TY - JOUR TI - Fast approximate STEM image simulations from a machine learning model AU - Combs, Aidan H. AU - Maldonis, Jason J. AU - Feng, Jie AU - Xu, Zhongnan AU - Voyles, Paul M. AU - Morgan, Dane PY - 2019 JO - Advanced Structural and Chemical Imaging VL - 5 IS - 1, 2 SP - 1-10 PB - Springer Science and Business Media LLC CY - Cham SN - 2198-0926 DO - 10.1186/s40679-019-0064-2 UR - https://fis.uni-bamberg.de/handle/uniba/97250 LA - eng AB - Accurate quantum mechanical scanning transmission electron microscopy image simulation methods such as the multislice method require computation times that are too large to use in applications in high-resolution materials imaging that require very large numbers of simulated images. However, higher-speed simulation methods based on linear imaging models, such as the convolution method, are often not accurate enough for use in these applications. We present a method that generates an image from the convolution of an object function and the probe intensity, and then uses a multivariate polynomial fit to a dataset of corresponding multislice and convolution images to correct it. We develop and validate this method using simulated images of Pt and Pt–Mo nanoparticles and find that for these systems, once the polynomial is fit, the method runs about six orders of magnitude faster than parallelized CPU implementations of the multislice method while achieving a 1 − R2 error of 0.010–0.015 and root-mean-square error/standard deviation of dataset being predicted of about 0.1 when compared to full multislice simulations. KW - Scanning transmission electron microscopy KW - Convolution KW - Frozen phonon multislice simulation KW - High-throughput KW - Linear regression ER -