The Kitchin Research Group

In this work, we tackled the challenge of accelerating catalyst discovery by focusing on transition state energy calculations. We show that a graph neural network potential, despite being trained on a different task, could accurately predict transition states—a crucial step in catalyst discovery—with a remarkable 28x speedup over traditional methods. To provide a benchmark for machine learning model performance in this area, we also curated the Open Catalyst 2020 Nudged Elastic Band (OC20NEB) dataset, which includes 932 DFT nudged elastic band calculations. To showcase the effectiveness of our approach, we applied it to two case studies: reaction mechanism search and ammonia synthesis. These demonstrations highlighted the significant potential of machine learning to enhance and speed up catalyst research.

@article{wander-2025-catts,
  author =       {Brook Wander and Muhammed Shuaibi and John R. Kitchin and
                  Zachary W. Ulissi and C. Lawrence Zitnick},
  title =        {{CatTsunami}: Accelerating Transition State Energy Calculations
                  With Pretrained Graph Neural Networks},
  journal =      {ACS Catalysis},
  volume =       {nil},
  number =       {nil},
  pages =        {5283-5294},
  year =         2025,
  doi =          {10.1021/acscatal.4c04272},
  url =          {http://dx.doi.org/10.1021/acscatal.4c04272},
  DATE_ADDED =   {Mon Mar 17 20:50:26 2025},
}

Copyright (C) 2025 by John Kitchin. See the License for information about copying.

org-mode source

Org-mode version = 9.8-pre

The ability to calculate energy Hessians has long been a cornerstone of understanding chemical reactions, but traditional methods like density functional theory (DFT) are computationally expensive. In this breakthrough study, researchers demonstrate how pretrained Graph Neural Network (GNN) models from the Open Catalyst Project (OCP) can effectively determine potential energy Hessians with remarkable accuracy. With an MAE of just 58 cm⁻¹, these machine-learned potentials enable efficient calculation of Gibbs free energy corrections, overcoming limitations of the harmonic approximation by incorporating translational entropy effects. Even in transition state searches, the models significantly improve convergence rates, making computational catalysis more accessible than ever. This research paves the way for AI-powered simulations to accelerate catalyst discovery and optimize reaction pathways at a fraction of the cost of traditional methods.

@article{wander-2025-acces-numer,
  author =       {Wander, Brook and Musielewicz, Joseph and Cheula, Raffaele and
                  Kitchin, John R.},
  title =        {Accessing Numerical Energy Hessians With Graph Neural Network
                  Potentials and Their Application in Heterogeneous Catalysis},
  journal =      {The Journal of Physical Chemistry C},
  volume =       0,
  number =       0,
  pages =        {null},
  year =         2025,
  doi =          {10.1021/acs.jpcc.4c07477},
  URL =          {https://doi.org/10.1021/acs.jpcc.4c07477},
  eprint =       { https://doi.org/10.1021/acs.jpcc.4c07477 },
}

Copyright (C) 2025 by John Kitchin. See the License for information about copying.

org-mode source

Org-mode version = 9.8-pre

Chemical engineering has undergone multiple modeling revolutions—from empirical correlations and manual analytical methods to computational techniques and, more recently, machine learning (ML). This article argues that we are now on the verge of a new era driven by differentiable programming, which merges domain knowledge with data-driven models in ways that can transform research, education, and industry.

Differentiable programming, powered by automatic differentiation (AD), allows models to compute derivatives efficiently, making optimization, uncertainty quantification, and hybrid physics-informed ML models more accessible. This capability enables engineers to integrate physical principles with data-driven methods, moving beyond “black-box” ML models. The article highlights the role of AD in process modeling, molecular simulations, and control systems, emphasizing its potential to refine chemical engineering calculations beyond traditional regression methods.

However, leveraging these tools requires a workforce skilled in ML, data science, and automation. The paper calls for educational reform, urging chemical engineering programs to integrate these technologies into curricula to prepare future engineers for this computational revolution.

@article{kitchin-2025-beyon-fourt,
  author =       {John R. Kitchin and Victor Alves and Carl D. Laird},
  title =        {Beyond the Fourth Paradigm of Modeling in Chemical
                  Engineering},
  journal =      {Nature Chemical Engineering},
  volume =       {nil},
  number =       {nil},
  pages =        {nil},
  year =         2025,
  doi =          {10.1038/s44286-024-00170-x},
  url =          {http://dx.doi.org/10.1038/s44286-024-00170-x},
  DATE_ADDED =   {Tue Jan 28 15:30:08 2025},
}

Copyright (C) 2025 by John Kitchin. See the License for information about copying.

org-mode source

Org-mode version = 9.8-pre

The future of hydrogen infrastructure demands seamless integration of technology performance, operating conditions, and system configuration. This paper introduces a Systems-to-Atoms (S2A) framework, bridging material, device, and system design to optimize hydrogen storage and transport. A case study on using liquid organic hydrogen carriers (LOHCs) for refueling stations demonstrates that high reaction pressures, while not enhancing catalyst performance, can significantly reduce system costs. Interestingly, less efficient catalysts like copper may be preferred over palladium in certain conditions due to cost and supply chain considerations. The study highlights the importance of cross-scale optimization for resilient, cost-effective hydrogen solutions.

@article{yuan-2025-integ-to,
  author =       {Mengyao Yuan and Giovanna Bucci and Tanusree Chatterjee and
                  Shyam Deo and John R. Kitchin and Carl D. Laird and Wenqin Li
                  and Thomas Moore and Corey Myers and Wenyu Sun and Ethan M.
                  Sunshine and Bo-Xun Wang and Matthew J. McNenly and Sneha A.
                  Akhade},
  title =        {Integrated <i>systems-To-Atoms</i> (S2A) Framework for
                  Designing Resilient and Efficient Hydrogen Infrastructure
                  Solutions},
  journal =      {Energy \& Fuels},
  volume =       {nil},
  number =       {nil},
  pages =        {nil},
  year =         2025,
  doi =          {10.1021/acs.energyfuels.4c05903},
  url =          {http://dx.doi.org/10.1021/acs.energyfuels.4c05903},
  DATE_ADDED =   {Mon Feb 3 15:30:53 2025},
}

Copyright (C) 2025 by John Kitchin. See the License for information about copying.

org-mode source

Org-mode version = 9.8-pre

In a push to optimize hydrogen storage and release, this study explores the multiscale optimization of formic acid dehydrogenation using linear model decision tree (LMDT) surrogates. This method tackles the challenge of integrating models that operate at different scales—ranging from atomistic catalyst design to reactor operation and techno-economic assessment. By replacing complex physics-based models with data-driven surrogates, the researchers created a framework that enables simultaneous optimization across these scales. The study finds that co-optimizing all three scales together leads to a 40% cost reduction compared to optimizing each part separately. The model also confirms economies of scale in hydrogen production, showing that optimal catalyst choice shifts from platinum to copper as production capacity increases. While the approach simplifies multiscale optimization, its accuracy depends on high-quality training data, and future work may integrate uncertainty quantification to improve reliability. This breakthrough highlights the power of surrogate models in chemical process optimization, paving the way for more cost-effective hydrogen energy solutions.

Ethan M. Sunshine, Giovanna Bucci, Tanusree Chatterjeed, Shyam Deod, Victoria M. Ehlinger, Wenqin Lib, Thomas Moore, Corey Myers, Wenyu Sun, Bo-Xun Wang, Mengyao Yuan, John R. Kitchin, Carl D. Laird, Matthew J. McNenly, Sneha A. Akhade, Multiscale Optimization of Formic Acid Dehydrogenation Process via Linear Model Decision Tree Surrogates, Computers and Chemical Engineering, 194, 108921 (2024) https://doi.org/10.1016/j.compchemeng.2024.108921.

@article{sunshine-2024-multis-optim,
  author =       {Ethan M. Sunshine and Giovanna Bucci and Tanusree Chatterjee
                  and Shyam Deo and Victoria M. Ehlinger and Wenqin Li and
                  Thomas Moore and Corey Myers and Wenyu Sun and Bo-Xun Wang and
                  Mengyao Yuan and John R. Kitchin and Carl D. Laird and Matthew
                  J. McNenly and Sneha A. Akhade},
  title =        {Multiscale Optimization of Formic Acid Dehydrogenation Process
                  Via Linear Model Decision Tree Surrogates},
  journal =      {Computers & Chemical Engineering},
  volume =       {194},
  pages =        108921,
  year =         2024,
  doi =          {10.1016/j.compchemeng.2024.108921},
  url =          {http://dx.doi.org/10.1016/j.compchemeng.2024.108921},
  DATE_ADDED =   {Mon Nov 25 14:10:36 2024},
}

Copyright (C) 2025 by John Kitchin. See the License for information about copying.

org-mode source

Org-mode version = 9.8-pre