Antonio Tadeu Azevedo Gomes
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Antonio Tadeu Azevedo Gomes has a total of 144 co-authors. Here are the top co-authors by number of shared publications:
- Artur Ziviani
National Laboratory for Scientific Computing (LNCC), Petrópolis, Brazil
15 shared publications - Frédéric Valentin
Laboratório Nacional de Computação Científica (LNCC)
6 shared publications Geoff Coulson
Lancaster University
4 shared publicationsMarkus Endler
Pontifical Catholic University of Rio de Janeiro (PUC-Rio)
4 shared publicationsAckbar Joolia
Computing Dept, Infolab21, Lancaster University, Lancaster LA1 4WA
3 shared publicationsArtur Ziviani
National Laboratory for Scientific Computing (LNCC), Brazil
3 shared publications
Latest Publications
- Open Access
- 4 January 2026
- Published by Wiley in Concurrency and Computation: Practice and Experience
- Vol. 38 (1)
Towards Workload‐Tailored Optimization of Job Scheduling Policies in HPC Environments
Supercomputers play a pivotal role in advancing research and development across diverse scientific and engineering domains. However, configuring job scheduling in these systems to ensure maximum productivity and cost‐effectiveness is a challenge. Workload simulation emerges as a crucial tool in this context, offering a mechanism to explore job scheduling configurations in the presence of expected user behaviors. In this paper, we focus on simulation‐based optimization applied to tuning job scheduling configurations. We introduce a discrete‐event simulator that utilizes two strategies to accommodate real workload traces under varying job scheduling policies: Job shaping and job splitting. Our findings from evaluating the proposed strategies on a real‐world case study suggest that they allow the effective accommodation of the real workload traces used as input to the simulation of incompatible policies. By plugging the simulator into an evolutionary optimization algorithm, we also demonstrate the flexibility of the proposed strategies in helping with the proper exploration of the job scheduling configuration space.
- 4 January 2026
- Published by Wiley in Concurrency and Computation: Practice and Experience
- Vol. 38 (1)
Open AccessTowards Workload‐Tailored Optimization of Job Scheduling Policies in HPC Environments
Supercomputers play a pivotal role in advancing research and development across diverse scientific and engineering domains. However, configuring job scheduling in these systems to ensure maximum productivity and cost‐effectiveness is a challenge. Workload simulation emerges as a crucial tool in this context, offering a mechanism to explore job scheduling configurations in the presence of expected user behaviors. In this paper, we focus on simulation‐based optimization applied to tuning job scheduling configurations. We introduce a discrete‐event simulator that utilizes two strategies to accommodate real workload traces under varying job scheduling policies: Job shaping and job splitting. Our findings from evaluating the proposed strategies on a real‐world case study suggest that they allow the effective accommodation of the real workload traces used as input to the simulation of incompatible policies. By plugging the simulator into an evolutionary optimization algorithm, we also demonstrate the flexibility of the proposed strategies in helping with the proper exploration of the job scheduling configuration space.
- Open Access
- 14 May 2025
- Published by Springer Nature in Research Square
Efficient Deployment of a 685B-Parameter Open-Source LLM on the Brazilian Santos Dumont Supercomputer
This brief communication presents the deployment of open-source large language models (LLMs) on the Santos Dumont Brazilian Supercomputer to support the Laborat´orio Nacional de Computa¸c˜ao Cient´ıfica (LNCC) academic community. The solution, named Carcar´a, leverages key optimizations such as dynamic quantization techniques to enable an accessible, scalable, and cost-effective implementation. As a result, a single-instance stateof- the-art independent model was deployed on a single node equipped with 4 × NVIDIA H100 GPUs. The quantized models fit entirely within the node’s VRAM, enabling horizontal scaling without the need for inter-node synchronization. In this way, we could provide access to these models to the entire LNCC academic community using only four computational nodes, demonstrating the efficiency and scalability of this approach. Crucially, our approach ensures data sovereignty, allowing LNCC researchers and postgraduate students to utilize AI for sensitive research topics with full control over their data, free from the privacy risks associated with proprietary solutions. This initiative strengthens national scientific autonomy while providing secure and efficient AI tools for academic and research advancements. The code to deploy this solution is openly available, and we encourage other institutions to adapt it and support their own communities. Discussions with the Brazilian Ministry of Science, Technology and Innovation are underway to expand this strategic solution to other research centers and universities in Brazil.
- 14 May 2025
- Published by Springer Nature in Research Square
Open AccessEfficient Deployment of a 685B-Parameter Open-Source LLM on the Brazilian Santos Dumont Supercomputer
This brief communication presents the deployment of open-source large language models (LLMs) on the Santos Dumont Brazilian Supercomputer to support the Laborat´orio Nacional de Computa¸c˜ao Cient´ıfica (LNCC) academic community. The solution, named Carcar´a, leverages key optimizations such as dynamic quantization techniques to enable an accessible, scalable, and cost-effective implementation. As a result, a single-instance stateof- the-art independent model was deployed on a single node equipped with 4 × NVIDIA H100 GPUs. The quantized models fit entirely within the node’s VRAM, enabling horizontal scaling without the need for inter-node synchronization. In this way, we could provide access to these models to the entire LNCC academic community using only four computational nodes, demonstrating the efficiency and scalability of this approach. Crucially, our approach ensures data sovereignty, allowing LNCC researchers and postgraduate students to utilize AI for sensitive research topics with full control over their data, free from the privacy risks associated with proprietary solutions. This initiative strengthens national scientific autonomy while providing secure and efficient AI tools for academic and research advancements. The code to deploy this solution is openly available, and we encourage other institutions to adapt it and support their own communities. Discussions with the Brazilian Ministry of Science, Technology and Innovation are underway to expand this strategic solution to other research centers and universities in Brazil.
- Open Access
- 20 January 2025
- Published by Brazilian Society for Computational and Applied Mathematics (SBMAC) in Proceeding Series of the Brazilian Society of Computational and Applied Mathematics
- Vol. 11 (1) , 1-7
Interplay of Physics-Informed Neural Networks and Multiscale Numerical Methods
Physics-Informed Neural Networks (PINNs) are machine learning tools that approximate the solution of general Partial Differential Equations (PDEs) by incorporating them as terms in the loss/cost function of a Neural Network (NN). However, they can present some difficulties when the PDEs present multiscale features. To mitigate these issues, we propose an approach to embed PINNs within the framework of the Multiscale Hybrid-Mixed (MHM) method. In the MHM method, multiscale basis functions are obtained on a coarse mesh by solving completely independent local problems. Here, we propose an approach to estimate these multiscale basis functions through PINN models. Thus, the model is adjusted to generate basis functions, adapting to the structure and characteristics provided by the local domain. Through numerical validations, we show the model’s capability to approximate multiscale basis functions for the Poisson problem.
- 20 January 2025
- Published by Brazilian Society for Computational and Applied Mathematics (SBMAC) in Proceeding Series of the Brazilian Society of Computational and Applied Mathematics
- Vol. 11 (1) , 1-7
Open AccessInterplay of Physics-Informed Neural Networks and Multiscale Numerical Methods
Physics-Informed Neural Networks (PINNs) are machine learning tools that approximate the solution of general Partial Differential Equations (PDEs) by incorporating them as terms in the loss/cost function of a Neural Network (NN). However, they can present some difficulties when the PDEs present multiscale features. To mitigate these issues, we propose an approach to embed PINNs within the framework of the Multiscale Hybrid-Mixed (MHM) method. In the MHM method, multiscale basis functions are obtained on a coarse mesh by solving completely independent local problems. Here, we propose an approach to estimate these multiscale basis functions through PINN models. Thus, the model is adjusted to generate basis functions, adapting to the structure and characteristics provided by the local domain. Through numerical validations, we show the model’s capability to approximate multiscale basis functions for the Poisson problem.
- Open Access
- 23 October 2024
- Published by Sociedade Brasileira de Computação
- p. 73-84
A job shaping strategy to accomodate workload traces under varying resource management policies
Supercomputers play a pivotal role in advancing research and development across diverse scientific and engineering domains. However, configuring resource management in these systems to ensure maximum productivity and cost-effectiveness is a challenge. Workload simulation emerges as a crucial tool in this context, offering a mechanism to explore resource management configurations in the presence of expected user behaviors. This paper focuses on a specific requirement for simulation-based optimization applied to tuning resource management configurations: the need for simulators that are both precise and efficient. This paper introduces a job shaping strategy to accommodate real workload traces under varying resource management policies in discrete-event RMS simulations. Our findings from evaluating the proposed strategy on a real-world case study suggest that job shaping allows effectively capturing changes in system behavior, regardless of whether some of the real workload traces used as input to the simulation are incompatible with the simulated policies.
- 23 October 2024
- Published by Sociedade Brasileira de Computação
- p. 73-84
Open AccessA job shaping strategy to accomodate workload traces under varying resource management policies
Supercomputers play a pivotal role in advancing research and development across diverse scientific and engineering domains. However, configuring resource management in these systems to ensure maximum productivity and cost-effectiveness is a challenge. Workload simulation emerges as a crucial tool in this context, offering a mechanism to explore resource management configurations in the presence of expected user behaviors. This paper focuses on a specific requirement for simulation-based optimization applied to tuning resource management configurations: the need for simulators that are both precise and efficient. This paper introduces a job shaping strategy to accommodate real workload traces under varying resource management policies in discrete-event RMS simulations. Our findings from evaluating the proposed strategy on a real-world case study suggest that job shaping allows effectively capturing changes in system behavior, regardless of whether some of the real workload traces used as input to the simulation are incompatible with the simulated policies.
- Open Access
A low-order locking-free multiscale finite element method for isotropic elasticity
The multiscale hybrid-mixed (MHM) method consists of a multi-level strategy to approximate the solution of boundary value problems with heterogeneous coefficients. In this context, we propose a family of low-order finite elements for the linear elasticity problem which are free from Poisson locking. The finite elements rely on face degrees of freedom associated with multiscale bases obtained from local Neumann problems with piecewise polynomial interpolations on faces. We establish sufficient refinement levels on the fine-scale mesh such that the MHM method is well-posed, optimally convergent under local regularity conditions, and locking-free. Two-dimensional numerical tests assess theoretical results.
Open AccessA low-order locking-free multiscale finite element method for isotropic elasticity
The multiscale hybrid-mixed (MHM) method consists of a multi-level strategy to approximate the solution of boundary value problems with heterogeneous coefficients. In this context, we propose a family of low-order finite elements for the linear elasticity problem which are free from Poisson locking. The finite elements rely on face degrees of freedom associated with multiscale bases obtained from local Neumann problems with piecewise polynomial interpolations on faces. We establish sufficient refinement levels on the fine-scale mesh such that the MHM method is well-posed, optimally convergent under local regularity conditions, and locking-free. Two-dimensional numerical tests assess theoretical results.
- Open Access
A MULTISCALE HYBRID METHOD
In this work we propose, analyze, and test a new multiscale finite element method called Multiscale Hybrid (MH) method. The method is built as a close relative to the Multiscale Hybrid Mixed (MHM) method, but with the fundamental difference that a novel definition of the Lagrange multiplier is introduced. The practical implication of this is that both the local problems to compute the basis functions, as well as the global problem, are elliptic, as opposed to the MHM method (and also other previous methods) where a mixed global problem is solved, and constrained local problems are solved to compute the local basis functions. The error analysis of the method is based on a hybrid formulation, and a static condensation process is done at the discrete level, so the final global system only involves the Lagrange multipliers. We tested the performance of the method by means of numerical experiments for problems with multiscale coefficients, and we carried out comparisons with the MHM method in terms of performance, accuracy, and memory requirements.
Open AccessA MULTISCALE HYBRID METHOD
In this work we propose, analyze, and test a new multiscale finite element method called Multiscale Hybrid (MH) method. The method is built as a close relative to the Multiscale Hybrid Mixed (MHM) method, but with the fundamental difference that a novel definition of the Lagrange multiplier is introduced. The practical implication of this is that both the local problems to compute the basis functions, as well as the global problem, are elliptic, as opposed to the MHM method (and also other previous methods) where a mixed global problem is solved, and constrained local problems are solved to compute the local basis functions. The error analysis of the method is based on a hybrid formulation, and a static condensation process is done at the discrete level, so the final global system only involves the Lagrange multipliers. We tested the performance of the method by means of numerical experiments for problems with multiscale coefficients, and we carried out comparisons with the MHM method in terms of performance, accuracy, and memory requirements.
- 21 December 2022
- Published by Springer Nature
- p. 90-102
Improving Boundary Layer Predictions Using Parametric Physics-Aware Neural Networks
- 21 December 2022
- Published by Springer Nature
- p. 90-102
Improving Boundary Layer Predictions Using Parametric Physics-Aware Neural Networks
- 3 August 2022
- Published by Oxford University Press (OUP) in IMA Journal of Numerical Analysis
- Vol. 43 (4) , 2265-2298
The MHM Method for Linear Elasticity on Polytopal Meshes
The multiscale hybrid-mixed (MHM) method consists of a multi-level strategy to approximate the solution of boundary value problems with heterogeneous coefficients. In this context, we propose a new family of finite elements for the linear elasticity equation defined on coarse polytopal partitions of the domain. The finite elements rely on face degrees of freedom associated with multiscale bases obtained from local Neumann problems with piecewise polynomial interpolations on faces. We establish sufficient conditions on the fine-scale interpolations such that the MHM method is well-posed and optimally convergent under local regularity conditions. Also, a multi-level error analysis demonstrates that the MHM method achieves convergence without refining the coarse partition. The upshot is that the Poincaré and Korn’s inequalities do not degenerate, and then the convergence arises on general meshes. Two- and three-dimensional numerical tests assess theoretical results and verify the robustness of the method on a multi-layer media case.
- 3 August 2022
- Published by Oxford University Press (OUP) in IMA Journal of Numerical Analysis
- Vol. 43 (4) , 2265-2298
The MHM Method for Linear Elasticity on Polytopal Meshes
The multiscale hybrid-mixed (MHM) method consists of a multi-level strategy to approximate the solution of boundary value problems with heterogeneous coefficients. In this context, we propose a new family of finite elements for the linear elasticity equation defined on coarse polytopal partitions of the domain. The finite elements rely on face degrees of freedom associated with multiscale bases obtained from local Neumann problems with piecewise polynomial interpolations on faces. We establish sufficient conditions on the fine-scale interpolations such that the MHM method is well-posed and optimally convergent under local regularity conditions. Also, a multi-level error analysis demonstrates that the MHM method achieves convergence without refining the coarse partition. The upshot is that the Poincaré and Korn’s inequalities do not degenerate, and then the convergence arises on general meshes. Two- and three-dimensional numerical tests assess theoretical results and verify the robustness of the method on a multi-layer media case.
- Open Access
Physics-Aware Neural Networks for Boundary Layer Linear Problems
Physics-Informed Neural Networks (PINNs) are machine learning tools that approximate the solution of general partial differential equations (PDEs) by adding them in some form as terms of the loss/cost function of a Neural Network. Most pieces of work in the area of PINNs tackle non-linear PDEs. Nevertheless, many interesting problems involving linear PDEs may benefit from PINNs; these include parametric studies, multi-query problems, and parabolic (transient) PDEs. The purpose of this paper is to explore PINNs for linear PDEs whose solutions may present one or more boundary layers. More specifically, we analyze the steady-state reaction-advection-diffusion equation in regimes in which the diffusive coefficient is small in comparison with the reactive or advective coefficients. We show that adding information about these coefficients as predictor variables in a PINN results in better prediction models than in a PINN that only uses spatial information as predictor variables. This finding may be instrumental in multiscale problems where the coefficients of the PDEs present high variability in small spatiotemporal regions of the domain, and therefore PINNs may be employed together with domain decomposition techniques to efficiently approximate the PDEs locally at each partition of the spatiotemporal domain, without resorting to different learned PINN models at each of these partitions.
Open AccessPhysics-Aware Neural Networks for Boundary Layer Linear Problems
Physics-Informed Neural Networks (PINNs) are machine learning tools that approximate the solution of general partial differential equations (PDEs) by adding them in some form as terms of the loss/cost function of a Neural Network. Most pieces of work in the area of PINNs tackle non-linear PDEs. Nevertheless, many interesting problems involving linear PDEs may benefit from PINNs; these include parametric studies, multi-query problems, and parabolic (transient) PDEs. The purpose of this paper is to explore PINNs for linear PDEs whose solutions may present one or more boundary layers. More specifically, we analyze the steady-state reaction-advection-diffusion equation in regimes in which the diffusive coefficient is small in comparison with the reactive or advective coefficients. We show that adding information about these coefficients as predictor variables in a PINN results in better prediction models than in a PINN that only uses spatial information as predictor variables. This finding may be instrumental in multiscale problems where the coefficients of the PDEs present high variability in small spatiotemporal regions of the domain, and therefore PINNs may be employed together with domain decomposition techniques to efficiently approximate the PDEs locally at each partition of the spatiotemporal domain, without resorting to different learned PINN models at each of these partitions.