Savoie Research Group

Team Philosophy

Our research is organized around the problems that you will find below. Our expertise lies in using computational methods to design and characterize organic molecules, materials, and reactions. A small team of students and postdocs works on each problem and the problems cycle over time as we generate new ideas, perceive a need, or conclude our studies. We use an array of methodologies and actively extend or invent new approaches as required by each problem. Combined with some of the best graduate students and postdocs on the planet, these ingredients create an extremely productive, cross-cutting, and creative research atmosphere. If you are interested in any of these problems, would like to collaborate with the group, or are interested in joining the group, please reach out to Brett.

All the images on this page have been generated by the marvelous DALL-E text-to-image generation model. The associated prompts are shown the description of each problem.

Predicting How Materials Degrade

Simulations play a crucial role in down-selecting and optimizing molecules and materials before more costly synthesis. However, the lacuna in the materials by design paradigm is that we still can’t predict the stability and degradation behaviors of otherwise promising materials. This results in a costly bottleneck where prospective materials must be empirically tested in make and break cycles and ultimately discarded at a late development stage if they do not meet requirements. Our group is developing computational methods that fill this gap to predict the degradation pathways, timescales, and products of prospective molecules and materials to various stressors.

Collaborators: Jianguo Mei (Purdue), Julia Laskin (Purdue), Olexandr Isayev (CMU), Geoffrey Hutchison (U. Pitt.)

DALL-E Prompt: A Hieronymus Bosch style painting of molecules being chased off a cliff

Post Li-Ion Battery Electrolytes

Every battery consists of two electrodes and an electrolyte. Out of these components, the electrolyte is generally the limiting factor in terms of performance, charging time, battery life, and safety. The success of conventional Li-ion batteries is built on top of decades of electrolyte engineering that is non-transferable to other battery chemistries. Thus, there is a major need to develop optimized electrolytes for prospective battery chemistries (Li-metal, fluoride, sodium, etc.) without waiting on decades of optimization. Our approach to this problem is developing new computational methods that will enable design rules elucidation and screening to occur on an accelerated timescale. Among these are methods to perform the first molecular simulations of the electrochemical processes occurring at prospective electrolyte/electrode interfaces and methods for predicting electrolyte degradation networks.

Collaborators: Vilas Pol (Purdue)

DALL-E Prompt: An expressive oil painting of molecules chained together on a plane with lightning above.

Molecular Passivation of Perovskites

Perovskite solar cells are potentially revolutionary in terms of cost and efficiency but are limited by their relative instability. In collaboration with the Dou group, we are developing a generic passivation strategy that involves depositing a molecularly-thin organic layer on the perovskite surface that preserves its outstanding optoelectronic properties. Our unique approach to this problem involves high-throughput data generation to create the largest database simulated 2D-perovskite layers in existence, developing ML-based screening tools for prospective passivation chemistries, and using simulations to assay halide and moisture diffusion associated with perovskite degradation. These strategies are the basis for our ongoing translational work on providing perovskite solar cells that achieve a >20 year lifespan.

Collaborators: Letian Dou (Purdue)

DALL-E Prompt: Futuristic devices being powered by the sun, digital art.

Developing Plastics with Tailored Degradation Profiles

The accumulation of plastics in the environment is the result of a mismatch between the lifetime and use time of these materials. To address this issue, we are developing new plastics that depolymerize on command while maintaining the mechanical properties that have made commodity plastics so useful. Our unique approach for this is to introduce susceptibilities into the polymer backbone that make depolymerization facile under tailored conditions. In collaboration with the Dou group, we have developed several novel depolymerization strategies that are unique for being compatible with plastics based on C-C backbones. Using a combination of these strategies we envision developing drop-in replacements for commodity polyolefins that exhibit (i) better alignment between environmental half-life and application need, (ii) facile recyclability from mixed plastics streams, and (iii) precursor sourcing from existing waste streams.

Collaborators: Letian Dou (Purdue)

DALL-E Prompt: A painting of plastics recycling in the mannerist style

Mixed Conduction in Soft Materials

Organic and hybrid organic/inorganic materials that exhibit both electronic and ionic conduction have disruptive potential for sensing and wearable electronics. Our group's work in this area is focused on designing new mixed conductors that exhibit the necessary stability and transference characteristics for end-use wearable applications. To date, extremely few materials have been developed for these applications due to the synthetic cost and incomplete understanding of how electrolyte media affect material morphology. To address this knowledge gap we are developing coarse-grained simulation models capable of characterizing mixed conduction in silico. In turn, this simulation data is forming the basis of novel generative machine learning models to identify new mixing conducting materials.

Collaborators: Bryan Boudouris (Purdue), Jianguo Mei (Purdue)

DALL-E Prompt: polymers in space, digital art

The Reaction Prediction Problem

The reaction prediction problem can be phrased as A→?, where the task is predicting the products knowing only the reactants and conditions. The state-of-the-art for discovering new reactions is slow, hypothesis-driven, experimental testing by organic chemists. Solving this problem would make in silico reaction discovery possible and lead to a paradigm shift where novel reactions are discovered algorithmically then verified experimentally. Our approach for addressing this is to recognize that the A→? problem can be rephrased as a search over A→B problems. Provided that the latter can be solved in parallel and at low cost then reinforcement learning can be used to learn an optimal policy for generating useful A→B subproblems. These strategies are the basis for our ongoing efforts to provide a generic solution to the reaction prediction problem.

Collaborators: Olexandr Isayev (CMU), Geoffrey Hutchison (U. Pitt.)

DALL-E Prompt: crayon drawing of a large network with molecules on dark paper

Machine Learning for Chemical Deduction Problems

Deduction is needed in underdetermined prediction scenarios or when there is a mixture of information sources. In contrast, the vast majority of ML in chemistry is inductive, learning from example, with a fixed set of features. Deductive bottlenecks face vast majority of proposals to automate chemical development workflows. For example, materials aging, soft materials property prediction, formulation development, and product identification all present information limitations that are typically filled by domain experts performing manual interpretation. Deductive ML models would fill such gaps and fundamentally expand the scope of ML in chemistry and materials research. Our approach for developing these models consists of combining task-specific inductive ML models within deductive super-networks and developing distinctive training protocols for deductive learning.

Savoie Research Group |

Purdue University

Research

Team Philosophy

All the images on this page have been generated by the marvelous DALL-E text-to-image generation model. The associated prompts are shown the description of each problem.