Rapid Variable Resolution Particle Initialization for Complex Geometries

Navaneet Villodi, Prabhu Ramachandran · 2025

Read Rapid Variable Resolution Particle Initialization for Complex Geometries by Navaneet Villodi, Prabhu Ramachandran (2025) 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-2508-12835.txt.

Introduction

Navaneet Villodi and Prabhu Ramachandran present a fast way to seed the starting particles for mesh-free flow simulations, in a 2025 paper. In methods like smoothed particle hydrodynamics, a fluid is modeled as a cloud of particles, and the quality of that first cloud drives the accuracy of everything after. Older seeding methods struggle to mix fine and coarse spacing, respect complex shapes, and pack particles evenly all at once.

Their method builds the cloud from the same basic tools the simulation itself uses. It seeds fluid and solid regions together, accepts shapes of any complexity, and avoids extra steps like gluing surface particles in place. In two- and three-dimensional tests, the clouds hugged boundaries well, stayed orderly, and held even density. The cost ran far below that of rival approaches.

Catalog record

CREATORNavaneet Villodi, Prabhu Ramachandran
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE49 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2508-12835

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Rapid Variable Resolution Particle Initialization for Complex Geometries Navaneet Villodi, Prabhu Ramachandran Abstract The accuracy of meshless methods like SPHis highly dependent on the quality of the particle distribution. Existing particle initialization techniques often struggle to simultaneously achieve adaptive resolution, handle intricate boundaries, and efficiently generate well-packed distributions inside and outside a boundary. This work presents a fast and robust particle initialization method that achieves these goals using standard SPHbuilding blocks. Our approach enables simultaneous initialization of fluid and solid regions, supports arbitrary geometries, and achieves high-quality, quasi-uniform particle arrangements without complex procedures like surface bonding. Extensive results in both 2D and 3D demonstrate that the obtained particle distributions exhibit good boundary conformity, low spatial disorder, and minimal density variation, all with significantly reduced computational cost compared to existing approaches. This work paves the way for automated particle initialization to accurately model flow in and around bodies with meshless methods, particularly with SPH. 1 Introduction Meshless methods require discretizing the domain into nodes. Unlike traditional mesh-based methods, these nodes are scattered without fixed connectivity. The nodes are required to be locally quasi-uniform and isotropic to ensure solution accuracy ( Van Der Sande and Fornberg 2021 ; Litvinov et al. 2015 ; Colagrossi et al. 2012 ) . In the context of SPH, these nodes are referred to as particles. Each particle is the representative of a (fuzzy) space surrounding itself. The particles carry physical properties such as mass, position, velocity, and density. Accurate mesh-free simulation of fluid flows involving boundaries require that the boundaries’ intricacies be captured accurately in terms of distributed particles. Most boundary treatment methods in SPHrequire that the particles be distributed on either side of the boundary as summarized by Negi and Ramachandran 2022a ; Villodi and Ramachandran 2024 . Getting an initial particle distribution for this purpose with zero-order consistency at the interface between bodies is challenging. Various authors have proposed methods to generate particle distributions. Persson and Strang 2004 asservate that most mesh generation methods tend to be complex and inaccessible. They propose a method in which an initial distribution of nodes is iteratively adjusted using Delaunay triangulations. This method is simple, albeit with some of the numerics delegated to MATLAB’s built-in functions. Colagrossi et al. 2012 proposed a novel particle packing algorithm, exclusively for initializing fluid particles. They state that with an incompatible initialization, particles may tend to resettle to the static solution predicted by the SPH scheme in-simulation, thus emphasising the importance of the algorithm being SPHnative to avoid such spurious motions. Xiong et al. 2013 proposed clipping a grid using the boundary. The slit cells are assigned partial masses while the untouched cells are assigned full masses. Jiang et al. 2015 proposed a method to generate particle distributions with variable resolution. They demonstrated their method in 3D, as well. Diehl et al. 2015 and Vela Vela et al. 2018 also proposed methods to generate particle distributions with variable resolution in 3D. These methods, however, cannot generate packed particle distributions for a solid and the surrounding fluid simultaneously. Fornberg and Flyer 2015 proposed an advancing front method for generating 2D particle distributions. This non-iterative method supported variable density and irregular boundaries. Van Der Sande and Fornberg 2021 built upon this work, extending it to 3D. This newer method utilises a background grid and introduces corrections to generate higher quality particle distributions. Fu and Ji 2019 proposed a method with particle relaxation based on the gradient of the pressure term of the momentum equation. The pressure is derived from an equation of state they proposed. Temporally reconstructed ghost particles based on level-set are used for dealing with boundaries. Fu et al. 2019 also proposed to constrain the motion of particles at the boundary surface and to project the near-surface particles to the boundary surface. Zhu et al. 2021 were able to demonstrate packing on various geometries with simple surface bonding, i.e., projecting near-surface particles to the boundary surface without restricting their motion or using ghost particles. Ji et al. 2021 introduced a correction term to account for the support at the boundary in lieu of the ghost particles. Yu et al. 2023 introduced yet another correction term based on a smoothed Heaviside function to account for the support at the boundary. Recently, Zhao et al. 2025 extended this lineage of work to support multi-body systems and simultaneous initialization of fluid and solid particles. Another method that supports simultaneous initialization of fluid and solid particles was proposed by Negi and Ramachandran 2021 . The hybrid method they proposed synthesizes the approaches of Colagrossi et al. 2012 and Jiang et al. 2015 with their improvements. This method also involves the projection of the near-surface particles to the boundary surface and constraining the motion of particles at the boundary surface. However, we find that their force based on Lennard-Jones like potential is a deterrent for introducing variable resolution support. To summarize, the following are the common pain points that one encounters when dealing with the present methods: 1. Simultaneous initialization: Separate initialization may lead to the formation of undesirable artifacts like gaps, especially around sharp features ( Zhao et al. 2025 ) . Most existing methods do not support simultaneous initialization of fluid and solid particles, except some newer methods such as that of Negi and Ramachandran 2021 ; Zhao et al. 2025 . 2. Adaptive resolution: Support for adaptive resolution is crucial for efficiently resolving intricate features. Some older methods support adaptive resolution, but the newer methods leave it to be desired. For example, the method Negi and Ramachandran 2021 has no support for adaptive resolution and support for adaptive resolution in Zhao et al. 2025 appears to be rudimentary. 3. Complexity: Static confinement ( Yu et al. 2023 ) , surface bonding ( Zhu et al. 2021 ; Negi and Ramachandran 2022a ; Zhao et al. 2025 ) , special treatment of sharp corners ( Negi and Ramachandran 2021 ) , restricted movement of particles at the boundary ( Fu et al. 2019 ; Negi and Ramachandran 2022a ) , alternate relaxation of fluid and solid particles ( Zhao et al. 2025 ) are essential to the existing methods. These procedures are complex, and the implementation is difficult. 4. Efficiency and Speed: Advancing front methods ( Fornberg and Flyer 2015 ; Van Der Sande and Fornberg 2021 ) have limited scope for parallelization. Moreover, stemming from the previous point, complexity often brings performance overhead. Therefore, relaxation-based methods that employ complex procedures are expected to be slow. We propose a method for particle initialization in SPH that addresses all the above pain points. It is very efficient, allowing one to generate variable resolution particle initializations simultaneously for solids and surrounding fluids with comparable accuracy to existing methods. Moreover, the method consists of pure SPHbuilding blocks. Therefore, it should be relatively easy for an SPHpractitioner to implement within any SPH framework (or standalone). In the following section, we describe the method in detail so that the reader can easily understand and implement it. 2 Methodology We employ a setup similar to Negi and Ramachandran 2021 with three sets of particles, free particles: interface particles, and frozen particles, as shown in fig. 1 . The interface particles are

Read the complete text