Development and application of new electronic structure methods and atomistic simulation tools in the broad area of catalysis

Leveraging community knowledge in machine learning models to accelerate the discovery of new catalysts and materials

Heather J. Kulik

Associate Professor, Departments of Chemical Engineering and Chemistry, Massachusetts Institute of Technology, Cambridge, MA 02139, USA

I will discuss our efforts to use machine learning (ML) to accelerate the computational tailoring and design of metal-organic framework (MOF) materials for catalysis. One limitation in a challenging materials space such as open shell transition metal chemistry present in the open metal sites of most catalytically active MOFs is that ML models and ML-accelerated high-throughput screening traditionally rely on density functional theory (DFT) for data generation, but DFT is both computationally demanding and prone to errors that limit its accuracy in predicting new MOFs. I will describe how we have curated a dataset of thousands of MOFs that have been experimentally synthesized and used this data to train ML models to predict experimentally reported measures of stability.1 These models predict experimental thermal stability and activation stability, which would be extremely difficult to predict using computational modeling. I will describe how we have leveraged these models to then screen for stable catalysts in the direct conversion of methane to methanol.2 I will also describe how we have used these models to accelerate the discovery of novel stable MOFs, creating a dataset of transition metal complexes enriched with stability and diversity 1-2 orders of magnitude beyond what is typically included in most hypothetical MOF datasets.3
In the second half of my talk, I will describe how we are overcoming limits of conventional DFT modeling. Machine learning (ML)-accelerated discovery of transition metal containing materials such as light-harvesting chromophores, phosphors, and other photoactive complexes holds great promise. Nevertheless, the open shell d electrons that impart many of the desirable properties to these systems also make them notoriously challenging to study with DFT. Thus, when ML is used to accelerate computational screening, it often inherits the biases of the underlying method used to generate training data. I will describe our recent efforts to overcome these limits through three complementary approaches. First, I will describe how we have developed machine learning models trained directly on experimental reports of iridium phosphors, leading to the development of ML models that can predict experimental emission energies and lifetimes with superior or equivalent performance to conventional methods such as time-dependent DFT but in a fraction of the computational time.4 Next, I will describe how we overcome limits of DFT uncertainty in screening for light harvesting chromophores with earth abundant 3d metals by incorporating method insensitivity into a multi-objective optimization workflow, requiring a consensus of functionals to agree on a property in order for a material to be selected as optimal.5 These workflows accelerate materials discovery by at least 1000-fold. Finally, I will describe our development of a density functional "recommender" that can identify which DFT functional is most predictive to obtain accurate vertical spin excitation energies in transition metal complexes.6

(1) Nandy, A.; Duan, C.; Kulik, H. J. Using Machine Learning and Data Mining to Leverage Community Knowledge for
the Engineering of Stable Metal-Organic Frameworks. Journal of the American Chemical Society 2021, 143, 17535–17547.

(2) Adamji, H.; Nandy, A.; Kevlishvili, I.; Roman-Leshkov, Y.; Kulik, H. J. Computational Discovery of Stable Metal-
Organic Frameworks for Methane-to-Methanol Catalysis. Journal of the American Chemical Society 2023, 145, 14365-14378.

(3) Nandy, A.; Yue, S.; Oh, C.; Duan, C.; Terrones, G. G.; Chung, Y. G.; Kulik, H. J. A Database of Ultrastable Mofs
Reassembled from Stable Fragments with Machine Learning Models. Matter 2023, 6, 1585-1603.

(4) Terrones, G. G.; Duan, C.; Nandy, A.; Kulik, H. J. Low-Cost Machine Learning Prediction of Excited State Properties
of Transition Metal Phosphors. Chem. Sci. 2023, 14, 1419-1433.

(5) Duan, C.; Nandy, A.; Terrones, G.; Kastner, D. W.; Kulik, H. J. Active Learning Exploration of Transition Metal
Complexes to Discover Method-Insensitive and Synthetically Accessible Chromophores. JACS Au 2023, 3, 391-401.

(6) Duan, C.; Nandy, A.; Meyer, R.; Arunachalam, N.; Kulik, H. J. A Transferable Recommender Approach for Selecting
the Best Density Functional Approximations in Chemical Discovery. Nature Computational Science 2023, 3, 38-47.