This dissertation presents a novel potential-based machine-learning method that can simulate drift-wave turbulence in fusion plasmas efficiently and precisely without explicitly resolving the critical inertial range. By using a neural network within a large eddy simulation framework, the method allows for reducing the Hasegawa-Wakatani model by 256x in space, while preserving the physical properties on average, spectrally, and statistically through retaining their statistical distributions closely.     «

This dissertation presents a novel potential-based machine-learning method that can simulate drift-wave turbulence in fusion plasmas efficiently and precisely without explicitly resolving the critical inertial range. By using a neural network within a large eddy simulation framework, the method allows for reducing the Hasegawa-Wakatani model by 256x in space, while preserving the physical properties on average, spectrally, and statistically through retaining their statistical distributions close...     »