Designing a hybrid CPU-GPU estimator for neutron transport: Advancing eco-efficient Monte Carlo simulations
Digital twins incorporating Monte Carlo simulation models are currently being developed for the design, operation, and decommissioning of nuclear facilities. These twins are capable of predicting physical quantities such as particle fluxes, gamma/neutron heating, and dose equivalent rates. However, the Monte Carlo method presents a major drawback: high computational time to achieve acceptable variance levels.
To enhance simulation efficiency, the eTLE estimator has been developed and integrated into the TRIPOLI-4® Monte Carlo code. Compared to the conventional TLE (Track Length Estimator), eTLE offers lower theoretical variance, particularly in highly absorbing media, by contributing to the detector response even when particles do not physically reach it. Nevertheless, its computational cost remains significant, especially when evaluating multiple detectors.
Two recent PhD works have proposed variants to overcome this limitation. The Forced Detection eTLE- (Guadagni, EPJ Plus 2021) employs preferential sampling that directs pseudo-particles toward the detector at each collision. It is particularly effective for small detectors and configurations with moderate shielding, especially for fast neutrons. The Split Exponential TLE (Hutinet & Antonsanti, EPJ Web 2024) is based on an asynchronous GPU approach, offloading straight-line particle transport to the graphics processor. Through multiple sampling, it maximizes GPU utilization and enables more efficient exploration of phase space.
The proposed thesis aims to combine these two approaches into a hybrid estimator named seTLE-DF. This new estimator could be used either directly or to generate importance maps without relying on auxiliary deterministic calculations. Its implementation will require dedicated GPU developments, particularly to optimize the geometry library and memory management in complex geometries.
This research topic aligns with green computing objectives, aiming to reduce the carbon footprint of high-performance computing. It relies on a hybrid CPU-GPU strategy, avoiding full porting of the Monte Carlo code to GPU. Solutions such as half-precision formats will be considered, and an energy impact assessment will be conducted before and after implementation. The future PhD student will be welcomed with the IRESNE Institute (CEA Cadarache)and will acquire strong expertise in neutron transport simulation, facilitating integration into major research institutions or companies within the nuclear sector.
Constrained geometric optimization of immersed boundaries for thermal-hydraulic simulations of turbulent flow in a finite-volume approach
The technical issue underpinning this thesis topic is the mitigation of the consequences of a loss of primary coolant accident in a pressurized water reactor with loops. It is of the utmost importance to minimize the flow of water leaving the vessel and to manage the available cold water reserves for safety injections as effectively as possible, in order to prevent or delay core flooding, overheating, and possible core degradation. To this end, the use of passive devices operating on the principle of hydraulic diodes, such as vessel flow limiters or advanced accumulators, is being considered. The subject of this thesis is the geometric optimization of this type of device, described by an immersed boundary, in order to maximize its service efficiency.
Several recent theses have shown how to introduce the Penalized Direct Forcing (PDF) immersed boundary method into the TRUST/TrioCFD software, under various spatial discretizations and for laminar and turbulent regimes. Similarly, they have ruled on the possibilities of deterministic geometric optimization in the finite-element context during simulations, based on the use of the PDF method.
After a bibliographic study of this kind of method, we will focus on the possibilities of implementation in finite volume discretization, the consideration of constraints, and the comparison to reference calculations. The latter will be carried out on academic and industrial configurations (accumulators and flow limiters).
The doctoral student will work in a R&D unit on innovative nuclear system within the IRESNE Institute (CEA Cadarache. He will develop skills in fluid mechanics and numerical methods.
Modeling of Critical Heat Flux Using Lattice Boltzmann Methods: Application to the Experimental Devices of the RJH
LBM (Lattice Boltzmann Methods) are numerical techniques used to simulate transport phenomena in complex systems. They allow modeling fluid behavior in terms of particles moving on a discrete grid (a "lattice"). Unlike classical methods, which solve the differential equations of fluids directly, LBM simulate the evolution of the fluid particle distribution functions in a discrete space using propagation and collision rules.
The choice of lattice in LBM is a crucial step, as it directly affects the accuracy, efficiency, and stability of the simulations. The lattice determines how fluid particles interact and move through space, as well as how the discretization of space and time is performed.
LBM methods exhibit a natural parallelism because the computations at each grid point are relatively independent. Compared to classical CFD methods, LBM can better capture certain complex phenomena (such as multiphase, turbulent, or porous media flows) because they rely on a mesoscopic modeling of the fluid, directly derived from particle kinetics, rather than on a macroscopic resolution of the Navier–Stokes equations. This approach allows for a finer representation of interfaces, nonlinear effects, and local interactions, which are often difficult to model accurately using classical CFD methods. LBM therefore enables the capture of complex phenomena at a lower computational cost. Recent studies have notably shown that LBM can reproduce the Nukiyama boiling curve (pool boiling) and, consequently, accurately calculate the critical heat flux. This flux corresponds to a bulk boiling, known as a boiling crisis, which results in a sudden degradation of heat transfer.
The critical heat flux is a crucial issue for the experimental devices (DEX) of the Jules Horowitz Reactor, as they are cooled by water either via natural convection (fuel capsule-type devices) or forced convection (loop-type devices). Thus, to ensure the proper cooling of the DEX and reactor safety, it is essential to verify that the critical heat flux is not reached within the studied parameter range. It must therefore be determined with precision. Previous studies conducted on a fuel-capsule-type DEX using the NEPTUNE-CFD code (classical CFD methods) have shown that modeling is limited to regions far from the critical heat flux. In general, flows with high void fractions (greater than 10%) cannot be easily resolved using classical CFD approaches.
The student will first define a lattice to apply LBM to a RJH device under natural convection. They will consolidate the results obtained for the critical heat flux on this configuration by comparing them with available data. Finally, exploratory calculations under forced convection (laminar to turbulent regime) will be conducted.
The student will be hosted at the IRESNE institute.
Modeling and dynamic studies of a space Nuclear Electric Propulsion system
Nuclear technology is key to enabling the establishment of scientific bases on the Moon or Mars, or for exploring deep space. Its use can take several forms (RTG, NTP among others), but this thesis focuses on Nuclear Electric Propulsion (NEP): heat produced by a nuclear reactor is converted into electricity to power an ionic propulsion engine. Various concepts have been studied in the past (PROMETHEUS, MEGAHIT and DEMOCRITOS, typically for Jupiter satellite exploration missions), while currently design studies are underway at CEA for a 100 kWe nuclear-electric NEP system.
The system of interest combines several specific design choices: uranium nitride fuel, direct gas cooling (helium-xenon mixture) and energy conversion system based on a Brayton cycle, as well as waste heat evacuation through thermal radiation. These choices address requirements to minimize mass and volume, and to ensure performance and reliability for the duration of the scientific mission. Analysis of the dynamic behavior of the nuclear-electric system is therefore crucial for project success. However, the issue of transient modeling of a complete spatial nuclear-electric system is very poorly addressed in the state of the art, especially for NEP.
The thesis objectives are therefore to research and develop physical models adapted to a NEP system, to propose an approach for their validation, and finally to implement them to analyze the dynamic behavior of the reactor and contribute to improving its design. Several mission phases will be studied: reactor startup in space, power variation transients for the ionic propulsion engine, reactor response in case of failure, and its potential shutdown with the problem of safe residual power evacuation.
The thesis will be conducted at IRESNE Institute (CEA Cadarache), in a stimulating scientific environment, and integrated into a team designing innovative nuclear reactors. CNES will also be involved in monitoring the work, particularly to define the ionic propulsion engine characteristics and exploration missions of interest for the nuclear-electric system. The thesis topic, combining modeling, fluid mechanics, thermodynamics, neutronics, and space mechanics, will lend itself to scientific communication and allow the development of key skills for an academic or industrial career.
Multi-physics modelling of a light water nuclear reactor operating under natural convection: study of innovative solutions for startup and power control
Among the most recent designs of water-moderated Small Modular Reactors (SMR), several concepts are characterized by natural convection in the primary circuit during normal and abnormal operation, with the aim of increasing the inherent safety of the design. The absence of primary pumps in this type of SMRs significantly complicates the start-up and power increase ramps. This requires the development of specific start-up procedures to heat up the primary water circuit and enable the reactor to reach its nominal conditions, in accordance with safety requirements. These kinds of procedures rely on simulations using validated models to understand the reactor behavior during these phases and define the accessible parameters domain.
The goal of this PhD project is to develop a numerical model capable of simulating the startup of an SMR operating in natural convection, and to contribute to the validation of this model. The PhD study also aims at developing a methodology for reactor control systems optimization, to attain a fast startup while remaining within the prescribed safety criteria.
The analysis of the reactor startup procedure entails two disciplines: thermal-hydraulics and neutronics, which requires the development of multi-physics coupled simulation tools. Three scientific calculation tools in particular will be coupled in the framework of the PhD study: CATHARE3 (reactor system thermal-hydraulics), FLICA5 (core thermal-hydraulics) and APOLLO3 (neutronics).
The PhD student will work in a team of neutron physicists and thermohydraulic engineers at the IRESNE Institute (CEA Cadarache). He/she will develop skills in nuclear reactor physics and modeling.