Yongseok Jang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Yongseok Jang has a total of 19 co-authors. Here are the top co-authors by number of shared publications:
- Simon Shaw
Brunel University
5 shared publications - Emeric Martin
ONERA, Université Paris-Saclay, Châtillon, France
3 shared publications A Preprint
2 shared publications- Carola Kruse
CERFACS, Toulouse, France
2 shared publications - Cédric Content
ONERA, Université Paris-Saclay, Châtillon, France
2 shared publications - Frank Hülsemann
EDF Lab Paris-Saclay Palaiseau France
2 shared publications
Latest Publications
- Open Access
- 19 September 2026
- Published by Elsevier in SSRN Electronic Journal
A stable DG-POD reduced-order method for dynamic linear viscoelasticity
We present a reduced-order modeling framework based on proper orthogonal decomposition for dynamic linear viscoelasticity governed by generalized Maxwell models. The hereditary constitutive law is reformulated using internal variables, and the resulting problem is discretized in space by a symmetric interior penalty discontinuous Galerkin method and in time by either the backward Euler or Crank--Nicolson scheme. The reduced model is obtained by projecting the fully discrete system onto low-dimensional spaces constructed from displacement snapshots. For the full-order discontinuous Galerkin formulation, we show well-posedness and derive an \textit{a priori} stability bound without using Gronwall's inequality. Consequently, the bound grows at most linearly, rather than exponentially, with the final time. We further derive spatial error estimates and an error decomposition that separates the effects of spatial discretization, temporal discretization, and proper orthogonal decomposition truncation. The reduced formulation replaces the large full-order systems with low-dimensional problems, enabling efficient long-time simulations. Numerical experiments verify the predicted convergence behavior, assess different inner products for constructing the reduced basis, and demonstrate accurate recovery of transient oscillations and viscoelastic relaxation with substantial computational savings, even after accounting for snapshot generation and basis construction.
- 19 September 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessA stable DG-POD reduced-order method for dynamic linear viscoelasticity
We present a reduced-order modeling framework based on proper orthogonal decomposition for dynamic linear viscoelasticity governed by generalized Maxwell models. The hereditary constitutive law is reformulated using internal variables, and the resulting problem is discretized in space by a symmetric interior penalty discontinuous Galerkin method and in time by either the backward Euler or Crank--Nicolson scheme. The reduced model is obtained by projecting the fully discrete system onto low-dimensional spaces constructed from displacement snapshots. For the full-order discontinuous Galerkin formulation, we show well-posedness and derive an \textit{a priori} stability bound without using Gronwall's inequality. Consequently, the bound grows at most linearly, rather than exponentially, with the final time. We further derive spatial error estimates and an error decomposition that separates the effects of spatial discretization, temporal discretization, and proper orthogonal decomposition truncation. The reduced formulation replaces the large full-order systems with low-dimensional problems, enabling efficient long-time simulations. Numerical experiments verify the predicted convergence behavior, assess different inner products for constructing the reduced basis, and demonstrate accurate recovery of transient oscillations and viscoelastic relaxation with substantial computational savings, even after accounting for snapshot generation and basis construction.
- Open Access
- 12 August 2026
- Published by Wiley in Numerical Linear Algebra with Applications
- Vol. 33 (4)
Mixed Precision Augmented GMRES
We aim to accelerate the restarted generalized minimal residual (GMRES) method for the solutions of linear systems by combining two types of techniques. On the one hand, mixed precision GMRES algorithms, which use lower precision in certain steps of the inner cycles, offer significant reductions in computational and memory costs. On the other hand, augmented GMRES algorithms, which recycle information on the eigenvalues of the matrix between restarts by incorporating additional vectors into the Krylov basis, can significantly speed up the convergence. In this work, we investigate how to combine mixed precision and augmentation, in order to cumulate the reduced per‐iteration cost of the former with the reduced number of iterations of the latter. We first explore the GMRES with deflated restarting (GMRES‐DR) variant, which we demonstrate possesses limited mixed precision opportunities. Indeed, GMRES‐DR can exploit a preconditioner constructed in low precision, but requires a flexible paradigm to also apply it in low precision; moreover, the matrix–vector product and orthonormalization steps must both be kept in high precision, as otherwise the method stagnates at low accuracy. We explain that this is because GMRES‐DR is based on some algebraic simplifications that are only valid in exact arithmetic, but fail to hold in finite precision. This observation leads us to investigate a more general augmented GMRES framework (AugGMRES) that avoids making these simplifications. AugGMRES is much more resilient to the use of low precision, does not require a flexible paradigm, and successfully converges to high accuracy even when low precision is used for all inner operations. Our numerical experiments illustrate the robustness and fast convergence of AugGMRES on a range of sparse matrices, including ill‐conditioned real‐life ones.
- 12 August 2026
- Published by Wiley in Numerical Linear Algebra with Applications
- Vol. 33 (4)
Open AccessMixed Precision Augmented GMRES
We aim to accelerate the restarted generalized minimal residual (GMRES) method for the solutions of linear systems by combining two types of techniques. On the one hand, mixed precision GMRES algorithms, which use lower precision in certain steps of the inner cycles, offer significant reductions in computational and memory costs. On the other hand, augmented GMRES algorithms, which recycle information on the eigenvalues of the matrix between restarts by incorporating additional vectors into the Krylov basis, can significantly speed up the convergence. In this work, we investigate how to combine mixed precision and augmentation, in order to cumulate the reduced per‐iteration cost of the former with the reduced number of iterations of the latter. We first explore the GMRES with deflated restarting (GMRES‐DR) variant, which we demonstrate possesses limited mixed precision opportunities. Indeed, GMRES‐DR can exploit a preconditioner constructed in low precision, but requires a flexible paradigm to also apply it in low precision; moreover, the matrix–vector product and orthonormalization steps must both be kept in high precision, as otherwise the method stagnates at low accuracy. We explain that this is because GMRES‐DR is based on some algebraic simplifications that are only valid in exact arithmetic, but fail to hold in finite precision. This observation leads us to investigate a more general augmented GMRES framework (AugGMRES) that avoids making these simplifications. AugGMRES is much more resilient to the use of low precision, does not require a flexible paradigm, and successfully converges to high accuracy even when low precision is used for all inner operations. Our numerical experiments illustrate the robustness and fast convergence of AugGMRES on a range of sparse matrices, including ill‐conditioned real‐life ones.
- 10 September 2025
- Published by SAGE Publications in Mathematics and Mechanics of Solids
- Vol. 31 (10) , 2049-2074
A priori and a posteriori analyses of the symmetric interior penalty Galerkin method for a quasi-static fractional-order viscoelasticity model
Power-law type stress relaxation leads to a fractional-order viscoelasticity model. We consider a quasi-static viscoelasticity model of a fractional order which is mathematically expressed as a Volterra integral of the second kind with a weakly singular kernel. By employing the symmetric interior penalty Galerkin (SIPG) method and linear interpolation in time, we derive a fully discrete problem. To mitigate temporal discretization errors from a physical perspective in the quasi-static state, we introduce an auxiliary discrete velocity using the Crank–Nicolson approximation. We present a priori stability and error analyses for both semi-discrete and fully discrete schemes. In addition, we provide a residual-based energy norm a posteriori error estimation. Finally, we conduct numerical experiments to validate the error estimate theorems and demonstrate their applicability using real material data.
- 10 September 2025
- Published by SAGE Publications in Mathematics and Mechanics of Solids
- Vol. 31 (10) , 2049-2074
A priori and a posteriori analyses of the symmetric interior penalty Galerkin method for a quasi-static fractional-order viscoelasticity model
Power-law type stress relaxation leads to a fractional-order viscoelasticity model. We consider a quasi-static viscoelasticity model of a fractional order which is mathematically expressed as a Volterra integral of the second kind with a weakly singular kernel. By employing the symmetric interior penalty Galerkin (SIPG) method and linear interpolation in time, we derive a fully discrete problem. To mitigate temporal discretization errors from a physical perspective in the quasi-static state, we introduce an auxiliary discrete velocity using the Crank–Nicolson approximation. We present a priori stability and error analyses for both semi-discrete and fully discrete schemes. In addition, we provide a residual-based energy norm a posteriori error estimation. Finally, we conduct numerical experiments to validate the error estimate theorems and demonstrate their applicability using real material data.
- Open Access
- 30 July 2025
- Published by Wiley in Numerical Linear Algebra with Applications
- Vol. 32 (4)
Randomized Orthogonalization Process With Reorthogonalization
The dimension reduction technique of random sketching is advantageous in significantly reducing computational complexity. In orthogonalization processes like the Gram‐Schmidt (GS) algorithm, incorporating random sketching results in a halving of computational costs compared to the classical/modified Gram‐Schmidt (CGS/MGS) algorithms, while maintaining numerical stability comparable to the MGS algorithm. The randomized Gram‐Schmidt (RGS) algorithm produces a set of sketched orthonormal vectors, and the loss of orthogonality in these vectors is linearly dependent on the condition number of the given matrix. We propose a new variant, RGS‐L2, with reorthogonalization to obtain a set of orthonormal vectors. A round‐off error analysis demonstrates that the loss of orthogonality is close to the unit round‐off level. Numerical experiments exhibit the benefits of our proposed algorithm. Furthermore, we apply the RGS‐L2 algorithm to the Generalized Minimal Residual Method (GMRES) and compare its numerical performance with other GMRES variants.
- 30 July 2025
- Published by Wiley in Numerical Linear Algebra with Applications
- Vol. 32 (4)
Open AccessRandomized Orthogonalization Process With Reorthogonalization
The dimension reduction technique of random sketching is advantageous in significantly reducing computational complexity. In orthogonalization processes like the Gram‐Schmidt (GS) algorithm, incorporating random sketching results in a halving of computational costs compared to the classical/modified Gram‐Schmidt (CGS/MGS) algorithms, while maintaining numerical stability comparable to the MGS algorithm. The randomized Gram‐Schmidt (RGS) algorithm produces a set of sketched orthonormal vectors, and the loss of orthogonality in these vectors is linearly dependent on the condition number of the given matrix. We propose a new variant, RGS‐L2, with reorthogonalization to obtain a set of orthonormal vectors. A round‐off error analysis demonstrates that the loss of orthogonality is close to the unit round‐off level. Numerical experiments exhibit the benefits of our proposed algorithm. Furthermore, we apply the RGS‐L2 algorithm to the Generalized Minimal Residual Method (GMRES) and compare its numerical performance with other GMRES variants.
- Open Access
Mixed Precision Augmented GMRES
We aim to accelerate the restarted generalized minimal residual (GMRES) method for the solution of linear systems by combining two types of techniques. On the one hand, mixed precision GMRES algorithms, which use lower precision in certain steps of the inner cycles, offer significant reductions to computational and memory costs. On the other hand, augmented GMRES algorithms, which recycle information on the eigenvalues of the matrix between restarts by incorporating additional vectors into the Krylov basis, can significantly speed up the convergence. In this work, we investigate how to combine mixed precision and augmentation, in order to cumulate the reduced per-iteration cost of the former with the reduced number of iterations of the latter. We first explore the GMRES with deflated restarting (GMRES-DR) variant, which we show to present limited mixed precision opportunities. Indeed, GMRES-DR can exploit a preconditioner constructed in low precision, but requires a flexible paradigm to also apply it in low precision; moreover, the matrix-vector product and orthonormalization steps must both be kept in high precision as otherwise the method stagnates at low accuracy. We explain that this is because GMRES-DR is based on some algebraic simplifications that are only valid in exact arithmetic, but fail to hold in finite precision. This observation leads us to investigate a more general augmented GMRES framework (AugGMRES) that avoids making these simplifications. AugGMRES is much more resilient to the use of low precision, does not require a flexible paradigm, and successfully converges to high accuracy even when low precision is used for all inner operations. Our numerical experiments illustrate the robustness and fast convergence of AugGMRES on a range of sparse matrices, including ill-conditioned real-life ones.
Open AccessMixed Precision Augmented GMRES
We aim to accelerate the restarted generalized minimal residual (GMRES) method for the solution of linear systems by combining two types of techniques. On the one hand, mixed precision GMRES algorithms, which use lower precision in certain steps of the inner cycles, offer significant reductions to computational and memory costs. On the other hand, augmented GMRES algorithms, which recycle information on the eigenvalues of the matrix between restarts by incorporating additional vectors into the Krylov basis, can significantly speed up the convergence. In this work, we investigate how to combine mixed precision and augmentation, in order to cumulate the reduced per-iteration cost of the former with the reduced number of iterations of the latter. We first explore the GMRES with deflated restarting (GMRES-DR) variant, which we show to present limited mixed precision opportunities. Indeed, GMRES-DR can exploit a preconditioner constructed in low precision, but requires a flexible paradigm to also apply it in low precision; moreover, the matrix-vector product and orthonormalization steps must both be kept in high precision as otherwise the method stagnates at low accuracy. We explain that this is because GMRES-DR is based on some algebraic simplifications that are only valid in exact arithmetic, but fail to hold in finite precision. This observation leads us to investigate a more general augmented GMRES framework (AugGMRES) that avoids making these simplifications. AugGMRES is much more resilient to the use of low precision, does not require a flexible paradigm, and successfully converges to high accuracy even when low precision is used for all inner operations. Our numerical experiments illustrate the robustness and fast convergence of AugGMRES on a range of sparse matrices, including ill-conditioned real-life ones.
- Open Access
Physics-informed neural networks for solving integro-differential equations with applications to viscoelasticity
We explore the application of physics-informed neural networks (PINNs) for solving integro-differential equations (IDEs), with a particular focus on Volterra integral equations. PINNs integrate physical laws directly into the neural network training process, enabling the solution of complex IDEs without relying on traditional discretization methods. Our framework incorporates both space-time differential and integral operators of IDEs, with neural networks approximating solutions that remain consistent with the underlying physics. As in the classical manner, the convolution integral terms of the IDEs can be computed using quadrature rules. When the kernel is defined as exponential, we introduce internal variables, which lead to a significant reduction in both time and memory requirements. The effectiveness of this approach is demonstrated through various numerical experiments, including viscoelastic models. Our results show that PINNs can provide accurate and efficient solutions for IDEs, even in the presence of memory effects, and present a highly effective method for inverse problems such as parameter identification in viscoelastic materials. This work highlights the potential of PINNs for solving a broad range of IDEs, contributing to the growing field of physics-informed machine learning.
Open AccessPhysics-informed neural networks for solving integro-differential equations with applications to viscoelasticity
We explore the application of physics-informed neural networks (PINNs) for solving integro-differential equations (IDEs), with a particular focus on Volterra integral equations. PINNs integrate physical laws directly into the neural network training process, enabling the solution of complex IDEs without relying on traditional discretization methods. Our framework incorporates both space-time differential and integral operators of IDEs, with neural networks approximating solutions that remain consistent with the underlying physics. As in the classical manner, the convolution integral terms of the IDEs can be computed using quadrature rules. When the kernel is defined as exponential, we introduce internal variables, which lead to a significant reduction in both time and memory requirements. The effectiveness of this approach is demonstrated through various numerical experiments, including viscoelastic models. Our results show that PINNs can provide accurate and efficient solutions for IDEs, even in the presence of memory effects, and present a highly effective method for inverse problems such as parameter identification in viscoelastic materials. This work highlights the potential of PINNs for solving a broad range of IDEs, contributing to the growing field of physics-informed machine learning.
- 13 December 2024
- Published by Springer Nature in Computational and Applied Mathematics
- Vol. 44 (2) , 1-36
Fast linear solvers for incompressible CFD simulations with compatible discrete operator schemes
- 13 December 2024
- Published by Springer Nature in Computational and Applied Mathematics
- Vol. 44 (2) , 1-36
Fast linear solvers for incompressible CFD simulations with compatible discrete operator schemes
- Open Access
Randomized orthogonalization process with reorthogonalization
The dimension reduction technique of random sketching is advantageous in significantly reducing computational complexity. In orthogonalization processes like the Gram-Schmidt (GS) algorithm, incorporating random sketching results in a halving of computational costs compared to the classical/modified Gram-Schmidt (CGS/MGS) algorithms, while maintaining numerical stability comparable to the MGS algorithm. The randomized Gram-Schmidt (RGS) algorithm produces a set of sketched orthonormal vectors, and the loss of orthogonality in these vectors is linearly dependent on the condition number of the given matrix. We propose a new variant, RGS2, with reorthogonalization to obtain a set of l2 orthonormal vectors. A round-off error analysis demonstrates that the loss of orthogonality is close to the unit round-off level. Numerical experiments exhibit the benefits of our proposed algorithm. Furthermore, we apply the RGS2 algorithm to the Generalized Minimal Residual Method (GMRES) and compare its numerical performance with other GMRES variants.
Open AccessRandomized orthogonalization process with reorthogonalization
The dimension reduction technique of random sketching is advantageous in significantly reducing computational complexity. In orthogonalization processes like the Gram-Schmidt (GS) algorithm, incorporating random sketching results in a halving of computational costs compared to the classical/modified Gram-Schmidt (CGS/MGS) algorithms, while maintaining numerical stability comparable to the MGS algorithm. The randomized Gram-Schmidt (RGS) algorithm produces a set of sketched orthonormal vectors, and the loss of orthogonality in these vectors is linearly dependent on the condition number of the given matrix. We propose a new variant, RGS2, with reorthogonalization to obtain a set of l2 orthonormal vectors. A round-off error analysis demonstrates that the loss of orthogonality is close to the unit round-off level. Numerical experiments exhibit the benefits of our proposed algorithm. Furthermore, we apply the RGS2 algorithm to the Generalized Minimal Residual Method (GMRES) and compare its numerical performance with other GMRES variants.
- Open Access
A priori and a posteriori analyses of a quasi-static fractional order viscoelasticity model
A power-law type stress relaxation leads to a fractional order viscoelasticity model. We consider the quasistatic fractional order viscoelasticity model which is mathematically expressed as a Volterra integral of the second kind with a weakly singular kernel. Employing the symmetric interior penalty discontinuous Galerkin (SIPG) method and linear interpolation in time, we derive the fully discrete problem. To mitigate temporal discretization errors from a physical perspective in the quasi-static state, we introduce an auxiliary discrete velocity using the Crank-Nicolson approximation. We present a priori stability and error analyses for both the semi-discrete and fully discrete schemes. Additionally, we provide a residual-based a posteriori energy error estimation. Finally, we conduct numerical experiments that validate the error estimate theorems and demonstrate their applicability using real material data.
Open AccessA priori and a posteriori analyses of a quasi-static fractional order viscoelasticity model
A power-law type stress relaxation leads to a fractional order viscoelasticity model. We consider the quasistatic fractional order viscoelasticity model which is mathematically expressed as a Volterra integral of the second kind with a weakly singular kernel. Employing the symmetric interior penalty discontinuous Galerkin (SIPG) method and linear interpolation in time, we derive the fully discrete problem. To mitigate temporal discretization errors from a physical perspective in the quasi-static state, we introduce an auxiliary discrete velocity using the Crank-Nicolson approximation. We present a priori stability and error analyses for both the semi-discrete and fully discrete schemes. Additionally, we provide a residual-based a posteriori energy error estimation. Finally, we conduct numerical experiments that validate the error estimate theorems and demonstrate their applicability using real material data.
- Open Access
- 14 October 2024
- Published by Elsevier in Computers & Mathematics with Applications
- Vol. 176, 165-178
Two-level dynamic load-balanced p-adaptive discontinuous Galerkin methods for compressible CFD simulations
- 14 October 2024
- Published by Elsevier in Computers & Mathematics with Applications
- Vol. 176, 165-178
Open AccessTwo-level dynamic load-balanced p-adaptive discontinuous Galerkin methods for compressible CFD simulations