Statistical Thermodynamics & Molecular Simulations (STMS) Seminar Series
These seminar series are aimed at providing a virtual platform for sharing scientific research in the area of statistical mechanics, molecular simulations, and computational materials science. In recent months, the coronavirus pandemic has stopped all large in-person scientific gatherings, including conferences and department seminars, and it is not clear that the situation will improve any time soon. STMS is aimed at filling this gap, and provide a venue for dissemination of research findings and exchange of ideas in the age of COVID. This model is being currently used by several other scientific communities, and can potentially continue even beyond the pandemic if successful.
Each seminar will be a 60-minute event and will comprise of a long-form (30-minute) talk by a principal investigator or a senior research scientists from academia or industry and a short-form (15-minute) presentation by a graduate student or a postdoc. The remainder of the event will be dedicated to Q&A (10 minutes for the PI, 5 minutes for the student/postdoc). Long-form speakers will be chosen by the STMS Organizing Committee, while we encourage suggestions from the community at large. Student and postdoctoral speakers, however, need to be nominated by their advisors. Seminars will take place on Fridays, from 11 AM-12 PM. During 2021, we expect to hold two seminar per month, at the last two Fridays of each month.This event's talks:
Finding truth in fiction: Discovering physical structure-property relationships from unphysical porous materials
Prof. Christopher Wilmer (University of Pittsburgh)
Abstract: There continues to be tremendous interest in studying and developing new porous materials for gas storage and separations applications, and molecular simulations have played an integral role in guiding these efforts. Traditionally, molecular simulations have been used to help understand adsorption behavior in already synthesized materials, but increasingly they are being used to screen hypothetical materials in advance of their synthesis. Here, we describe a yet new role for molecular simulations: to discover the bounding behavior of gas adsorption in porous materials by screening practically all conceivable materials – a task made easier by ignoring whether materials are physically realizable or not.
Our approach is to randomly generate a set of porous materials, simulate their gas adsorption behavior, and then generate subsequent populations by mutating those materials whose structure-property combinations were rare. We continue iteratively generating and mutating randomly generated materials until all structure-property combinations are equally represented (at which point the method has converged).
The resulting structure-property data provides a clear picture as to what gas adsorption behavior can practically be discovered in the universe of possible porous materials. We anticipate this approach to have an important impact in determining various “best-case” scenarios related to important applications such as hydrogen storage, carbon-dioxide capture, and various industrial separation processes.
Speaker Bio: Chris was born in Canada to Polish parents who immigrated to Canada shortly before Poland went under martial law in 1981. Spurred by nanotechnology-driven visions of the future penned by writers Erik Drexler and Ray Kurzweil, Chris acquired a B.A.Sc. degree from the University of Toronto’s Engineering Science—Nanoengineering program. Coming to the United States to pursue a Ph.D. in Chemical Engineering at Northwestern under the mentorship of Prof. Randall Q. Snurr, he took an interest in the American way of developing new technologies—through entrepreneurship. While still a student, he co-founded, NuMat Technologies, which develops commercial gas storage solutions using MOFs, for which he was named to the Forbes Top 30-Under-30 list in Energy Innovation. At the University of Pittsburgh, he directs the Hypothetical Materials Lab, whose research focuses on advanced uses of porous crystals, such as in developing artificial noses or storing oxygen. His lab recently spun-out Aeronics, which manufactures inexpensive oxygen storage containers for people (and pets!) with decreased lung function. In his ample spare time (that is a joke) he is the managing editor of Ledger, the world's first scholarly journal focused on blockchain research, the technology that powers Bitcoin (that is not a joke).
Hybrid unsupervised-supervised machine learning models for materials science
Dr. Rose Cersonsky (École Polytechnique Fédérale de Lausanne)
Abstract: In recent years, the introduction of machine learning methods into the field of atomistic modeling has greatly accelerated the discovery and simulation of novel materials. Most machine learning algorithms fall into two broad categories: unsupervised and supervised learning. In unsupervised learning, we aim to understand the landscape and variance of the system of interest, particularly as it applies to data clustering and compression. In supervised learning, we aim to correlate given materials to some target quantity, as is the case in property regression. Data analyses based on linear methods constitute the simplest, most robust and transparent approaches to the automatic processing of large amounts of data for building supervised or unsupervised machine learning models. Principal covariates regression (PCovR), a method originating in the psychology field, interpolates between two popular unsupervised and supervised methods, namely principal component analysis and linear regression, and can be used to conveniently reveal structure-property relations in terms of simple-to-interpret, low-dimensional maps. Here I will introduce a kernelized version of PCovR and demonstrate the performance of this approach in revealing and predicting structure-property relations in chemistry and materials science. Additionally, I will demonstrate the improved performance resulting from incorporating PCovR into two popular data selection methodologies: CUR and Farthest Point Sampling, which iteratively identify the most diverse samples and discriminating features. I will conclude by discussing the implementation of these methods for materials informatics and simulation.
Speaker Bio: Rose K. Cersonsky received her Bachelor of Science degree in Materials Science and Engineering from the University of Connecticut in 2014, serving as commencement speaker and earning the School of Engineering’s Award for Outstanding Academic Achievement. She then went on to obtain her Ph.D. in Macromolecular Science and Engineering from the University of Michigan in 2019 under Professor Sharon C. Glotzer, the John Werner Cahn Distinguished University Professor of Engineering, where Rose’s doctoral thesis was titled “Designing Nanoparticles for Self-Assembly of Novel Materials.” Her doctoral work was awarded the ACS Colloids Victor K. LaMer Dissertation Award, as well as University of Michigan distinctions, including the Charles G. Overberger Award for Excellence in Research and the Biointerfaces Institute Student Innovator Award. She is currently working as a postdoctoral researcher in the Laboratory of Computational Science and Modeling (COSMO) at École Polytechnique Fédérale de Lausanne (EPFL) in Lausanne, Switzerland.
In addition to research, Rose has devoted herself to scientific service, leading and coordinating multiple outreach programs at both the University of Connecticut and the University of Michigan, and publishing work focused on community engagement in educational journals. In her spare time, she enjoys running, rock climbing, and singing.