(4fd) Accelerating the Design Cycle of Materials for Energy Applications: Harnessing Data to Bridge the Gap between Prototypes and Synthesis

Materials play a pivotal role in achieving energy and sustainability goals. From enabling non-thermal chemical separations (accounting for a potential reduction of 15% in the energy consumption in the US) to storing, transporting and delivering energy vectors as hydrogen (central in the DOE clean energy roadmap), among others. Therefore, the discovery of new materials to enable the deployment of technologies that help to achieve these energy and sustainability goals is a paramount challenge. Traditionally, new materials have been discovered through trial-and-error synthesis of potential structures. However, to synthesize high-performing materials tailored for specific applications, we must navigate a vast design space where the traditional approach becomes impractical. Computational materials discovery offers a promising alternative to explore this complex design space by allowing researchers to generate computer prototype materials and use molecular simulations to evaluate their performance. This approach enables efficient screenings of an enormous number of promising candidates at a fraction of the time and cost required by trial-and-error methods, this approach is known as high-throughput screening (HTPS). However, there is a bottleneck between the generation of prototype materials for specific energy applications and the synthesis of the best candidate.

Harnessing data offers solutions that can help to bridge the gap between the proposition of prototypes and the synthesis of a functional material. Evaluating materials, at the required level of detail, to assess their synthesizability-likelihood using traditional molecular simulation methods is prohibitively expensive, especially in high-throughput scenarios. Every day, vast amounts of experimental and computational data on materials are generated, covering structural descriptions, synthesis protocols, and performance properties. By no means this data is enough; However, if we combine this information, with the generation of databases of prototype materials, virtual hierarchical high-throughput screenings on application-relevant and computationally “cheap” properties, machine learning strategies to predict more complex and computationally expensive properties from existing data, it is possible to create a workflow that allow us not only to identify synthesizable materials, but also to propose “design rules” from structure-properties relationships.

Interested in the potential of metal-organic frameworks (MOFs) to impact energy consumption, I studied their synthesizability likelihood using the estimation of vacuum free energy, and solvation free energies as surrogate synthesizability criteria. To this end, I built a virtual pipeline that enables the high-throughput estimation of solvation free energy for MOFs with four solvents (N-N dimethylformamide, water, methanol and hexane). I studied the impact of functionalization on their synthesizability likelihood and the role of solvent and functionalization on the polymorph selection. Moreover, and motivated by the implementation of machine learning strategies to efficiently and inexpensively predict application-relevant properties in MOFs, I developed and tested physics-informed two-dimensional histograms as MOF representations for adsorption prediction. Using this approach, I evaluated the efficiency of this representation on multiple machine learning strategies to predict the adsorption of multiple molecules. This effort is a fundamental part of the development of a universal adsorption machine learning model that potentially will speed-up the development of structure-properties relationships for the rational design of molecule specific adsorbents.

Research Interests

My research vision is to use a combination of theory, molecular simulation, and machine learning to accelerate the discovery of tailored-made materials for energy applications. In the near future, I want to include autonomous experimentation as part of this materials discovery framework.

To this end I’ve developed knowledge in areas such as:

• Structure-Properties relationships in materials
• High-throughput molecular simulation
• Machine learning strategies
• Porous Materials / Surface Science

And in the near future, I am interested in developing expertise in:

• Generative models for prototype materials database creation
• Large Language Models for synthesis protocol prediction
• Machine learning potentials for fast calculation of complex properties in materials.