ELMFIRE is a full-f gyrokinetic plasma simulation code developed as a cooperation between TKK and VTT. The objective of this project is make ELMFIRE capable of simulating the L-H transition of fusion plasmas. This objective is to be fulfilled by first of all identifying the key issues for simulating the transition, and then by developing, implementating and testing computational algorithms that provide such simulation capabilities. Kinetic simulations of plasmas are computationally costful, and ELMFIRE is based on certain simplifications that allows for faster calculations of plasma evolutions. However, the way plasma edge is modeled in ELMFIRE may be insufficient to reproduce the physical processes involved in the plasma dynamics of that region.

Plasma transition from low to high confinement states is a process intrincately related to heated plasma dynamics near the edge, specially the SOL (scrape off layer). This process is of the highest importante to magnetic fusion physics since it provides enhanced confinement capabilities that may increase reactor performance.

ELMFIRE has been used to reproduce the LH-transition results obtained in the ASDEX-U tokamak without achieving successful results. By analyzing the results it was made clear and ELMFIRE current models for the SOL did not reproduce the dynamics of the actual plasma and several improvements were devised in order to make ELMFIRE accurate in the plasma edge.

From the computational point of view, these developments demand a restructuration of the algorithms that deal with geometry, and that so far were bounded to a purely toroidal geometry of concentric flux surfaces. The implementation of all changes to ELMFIRE is a project that spans for longer time than the duration of this project and this is why the objective during the financiation of HPC-Europa2 will consist of the restructuration of interpolation schemes and implicit drift operators to work with arbitrary geometry.

Following the line drafted in the objectives, several changes have been introduced in the ELMFIRE code in order to get a version suitable to work with arbitrary geometry.

Boundary conditions for the calculation of electrostatic potential have been unified by including them as additional equations in the gyrokinetic Poisson equation. The gyrokinetic Poisson equation contains nonlocal operators, which produces a populated stencil which demands boundary conditions in deep. The set of boundary conditions and extrapolations have been unified in the code in order to increase modularity.

Consistent unification of the implicit and explicit operators for particle drifts. ELMFIRE solves for the electrostatic potential by inverting the polarization and electron parallel acceleration operators. The construction of this operators is now performed through interpolation techniques independent of geometry, and using the same subroutines as the explicit computation of the same operators (which have to be applied after solving the potential). This unification and abstraction through call to subroutines has two immediate benefits:

This achievements by themselves have produced a 20% computation speed increase by reducing the order of accuracy of the interpolation schemes, while keeping the quality of the solution. The optimal combination of interpolation schemes for the implicit operators is still under analysis.

Not being the crucial objectives of this project, the following issues have also been treated and improved during this visit:

We have studied trends in reactivity and competition between SN2, anti-E and syn-E mechanisms of X– + CH3CH2Y model reactions (X, Y = F, Cl, Br, I, At) using relativistic density functional theory (DFT) at ZORA-OLYP/TZ2P. The effect of solvation in water has been simulated using the conductor-like screening model (COSMO).

In conclusion, solvation slows down all of our model reactions. Furthermore, it shifts the competition in the case of F–-induced reactions from E2 into the direction of SN2. For the heavier halides, SN2 substitution is anyway preferred over E2 elimination, both in the gas phase and in solution. This follows from our relativistic DFT study of X– + CH3CH2Y model reactions (X, Y = F, Cl, Br, I, At) at ZORA-OLYP/TZ2P in which we employed the COSMO model to simulate the effect of solvation in water.

Performance is an open issue in data intensive applications in large-scale systems. Finding the best implementation solution and influential performance factors between hardware and software platforms requires trial and error. However, it is very difficult and costly to perform the trials using in real large-scale environment. In this paper we use a generic simulation framework SIMCAN to simulate hardware and software platforms for data intensive applications to investigate the influential performance factors, find the best implementation solution, and thereby provide suggestion for decision making on the architectures of data intensive applications. We have employed a use case of data-mining application that has been proposed with a pipeline model. We have simulated various scenarios to investigate factors that can influence the system performance 

In this work we expose the convenience of using simulation models during the design and the optimization of data intensive applications. For that, we need a simulator framework easily configurable and scalable. We use a typical image pattern recognition task as a use of case, which has been implemented using a pipeline model. The results show that the computing times are the main bottleneck in most scenarios followed by the I/O system. In future works we will test different and more complex uses of case also with different architectures In the end we will focus in obtain a general strategy to employ simulation models to aid the design of data intensive applications.