Deep neural networks to correct sub-precision errors in CFD

Information loss in numerical physics simulations can arise from various sources when solving discretized partial differential equations. In particular, errors related to precision ("sub-precision errors") accumulate the quantities of interest are performed using low-precision 16-bit floating-point arithmetic compared an equivalent 64-bit simulation. On other hand, computation is less resource intensive than high-precision computation. Several machine learning techniques proposed recently have been successful correcting due coarse spatial discretization. this work, we extend these improve CFD with low precision. We quantify precision-related accumulated a Kolmogorov forced turbulence test case. Subsequently, employ Convolutional Neural Network together fully differentiable solver performing learn tightly-coupled ML-CFD hybrid solver. Compared solver, demonstrate efficacy towards improving metrics pertaining statistical and pointwise accuracy