Searcharxiv⌕ Search

arXiv subjects

Teng-Yuan Chang

Publications and source records attributed to Teng-Yuan Chang.

4 recordsLinked to original sources

Singular layer neural networks for the Navier-Stokes equations at small viscosity

We develop neural network methods for the two-dimensional incompressible Navier-Stokes equations at a small viscosity $\varepsilon > 0$, whose accuracy is uniform in $\varepsilon$ and whose residual is evaluated on a grid that does not resolve the vorticity boundary layer. In the vorticity-stream function formulation with the slip boundary conditions, a vorticity layer of width $\sqrt{\varepsilon}$ forms along the walls. The trial function of the singular layer Physics-Informed Neural Network (sl-PINN) contains the corrector of the asymptotic analysis, together with the law that fixes its amplitude at the wall. The amplitude is tied to the wall trace of the network itself, so that the method requires no a priori knowledge of the limit solution. The sl-PINN keeps the error uniform in $\varepsilon$ over six decades, for the Stokes and the Navier-Stokes equations alike, although in the latter the wall trace travels with the flow and is not known in advance. An enrichment that uses the thin scale but leaves the profile and the amplitude of the layer free deteriorates as $\varepsilon\to0$. Since the corrector carries the layer analytically, the Laplacian in the residual can be discretized by a finite volume flux difference on a coarse grid. The resulting singular layer Finite Volume Neural Network (sl-FVNN) attains the same uniform accuracy as the sl-PINN, to within the scatter of the random seed, at about half the cost of a training step. The combination of the asymptotic corrector with the coarse-grid finite volume discretization is the main result of this article.

physics.flu-dyn↗

A data free neural operator enabling fast inference of 2D and 3D Navier Stokes equations

Ensemble simulations of high-dimensional flow models (e.g., Navier Stokes type PDEs) are computationally prohibitive for real time applications. Neural operators enable fast inference but are limited by costly data requirements and poor generalization to 3D flows. We present a data-free operator network for the Navier Stokes equations that eliminates the need for paired solution data and enables robust, real time inference for large ensemble forecasting. The physics-grounded architecture takes initial and boundary conditions as well as forcing functions, yielding solutions robust to high variability and perturbations. Across 2D benchmarks and 3D test cases, the method surpasses prior neural operators in accuracy and, for ensembles, achieves greater efficiency than conventional numerical solvers. Notably, it delivers accurate solutions of the three dimensional Navier Stokes equations, a regime not previously demonstrated for data free neural operators. By uniting a numerically grounded architecture with the scalability of machine learning, this approach establishes a practical pathway toward data free, high fidelity PDE surrogates for end to end scientific simulation and prediction.

cs.LG↗

Singular layer PINN methods for Burgers' equation

In this article, we present a new learning method called sl-PINN to tackle the one-dimensional viscous Burgers problem at a small viscosity, which results in a singular interior layer. To address this issue, we first determine the corrector that characterizes the unique behavior of the viscous flow within the interior layers by means of asymptotic analysis. We then use these correctors to construct sl-PINN predictions for both stationary and moving interior layer cases of the viscous Burgers problem. Our numerical experiments demonstrate that sl-PINNs accurately predict the solution for low viscosity, notably reducing errors near the interior layer compared to traditional PINN methods. Our proposed method offers a comprehensive understanding of the behavior of the solution near the interior layer, aiding in capturing the robust part of the training solution.

math.NA↗

Singular layer Physics Informed Neural Network method for Plane Parallel Flows

We construct in this article the semi-analytic Physics Informed Neural Networks (PINNs), called {\em singular layer PINNs} (or {\em sl-PINNs}), that are suitable to predict the stiff solutions of plane-parallel flows at a small viscosity. Recalling the boundary layer analysis, we first find the corrector for the problem which describes the singular behavior of the viscous flow inside boundary layers. Then, using the components of the corrector and its curl, we build our new {\em sl-PINN} predictions for the velocity and the vorticity by either embedding the explicit expression of the corrector (or its curl) in the structure of PINNs or by training the implicit parts of the corrector (or its curl) together with the PINN predictions. Numerical experiments confirm that our new {\em sl-PINNs} produce stable and accurate predicted solutions for the plane-parallel flows at a small viscosity.

math.NA↗