Ibtihel Ben Gharbia
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Ibtihel Ben Gharbia has a total of 19 co-authors. Here are the top co-authors by number of shared publications:
- Quang Huy Tran
IFP Energies nouvelles
2 shared publications - Abdoulaye Koroko
IFP Energies nouvelles, 1 et 4 avenue de Bois Préau, 92852 Rueil-Malmaison cedex, France
1 shared publication - Ani Anciaux-Sedrakian
IFP Energies nouvelles, 1 et 4 avenue de Bois Préau, 92852 Rueil-Malmaison cedex, France
1 shared publication - Antonio Sciarretta
IFP Energies nouvelles
1 shared publication - Clément Cancès
Université de Lille, France
1 shared publication - Giovanni De Nunzio
IFP Energies nouvelles
1 shared publication
Latest Publications
- Open Access
Reactive flash for ideal multiphase mixtures: Unified formulation and efficient computation
Multiphase chemical equilibrium problems lead to nonlinear systems with complementarity constraints, which become particularly challenging when phases may vanish. We introduce a new algebraic formulation of the equilibrium problem based on extended mole fractions, derived from the subdifferential of the Gibbs free energy, and establish its equivalence with the classical minimization problem. Our analysis provides new conditions ensuring the uniqueness of solutions, even when some phases disappear. Building on this formulation, we propose two parametrized Newton-based strategies: one reformulates the relation between species quantities and chemical potentials, while the other parametrizes the complementarity conditions directly. Numerical experiments on a system with 72 species and 22 phases confirm the robustness and efficiency of the proposed methods. In tests with randomized inputs, both strategies achieve success rates above 90% with moderate iteration counts, outperforming established approaches such as the Newton-min and Fischer-Burmeister complementarity functions, and interior-point methods.
Open AccessReactive flash for ideal multiphase mixtures: Unified formulation and efficient computation
Multiphase chemical equilibrium problems lead to nonlinear systems with complementarity constraints, which become particularly challenging when phases may vanish. We introduce a new algebraic formulation of the equilibrium problem based on extended mole fractions, derived from the subdifferential of the Gibbs free energy, and establish its equivalence with the classical minimization problem. Our analysis provides new conditions ensuring the uniqueness of solutions, even when some phases disappear. Building on this formulation, we propose two parametrized Newton-based strategies: one reformulates the relation between species quantities and chemical potentials, while the other parametrizes the complementarity conditions directly. Numerical experiments on a system with 72 species and 22 phases confirm the robustness and efficiency of the proposed methods. In tests with randomized inputs, both strategies achieve success rates above 90% with moderate iteration counts, outperforming established approaches such as the Newton-min and Fischer-Burmeister complementarity functions, and interior-point methods.
- Open Access
- 30 August 2023
- Published by EDP Sciences in ESAIM: Proceedings and Surveys
- Vol. 73, 218-237
Efficient approximations of the fisher matrix in neural networks using kronecker product singular value decomposition
We design four novel approximations of the Fisher Information Matrix (FIM) that plays a central role in natural gradient descent methods for neural networks. The newly proposed approximations are aimed at improving Martens and Grosse’s Kronecker-factored block diagonal (KFAC) one. They rely on a direct minimization problem, the solution of which can be computed via the Kronecker product singular value decomposition technique. Experimental results on the three standard deep auto-encoder benchmarks showed that they provide more accurate approximations to the FIM. Furthermore, they outperform KFAC and state-of-the-art first-order methods in terms of optimization speed.
- 30 August 2023
- Published by EDP Sciences in ESAIM: Proceedings and Surveys
- Vol. 73, 218-237
Open AccessEfficient approximations of the fisher matrix in neural networks using kronecker product singular value decomposition
We design four novel approximations of the Fisher Information Matrix (FIM) that plays a central role in natural gradient descent methods for neural networks. The newly proposed approximations are aimed at improving Martens and Grosse’s Kronecker-factored block diagonal (KFAC) one. They rely on a direct minimization problem, the solution of which can be computed via the Kronecker product singular value decomposition technique. Experimental results on the three standard deep auto-encoder benchmarks showed that they provide more accurate approximations to the FIM. Furthermore, they outperform KFAC and state-of-the-art first-order methods in terms of optimization speed.
- Open Access
A General Constrained Optimization Framework for the Eco-Routing Problem: Comparison and Analysis of Solution Strategies for Hybrid Electric Vehicles
Vehicles electrification marks a very important step towards sustainable mobility. However, energy efficiency and driving range of electrified vehicles are nowadays a major concern. From an algorithmic perspective, eco-routing opens up new possibilities regarding the strategies and tools aimed at improving energy efficiency by finding an energy-minimal route under different constraints coming from vehicle characteristics (powertrain, battery capacity, etc.) and user preferences (travel time, etc.). In this work, a powertrain-independent speed prediction model is presented. This model is then used to derive a fast numerical solution of the powertrain energy management for hybrid electric vehicles. Furthermore, a new general formulation is derived for the minimum-energy navigation problem, with a focus on the specific complexity introduced by electrified vehicles. The general constrained optimization problem is reformulated in several alternative ways in order to achieve a solution in limited computation time. The most commonly used approaches nowadays in the literature (integer programming and shortest path algorithms on directed graphs) are compared and benchmarked in terms of solution accuracy and computational effort. The objective is to identify best-practices in accurately and efficiently solving the constrained eco-routing problem.
Open AccessA General Constrained Optimization Framework for the Eco-Routing Problem: Comparison and Analysis of Solution Strategies for Hybrid Electric Vehicles
Vehicles electrification marks a very important step towards sustainable mobility. However, energy efficiency and driving range of electrified vehicles are nowadays a major concern. From an algorithmic perspective, eco-routing opens up new possibilities regarding the strategies and tools aimed at improving energy efficiency by finding an energy-minimal route under different constraints coming from vehicle characteristics (powertrain, battery capacity, etc.) and user preferences (travel time, etc.). In this work, a powertrain-independent speed prediction model is presented. This model is then used to derive a fast numerical solution of the powertrain energy management for hybrid electric vehicles. Furthermore, a new general formulation is derived for the minimum-energy navigation problem, with a focus on the specific complexity introduced by electrified vehicles. The general constrained optimization problem is reformulated in several alternative ways in order to achieve a solution in limited computation time. The most commonly used approaches nowadays in the literature (integer programming and shortest path algorithms on directed graphs) are compared and benchmarked in terms of solution accuracy and computational effort. The objective is to identify best-practices in accurately and efficiently solving the constrained eco-routing problem.
- 10 January 2020
- Published by Springer Nature in Computational Geosciences
- Vol. 24 (3) , 1031-1055
A posteriori error estimates for a compositional two-phase flow with nonlinear complementarity constraints
- 10 January 2020
- Published by Springer Nature in Computational Geosciences
- Vol. 24 (3) , 1031-1055
A posteriori error estimates for a compositional two-phase flow with nonlinear complementarity constraints
- Open Access
- 19 October 2018
- Published by EDP Sciences in ESAIM: Proceedings and Surveys
- Vol. 63, 135-151
Fast Solution of Boundary Integral Equations for Elasticity Around a Crack Network: A Comparative Study
Because of the non-local nature of the integral kernels at play, the discretization of boundary integral equations leads to dense matrices, which would imply high computational complexity. Acceleration techniques, such as hierarchical matrix strategies combined with Adaptive Cross Approximation (ACA), are available in literature. Here we apply such a technique to the solution of an elastostatic problem, arising from industrial applications, posed at the surface of highly irregular cracks networks. La discrétisation des équations intégrales fait apparaître des matrices denses du fait de la nature non-locale des noyaux intégraux mis en jeu, ce qui implique un coût important des calculs. Il existe différentes techniques dans la littérature permettant de réduire ces coûts, notamment l’utilisation de matrices hiérarchiques associées à l’Adaptative Cross Approximation (ACA). Nous appliquons cette technique pour résoudre des problèmes d’élasto-statique issus d’applications industrielles et posés à la surface de réseaux complexes de fractures.
- 19 October 2018
- Published by EDP Sciences in ESAIM: Proceedings and Surveys
- Vol. 63, 135-151
Open AccessFast Solution of Boundary Integral Equations for Elasticity Around a Crack Network: A Comparative Study
Because of the non-local nature of the integral kernels at play, the discretization of boundary integral equations leads to dense matrices, which would imply high computational complexity. Acceleration techniques, such as hierarchical matrix strategies combined with Adaptive Cross Approximation (ACA), are available in literature. Here we apply such a technique to the solution of an elastostatic problem, arising from industrial applications, posed at the surface of highly irregular cracks networks. La discrétisation des équations intégrales fait apparaître des matrices denses du fait de la nature non-locale des noyaux intégraux mis en jeu, ce qui implique un coût important des calculs. Il existe différentes techniques dans la littérature permettant de réduire ces coûts, notamment l’utilisation de matrices hiérarchiques associées à l’Adaptative Cross Approximation (ACA). Nous appliquons cette technique pour résoudre des problèmes d’élasto-statique issus d’applications industrielles et posés à la surface de réseaux complexes de fractures.