Accelerating droplet-laden Stokes flow simulations with hierarchical surrogate modeling
Abstract
We present a surrogate modeling strategy for Stokes flows with liquid droplets suspended in a carrier fluid. Our approach is based on a multi-fidelity framework. At the lowest fidelity, droplets are treated as passive tracers, neglecting their influence on the ambient flow field. Building on this approximation, we derive a PDE that represents the current modeling error. This error equation is then solved approximately to correct the flow field and the procedure is iterated. Two fidelities are employed in an alternating fashion: Stokes flow in the absence of droplets and flow around a single droplet in free space. By systematically combining these models, the method captures droplet-flow, droplet-boundary, and droplet-droplet interactions. In this work, the framework is developed and validated for circular, non-deforming droplets in two spatial dimensions. The geometric self-similarity of the droplets allows us to construct an efficient offline-online strategy based on the reuse of precomputed single-droplet solutions. Extensions to deformable droplets are also discussed. Numerical experiments demonstrate the accuracy and efficiency of the proposed surrogate in a variety of tests, including scenarios with up to 104 droplets. Notably, we show that the proposed surrogate achieves substantially reduced computational cost compared to fully resolved multi-fluid simulations with state-of-the-art software.
Versions
Related
- A Review of Bayesian Surrogate Modeling for High-Dimensional Computer Simulations
- Accelerating phase-field simulations of grain growth via tensor decomposition
- Mask Reconstruction via Kernel-Based Surrogate Modeling for Automated Circulator Design
- A hierarchy of physics-embedded neural operators for structural dynamic surrogate modelingA hierarchy of physics-embedded neural operators for structural dynamic surrogate modeling
- Numerical Modeling of a Metallic Alloy Droplet Impact, Solidification, and Macrosegregation
BibTeX
@misc{pradovera2026accelerating,
title = {{Accelerating droplet-laden Stokes flow simulations with hierarchical surrogate modeling}},
author = {Davide Pradovera and Thomas Frachon and Sara Zahedi},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2607.03301},
eprint = {2607.03301},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2607.03301},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Accelerating droplet-laden Stokes flow simulations with hierarchical surrogate modeling
AU - Davide Pradovera
AU - Thomas Frachon
AU - Sara Zahedi
PY - 2026
DA - 2026/10/06
PB - arXiv
DO - 10.48550/arxiv.2607.03301
UR - https://arxiv.org/abs/2607.03301
AB - We present a surrogate modeling strategy for Stokes flows with liquid droplets suspended in a carrier fluid. Our approach is based on a multi-fidelity framework. At the lowest fidelity, droplets are treated as passive tracers, neglecting their influence on the ambient flow field. Building on this approximation, we derive a PDE that represents the current modeling error. This error equation is then solved approximately to correct the flow field and the procedure is iterated. Two fidelities are employed in an alternating fashion: Stokes flow in the absence of droplets and flow around a single droplet in free space. By systematically combining these models, the method captures droplet-flow, droplet-boundary, and droplet-droplet interactions. In this work, the framework is developed and validated for circular, non-deforming droplets in two spatial dimensions. The geometric self-similarity of the droplets allows us to construct an efficient offline-online strategy based on the reuse of precomputed single-droplet solutions. Extensions to deformable droplets are also discussed. Numerical experiments demonstrate the accuracy and efficiency of the proposed surrogate in a variety of tests, including scenarios with up to 10^4 droplets. Notably, we show that the proposed surrogate achieves substantially reduced computational cost compared to fully resolved multi-fluid simulations with state-of-the-art software.
ER -