Fourier pseudo-spectral method on equispaced grid is one of the approaches in turbulence simulation, to compute derivative of discrete data, using fast Fourier Transform (FFT) and gives low dispersion and dissipation errors. In many turbulent flows the dynamically important scales of motion are concentrated in certain regions which requires a coarser grid for higher accuracy. A coarser grid in other regions minimizes the memory requirement. This requires the use of Non-equispaced Fast Fourier Transform (NFFT) to compute the Fourier transform, by solving a system of linear equations. To achieve similar accuracy, the NFFT needs to return more Fourier coefficients than the number of non-equispaced gridpoints, making it an under-determined system. The minimum L2 norm solution of the system is refined using an iterative reconstruction algorithm, FOCUSS. The NFFT and FOCUSS algorithms yield accurate results with smaller test case of a Direct Numerical Simulation on a grid of 64 gridpoints in each dimension, using Taylor green initial condition. The computational speed for this case was found to be unacceptably slow and few methods to improve the performance have been discussed. The approach of NFFT and FOCUSS was tested on a line extracted from 3-dimensional turbulent flow field. Fourier transform of the extracted line, sampled on 1024 non-equispaced gridpoints, computed for 2048 coefficients and the corresponding numerical derivative are found to be inaccurate. It can be observed that the NFFT and FOCUSS approach works for sparse Fourier transform, but not for turbulent fields having a wideband Fourier transform.

On présente ici un ensemble de méthodes numériques performantes pour lasimulation micromagnétique 3D reposant sur l’équation de Landau-Lifchitz-Gilbert, constituantun code nommé feeLLGood. On a choisi l’approche éléments finis pour sa flexibilitégéométrique. La formulation adoptée respecte la contrainte d’orthogonalité entre l’aimantationet sa dérivée temporelle, contrairement à la formulation classique sur-dissipative.On met au point un schéma de point milieu pour l’équation Landau-Lifchitz-Gilbert quiest stable et d’ordre deux en temps. Cela permet de prendre, à précision égale, des pas detemps beaucoup plus grands (typiquement un ordre de grandeur) que les schémas classiques.Un véritable enjeu numérique est le calcul du champ démagnétisant, non local. Oncompare plusieurs techniques de calcul rapide pour retenir celles, inédites dans le domaine,des multipôles rapides (FMM) et des transformées de Fourier hors-réseau (NFFT). Aprèsavoir validé le code sur des cas-tests et établi son efficacité, on présente les applications àla simulation des nanostructures : sélection de chiralité et résonance ferromagnétique d’unplot monovortex de cobalt, hystérésis des chapeaux de Néel dans un plot allongé de fer.Enfin, l’étude d’un oscillateur spintronique prouve l’évolutivité du code. / Here is presented a set of efficient numerical methods for 3D micromagneticsimulation based on the Landau-Lifchitz-Gilbert equation, making up a code named feeLLGood.The finite element approach was chosen for its geometrical flexibility. The adoptedformulation meets the orthogonality constraint between the magnetization and its time derivative,unlike the over-dissipative classical formulation. A midoint rule was developed forthe Landau-Lifchitz-Gilbert equation which is stable and second order in time. This allowsfor much bigger time steps (typically an order of magnitude) than classical schemes at thesame precision. Computing the nonlocal demagnetizing interaction is a real numerical challenge.Several fast computation techniques are compared. Those selected are novel to thefield : the Fast Multipole Method (FMM) and Non-uniform Fast Fourier Transforms (NFFT).After the code is validated on test cases and its efficiency established, applications to the simulationof nanostructures are presented : chirality selection and ferromagnetic resonanceof a cobalt monovortex dot, Neel caps hysteresis in an iron dot. Finally, the study of a spintronicoscillator proves the code’s upgradability.