Jeff Candy, Emily Belli, and Orso Meneghini traveled to Oak Ridge National Laboratory this week to work on CGYRO optimization, particularly for running on GPU's on the ORNL Titan machine. Work was also done implementing and optimizing the OMFIT SOLPS module.

Jeff Candy and Phil Snyder attended the FES / ASCR Exascale Requirements Review for Fusion Energy Sciences in Gaithersburg, MD January 27-29, 2016.

Studies of self-consistent core-pedestal optimization of the 15MA D-T ITER baseline scenario have been carried out using the OMFIT framework to couple the EPED pedestal model to TGLF and NEO in the core, using TRANSP and EFIT to calculate sources and the current profile. The optimization study has identified the pedestal density neped as well as effective ion charge Zeffped as strong actuators for the ITER fusion performance. In fact, as a consequence of core transport stiffness, the core pressure closely correlates with the pedestal height. A systematic optimization in neped and Zeffped shows that the cases with the highest fusion gain share similar pedestal pressure and current profiles, and they all lie on a surface of constant Zeffped (neped)2. These observations can be understood in terms of the collisional scaling of the bootstrap current in the edge barrier region. The loss of performance at lower and higher collisionality is attributed to the effect of primarily current and pressure driven peeling-ballooning modes, respectively. The strong dependence of the fusion gain with respect to neped and Zeffped points to the need for coupling to a Scrape-Off-Layer (SOL) in the iterative workflow; this is an important future development that is under way as part of the AToM SciDAC project. Also, this study suggests that control of neped and Zeffped during ITER operations is important for optimization of fusion performance.

A highly efficient numerical tool for predicting the pedestal height and width for tokamaks capable of real time application during H-mode operations has been developed. The new tool, named NEUPED, is constructed from a regularized non-linear regression of EPED1 simulations. The regression is performed with a multilayer neural-network. A single neural-network is trained to capture the EPED1 predictions across an input parameter range that spans multiple devices. The multi-device database of EPED1 runs that has been used for training the neural-network was generated by leveraging High Performance Computing capabilities enabled by the AToM project. To date the database comprises more than 18000 EPED1 predictions using inputs from the parameter space characteristic of DIII-D (3000 cases), KSTAR (700 cases), JET (200 cases) and ITER (15000 cases). Handling of the EPED1 results database and training of the neural network is performed within the OMFIT framework. The trained neural network can be used both within OMFIT, and as a portable standalone code. Because NEUPED is millions of times faster than running EPED1 directly, and provides a good approximation to EPED1 predictions, it enables numerous applications, including incorporation in control algorithms and highly efficient integrated modeling.

The NIMROD code has been modified to correctly account for the inertia of the neutral gas source during rotating Massive Gas Injection (MGI) simulations. Because the MGI source initially has no toroidal angular momentum, this problem was previously sidestepped by injecting the source into regions with an artificially enforced no-flow condition. This resulted in under-prediction of toroidal impurity transport and required imposing the no-flow condition on a significantly larger region than the localized injection source. Alternatively, when the constraint on the flow was removed entirely, significant angular momentum was artificially added to the system by the neutral source. The newly added term that correctly accounts for the source inertia, with no artificial constraints, has been tested and is observed to locally suppress flows at the source injection location while allowing some flow in the edge region farther away. Complete DIII-D MGI simulations with the newly added term are in progress.