A Robust Compressible APIC–FLIP Particle–Grid Method with Conservative Resampling and Adaptive APIC–PIC Blending
Jiansheng Yao, Yingkui Zhao · 2026
Read A Robust Compressible APIC–FLIP Particle–Grid Method with Conservative Resampling and Adaptive APIC–PIC Blending by Jiansheng Yao, Yingkui Zhao (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-03860.txt.
Introduction
Jiansheng Yao and Yingkui Zhao present a particle-grid simulation method for inviscid compressible flows with shocks and strong vortices. Such flows are hard for particle-grid methods because moving discontinuities, cell-crossing noise, and quadrature damage under deformation all cause trouble. Working from a combined FLIP and APIC framework with vorticity-aware artificial viscosity, the authors identify a long-time failure in Rayleigh-Taylor instability tests. Particles drain from spike heads, quadrature degrades, and nonphysical void-like dents appear.
Their fix adds conservative split resampling, which replenishes depleted cells while exactly conserving mass, momentum, and internal energy. It also adds a soft switch that weakens only the affine part of the update where particle support is thin. Tests on the Sod shock tube and Rayleigh-Taylor cases remove the voids while preserving vortex roll-up and matching reference growth measures.
Catalog record
| CREATOR | Jiansheng Yao, Yingkui Zhao |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 55840 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.03860
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A Robust Compressible APIC–FLIP Particle–Grid Method with Conservative Resampling and Adaptive APIC–PIC Blending Jiansheng Yao, Yingkui Zhao Abstract Modeling inviscid compressible flows with shocks and vortex-dominated dynamics remains challenging for particle–grid methods due to moving discontinuities, cell-crossing noise, and quadrature degradation under strong deformation. Building on a FLIP–APIC framework with vorticity-aware tensor artificial viscosity, we identify a long-time RTI failure mode: particle depletion at spike heads degrades quadrature and particle–grid coupling, producing nonphysical, void-like dents. Standard mitigations (CPDI-lite and subcell-jittered seeding) reduce but do not eliminate this artifact. We therefore add two sampling-aware controls: (i) conservative split resampling that replenishes depleted cells while exactly conserving mass, momentum, and internal energy; and (ii) a soft-switch that attenuates only the APIC affine term when local support is insufficient. Tests on the Sod shock tube and single-/multi-mode RTI show that the method removes spike-head voids in long-time RTI while preserving vortex roll-up, and matches reference Euler growth metrics. keywords compressible flow, particle–grid methods, FLIP, APIC, tensor artificial viscosity, conservative resampling, Rayleigh–Taylor instability 1 Introduction Robust discontinuity control and shock capturing are central requirements in compressible-flow simulation, particularly for problems involving shock propagation, multi-material interfaces, and vortical dynamics. Conventional Eulerian schemes have achieved great success; however, fixed spatial discretizations may diffuse shocks and interfaces unless high-resolution shock-capturing techniques (e.g., Riemann solvers, limiters, or adaptive refinement) are employed, and convection-dominated regimes typically rely on upwinding for stability. WoodwardColella1984 ; Roe1986 ; ShuOsher1988 In contrast, Lagrangian descriptions follow fluid parcels, reducing convective transport error and helping to preserve material interfaces, although additional stabilization remains necessary in the presence of shocks and moving discontinuities. In this context, particle-based approaches are attractive because they avoid severe mesh distortion and remeshing. Among them, the material point method (MPM) combines Lagrangian material points with a background grid: particles act as moving quadrature points, while the grid provides a convenient numerical workspace for force evaluation and updates. Sulsky1994MPM ; deVaucorbeil2020MPMReview MPM is closely related to classical particle-in-cell (PIC) schemes Harlow1964PIC , and inherits low-dissipation transfer variants such as FLIP BrackbillRuppel1986FLIP ; BurgessSulskyBrackbill1992FLIPMassMatrix , weighted least-squares PIC transfers WallstedtGuilkey2011WLS , and APIC Jiang2015APIC . To mitigate cell-crossing noise and quadrature degradation under large deformation, widely used extensions include GIMP BardenhagenKober2004GIMP and CPDI/CPDI2 Sadeghirad2011CPDI ; Sadeghirad2013CPDI2 , as well as improved particle updates and accuracy analyses Steffen2008Quadrature ; HammerquistNairn2017Noise ; NairnHammerquist2021MassMatrixInv ; Stefan2023StrongShock . Reproducing-kernel variants further enhance consistency in particle–grid coupling Liu1995RKPM ; Huang2019RKPM2D . Canonical strong-shock benchmarks for validating compressible formulations include the Sod shock tube and the Sedov blast wave Sod1978 ; Sedov1993Similarity . Despite these advantages, compressible MPM must balance robustness near shocks with fidelity in vortical and shear-dominated regions. Moving discontinuities and shock-driven transients can trigger oscillations unless sufficient numerical dissipation is introduced. Artificial viscosity (AV) is therefore widely used to spread shocks over a few grid cells, damp spurious oscillations, and improve robustness; the quadratic AV term controls shock thickness through its nonlinear dependence on velocity divergence. However, classical von Neumann–Richtmyer–Landshoff AV exposes a trade-off between stability and accuracy in multi-dimensional flows, VonNeumannRichtmyer1950 ; Landshoff1955 ; Wilkins1980 ; Noh1987 ; KolevRieben2009 especially when shocks coexist with shear layers and vortical roll-up. Peddavarapu and Huang (2025) PeddavarapuHuang2025 identify two key limitations: (i) a lack of deviatoric/shear contributions and (ii) an overly restrictive linear term in vortical regimes. They propose a vorticity-aware tensor-viscosity formulation that improves stability while better preserving vortical structures. Building on this line of work, we target a complementary but practically critical failure mode in compressible particle–grid solvers: under-sampling (quadrature deficiency) in strongly deforming regions. This pathology is prominent in long-time Rayleigh–Taylor instability (RTI) Taylor1950RTI ; Kull1991RTITheory , where the spike head can become transiently particle-depleted under severe stretching. The resulting loss of particle support degrades local quadrature and particle–grid coupling, manifesting as nonphysical pressure/density depressions and, in severe cases, pit-like voids at the spike head. Moreover, while higher-order particle–grid transfers such as APIC are desirable for resolving vortex-rich shear layers, poorly supported affine modes can become ill-conditioned in depleted cells and inject spurious energy, further amplifying spike-head artifacts. To address these coupled issues, we propose a robust compressible particle–grid framework that extends the FLIP–APIC + vorticity-aware tensor-AV baseline of Peddavarapu and Huang (2025) with sampling-aware controls. We first incorporate two standard mitigations against sampling bias: CPDI-lite transfers to reduce quadrature error under imperfect particle distributions Sadeghirad2011CPDI , and a small subcell jitter in the initial particle seeding to break phase locking and suppress grid imprinting Steffen2008Quadrature ; HammerquistNairn2017Noise . While these measures substantially reduce the artifact, small residual dents may still appear in long-time RTI runs when the spike head becomes transiently depleted; once triggered, such dents can evolve non-physically and deepen into pit-like voids, compromising the late-time morphology. We therefore introduce two additional remedies that directly address particle-support loss and affine instability: (i) a conservative resampling strategy that splits particles in depleted cells while exactly conserving mass, momentum, and internal energy, thereby restoring local quadrature quality; and (ii) a cell-wise soft-switch that continuously down-weights only the APIC affine contribution toward PIC-like behavior when local particle support falls below prescribed thresholds, preventing spurious affine energy injection precisely where particle deficiency is most severe. Together, these components retain the shock/discontinuity robustness of tensor AV and the strong interface-advection properties of FLIP–APIC in well-sampled regions, while eliminating sampling-driven spike-head void formation and preserving physically meaningful vortex roll-up in challenging long-time RTI regimes. The paper is organized as follows. Section 2 reviews governing equations and notation. Section 3 summarizes the baseline behavior and highlights the spike-head failure mode. Section 4 details the proposed transfers, viscosity model, sampling-aware soft-switch, and conservative resampling. Section 5 presents validation and ablation studies on standard benchmarks and single-/multi-mode RTI, followed by discussion and conclusions in Section 6. 2 Basic equations 2.1 Governing equations We consider inviscid, compressible flow governed by the Euler equations in conservative form: \frac{\partial\rho}{\partial t}+\nabla\cdot(\rho\mathbf{u})=0, (1) \frac{\partial(\rho\mathbf{u})}{\partial t}+\nabla\cdot\!\left(\rho\mathbf{u}\otimes\mathbf{u}+p\mathbf{I}\right)=\rho\mathbf{g}, (2) \frac{\partial(\rho E)}{\partial t}+\nabla\cdot\!\left((\rho E+p)\mathbf{u}\right)=\rho\,\mathbf{g}\cdot\mathbf{u}. (3) Here \rho is density, \mathbf{u} is velocity, p is pressure, \mathbf{I} is the identity tensor, \mathbf{g} is body acceleration, and E=e+\tfrac{1}{2}\lVert\mathbf{u}\rVert^{2} is the total specific energy with internal energy e . We close the system with the ideal-gas equation of state: p=(\gamma-1)\rho e. (4) The local sound speed is c=\sqrt{\gamma p/\rho} . We adopt a standard particle–grid splitting in which Lagrangian particles (material points) carry state variables and act as moving quadrature points, while a background grid is used for spatial discretization and time stepping. Throughout the remainder of this paper, the subscript p denotes particles (material points) and the subscript i denotes grid nodes. Each particle p stores position \mathbf{x}_{p} , velocity \mathbf{v}_{p} , mass m_{p} , specific internal energy e_{p} , density \rho_{p} , and volume V_{p} . In APIC mode, each particle also stores an affine matrix (e.g., \mathbf{C}_{p} ) representing a local velocity-gradient proxy. Grid node i stores lumped mass m_{i} and velocity \mathbf{v}_{i} . At each time step, we (i) reset grid fields, (ii) transfer particle mass and momentum to the grid (P2G), (iii) apply body forces and internal forces (pressure and AV) to update grid velocities, (iv) transfer updated grid velocities back to particles (G2P) using a PIC/FLIP/APIC blend, (v) advect particles, and (vi) optionally apply conservative resampling every N_{\mathrm{res}}