Introduction

Neutral particles (atoms and molecules) are always present in thermonuclear laboratory plasmas such as those encountered in magnetic fusion experiments, especially in regions near material surfaces. The majority of these neutrals are the direct result of particle recycling, i.e., plasma ions striking material walls and reflected back as neutrals, but they can also arise from external injection of neutral atoms into the system for fueling or heating purposes or be created by recombining plasma ions.

These neutrals can have an important effect on the local plasma particle and power balance. They affect the en- ergy and particle fluxes to the first wall and divertor plates (therefore, playing a major role in processes such as wall erosion and intrinsic impurity production), as well as on the performance of important reactor engineering compo- nents such as the fueling and pumping systems. In addition, experimental observations and theoretical predictions suggest that neutrals play an important role in the overall performance of the core plasma, since they can affect the attainment of various improved confinement regimes, induce density limiting thermal instabilities in the plasma edge, etc. Therefore, the modeling of transport of neutral particles at the edge of magnetically confined plasmas is very important for the interpretation of present day fusion experiments and for the design of next generation fusion reactors, and computer codes that perform such simulations are indispensable tools for plasma modelers. The modeling of neutral transport in fusion plasmas is challenging, since the highest neutral concentrations occur in regions characterized by considerable geometrical complexity (divertors, baffles, pumps, plenums, etc.), widely varying neutral mean-free-paths and background plasmas with densities and temperatures characterized by strong gradients.

Most state-of-the art codes for neutral particle transport are based on the Monte Carlo method [1–3], although methods based on alternative concepts (diffusion [4,5], discrete ordinates, various forms of integral transport [6]) have also been considered. Monte Carlo methods are capable of representing the complex geometries encountered at the plasma edge exactly, can treat the complex atomic and molecular physical processes characterizing the plasma edge region efficiently and can achieve very good accuracy if a sufficient number of particle histories are run. The most serious disadvantage of Monte Carlo-based neutral transport codes is their computational speed. They are computationally expensive, requiring a large number of particle histories in order to yield acceptable statistics. While this may not be a serious problem for stand-alone simulations with fixed background plasma, it becomes much more limiting in coupled plasma-neutrals simulations where a large number of iterations may be required until the two-dimensional (2-D) plasma fluid calculation (which is usually computationally demanding itself) and the neutrals calculation converge. In addition, the numerical noise inherently present in Monte Carlo simulations makes convergence even more difficult. Such coupled edge plasma-neutrals simulations are becoming increasingly more common and the need for a faster alternative to traditional Monte Carlo codes has been recognized by the international fusion community.

The Georgia Tech Neutral Transport code GTNEUT described in this paper is such an alternative. GTNEUT is a computationally efficient and accurate tool for the calculation of neutral transport at the edge of thermonuclear plasmas based on the transmission and escape probability method [7]. The code has been benchmarked extensively against Monte Carlo and experiment [8–10].

The present version of the code has reached a level of maturity and stability and several research groups have expressed an interest in using it for their neutral simulation needs. We therefore feel that an extensive description of the code and the methodology is warranted in order to facilitate its use by the wider fusion community. This paper is organized as follows: The basic assumptions and equations of the code are summarized in Section 2; the details of the code implementation, including a description of the input preparation, a discussion of the solution methodology and the overall structure of the code is presented in Section 3; two test problems, included in the code distribution, are presented and discussed in Section 4; conclusions and a brief discussion, including plans for future development, follow in Section 5; and, finally, a complete list of the input variables is included in Appendix A.