Colloquium: Dr. Fernando Reboredo, Oak Ridge National Labor.
Wednesday, May 8, 2019 · 3:30 - 4:30 PM
TITLE: Computer power and its impact in current theoretical research in materials
ABSTRACT:
Our knowledge of the fundamental physical laws that rule the electronic, optical and structural properties of matter has changed slowly in the last 50 years. We still believe that many-body quantum mechanics is an accurate theoretical description of matter. Fifty years ago, however, slide rulers were still being used. Those rulers were first replaced by calculators, which in turn were quickly replaced by computers. The graphical methods used in engineering and science became obsolete as linear algebra solutions became practical. In the last 30 years, the computer power available to solve these fundamental equations has increased 20 orders of magnitude, adding memory and speed. Linear algebra methods are becoming cumbersome for the new supercomputer architectures. Some questions that arise today are: What kind of Physics or Chemistry can we tackle theoretically with this huge computational power. Which methods have become practical and which could become obsolete in the near term. How we can best prepare as scientist or students for future resources that will continue to change. Those are in general very difficult questions to answer, particularly the last one. However, we can discuss some examples of the new methods and solutions that are available today, and which skills I found useful for me to be able to adapt to 30 years of continuous change.
For almost 30 years, the theoretical research of materials from the ab-initio point of view has been primarily addressed using approximations of density functional theory (DFT). These approximations usually fail for a broad class of interesting materials such as those involving oxides of transition metals. In general, highly inhomogeneous electronic densities result in large errors for most DFT approximations. In our program we have tackled these class of materials with a fundamentally different approach, Diffusion Monte Carlo (DMC), taking maximum advantage of the computer power available in Titan and Summit. I will discuss the fundamental theory of Diffusion Monte Carlo, why is better for current computers, and some of the results of our program that vary from simple binary oxides to complex perovskites, interphases and superlattices.