(479a) Formulating a General Optimization-Based Materials Design Problem

To allow materials to rapidly meet certain specifications and to develop materials with enhanced features compared to traditional materials, new ways of development and manufacturing of materials using molecular-level analysis may be required [1]. Materials development has traditionally been strongly dependent on engineering experience and experimentation, which has the potential to be costly [2]. Computational techniques have been of interest for understanding and creating desired properties of matter as they enable a wide variety of assessments of materials in silico. Particularly, materials discovery and design using machine learning have been receiving increasing attention due to improvements in terms of prediction accuracy and time efficiency [3]. In addition, mathematical optimization for materials design has been suggested in the literature [4], in which structure-function relationships are used to predict the properties of the material as functions of the design variables, and these design variables can be optimized to achieve product goals. However, these advances open the question of how to select an optimal material via an optimization problem that is aware of the material model, while reducing the number of assumptions required in the materials models with regard to atoms/molecules or materials structure to enable more comprehensive searches for the optimal materials which meet a given target. An optimization problem that uses a model of materials based on first-principles (e.g., atomic-level considerations) to make comprehensive predictions of material behavior [5] would be expected to accelerate materials design efforts. Nevertheless, determining the best way to formulate a general optimization-based materials design problem that incorporates a molecular-level modeling framework for different modeling techniques, and how to deal with the resulting computation time, constitute open challenges.

In this work, motivated by the above, we explore how the optimization-based materials design problem would be formulated using a variety of molecular-level simulation techniques that allow predictions of material characteristics and look at the advantages and disadvantages of each method (e.g., density functional theory and molecular dynamics). The most appropriate objective functions, decision variables, and constraints for different model types are investigated in the formulation of the optimization-based materials design problem. Furthermore, since computational tractability is critical for the optimization-based materials design problem but can be problematic when using fundamental molecular-level modeling approaches, we explore a number of techniques to attempt to reduce the computation time for solving the conceptualized optimization-based materials design problem, including data-driven modeling approaches, by analyzing their impacts on computation effort and model accuracy. Finally, we use the insights obtained by the molecular-level modeling frameworks and computation time reduction techniques to elucidate a number of challenges that remain to be addressed. These investigations make progress toward our long-term vision of systematic materials development that can incorporate considerations related both to product end use as well as to process design/manufacturability in the materials selection process.

References:

[1] Alberi, K., Nardelli, M.B., Zakutayev, A., Mitas, L., Curtarolo, S., Jain, A., Fornari, M., Marzari, N., Takeuchi, I., Green, M.L. & Kanatzidis, M. (2018). The 2019 materials by design roadmap. Journal of Physics D: Applied Physics, 52(1), p.013001.

[2] Olson, G. B. (2000). Designing a new material world. Science, 288(5468), 993-998.

[3] Liu, Y., Zhao, T., Ju, W., & Shi, S. (2017). Materials discovery and design using machine learning. Journal of Materiomics, 3(3), 159-177.

[4] Hanselman, C. (2019). An Optimization Framework for Nanomaterials Design (Doctoral dissertation, Carnegie Mellon University).

[5] Akimov, A. V., & Prezhdo, O. V. (2015). Large-scale computations in chemistry: a bird’s eye view of a vibrant field. Chemical reviews, 115(12), 5797-5890.