arXiv subjects

Seung Whan Chung

Publications and source records attributed to Seung Whan Chung.

17 recordsLinked to original sources

arXiv2026

Hard-constrained Physics-informed Neural Networks for Interface Problems

Physics-informed neural networks (PINNs) have emerged as a flexible framework for solving partial differential equations, but interface problems remain challenging because continuity and flux conditions are typically imposed through soft penalty terms, leading to imperfect interface enforcement and degraded accuracy near interfaces. We introduce two ansatz-based hard-constrained PINN formulations for interface problems that embed the interface physics into the solution representation, decoupling interface enforcement from PDE residual minimization. The windowing approach constructs the trial space from compactly supported windowed subnetworks so that interface continuity and flux balance are satisfied by design. The buffer approach discretely hard-constrains subnetworks with additive buffer functions that enforce boundary and interface constraints at sampled points through a lightweight correction. We study both formulations on one- and two-dimensional elliptic interface benchmarks against soft-constrained baselines. A statistical study in one dimension shows that the windowing approach occasionally attains very high accuracy (as low as $O(10^{-9})$) on simple structured cases, but its rigid ansatz makes training stiff, so its median errors are often worse than the baselines. In contrast, the buffer approach achieves the smallest median error (as low as $O(10^{-5})$) across a wider range of source terms and interface configurations. In two dimensions, the buffer formulation is more robust for general geometries, while the windowing construction is more sensitive to overlap and corner effects and over-constrains the problem. These results position the buffer method as a straightforward and promising approach for interface problems, with the search for an optimal buffer function form left for future work.

math.NA↗

arXiv2026

Latent Space Dynamics Identification for Interface Tracking with Application to Shock-Induced Pore Collapse

Capturing sharp, evolving interfaces remains a central challenge in reduced-order modeling, especially when data is limited and the system exhibits localized nonlinearities or discontinuities. We propose LaSDI-IT (Latent Space Dynamics Identification for Interface Tracking), a data-driven framework that combines low-dimensional latent dynamics learning with explicit interface-aware encoding to enable accurate and efficient modeling of physical systems involving moving material boundaries. At the core of LaSDI-IT is a revised auto-encoder architecture that jointly reconstructs the physical field and an indicator function representing material regions or phases, allowing the model to track complex interface evolution without requiring detailed physical models or mesh adaptation. The latent dynamics are learned through linear regression in the encoded space and generalized across parameter regimes using Gaussian process interpolation with greedy sampling. We demonstrate LaSDI-IT on the problem of shock-induced pore collapse in high explosives, a process characterized by sharp temperature gradients and dynamically deforming pore geometries. The method achieves relative prediction errors below 9% across the parameter space, accurately recovers key quantities of interest such as pore area and hot spot formation, and matches the performance of dense training with only half the data. This latent dynamics prediction was 106 times faster than the conventional high-fidelity simulation, proving its utility for multi-query applications. These results highlight LaSDI-IT as a general, data-efficient framework for modeling discontinuity-rich systems in computational physics, with potential applications in multiphase flows, fracture mechanics, and phase change problems.

physics.comp-ph↗

arXiv2026

mLaSDI: Multi-stage latent space dynamics identification

Accurately solving partial differential equations (PDEs) is essential across many scientific disciplines. However, high-fidelity solvers can be computationally prohibitive, motivating the development of reduced-order models (ROMs). Recently, Latent Space Dynamics Identification (LaSDI) was proposed as a data-driven, non-intrusive ROM framework. LaSDI compresses the training data via an autoencoder and learns user-specified ordinary differential equations (ODEs), governing the latent dynamics, enabling rapid predictions for unseen parameters. While LaSDI has produced effective ROMs for numerous problems, the autoencoder must simultaneously reconstruct the training data and satisfy the imposed latent dynamics, which are often competing objectives that limit accuracy, particularly for complex or high-frequency phenomena. To address this limitation, we propose multi-stage Latent Space Dynamics Identification (mLaSDI). With mLaSDI, we train LaSDI sequentially in stages. After training the initial autoencoder, we train additional decoders which map the latent trajectories to residuals from previous stages. This staged residual learning, combined with periodic activation functions, enables recovery of high-frequency content without sacrificing interpretability of the latent dynamics. We further provide an error decomposition separating autoencoder and latent dynamics contributions, and prove that additional training stages cannot increase the training residual. Numerical experiments on a multiscale oscillating system, unsteady wake flow, and the 1D-1V Vlasov equation demonstrate that mLaSDI achieves significantly lower reconstruction and prediction errors, often by an order of magnitude, while requiring less training time and reduced hyperparameter tuning compared to standard LaSDI.

cs.LG↗

arXiv2026

Latent Space Element Method

How can we build surrogate solvers that train on small domains but scale to larger ones without intrusive access to PDE operators? Inspired by the Data-Driven Finite Element Method (DD-FEM) framework for modular data-driven solvers, we propose the Latent Space Element Method (LSEM), an element-based latent surrogate assembly approach in which a learned subdomain ("element") model can be tiled and coupled to form a larger computational domain. Each element is a LaSDI latent ODE surrogate trained from snapshots on a local patch, and neighboring elements are coupled through learned directional interaction terms in latent space, avoiding Schwarz iterations and interface residual evaluations. A smooth window-based blending reconstructs a global field from overlapping element predictions, yielding a scalable assembled latent dynamical system. Experiments on the 1D Burgers and Korteweg-de Vries equations show that LSEM maintains predictive accuracy while scaling to spatial domains larger than those seen in training. LSEM offers an interpretable and extensible route toward foundation-model surrogate solvers built from reusable local models.

math.DS↗

arXiv2025

Sequential decoder training for improved latent space dynamics identification

Accurate numerical solutions of partial differential equations are essential in many scientific fields but often require computationally expensive solvers, motivating reduced-order models (ROMs). Latent Space Dynamics Identification (LaSDI) is a data-driven ROM framework that combines autoencoders with equation discovery to learn interpretable latent dynamics. However, enforcing latent dynamics during training can compromise reconstruction accuracy of the model for simulation data. We introduce multi-stage LaSDI (mLaSDI), a framework that improves reconstruction and prediction accuracy by sequentially learning additional decoders to correct residual errors from previous stages. Applied to the 1D-1V Vlasov equation, mLaSDI consistently outperforms standard LaSDI, achieving lower prediction errors and reduced training time across a wide range of architectures.

cs.LG↗

arXiv2025

Theory and numerics of subspace approximation of eigenvalue problems

Large-scale eigenvalue problems arise in various fields of science and engineering and demand computationally efficient solutions. In this study, we investigate the subspace approximation for parametric linear eigenvalue problems, aiming to mitigate the computational burden associated with high-fidelity systems. We provide general error estimates under non-simple eigenvalue conditions, establishing some theoretical foundations for understanding the convergence behavior of subspace approximations. Numerical examples, including problems with one-dimensional to three-dimensional spatial domain and one-dimensional to two-dimensional parameter domain, are presented to demonstrate the efficacy of reduced basis method in handling parametric variations in boundary conditions and coefficient fields to achieve significant computational savings while maintaining high accuracy, making them promising tools for practical applications in large-scale eigenvalue computations.

math.NA↗

arXiv2025

Defining Foundation Models for Computational Science: A Call for Clarity and Rigor

The widespread success of foundation models in natural language processing and computer vision has inspired researchers to extend the concept to scientific machine learning and computational science. However, this position paper argues that as the term "foundation model" is an evolving concept, its application in computational science is increasingly used without a universally accepted definition, potentially creating confusion and diluting its precise scientific meaning. In this paper, we address this gap by proposing a formal definition of foundation models in computational science, grounded in the core values of generality, reusability, and scalability. We articulate a set of essential and desirable characteristics that such models must exhibit, drawing parallels with traditional foundational methods, like the finite element and finite volume methods. Furthermore, we introduce the Data-Driven Finite Element Method (DD-FEM), a framework that fuses the modular structure of classical FEM with the representational power of data-driven learning. We demonstrate how DD-FEM addresses many of the key challenges in realizing foundation models for computational science, including scalability, adaptability, and physics consistency. By bridging traditional numerical methods with modern AI paradigms, this work provides a rigorous foundation for evaluating and developing novel approaches toward future foundation models in computational science.

cs.LG↗

arXiv2025

Local Reduced-Order Modeling for Electrostatic Plasmas by Physics-Informed Solution Manifold Decomposition

Despite advancements in high-performance computing and modern numerical algorithms, computational cost remains prohibitive for multi-query kinetic plasma simulations. In this work, we develop data-driven reduced-order models (ROMs) for collisionless electrostatic plasma dynamics, based on the kinetic Vlasov-Poisson equation. Our ROM approach projects the equation onto a linear subspace defined by the proper orthogonal decomposition (POD) modes. We introduce an efficient tensorial method to update the nonlinear term using a precomputed third-order tensor. We capture multiscale behavior with a minimal number of POD modes by decomposing the solution manifold into multiple time windows and creating temporally local ROMs. We consider two strategies for decomposition: one based on the physical time and the other based on the electric field energy. Applied to the 1D1V Vlasov-Poisson simulations, that is, prescribed E-field, Landau damping, and two-stream instability, we demonstrate that our ROMs accurately capture the total energy of the system both for parametric and time extrapolation cases. The temporally local ROMs are more efficient and accurate than the single ROM. In addition, in the two-stream instability case, we show that the energy-windowing reduced-order model (EW-ROM) is more efficient and accurate than the time-windowing reduced-order model (TW-ROM). With the tensorial approach, EW-ROM solves the equation approximately 90 times faster than Eulerian simulations while maintaining a maximum relative error of 7.5% for the training data and 11% for the testing data.

math.NA↗

arXiv2024

Scalable nonlinear manifold reduced order model for dynamical systems

The domain decomposition (DD) nonlinear-manifold reduced-order model (NM-ROM) represents a computationally efficient method for integrating underlying physics principles into a neural network-based, data-driven approach. Compared to linear subspace methods, NM-ROMs offer superior expressivity and enhanced reconstruction capabilities, while DD enables cost-effective, parallel training of autoencoders by partitioning the domain into algebraic subdomains. In this work, we investigate the scalability of this approach by implementing a "bottom-up" strategy: training NM-ROMs on smaller domains and subsequently deploying them on larger, composable ones. The application of this method to the two-dimensional time-dependent Burgers' equation shows that extrapolating from smaller to larger domains is both stable and effective. This approach achieves an accuracy of 1% in relative error and provides a remarkable speedup of nearly 700 times.

math.NA↗

arXiv2024

Scaled-up prediction of steady Navier-Stokes equation with component reduced order modeling

Scaling up new scientific technologies from laboratory to industry often involves demonstrating performance on a larger scale. Computer simulations can accelerate design and predictions in the deployment process, though traditional numerical methods are computationally intractable even for intermediate pilot plant scales. Recently, component reduced order modeling method is developed to tackle this challenge by combining projection reduced order modeling and discontinuous Galerkin domain decomposition. However, while many scientific or engineering applications involve nonlinear physics, this method has been only demonstrated for various linear systems. In this work, the component reduced order modeling method is extended to steady Navier-Stokes flow, with application to general nonlinear physics in view. Large-scale, global domain is decomposed into combination of small-scale unit component. Linear subspaces for flow velocity and pressure are identified via proper orthogonal decomposition over sample snapshots collected at small scale unit component. Velocity bases are augmented with pressure supremizer, in order to satisfy inf-sup condition for stable pressure prediction. Two different nonlinear reduced order modeling methods are employed and compared for efficient evaluation of nonlinear advection: 3rd-order tensor projection operator and empirical quadrature procedure. The proposed method is demonstrated on flow over arrays of five different unit objects, achieving $23$ times faster prediction with less than $4\%$ relative error up to $256$ times larger scale domain than unit components. Furthermore, a numerical experiment with pressure supremizer strongly indicates the need of supremizer for stable pressure prediction. A comparison between tensorial approach and empirical quadrature procedure is performed, which suggests a slight advantage for empirical quadrature procedure.

math.NA↗

arXiv2024

Scalable physics-guided data-driven component model reduction for steady Navier-Stokes flow

Computational physics simulation can be a powerful tool to accelerate industry deployment of new scientific technologies. However, it must address the challenge of computationally tractable, moderately accurate prediction at large industry scales, and training a model without data at such large scales. A recently proposed component reduced order modeling (CROM) tackles this challenge by combining reduced order modeling (ROM) with discontinuous Galerkin domain decomposition (DG-DD). While it can build a component ROM at small scales that can be assembled into a large scale system, its application is limited to linear physics equations. In this work, we extend CROM to nonlinear steady Navier-Stokes flow equation. Nonlinear advection term is evaluated via tensorial approach or empirical quadrature procedure. Application to flow past an array of objects at moderate Reynolds number demonstrates $\sim23.7$ times faster solutions with a relative error of $\sim 2.3\%$, even at scales $256$ times larger than the original problem.

math.NA↗

arXiv2024

Divide And Conquer: Learning Chaotic Dynamical Systems With Multistep Penalty Neural Ordinary Differential Equations

Forecasting high-dimensional dynamical systems is a fundamental challenge in various fields, such as geosciences and engineering. Neural Ordinary Differential Equations (NODEs), which combine the power of neural networks and numerical solvers, have emerged as a promising algorithm for forecasting complex nonlinear dynamical systems. However, classical techniques used for NODE training are ineffective for learning chaotic dynamical systems. In this work, we propose a novel NODE-training approach that allows for robust learning of chaotic dynamical systems. Our method addresses the challenges of non-convexity and exploding gradients associated with underlying chaotic dynamics. Training data trajectories from such systems are split into multiple, non-overlapping time windows. In addition to the deviation from the training data, the optimization loss term further penalizes the discontinuities of the predicted trajectory between the time windows. The window size is selected based on the fastest Lyapunov time scale of the system. Multi-step penalty(MP) method is first demonstrated on Lorenz equation, to illustrate how it improves the loss landscape and thereby accelerates the optimization convergence. MP method can optimize chaotic systems in a manner similar to least-squares shadowing with significantly lower computational costs. Our proposed algorithm, denoted the Multistep Penalty NODE, is applied to chaotic systems such as the Kuramoto-Sivashinsky equation, the two-dimensional Kolmogorov flow, and ERA5 reanalysis data for the atmosphere. It is observed that MP-NODE provide viable performance for such chaotic systems, not only for short-term trajectory predictions but also for invariant statistics that are hallmarks of the chaotic nature of these dynamics.

cs.LG↗

arXiv2024

Characterization of uncertainties in electron-argon collision cross sections

The predictive capability of a plasma discharge model depends on accurate representations of electron-impact collision cross sections, which determine the key reaction rates and transport properties of the plasma. Although many cross sections have been identified through experiments and quantum mechanical simulations, their uncertainties are not well-investigated. We characterize the uncertainties in electron-argon collision cross sections using a Bayesian framework. Six collision processes -- elastic momentum transfer, ionization, and four excitations -- are characterized with semi-empirical models, whose parametric uncertainties effectively capture the features important to the macroscopic properties of the plasma, namely transport properties and chemical reaction rates. The method is designed to capture the effects of systematic errors that lead to large discrepancies between some data sets. Specifically, for the purposes of Bayesian inference, each of the parametric cross section models is augmented with a Gaussian process representing systematic measurement errors as well as model inadequacies in the parametric form. The results show that the method is able to capture scatter in the data between the electron-beam experiments and ab-initio quantum simulations. The calibrated cross section models are further validated against measurements from swarm-parameter experiments.

physics.plasm-ph↗

arXiv2024

Improved deep learning of chaotic dynamical systems with multistep penalty losses

Predicting the long-term behavior of chaotic systems remains a formidable challenge due to their extreme sensitivity to initial conditions and the inherent limitations of traditional data-driven modeling approaches. This paper introduces a novel framework that addresses these challenges by leveraging the recently proposed multi-step penalty (MP) optimization technique. Our approach extends the applicability of MP optimization to a wide range of deep learning architectures, including Fourier Neural Operators and UNETs. By introducing penalized local discontinuities in the forecast trajectory, we effectively handle the non-convexity of loss landscapes commonly encountered in training neural networks for chaotic systems. We demonstrate the effectiveness of our method through its application to two challenging use-cases: the prediction of flow velocity evolution in two-dimensional turbulence and ocean dynamics using reanalysis data. Our results highlight the potential of this approach for accurate and stable long-term prediction of chaotic dynamics, paving the way for new advancements in data-driven modeling of complex natural phenomena.

cs.LG↗

arXiv2024

A time-parallel multiple-shooting method for large-scale quantum optimal control

Quantum optimal control plays a crucial role in quantum computing by providing the interface between compiler and hardware. Solving the optimal control problem is particularly challenging for multi-qubit gates, due to the exponential growth in computational complexity with the system's dimensionality and the deterioration of optimization convergence. To ameliorate the computational complexity of time-integration, this paper introduces a multiple-shooting approach in which the time domain is divided into multiple windows and the intermediate states at window boundaries are treated as additional optimization variables. This enables parallel computation of state evolution across time-windows, significantly accelerating objective function and gradient evaluations. Since the initial state matrix in each window is only guaranteed to be unitary upon convergence of the optimization algorithm, the conventional gate trace infidelity is replaced by a generalized infidelity that is convex for non-unitary state matrices. Continuity of the state across window boundaries is enforced by equality constraints. A quadratic penalty optimization method is used to solve the constrained optimal control problem, and an efficient adjoint technique is employed to calculate the gradients in each iteration. We demonstrate the effectiveness of the proposed method through numerical experiments on quantum Fourier transform gates in systems with 2, 3, and 4 qubits, noting a speedup of 80x for evaluating the gradient in the 4-qubit case, highlighting the method's potential for optimizing control pulses in multi-qubit quantum systems.

quant-ph↗

arXiv2023

Train Small, Model Big: Scalable Physics Simulators via Reduced Order Modeling and Domain Decomposition

Numerous cutting-edge scientific technologies originate at the laboratory scale, but transitioning them to practical industry applications is a formidable challenge. Traditional pilot projects at intermediate scales are costly and time-consuming. An alternative, the E-pilot, relies on high-fidelity numerical simulations, but even these simulations can be computationally prohibitive at larger scales. To overcome these limitations, we propose a scalable, physics-constrained reduced order model (ROM) method. ROM identifies critical physics modes from small-scale unit components, projecting governing equations onto these modes to create a reduced model that retains essential physics details. We also employ Discontinuous Galerkin Domain Decomposition (DG-DD) to apply ROM to unit components and interfaces, enabling the construction of large-scale global systems without data at such large scales. This method is demonstrated on the Poisson and Stokes flow equations, showing that it can solve equations about $15 - 40$ times faster with only $\sim$ $1\%$ relative error. Furthermore, ROM takes one order of magnitude less memory than the full order model, enabling larger scale predictions at a given memory limitation.

cs.CE↗

arXiv2023

Accelerating Flow Simulations using Online Dynamic Mode Decomposition

We develop an on-the-fly reduced-order model (ROM) integrated with a flow simulation, gradually replacing a corresponding full-order model (FOM) of a physics solver. Unlike offline methods requiring a separate FOM-only simulation prior to model reduction, our approach constructs a ROM dynamically during the simulation, replacing the FOM when deemed credible. Dynamic mode decomposition (DMD) is employed for online ROM construction, with a single snapshot vector used for rank-1 updates in each iteration. Demonstrated on a flow over a cylinder with Re = 100, our hybrid FOM/ROM simulation is verified in terms of the Strouhal number, resulting in a 4.4 times speedup compared to the FOM solver.

physics.flu-dyn↗