Search

The interaction between a plasma and a solid surface is studied in a (1D-1V) kinetic framework using a localized particle and convective energy source. Matching the quasineutral plasma region and sheath horizon is addressed in the fluid framework with a zero heat flux closure. It highlights non-polytropic nature of the physics of parallel transport. Shortfalls of this approach compared to a reference kinetic simulation, highlighting the importance of the heat flux, leads to addressing the sound velocity, non-collisional closure and higher moment closure. No gain in the predictive capability is obtained. The kinetic constraint at the sheath horizon is discussed and modified to account for conditions that are actually met in simulations, namely quasineutrality with a small but finite charge density. Analyzing the distribution functions shows that collisional transfer is mandatory to achieve steady-state self-organization on the open field lines.

We complete the 2D 2-fields turbulence model previously used with an interchange-like instability by slightly modifying the parallel loss terms to drive drift wave instabilities. We show that the instability driven by temperature fluctuations of the sheath losses is identical to that of the drift wave turbulence. The linear analysis is performed and used to select control parameters that yield identical maximum growth rates for the interchange alone and drift wave alone instability. Combining the two instabilities doubles the maximum growth rate. The non-linear simulations are used to analyse the SOL width. The simulations allow one to identify a low field side SOL region where interchange and drift wave are unstable and a high field side SOL region where only the drift wave is unstable. The SOL profiles appear exponential in the region close to the source but depart from a simple exponential fall-off in the far SOL. The low field side SOL width is found to be larger in the interchange alone case, slightly smaller when both instabilities are present and finally narrower when only the drift waves. For the high field side SOL, without interchange, the drift wave SOL width is observed to be identical to that on the low field side and larger than that when both instabilities at play. The Sherwood dimensionless parameter, ratio of convective particle flux divided by the diffusive particle flux, is used to compare the efficiency of turbulent transport. The profiles of the Sherwood parameter for time and flux surface averaged transport indicate that turbulent transport is dominant close to the separatrix but is less effective towards the far SOL. The Sherwood parameter evolution, determined with the flux-surface averaged transport, indicates that outward avalanche transport with corrugations governs the case with interchange only. When combining the two instabilities, outward avalanche transport is less pronounced and inward avalanche transport is observed, reducing the overall turbulent transport efficiency. The avalanche transport with drift waves only compared to interchange only is found to be inhibited.

International audience

International audience

We present the theoretical background of the PoPe and iPoPe verification scheme. The verification that is performed uses the output of actual simulations of production runs. With a small computing overhead it is possible to check that the problem that is solved numerically is consistent with the equations that are to be addressed. In fact, one shows that the numerical error determined by both procedures can be split into a part proportional to the existing operators of the equations, thus modifying their control parameters, completed by a residual error orthogonal to these operators. The accuracy of the numerical solution can be tested on the error as well as on the modification of the control parameters. To illustrate the method, the evolution equation of a simple mechanical system with two conjugate degrees of freedom is used as simulation test bed. Importantly, although dissipative, the trajectory equations evolve towards a chaotic attractor, a strange attractor, characterised by a positive Lyapunov exponent and therefore sensitivity to initial conditions. It is shown that the chaotic state cannot be verified with the standard Method of Manufactured Solution. We present different facets of the PoPe verification method applied to this test case. We show that the evaluation of the accuracy is case dependent for two reasons. First, the error that is generated depends on the values of the control parameter and not only on the numerical scheme. Second, the target accuracy will depend on the problem one wants to address. In a case characterised by bifurcations between different states, the accuracy is determined by the level of detail of the bifurcation phenomena one wants to achieve. A unique verification index is proposed to characterise the accuracy, and consequently the verification, of any given simulation in the production runs. This PoPe index then gives a level of confidence of each simulation. A PoPe index of zero characterises a situation with 100% error level. One finds that although the accuracy is poor the robust features of the solution can still be recovered. The maximum PoPe index is determined by machine precision, typically in the range of 12 to 14. As an illustration this PoPe index is used to choose between a high order integration scheme and a reduced order integration scheme that is less precise but requires less operations. For the chosen example the PoPe index indicates that the high order scheme leads to a reduction of computer resources up to a factor 4 at given accuracy.

Performance portability is considered to be an inevitable requirement in the exascale era. We explore a performance portable approach for fusion plasma turbulence simulation code employing kinetic model, namely the GYSELA code. For this purpose, we extract the key features of GYSELA such as the high dimensionality and the semi- Lagrangian scheme, and encapsulate them into a mini-application which solves the similar but a simplied Vlasov-Poisson system. We implement the mini-app with a mixed OpenACC/OpenMP and Kokkos implementation, where we suppress unnecessary duplications of code lines. For a reference case with the problem size of 1284, the Skylake (Kokkos), Nvidia Tesla P100 (OpenACC), and P100 (Kokkos) versions achieve an acceleration of 1.45, 12.95, and 17.83, respectively, with respect to the baseline OpenMP version on Intel Skylake. In addition to the performance portability, we discuss the code readability and productivity of each implementation. Based on our experience, Kokkos can oer a readable and productive code at the cost of initial porting eorts, which would be enormous for a large scale simulation code like GYSELA., SC19

International audience

International audience

International audience

International audience; The shear of the radial electric field at the edge is widely accepted to be responsible for turbulence reductionin edge transport barriers and thought as a key ingredient of the improved confinement of H-mode plasmas.Nevertheless, a full understanding of how its profile builds up at the edge is still lacking. It can either beformulated as the result of a competition between several mechanisms that generate and damp flows or as theresult of a non-ambipolar particle flux, which enhances radial charge separation imposing a radial electric fieldprofile. Among possible mechanisms, one can think of turbulence generated flow via Reynolds stress, ion orbitlosses, toroidal magnetic ripple and the effect of neutral friction at the edge. Based on previous results obtainedon Tore Supra, the radial profile of the perpendicular flow expected in WEST can be separated in three spatialareas. Inside ρ = 0.8, the radial electric field is dominated by losses of thermal ions in the magnetic ripple while between 0.7 < ρ < 0.95, a competition between this latter and the generation of large scale flows byturbulence appears as a possible explanation of the measured poloidal asymmetry of the mean perpendicularvelocity. In addition, edge conditions such as contact points and parallel dynamics in the scrape-off-layer(SOL) influence the edge profiles beyond ρ = 0.9 (3). This contribution presents unexpected differences in theradial electric field profile observed in WEST between Lower Single Null (LSN) and Upper Single Null (USN)configurations and a study of the competition between specific mechanisms playing a role in the formationof this profile.

Communication and computation overlapping techniques have been introduced in the five‐dimensional gyrokinetic codes GYSELA and GKV. In order to anticipate some of the exa‐scale requirements, these codes were ported to the modern accelerators, Xeon Phi KNL and Tesla P 100 GPU. On accelerators, a serial version of GYSELA on KNL and GKV on GPU are respectively 1.3× and 7.4× faster than those on a single Skylake processor (a single socket). For the scalability, we have measured GYSELA performance on Xeon Phi KNL from 16 to 512 KNLs (1024 to 32k cores) and GKV performance on Tesla P 100 GPU from 32 to 256 GPUs. In their parallel versions, transpose communication in semi‐Lagrangian solver in GYSELA or Convolution kernel in GKV turned out to be a main bottleneck. This indicates that in the exa‐scale, the network constraints would be critical. In order to mitigate the communication costs, the pipeline and task‐based overlapping techniques have been implemented in these codes. The GYSELA 2D advection solver has achieved a 33% to 92% speed up, and the GKV 2D convolution kernel has achieved a factor of 2 speed up with pipelining. The task‐based approach gives 11% to 82% performance gain in the derivative computation of the electrostatic potential in GYSELA. We have shown that the pipeline‐based approach is applicable with the presence of symmetry, while the task‐based approach can be applicable to more general situations.

プラズマ乱流の運動論モデルは、4次元以上の高次元性と各次元の低い解像度に特徴付けられる。これらのコードを、性能可搬性を維持しつつ、GPUによって加速するため、OpenACCとKokkosなどのフレームワークを用いて実装を行った。性能可搬性は、Nvidia P100、Nvidia V100、Marvell Thunder X2、Intel Xeon Goldなど最新のCPU、GPU上で検証した。発表では、OpenACCとKokkosの利点、欠点を性能、可搬性、抽象化などの観点から論じる。, 5th US-Japan JIFT Exascale Computing Collaboration

乱流が非線型結合によって駆動するconvective cellモードに対し数値フィルタを適用する場合としない場合の比較を通じ、その輸送への影響評価を行う。フィルタの有無で乱流スペクトルなどに大きな変化は見られなかったが、フィルタはconvective cellモードを介して間接的に分布関数へ作用し、その分布関数の応答によって新古典熱輸送が変化する場合があることを明らかにした。, 10th Festival de Théorie