Session C62: Machine Learning Applications in Experimental Quantum Materials Research

Learning Quantum Emergence with AI.

In recent years, enormous data sets have begun to appear in real-space visualizations (scanning probes) and reciprocal-space visualizations (scattering probes) of electronic quantum matter. Increasing volume and variety of such data present new challenges and opportunities that are ripe for a new approach: machine learning. However, the scientific questions in the field of electronic quantum matter require fundamentally new approaches to data science for two reasons: (1) quantum mechanical imaging of electronic behavior is probabilistic, (2) inference from data should be subject to fundamental laws governing microscopic interactions. In this talk, I will review the aspects of machine learning that are appealing for dealing with quantum complexity and present how we implemented a machine learning approach to the analysis of scanning tunneling spectroscopy data and X-ray scattering data.   

Electron Microscopy of Quantum Materials: From Learning Physics to Atomic Manipulation

Atomically-resolved imaging of materials has become the mainstay of modern materials science, as enabled by advent of aberration corrected scanning transmission electron microscopy (STEM). In this talk, I will present the new opportunities enabled by physics-informed big data and machine learning technologies to extract physical information from static and dynamic STEM images. The deep learning models trained on theoretically simulated images or labeled library data demonstrate extremely high efficiency in extracting atomic coordinates and trajectories, converting massive volumes of statistical and dynamic data into structural descriptors. I further present a method to take advantage of atomic-scale observations of chemical and structural fluctuations and use them to build a generative model (including near-neghbour interactions) that can be used to predict the phase diagram of the system in a finite temperature and composition space. Similar approach is applied to probe the kinetics of solid-state reactions on a single defect level and defect formation in solids via atomic-scale observations. Finally, synergy of deep learning image analytics and real-time feedback further allows harnessing beam-induced atomic and bond dynamics to enable direct atom-by-atom fabrication. Examples of direct atomic motion over mesoscopic distances, engineered doping at selected lattice site, and assembly of multiatomic structures will be demonstrated. These advances position STEM towards transition from purely imaging tool for atomic-scale laboratory of electronic, phonon, and quantum phenomena in atomically-engineered structures.   

Bridging simulations and theories of correlated electron materials using ideas from machine learning

There has historically been a tension between first principles calculations, which attempt to solve a realistic model while controlling approximations, and effective Hamiltonians, which attempt to condense the important physics into a simple enough model to solve without approximation. I will report on a research program that has, without approximation, recast the effective Hamiltonian as a compressed representation of the first principles Hamiltonian in a subspace, in a similar way that a JPEG file achieves a parsimonious description of a picture. Just as in the case of images, it turns out that concepts from data science and machine learning can help us describe complex effective Hamiltonians and choose, from first principles, the most accurate and parsimonious effective Hamiltonians.
  

Autonomous Quantum Materials Research: Phase Mapping

The last few decades have seen significant advancements in materials research tools, allowing researchers to rapidly synthesis and characterize large numbers of samples - a major step toward high-throughput materials discovery. Machine learning has been tasked to aid in converting the collected materials property data into actionable knowledge, and more recently it has been used to assist in experiment design. In this talk we present the next step in machine learning for materials research - autonomous materials research systems. We first demonstrate autonomous measurement systems for phase mapping, followed by a discussion of ongoing work in building fully autonomous systems. For the autonomous measurement systems, machine learning controls X-ray diffraction measurement equipment both in the lab and at the beamline to identify phase maps from composition spreads with a minimum number of measurements. The algorithm also capitalizes on prior knowledge in the form of physics theory and external databases, both theory-based and experiment-based, to more rapidly hone in on the optimal results.   

Machine learning modeling of superconducting critical temperature

Machine learning has emerged as a powerful new research tool that can be used to answer many scientific questions in unconventional ways. In this talk I will discuss how it can help us address one of the most challenging problems in the study of quantum matter – finding connection between superconductivity – in particular critical temperature T_c – and chemical/structural properties of materials. I will present several recently developed machine learning methods for modeling T_c of the 12,000+ known superconductors available via the SuperCon database. These models use coarse-grained predictors based only on the chemical composition of the materials. They demonstrate good performance and strong predictive power, with learned predictors offering insights into the mechanisms behind superconductivity in different families. The models can be combined into a single pipeline and employed to search for potential new superconductors. Searching the entire Inorganic Crystallographic Structure Database led to the identification of 35 compounds as candidate high-T_c materials. I will also discuss how machine learning can be used to guide and accelerate the experimental process.