What the study found
A data-driven surrogate framework was developed to estimate hydrodynamic responses of rigid, non-spherical particles in Stokes flow, including the stresslet, angular velocity, and, for helicoidal particles, chiral thrust. The surrogate was trained on data generated by a validated boundary element method and was tested across random orientations and flow types.
Why the authors say this matters
The authors conclude that combining validated data generation with fast surrogate inference offers a practical route to including complex particle shapes in mesoscale solvers such as the force-coupling method. They say this could support large-ensemble studies of microstructure and suspension rheology.
What the researchers tested
The researchers used a regularized-Stokeslet boundary element method, a numerical method for solving flow around particles, to compute hydrodynamic responses in canonical linear flows. For spheroids, they compared the solver with analytical benchmarks; for helicoidal particles, they used self-convergence and additional checks of linearity, frame objectivity, and chirality-dependent symmetries. They then trained a neural-operator surrogate from these datasets.
What worked and what didn't
For spheroids, the boundary element solver agreed with available analytical benchmarks, and parameter choices were selected through convergence studies. For helicoidal particles, no analytical solutions were available, so accuracy was assessed by self-convergence and consistency tests. The surrogate achieved median relative errors below 1% for the deviatoric stresslet and 95th-percentile errors below 3%, with comparable accuracy for angular velocity and thrust.
What to keep in mind
The abstract does not describe limitations beyond the absence of analytical solutions for helicoidal particles. The reported performance applies to the independent test sets described in the abstract, spanning random orientations and flow types.
Key points
- A neural-operator surrogate was trained to predict stresslet, angular velocity, and chiral thrust for complex-shaped particles.
- The training data came from a regularized-Stokeslet boundary element method in canonical linear flows.
- The boundary element solver was validated against analytical benchmarks for spheroids.
- For helicoidal particles, accuracy was checked with self-convergence and symmetry tests because no analytical solution was available.
- The surrogate achieved median relative errors below 1% for the deviatoric stresslet and 95th-percentile errors below 3%.
Disclosure
- Research title:
- Neural-network surrogate predicts particle hydrodynamic responses accurately
- Authors:
- Marco Laudato
- Publication date:
- 2026-04-23
- OpenAlex record:
- View
Get the weekly research newsletter
Stay current with peer-reviewed research without reading academic papers — one filtered digest, every Friday.