MAE Seminar – Data-driven Discovery of Stable Surfaces with Tailored Properties via Equivariant Graph Neural N

Details

Dear Undergraduate and Graduate Students, Faculty, and Staff,
You are invited! UF Department of Mechanical and Aerospace Engineering’s Seminar Series
This is a perfect opportunity to enjoy some coffee, cookies, and captivating talks! These sessions feature amazing guest speakers, from academic trailblazers and industry movers to our very own faculty candidates showing off their expertise and fresh perspectives.
Come for the treats, stay for the engaging discussions, and connect with fellow MAE enthusiasts. Everyone is welcome!

Data-driven Discovery of Stable Surfaces with Tailored Properties via Equivariant Graph Neural Networks and Foundational Interatomic Potentials

November 13, 2025, at 12:50pm
Location: MAE-A 303

Dr. Peter Schindler
Assistant Professor at Northeastern University

Abstract
Accurately assessing the properties of materials’ surfaces and their stability is critical for diverse applications such as heterogeneous catalysis, electron emission technologies, and interface engineering in semiconductors and batteries. The stability of a surface with a particular Miller index and termination is governed by its cleavage or surface energy. This property also governs the Wulff construction, which defines the equilibrium shapes of nanoparticles. Another crucial surface property is the work function (i.e., the energy required to extract an electron from the surface) that determines the contact barrier at interfaces. While first-principles calculations provide reliable predictions of these surface properties, their computational cost severely limits the exploration of the vast space of possible surfaces.

Equivariant graph neural networks that enforce symmetry of three-dimensional Euclidean space (E3GNNs) have demonstrated accurate predictions of structure-property relationships of materials and molecules. Instead of relying on invariant feature engineering, E3GNNs incorporate physical symmetries directly into the neural network layers. The features of equivariant networks are thus more expressive and better able to capture local symmetries, promising greater transferability and training data efficiency.

Another recent advance is the emergence of universal machine learning interatomic potentials (uMLIPs, or recently termed “foundational interatomic potentials”) that promise applicability across the entire periodic table and structural types. They enable rapid zero-shot predictions of materials and molecular properties that can be derived from the potential energy surface or its derivative, with little to no need for additional training.

In this talk, I will discuss (1) a novel E3GNN architecture that incorporates symmetry-breaking along the surface normal, for accurately predicting both the stability and work function of surfaces, and (2) a benchmarking study of cleavage energy predictions for a wide range of state-of-the-art uMLIPs. Finally, I will demonstrate the capability of these approaches to speed up screening for stable surfaces with tailored properties by orders of magnitude.

Biography
Dr. Peter Schindler is an Assistant Professor at Northeastern University leading the Data-Driven Renewables Research (D2R2) group that seeks to discover novel materials for renewable energy applications using high-throughput, quantum chemistry calculations, and data-driven materials property predictions. He is a scientific advisory board member (and former senior scientist) at Aionics Inc., a startup that has become a recognized leader in the field of battery informatics. His work has been featured on the covers of Advanced Materials, ACS Energy Letters, and Digital Discovery. During his postdoctoral research at the University of Vienna and Stanford University, his research established renewable energy materials both computationally and experimentally, for which he was awarded the Erwin-Schrödinger fellowship by the Austrian Science Fund (FWF). He received his Ph.D. in physics from the University of Vienna, working on thin-film semiconductor synthesis and characterization. His expertise in both experimental synthesis and computational materials simulations enables him to carry out cutting-edge research at the intersection of the two fields.

MAE Faculty Host: Dr. Jaeyun Moon