סמינר מחלקה של אלעד תדמור - מתי אין לסמוך על חיזוי של סימולציות מולקולריות
School of Mechanical Engineering Seminar
Wednesday, June 22 2022, at 14:00
Wolfson Building of Mechanical Engineering, Room 206
" When not to trust the predictions of molecular simulations"
Ellad B. Tadmor
Department of Aerospace Engineering and Mechanics
University of Minnesota
Molecular simulations are increasingly being used to predict material behavior. This trend has been accelerated by machine learning interatomic potentials (IPs) that can provide accuracy close to that of first-principles methods, such as density functional theory (DFT), at a fraction of the computational cost. This greatly extends the scope of accurate molecular simulations, providing opportunities for quantitative design of materials and devices on scales hitherto unreachable by DFT methods. However, machine learning IPs have a basic limitation in that they lack a physical model for the phenomena being predicted and therefore have unknown accuracy when extrapolating outside their training set. It is therefore vital to accompany the predictions of a molecular simulation with uncertainty estimates. This issue is discussed within the context of both physics-based and machine learning IPs. In particular, an approach for estimating uncertainty based on neural network dropout regularization is described. This method can be used to detect configurations outside the training set, and in some cases, can also serve as an estimate for the accuracy of a calculation.
Host: Prof. Dov Sherman
Date: Wednesday, 22 June 2022, 14:00, Room 206, Wolfson Building.
Join Zoom Meeting