Communication Dans Un Congrès Année : 2025

Accelerating iterative solvers in the discrete dipole approximation using dedicated initial guesses

Résumé

The Discrete Dipole Approximation (DDA) [1] is a versatile and widely used method for simulating light scattering by particles with arbitrarily shapes and internal structure, ranging in sizes from much smaller than to several tenths of a wavelength. Its flexibility, coupled with open-source implementations like ADDA, DDSCAT, and IF-DDA, has enabled applications in biology, nanotechnology, and climate studies. However, the high computational cost of solving the underlying large-scale linear systems remains a significant limitation. This work focuses on accelerating the iterative solvers within the DDA by improving their initialization strategies. Previous works on DDA implementations, have explored preconditioning, block iterative methods, and initial guesses to enhance efficiency. While these approaches have yielded modest gains, achieving acceleration factors of 2–3 for block methods and up to 50% improvement for initial guesses, they leave significant room for enhancement. We aim to optimize initial guesses to improve solver convergence, particularly for soft particles. We build on existing studies [2,3] that demonstrated promising results using scalar solutions or approximate methods as initial guesses and compared several iterative solver methods. Our approach aims to extend these strategies by incorporating multiple approximations, theory, and geometric optics. For specific particle shapes, such as spheres and spheroids with refractive indices and sizes consistent with previous study [4], these advanced approximations are expected to yield substantial improvements. We also plan to benchmark these advanced initial guess strategies by directly integrating initial guesses produced by IF-DDA [2,3], such as scalar fields (e.g., uGu), into the iterative solvers available in ADDA. This approach will allow us to compare the performance of iterative methods across both codes, providing a comprehensive evaluation of their efficiency under different configurations. Additionally, by exploring the interoperability between ADDA and IF-DDA, we aim to bridge their formulations, contributing to open-source development and fostering broader adoption within the DDA community. Our results aim to provide a comparative analysis of initial guesses used with already established optimized iterative solvers, highlighting scenarios where these advanced strategies significantly reduce computational costs. Furthermore, initial guesses can be combined with preconditioning techniques and block iterative methods to enhance solver performance. These approaches are also extendable to particles with more complex geometries, broadening the applicability of our methods across diverse scattering problems. By refining solver initialization, this study paves the way for more efficient DDA simulations across diverse applications. [1] Purcell, E. M. and Pennypacker, C. R., Astrophys. J., 186, p. 705, 1973. [2] Chaumet, P. C., J. Quant. Spectrosc. Radiat. Transfer, 312, 108816, 2024. [3] Chaumet, P. C., Maire, G., and Sentenac, A., J. Quant. Spectrosc. Radiat. Transfer, 298, 108505, 2023. [4] Inzhevatkin, K. G. and Yurkin, M. A., J. Quant. Spectrosc. Radiat. Transfer, 277, 107965, 2022.

Fichier non déposé

Dates et versions

hal-05142441 , version 1 (23-07-2025)

Identifiants

  • HAL Id : hal-05142441 , version 1

Citer

Clément Argentin, Maxim A. Yurkin. Accelerating iterative solvers in the discrete dipole approximation using dedicated initial guesses. The 21th Electromagnetic and Light Scattering Conference, University of Messina, Jun 2025, Milazzo, Italy. pp.50. ⟨hal-05142441⟩
200 Consultations
3 Téléchargements

Partager

  • More