad\mathbf{F}_{*,\mathrm{sh}}=\begin{pmatrix}1&\gamma\\0&1\end{pmatrix},\qquad\mathbf{F}_{*,\mathrm{ss}}=\begin{pmatrix}1.15&\kappa\\0&1/1.15\end{pmatrix}. \tag*{(40)}

They represent uniaxial stretch, area-preserving stretch, simple shear, and combined stretch–shear. Here λs\lambda_s is the scalar stretch parameter, γ\gamma is the simple-shear parameter, and κ\kappa is the shear parameter added to the fixed area-preserving stretch with axial stretch 1.15. In the stress-response curves, λs∈[1,1.45]\lambda_s\in[1,1.45] for uniaxial stretch, λs∈[1,1.40]\lambda_s\in[1,1.40] for area-preserving stretch, and γ,κ∈[0,0.80]\gamma,\kappa\in[0,0.80]. The material set consists of SVK, compressible neo-Hookean, logarithmic neo-Hookean, Mooney–Rivlin, Yeoh, and Gent solids, all evaluated through the same pointwise map P(F)\mathbf{P}(\mathbf{F}).

Figure 6 compares analytical stress-response curves with numerical affine-patch samples. The selected scalar response is P11P_{11} for uniaxial stretch, P11−P22P_{11}-P_{22} for area-preserving stretch, P12P_{12} for simple shear, and ∥P∥F\lVert\mathbf{P}\rVert_{\mathrm{F}} for the combined stretch–shear path. The numerical markers lie on the corresponding analytical curves across the six material laws. Thus the same lattice update follows distinct finite-strain stress responses through the local evaluation of P(F)\mathbf{P}(\mathbf{F}).

The corresponding tensor errors are shown in Figure 7. The refinement study uses n=20,40,80,160n=20,40,80,160, c=5c=5, second-order initialization, and the same refinement-scaled relaxation used in the external benchmarks. The deformation-gradient error decreases systematically for all four paths, with observed slopes between approximately 1.5 and 1.7 over the grid sequence. On the 1602160^2 grid, the largest relative errors over all material–path combinations are 8.34×10−68.34\times10^{-6} in F\mathbf{F}, 4.34×10−54.34\times10^{-5} in P\mathbf{P}, and 4.36×10−54.36\times10^{-5} in σ\boldsymbol{\sigma}. The stress errors therefore follow the directly evolved deformation gradient, confirming that the post-processed Piola and Cauchy stresses remain consistent with the affine exact solution.

Figure 7. Affine finite-strain patch errors. Panel (a) reports grid refinement of the deformation-gradient error for the four affine paths. Panel (b) summarizes the n=160n = 160 relative errors in FF, PP, and σ\sigma, taking the maximum over the six material laws for each path.

Wave speeds about predeformed states

The next case probes the tangent dynamics about finite homogeneous states. Let F0\mathbf{F}_0 be a constant predeformation and let a small displacement perturbation travel in the material direction n\mathbf{n}. Using the acoustic tensor introduced in Eq. (9), the analytical phase speeds are obtained from

Q(n;F0)q=cac2q,equivalentlyQik(n;F0)qk=cac2qi,(41)Q(\mathbf{n};\mathbf{F}_0)\mathbf{q}=c_{\mathrm{ac}}^2\mathbf{q}, \qquad\text{equivalently} \qquad Q_{ik}(\mathbf{n};\mathbf{F}_0)q_k=c_{\mathrm{ac}}^2q_i, \tag*{(41)}

where q=(q1,q2)T\mathbf{q}=(q_1,q_2)^{T} is the polarization vector, qiq_i denotes its ii-th component, and cacc_{\mathrm{ac}} is the tangent wave speed. The prescribed finite predeformations are then sampled directly through periodic wave propagation.

For a small perturbation amplitude ϵ\epsilon and wavenumber kk, the corresponding linearized reference field is

u′(X,t)=ϵqsin⁡(kn⋅X−kcact),v′(X,t)=−ϵkcacqcos⁡(kn⋅X−kcact),(42)\mathbf{u}'(\mathbf{X},t)=\epsilon\mathbf{q}\sin(k\mathbf{n}\cdot\mathbf{X}-kc_{\mathrm{ac}}t), \qquad\mathbf{v}'(\mathbf{X},t)=-\epsilon kc_{\mathrm{ac}}\mathbf{q}\cos(k\mathbf{n}\cdot\mathbf{X}-kc_{\mathrm{ac}}t), \tag*{(42)}

and

F(X,t)=F0+ϵkq⊗ncos⁡(kn⋅X−kcact)+O(ϵ2).(43)\mathbf{F}(\mathbf{X},t)=\mathbf{F}_0+\epsilon k\mathbf{q}\otimes\mathbf{n}\cos(k\mathbf{n}\cdot\mathbf{X}-kc_{\mathrm{ac}}t)+O(\epsilon^2). \tag*{(43)}

The phase-speed comparison below uses the material direction n=(1,0)T\mathbf{n}=(1,0)^T, k=2πk=2\pi, and ϵ=10−4\epsilon=10^{-4}. The computations use periodic domains, n=80n=80, c=5c=5, and ω=2\omega=2. Two predeformations are considered,

F0=(1.20001/1.20)andF0=(10.3001),\mathbf{F}_0= \begin{pmatrix} 1.20 & 0\\ 0 & 1/1.20 \end{pmatrix} \qquad\text{and}\qquad \mathbf{F}_0= \begin{pmatrix} 1 & 0.30\\ 0 & 1 \end{pmatrix},

corresponding to area-preserving stretch and simple shear. For each material and each F0\mathbf{F}_0, the slow and fast acoustic branches are initialized with their eigenvectors. The numerical phase speed is then measured from the phase shift of the projected velocity field. Specifically, the simulated velocity is projected onto the polarization,

vq(X,t)=q⋅v(X,t),v_q(\mathbf{X},t)=\mathbf{q}\cdot\mathbf{v}(\mathbf{X},t),

and the fundamental Fourier mode is fitted as

ac(t)=2⟨vq(⋅,t)cos⁡(kX)⟩,as(t)=2⟨vq(⋅,t)sin⁡(kX)⟩,ϕ(t)=atan2⁡(as(t),ac(t)).a_c(t)=2\langle v_q(\cdot,t)\cos(kX)\rangle,\qquad a_s(t)=2\langle v_q(\cdot,t)\sin(kX)\rangle,\qquad\phi(t)=\operatorname{atan2}(a_s(t),a_c(t)).

Here atan2⁡\operatorname{atan2} denotes the two-argument arctangent. After unwrapping the phase relative to the initial value, the measured lattice phase speed is cnum=Δϕ/(kt)c_{\mathrm{num}}=\Delta\phi/(kt).

Figure 8 shows the comparison between acoustic-tensor speeds and lattice phase speeds. Across the six material laws, two predeformations, and both branches, the maximum relative speed difference is 2.18×10−42.18\times10^{-4}. The agreement demonstrates that the finite predeformation enters the measured dynamics through the current tangent moduli.

Figure 8. Wave speeds about predeformed states. Analytical speeds are obtained from the acoustic tensor Q(n;F0)Q(\mathbf{n}; \mathbf{F}_0), and open symbols denote phase speeds measured from D2Q4×\times6 simulations. The two branches correspond to the slow and fast eigenmodes of Eq. (41).

Finite-amplitude periodic shear wave

The acoustic-wave calculation uses an infinitesimal perturbation. A complementary periodic reference is available for a finite-amplitude shear wave in the compressible neo-Hookean and logarithmic neo-Hookean laws. Let

u1=0,u2(X,t)=Asin⁡(kX−cskt),cs=μ,(44)u_1=0,\qquad u_2(X,t)=A\sin(kX-c_sk t),\qquad c_s=\sqrt{\mu}, \tag*{(44)}

where AA is the displacement amplitude, kk is the wavenumber, and csc_s is the shear-wave speed. The calculations below use the unit periodic domain with k=4πk=4\pi; the reported shear amplitude is the product AkAk. This gives

F=(10Akcos⁡(kX−cskt)1),J=1.(45)\mathbf{F}= \begin{pmatrix} 1 & 0\\ Ak\cos(kX-c_sk t) & 1 \end{pmatrix}, \qquad J=1. \tag*{(45)}

The nonzero velocity component is

v2(X,t)=−Acskcos⁡(kX−cskt).(46)v_2(X,t)=-Ac_sk\cos(kX-c_sk t). \tag*{(46)}

Along this path the volumetric contribution vanishes and both constitutive laws give

P21(X,t)=μF21(X,t)=μAkcos⁡(kX−cskt).(47)P_{21}(X,t)=\mu F_{21}(X,t)=\mu A_k\cos(kX-c_skt). \tag*{(47)}

The field in Eq. (44) is therefore an exact travelling-wave solution on a periodic domain.

Figure 9 compares the exact and numerical profiles for μ=2\mu=2, n=80n=80, c=5c=5, ω=2\omega=2, and shear amplitudes Ak=0.2A_k=0.2 and Ak=0.5A_k=0.5. The figure reports the displacement, velocity, deformation-gradient, and Piola-stress components that participate in the wave. The largest relative L2L^2 error over the plotted cases is 7.36×10−47.36\times10^{-4}.

Figure 9. Finite-amplitude periodic shear wave for compressible neo-Hookean and logarithmic neo-Hookean materials. Solid curves are the exact travelling-wave solution; open symbols are D2Q4×6 results for shear amplitudes Ak=0.2A_k = 0.2 and Ak=0.5A_k = 0.5.

The same level of agreement is obtained for the neo-Hookean and logarithmic neo-Hookean closures, as expected from their identical response on this isochoric shear path with J=1J=1 and finite shear component F21F_{21}.

Cantilever bending-wave benchmark

The final example considers a bounded-domain wave problem on a cantilever beam. The reference domain is [0,4]×[0,1][0,4] \times[0,1], discretized by 160×40160 \times40 lattice nodes. The left edge is clamped, the top and bottom edges are traction free, and the right edge receives the vertical nominal traction

Tˉ(Y,t)=(0,pR(t)w(Y))T,pR(t)=2p03sπ1/4(1−ζ2)e−ζ2/2,ζ=t−t0s.\bar{\mathbf{T}}(Y,t) = (0,p_R(t)w(Y))^{T}, \qquad p_R(t) = \frac{2p_0}{\sqrt{3}s\pi^{1/4}}(1-\zeta^2)e^{-\zeta^2/2}, \qquad\zeta= \frac{t-t_0}{s}.

for 0≤t≤20 \le t \le2, and Tˉ=0\bar{\mathbf{T}} = 0 afterwards. Here p0=0.15p_0 = 0.15, s=0.215s = 0.215, and t0=1t_0 = 1 are the Ricker-pulse amplitude, width, and centre time. The factor w(Y)w(Y) is a smooth right-edge taper that vanishes over eight lattice nodes near the loaded-end corners and is normalized to preserve the resultant load. The material is compressible neo-Hookean with μ=1\mu= 1 and ν=0.20\nu= 0.20. The calculation uses c=10c = 10, τ=0.5025\tau= 0.5025, equivalently ω=1.99005\omega= 1.99005, where τ=1/ω\tau= 1/\omega is the BGK relaxation time, and second-order initialization.

Figure 10 shows the deformed beam at t=2,5,8,12t = 2, 5, 8, 12, colored by the velocity magnitude. The initial pulse enters from the loaded end, reflects within the beam, and produces a bending-wave response over the full integration interval. At t=12t = 12, the simulation remains finite with 0.949≤J≤1.0430.949 \le J \le1.043. The maximum absolute tip displacement in the vertical direction is 1.54×10−21.54 \times10^{-2}, while the maximum displacement magnitude over the beam is 2.81×10−22.81 \times10^{-2}. These values indicate a small but clearly resolved finite-strain wave response on a nonperiodic domain with mixed boundary conditions.

Figure 10. Transient bending waves in the 160 × 40 neo-Hookean cantilever. The panels show the deformed configuration at t=2,5,8,12t=2,5,8,12, colored by the velocity magnitude.

For quantitative assessment, the same cantilever calculation is compared with an independent Q1 finite-element reference. The comparison uses the E2E_2 and E∞E_{\infty} measures introduced in § 4 for the external benchmarks, evaluated over the full lattice domain.

Figure 11 reports the time variation of E∞E_{\infty} and E2E_2 up to t=12t=12, together with the τ=0.55\tau=0.55 results of Müller et al. [22]. With τ=0.5025\tau=0.5025 and c=10c=10, the maximum errors are E2=5.55×10−5E_2=5.55\times10^{-5} and E∞=3.66×10−2E_{\infty}=3.66\times10^{-2}. A companion run at τ=0.55\tau=0.55 gives larger maxima, E2=1.44×10−4E_2=1.44\times10^{-4} and E∞=8.92×10−2E_{\infty}=8.92\times10^{-2}, showing the sensitivity of this corner-loaded wave problem to the relaxation parameter. Together with the deformed wave fields in Figure 10, this comparison supports the use of the same boundary reconstruction for longer-time bending-wave propagation.

Conclusions and outlook

This work has introduced a total-Lagrangian vectorial lattice Boltzmann formulation for two-dimensional finite-strain hyperelastic dynamics. The governing equations were written as a conservative first-order system for the material velocity and the full deformation gradient, allowing the lattice to remain fixed in the reference configuration while the first Piola–Kirchhoff stress is evaluated locally from F\mathbf{F}. A D2Q4 lattice with six-component vector populations was then used to match the macroscopic state and the two material-coordinate fluxes. In this way, finite-strain material nonlinearity enters only through nonlinear flux moments, whereas the method retains the local collide–stream structure of LBM. The formulation further includes trapezoidally centered forcing, second-order population initialization, displacement recovery by velocity quadrature, and half-way reconstructions for velocity Dirichlet and nominal-traction Neumann boundaries on grid-aligned domains.

The numerical results support the accuracy and applicability of the construction. Periodic manufactured solutions show second-order convergence for displacement, first Piola stress, and Cauchy stress when the second-order initialization is used. Boundary manufactured solutions confirm the expected behaviour for Dirichlet data and show that mixed traction boundaries mainly introduce localized stress errors, while the interior remains governed by the second-order bulk discretization. The uniaxial-tension and simple-shear benchmarks further demonstrate that the same flux and boundary reconstructions carry over to finite-strain benchmark configurations. The method also reproduces homogeneous affine stress responses for several hyperelastic laws, acoustic-tensor wave speeds about predeformed states, finite-amplitude periodic shear waves, and a bounded-domain cantilever bending wave.

Figure 11. Cantilever response compared with a finite-element reference. The present D2Q4×\times6 errors are measured against an independent Q1 reference using the benchmark definitions of E∞E_{\infty} and E2E_2, and are shown together with the τ=0.55\tau= 0.55 results of Müller et al. [22].

These results indicate that the proposed total-Lagrangian vectorial LBM can represent both nonlinear constitutive response and elastodynamic wave propagation within one explicit local framework. The formal consistency argument in Appendix § 9 supports this observation for smooth solutions: under acoustic scaling, ω=2\omega= 2, second-order initialization, and sufficiently accurate boundary data, the leading macroscopic field satisfies the target first-order hyperelastic system. Compared with the linear vectorial formulation, the essential change is the use of state-dependent Piola fluxes and tangent moduli; the moment-matching mechanism itself remains unchanged.

Several limitations remain and define natural directions for future work. The present study is restricted to two dimensions, uniform Cartesian lattices, and grid-aligned boundaries. A natural next step is to carry the same construction to three dimensions, where the state would contain three velocity components and the full 3×33 \times3 deformation gradient. The treatment of boundaries will also need to move beyond the present grid-aligned setting. For practical solid-mechanics applications, the method should be able to handle curved surfaces, cut cells, corners, and mixed displacement–traction data on arbitrary geometries, while preserving as much as possible the locality that makes LBM attractive. Another promising extension is anisotropic hyperelasticity. In a total-Lagrangian formulation, material directions are fixed in the reference configuration and can therefore be passed directly to the local constitutive update, making orthotropic and fiber-reinforced solids a natural target.

Data availability

The data underlying this article are available in Zenodo at https://doi.org/10.5281/zenodo.20376790.

Constitutive laws used in the implementation

Let

C=FTF,E=C−I2,J=det⁡F,H=F−T.(48)C = F^{T}F,\qquad E = \frac{C-I}{2},\qquad J = \det F,\qquad H = F^{-T}. \tag*{(48)}

For the invariant-based Mooney–Rivlin, Yeoh, and Gent closures, define

I1=tr⁡(C)+1,I2=J2+tr⁡(C),Iˉ1=J−2/3I1,Iˉ2=J−4/3I2,A=Iˉ1−3,(49)I_{1} = \operatorname{tr}(C)+1,\qquad I_{2} = J^{2}+\operatorname{tr}(C),\qquad\bar{I}_{1} = J^{-2/3}I_{1},\qquad\bar{I}_{2} = J^{-4/3}I_{2},\qquad A = \bar{I}_{1}-3, \tag*{(49)}

and

G1=∂Iˉ1∂F=J−2/3(2F−23I1H),G2=∂Iˉ2∂F=J−4/3[2F+(2J2−43I2)H].(50)G_{1} = \frac{\partial\bar{I}_{1}}{\partial F} = J^{-2/3}\left(2F-\frac{2}{3}I_{1}H\right),\qquad G_{2} = \frac{\partial\bar{I}_{2}}{\partial F} = J^{-4/3}\left[2F+\left(2J^{2}-\frac{4}{3}I_{2}\right)H\right]. \tag*{(50)}

The strain energies and first Piola stresses are

WSVK=λ2[tr⁡(E)]2+μE:E,W_{\mathrm{SVK}} = \frac{\lambda}{2}[\operatorname{tr}(E)]^{2}+\mu E:E,
PSVK=F[λtr⁡(E)I+2μE],(51)P_{\mathrm{SVK}} = F[\lambda\operatorname{tr}(E)I+2\mu E], \tag*{(51)}
WNH=μ2(tr⁡(C)−2−2log⁡J)+λ4(J2−1−2log⁡J),W_{\mathrm{NH}} = \frac{\mu}{2}\left(\operatorname{tr}(C)-2-2\log J\right)+\frac{\lambda}{4}\left(J^{2}-1-2\log J\right),
PNH=μ(F−H)+λ2(J2−1)H,(52)P_{\mathrm{NH}} = \mu(F-H)+\frac{\lambda}{2}(J^{2}-1)H, \tag*{(52)}
Wlog⁡NH=μ2(tr⁡(C)−2−2log⁡J)+λ2(log⁡J)2,W_{\log\mathrm{NH}} = \frac{\mu}{2}\left(\operatorname{tr}(C)-2-2\log J\right)+\frac{\lambda}{2}(\log J)^{2},
Plog⁡NH=μ(F−H)+λlog⁡JH,(53)P_{\log\mathrm{NH}} = \mu(F-H)+\lambda\log JH, \tag*{(53)}
WMR=c10(Iˉ1−3)+c01(Iˉ2−3)+λ2(log⁡J)2,W_{\mathrm{MR}} = c_{10}(\bar{I}_{1}-3)+c_{01}(\bar{I}_{2}-3)+\frac{\lambda}{2}(\log J)^{2},
PMR=c10G1+c01G2+λlog⁡JH,(54)P_{\mathrm{MR}} = c_{10}G_{1}+c_{01}G_{2}+\lambda\log JH, \tag*{(54)}
WYeoh=c1A+c2A2+c3A3+λ2(log⁡J)2,W_{\mathrm{Yeoh}} = c_{1}A+c_{2}A^{2}+c_{3}A^{3}+\frac{\lambda}{2}(\log J)^{2},
PYeoh=(c1+2c2A+3c3A2)G1+λlog⁡JH,(55)P_{\mathrm{Yeoh}} = (c_{1}+2c_{2}A+3c_{3}A^{2})G_{1}+\lambda\log JH, \tag*{(55)}
WGent=−μJm2log⁡(1−AJm)+λ2(log⁡J)2,W_{\mathrm{Gent}} = -\frac{\mu J_{m}}{2}\log\left(1-\frac{A}{J_{m}}\right)+\frac{\lambda}{2}(\log J)^{2},
PGent=μ2(1−A/Jm)G1+λlog⁡JH.(56)P_{\mathrm{Gent}} = \frac{\mu}{2(1-A/J_{m})}G_{1}+\lambda\log JH. \tag*{(56)}

When material parameters are specified through μ\mu and Poisson’s ratio ν\nu, the implementation uses λ=2μν/(1−2ν)\lambda=2\mu\nu/(1-2\nu). It sets c10+c01=μ/2c_{10}+c_{01}=\mu/2 for Mooney–Rivlin and c1=μ/2c_{1}=\mu/2 for Yeoh unless these parameters are supplied explicitly. Admissible states satisfy J>0J>0, with the additional Gent restriction A<JmA<J_{m}.

Formal second-order consistency

This appendix gives the Chapman–Enskog/Taylor expansion behind the second-order statement used in § 3. All lattice moments, fluxes, source terms, and boundary reconstructions are those already defined in the main text. The expansion assumes smooth fields, acoustic scaling with fixed c=Δx/Δtc=\Delta x/\Delta t, the nondissipative choice ω=2\omega=2, the initialization in Eq. (16), and half-way boundary data that are consistent to the order stated below.

Bulk Chapman–Enskog expansion

Use Δt\Delta t as the small parameter and write

fq=∑m≥0Δtmfq(m),Unum=∑m≥0ΔtmU(m),fqeq(Unum)=∑m≥0Δtmeq(m).(57)f_q=\sum_{m\ge0}\Delta t^m f_q^{(m)},\qquad U^{\mathrm{num}}=\sum_{m\ge0}\Delta t^m U^{(m)},\qquad f_q^{\mathrm{eq}}(U^{\mathrm{num}})=\sum_{m\ge0}\Delta t^m e_q^{(m)}. \tag*{(57)}

The equilibrium expansion is

eq(0)=fqeq(U(0)),e_q^{(0)}=f_q^{\mathrm{eq}}(U^{(0)}),
eq(1)=(fqeq)′(U(0))U(1),e_q^{(1)}=\left(f_q^{\mathrm{eq}}\right)'(U^{(0)})U^{(1)},
eq(2)=(fqeq)′(U(0))U(2)+12(fqeq)′′(U(0))[U(1),U(1)].(58)e_q^{(2)}=\left(f_q^{\mathrm{eq}}\right)'(U^{(0)})U^{(2)}+\frac{1}{2}\left(f_q^{\mathrm{eq}}\right)''(U^{(0)})[U^{(1)},U^{(1)}]. \tag*{(58)}

The Hessian term is present because the stress fluxes are nonlinear. It does not produce an additional macroscopic consistency term: the zeroth moment identity in Eq. (11) implies ∑qeq(m)=U(m)\sum_q e_q^{(m)}=U^{(m)} order by order.

From the source-shifted state recovery in Eq. (10),

∑qfq(0)=U(0),∑qfq(1)=U(1)−12B,∑qfq(m)=U(m)(m≥2).(59)\sum_q f_q^{(0)}=U^{(0)},\qquad\sum_q f_q^{(1)}=U^{(1)}-\frac{1}{2}B,\qquad\sum_q f_q^{(m)}=U^{(m)}\quad(m\ge2). \tag*{(59)}

For ω=2\omega=2, the explicit source term in Eq. (14) vanishes and the combined collide–stream equation becomes

fq(X+cqΔt,t+Δt)=2fqeq(Unum)−fq(X,t).(60)f_q(X+c_q\Delta t,t+\Delta t)=2f_q^{\mathrm{eq}}(U^{\mathrm{num}})-f_q(X,t). \tag*{(60)}

Expanding the left-hand side along the lattice characteristic gives

fq+ΔtDqfq+Δt22Dq2fq+O(Δt3)=2fqeq(Unum)−fq,(61)f_q+\Delta tD_qf_q+\frac{\Delta t^2}{2}D_q^2f_q+O(\Delta t^3)=2f_q^{\mathrm{eq}}(U^{\mathrm{num}})-f_q, \tag*{(61)}

with DqD_q defined in Eq. (16). Substitution of Eq. (B.1) into Eq. (B.5) gives the following balances.

At order Δt0\Delta t^0,

fq(0)=eq(0)=fqeq(U(0)).(62)f_q^{(0)}=e_q^{(0)}=f_q^{\mathrm{eq}}(U^{(0)}). \tag*{(62)}

At order Δt1\Delta t^1,

Dqfq(0)=2(eq(1)−fq(1)),fq(1)=eq(1)−12Dqeq(0).(63)D_qf_q^{(0)}=2\left(e_q^{(1)}-f_q^{(1)}\right),\qquad f_q^{(1)}=e_q^{(1)}-\frac{1}{2}D_qe_q^{(0)}. \tag*{(63)}

Taking the zeroth moment of Eq. (B.7) gives

∑qDqfq(0)=∂t∑qfq(0)+∂X∑qcifq(0)+∂Y∑qcjfq(0)=∂tU(0)+∂XΦX(U(0))+∂YΦY(U(0)),(64)\begin{aligned} \sum_q D_qf_q^{(0)}=\partial_t\sum_q f_q^{(0)}+\partial_X\sum_q c_i f_q^{(0)}+\partial_Y\sum_q c_j f_q^{(0)} \\ &=\partial_tU^{(0)}+\partial_X\Phi_X(U^{(0)})+\partial_Y\Phi_Y(U^{(0)}), \tag*{(64)} \end{aligned}

where Eq. (11) has been used with Eq. (B.6). The right-hand side is

2(∑qeq(1)−∑qfq(1))=2(U(1)−U(1)+12B)=B.(65)2\left(\sum_q e_q^{(1)}-\sum_q f_q^{(1)}\right)=2\left(U^{(1)}-U^{(1)}+\frac{1}{2}B\right)=B. \tag*{(65)}

Thus the leading field satisfies

∂tU(0)+∂XΦX(U(0))+∂YΦY(U(0))=B.(66)\partial_tU^{(0)}+\partial_X\Phi_X(U^{(0)})+\partial_Y\Phi_Y(U^{(0)})=B. \tag*{(66)}

At order Δt2\Delta t^2,

Dqfq(1)+12Dq2fq(0)=2(eq(2)−fq(2)).(67)D_qf_q^{(1)}+\frac{1}{2}D_q^2f_q^{(0)}=2\left(e_q^{(2)}-f_q^{(2)}\right). \tag*{(67)}

Using Eqs. (B.6) and (B.7), the left-hand side reduces exactly:

Dqfq(1)+12Dq2fq(0)=Dq(eq(1)−12Dqeq(0))+12Dq2eq(0)=Dqeq(1).(68)D_q f_q^{(1)} + \frac{1}{2}D_q^2 f_q^{(0)} = D_q\left(e_q^{(1)} - \frac{1}{2}D_q e_q^{(0)}\right) + \frac{1}{2}D_q^2 e_q^{(0)} = D_q e_q^{(1)}. \tag*{(68)}

Taking the zeroth moment of Eq. (B.11) therefore gives

∑qDqeq(1)=2(∑qeq(2)−∑qfq(2))=0.(69)\sum_q D_q e_q^{(1)} = 2\left(\sum_q e_q^{(2)}-\sum_q f_q^{(2)}\right)=0. \tag*{(69)}

The differentiated moment identities following from Eq. (13) give

∑qeq(1)=U(1),∑qcieq(1)=AX(U(0))U(1),∑qcjeq(1)=AY(U(0))U(1).(70)\sum_q e_q^{(1)}=U^{(1)},\qquad\sum_q c_i e_q^{(1)}=A_X(U^{(0)})U^{(1)},\qquad\sum_q c_j e_q^{(1)}=A_Y(U^{(0)})U^{(1)}. \tag*{(70)}

Consequently,

∂tU(1)+∂X(AX(U(0))U(1))+∂Y(AY(U(0))U(1))=0.(71)\partial_t U^{(1)}+\partial_X\left(A_X(U^{(0)})U^{(1)}\right)+\partial_Y\left(A_Y(U^{(0)})U^{(1)}\right)=0. \tag*{(71)}

This is the homogeneous equation for the first correction. The only finite-strain change relative to the linear vectorial calculation is that the Jacobians in Eq. (B.15) are evaluated along U(0)U^{(0)}; the nonlinear second derivative in Eq. (B.2) cancels from the zeroth-moment equation through the exact equilibrium moment identity.

Initialization and half-way boundaries

The initialization in Eq. (16) removes the first-order initial error. Comparing it with Eqs. (B.6)–(B.7) at t=0t=0 gives

(fqeq)′(U0)U(1)(X,0)=0.(72)\left(f_q^{eq}\right)'(U_0)U^{(1)}(X,0)=0. \tag*{(72)}

Summation over qq, together with the zeroth differentiated moment in Eq. (13), yields

U(1)(X,0)=0.(73)U^{(1)}(X,0)=0. \tag*{(73)}

For a missing direction d=−Nd=-N at a half-way boundary, Eq. (12) gives the pair identities

fdeq+f−deq=12U,fdeq−f−deq=1cdAΦA(U).(74)f_d^{eq}+f_{-d}^{eq}=\frac{1}{2}U,\qquad f_d^{eq}-f_{-d}^{eq}=\frac{1}{c}d_A\Phi_A(U). \tag*{(74)}

Thus anti-bounce-back imposes a state component at leading order, whereas bounce-back with a correction imposes the corresponding normal flux component.

For the velocity Dirichlet rule in Eq. (25), the anti-bounce-back velocity entries give

12v(0)=12vD,v(1)=0(75)\frac{1}{2}v^{(0)}=\frac{1}{2}v_D,\qquad v^{(1)}=0 \tag*{(75)}

at the boundary. The bounce-back deformation-gradient entries impose the kinematic normal flux because

1cdAΦAFiB(U(0))=NBvi(0)c,(76)\frac{1}{c}d_A\Phi_A^{F_{iB}}(U^{(0)})=\frac{N_Bv_i^{(0)}}{c}, \tag*{(76)}

which is exactly the correction used in Eq. (25). The first-order part is therefore homogeneous for the error equation.

For the Neumann rule in Eq. (26), the bounce-back velocity entries impose

1cdAΦAvi(U(0))=1cPiA(F(0))NA=Tˉic.(77)\frac{1}{c}d_A\Phi_A^{v_i}(U^{(0)})=\frac{1}{c}P_{iA}(F^{(0)})N_A=\frac{\bar{T}_i}{c}. \tag*{(77)}

The anti-bounce-back deformation-gradient entries impose the boundary value FbF^b. Hence this boundary is second-order consistent provided

Fb=F(Xb,t+Δt/2)+O(Δt2)(78)F^b = F(X_b,t+\Delta t/2) + O(\Delta t^2) \tag*{(78)}

and the local traction inversion defining FbF^b is solved to the same order. Under this condition the O(Δt)O(\Delta t) Neumann error data are homogeneous, namely the linearized traction condition δP N=0\delta P\,N=0 together with the corresponding state condition on the reconstructed deformation-gradient entries. If the tangential column extrapolation or local Newton solve is only first order, this boundary part no longer supplies the zero first-order data required by the formal second-order boundary argument, although the interior expansion above is unchanged.

With Eq. (B.17) and homogeneous first-order boundary data, uniqueness of the linearized homogeneous problem in Eq. (B.15) gives U(1)=0U^{(1)}=0. The leading field U(0)U^{(0)} satisfies the target first-order system by Eq. (B.10); therefore, for smooth solutions,

Unum=U(0)+ΔtU(1)+O(Δt2)=U+O(Δt2)=U+O(Δx2).(79)U^{\mathrm{num}}=U^{(0)}+\Delta t U^{(1)}+O(\Delta t^2)=U+O(\Delta t^2)=U+O(\Delta x^2). \tag*{(79)}

References

  1. [1]O. Boolakee, M. Geier, L. De Lorenzis, Lattice Boltzmann for linear elastodynamics: Periodic problems and Dirichlet boundary conditions, Comput. Methods Appl. Mech. Engrg. 433 (2025) 117469.DOI
  2. [2]G.R. McNamara, G. Zanetti, Use of the Boltzmann equation to simulate lattice-gas automata, Phys. Rev. Lett. 61 (20) (1988) 2332–2335.
  3. [3]S. Succi, The Lattice Boltzmann Equation for Fluid Dynamics and Beyond, Clarendon Press, Oxford, 2001.
  4. [4]T. Krüger, H. Kusumaatmaja, A. Kuzmin, O. Shardt, G. Silva, E.M. Viggen, The Lattice Boltzmann Method: Principles and Practice, Springer, Cham, 2017.
  5. [5]I.W. Haslam, R.S. Crouch, M. Seaïd, Coupled finite element–lattice Boltzmann analysis, Comput. Methods Appl. Mech. Engrg. 197 (51–52) (2008) 4505–4511.
  6. [6]H. Zhou, G. Mo, F. Wu, J. Zhao, M. Rui, K. Cen, GPU implementation of lattice Boltzmann method for flows with curved boundaries, Comput. Methods Appl. Mech. Engrg. 225–228 (2012) 65–73.
  7. [7]W. Verdier, P. Kestener, A. Cartalade, Performance portability of lattice Boltzmann methods for two-phase flows with phase change, Comput. Methods Appl. Mech. Engrg. 370 (2020) 132666.
  8. [8]P. Zhang, S. Sun, Y. Chen, S.A. Galindo-Torres, W. Cui, Coupled material point lattice Boltzmann method for modeling fluid–structure interactions with large deformations, Comput. Methods Appl. Mech. Engrg. 385 (2021) 110440.
  9. [9]J.-S. Feng, J.-C. Min, Lattice Boltzmann method simulation of two-phase flow in horizontal channel, Acta Phys. Sin. 72 (8) (2023) 084701.DOI
  10. [10]Y. Liu, J. Feng, J. Min, X. Zhang, Homogenized color-gradient lattice Boltzmann model for immiscible two-phase flow in multiscale porous media, J. Appl. Phys. 135 (18) (2024) 184701.
  11. [11]J. Feng, C. Wang, Y. Guo, Y. Liu, J. Min, M. Wang, K. Xu, Helmholtz-guided suppression of spurious currents in multicomponent pseudopotential model for high-fidelity immiscible flow in porous media, Phys. Fluids 38 (2) (2026) 026611.
  12. [12]J. Feng, J. Leng, J. Jiang, X. Chu, Entropic lattice Boltzmann method for general anisotropic advection–diffusion, arXiv:2605.01774, 2026.arxiv.org/abs/2605.01774
  13. [13]X. Yin, G. Yan, T. Li, Direct simulations of the linear elastic displacements field based on a lattice Boltzmann model, Internat. J. Numer. Methods Engrg. 107 (3) (2016) 234–251.
  14. [14]O. Boolakee, M. Geier, L. De Lorenzis, A new lattice Boltzmann scheme for linear elastic solids: periodic problems, Comput. Methods Appl. Mech. Engrg. 404 (2023) 115756.
  15. [15]O. Boolakee, M. Geier, L. De Lorenzis, Dirichlet and Neumann boundary conditions for a lattice Boltzmann scheme for linear elastic solids on arbitrary domains, Comput. Methods Appl. Mech. Engrg. 415 (2023) 116225.
  16. [16]S. Marconi, B. Chopard, A lattice Boltzmann model for a solid body, Internat. J. Modern Phys. B 17 (1–2) (2003) 153–156.
  17. [17]G.S. O’Brien, T. Nissen-Meyer, C.J. Bean, A lattice Boltzmann method for elastic wave propagation in a Poisson solid, Bull. Seismol. Soc. Am. 102 (3) (2012) 1224–1234.
  18. [18]J.S.N. Murthy, P.K. Kolluru, V. Kumaran, S. Ansumali, Lattice Boltzmann method for wave propagation in elastic solids, Commun. Comput. Phys. 23 (4) (2018) 1223–1240.
  19. [19]M. Escande, P.K. Kolluru, L.M. Cléon, P. Sagaut, Lattice Boltzmann method for wave propagation in elastic solids with a regular lattice: theoretical analysis and validation, arXiv:2009.06404, 2020.arxiv.org/abs/2009.06404
  20. [20]A. Schlüter, C. Kuhn, R. Müller, Lattice Boltzmann simulation of antiplane shear loading of a stationary crack, Comput. Mech. 62 (5) (2018) 1059–1069.
  21. [21]E. Faust, A. Schlüter, H. Müller, F. Steinmetz, R. Müller, Dirichlet and Neumann boundary conditions in a lattice Boltzmann method for elastodynamics, Comput. Mech. 73 (2) (2024) 317–339.
  22. [22]H. Müller, E. Faust, A. Schlüter, R. Müller, Extending the lattice Boltzmann method to non-linear elastodynamics, Comput. Methods Appl. Mech. Engrg. 443 (2025) 118076.
  23. [23]P.J. Dellar, Moment equations for magnetohydrodynamics, J. Stat. Mech. Theory Exp. 2009 (6) (2009) P06003.
  24. [24]S. Jin, Z. Xin, The relaxation schemes for systems of conservation laws in arbitrary space dimensions, Comm. Pure Appl. Math. 48 (3) (1995) 235–276.
  25. [25]B. Graille, Approximation of mono-dimensional hyperbolic systems: A lattice Boltzmann scheme as a relaxation method, J. Comput. Phys. 266 (2014) 74–88.
  26. [26]F. Dubois, Simulation of strong nonlinear waves with vectorial lattice Boltzmann schemes, Internat. J. Modern Phys. C 25 (12) (2014) 1441014.
  27. [27]J. Zhao, W.A. Yong, Vectorial finite-difference-based lattice Boltzmann method: Consistency, boundary schemes and stability analysis, J. Comput. Appl. Math. 441 (2024) 115677.
  28. [28]P.L. Bhatnagar, E.P. Gross, M. Krook, A model for collision processes in gases. I. Small amplitude processes in charged and neutral one-component systems, Phys. Rev. 94 (1954) 511–525.
  29. [29]R.W. Ogden, Non-Linear Elastic Deformations, Dover Publications, Mineola, 1997.
  30. [30]G.A. Holzapfel, Nonlinear Solid Mechanics: A Continuum Approach for Engineering, Wiley, Chichester, 2000.
  31. [31]J. Bonet, R.D. Wood, Nonlinear Continuum Mechanics for Finite Element Analysis, second ed., Cambridge University Press, Cambridge, 2008.
  32. [33]M. Junk, A. Klar, L.S. Luo, Asymptotic analysis of the lattice Boltzmann equation, J. Comput. Phys. 210 (2) (2005) 676–704.

Paper details

vlnc:W_avy9gb36
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents