SearcharxivSearch

arXiv subjects

William De Deyn

Publications and source records attributed to William De Deyn.

2 recordsLinked to original sources

A noise-robust Monte Carlo method for electric field calculations in EMC3

EMC3 is a state-of-the-art 3D Monte Carlo code for plasma edge transport in stellarator configurations, but it does not yet treat the $E \times B$ drift self-consistently. Computing the drift requires the electric field $E = -\nabla φ$, and hence an accurate gradient of the electric potential $φ$. Because the plasma fields produced by EMC3 are inherently noisy, the finite difference approximation used previously amplifies this noise, increasingly so as the grid is refined. We extend the Monte Carlo Gradient Approximation method, originally developed for 1D Fokker--Planck equations, to a 2D setting and apply it to the electric potential. For an isotropic diffusion coefficient, we derive a PDE governing the evolution of the electric field itself, which allows $E$ to be approximated directly by a Monte Carlo simulation, avoiding finite differences altogether. A numerical experiment based on manufactured solutions demonstrates the accuracy of the method and shows that the variance grows substantially more slowly under grid refinement than for finite differences. The resulting formulation contains a source term involving the second poloidal derivative of the potential, which we neglect. This is admissible only when the radial scale of the plasma edge is thin compared to the poloidal variation scale, an ordering expected to fail near X-points, during detachment, and in island divertors such as W7-X. The present work should therefore be regarded as a proof of concept rather than a general electric field solver for EMC3.

math.NA

Mean-Field Model for Two-Layer Neural Networks Trained with Consensus-Based Optimization

We study Consensus-Based Optimization (CBO) for two-layer neural network training. We compare the performance of CBO against Adam on two test cases and demonstrate how a hybrid approach, combining CBO with Adam, provides faster convergence than CBO. Additionally, in the context of multi-task learning, we recast CBO into a formulation that offers less memory overhead. The CBO method allows for a mean-field model formulation, which we couple with the mean-field model of the neural network. To this end, we first reformulate CBO within the optimal transport framework. As the number of particles tends to infinity, we lift the corresponding dynamics to the Wasserstein-over-Wasserstein space and show that the variance decreases monotonically. We confirm numerically that both mean-field models converge.

cs.LG