Abstract
Quantum chemical calculations on atomistic systems have evolved into a standard approach to study molecular matter. These calculations often involve a significant amount of manual input and expertise although most of this effort could be automated, which would alleviate the need for expertise in software and hardware accessibility. Here, we present the AutoRXN workflow, an automated workflow for exploratory high-throughput electronic structure calculations of molecular systems, in which (i) density functional theory methods are exploited to deliver minimum and transition-state structures and corresponding energies and properties, (ii) coupled cluster calculations are then launched for optimized structures to provide more accurate energy and property estimates, and (iii) multi-reference diagnostics are evaluated to back check the coupled cluster results and subject them to automated multi-configurational calculations for potential multi-configurational cases. All calculations are carried out in a cloud environment and support massive computational campaigns. Key features of all components of the AutoRXN workflow are autonomy, stability, and minimum operator interference. We highlight the AutoRXN workflow at the example of an autonomous reaction mechanism exploration of the mode of action of a homogeneous catalyst for the asymmetric reduction of ketones.
Introduction
Elucidating chemical reaction mechanisms is a cornerstone of modern chemistry, yet it remains a labor-intensive process. Traditionally, researchers rely on intuition to propose pathways and manually configure individual simulations to test their feasibility. This "expert-in-the-loop" model is not only slow but also prone to human bias, often overlooking critical side reactions or decomposition pathways. Furthermore, while Density Functional Theory (DFT) provides a computationally accessible starting point, its accuracy is often insufficient for high-energy transition states. More precise methods, such as Coupled Cluster (CC) or multi-reference (MR) theories, require significant expertise to set up and massive computational resources to execute.
The AutoRXN workflow addresses these bottlenecks by automating the decision-making process for high-level quantum chemistry and leveraging cloud infrastructure to handle high-throughput screening.
The AutoRXN Workflow
The AutoRXN framework operates as a three-stage automated pipeline designed for autonomy and stability.
- Reaction Exploration: Utilizing the
SCINE Chemotonframework, the system autonomously explores potential reaction pathways by simulating molecular interactions in various orientations. Initial structures and transition states are identified using DFT (PBE-D3BJ), constructing a comprehensive reaction network graph. - Coupled-Cluster Refinement: To improve energy estimates, the workflow automatically triggers
CCSD(T)calculations—the "gold standard" of single-reference quantum chemistry—for all identified structures. This stage is managed byAiiDA, ensuring robust job scheduling and data provenance. - Automated Multi-reference Validation: For complex systems, particularly those involving transition metals, single-reference methods may fail. AutoRXN employs
autoCASto detect these cases using orbital entanglement entropies. If a system is flagged, the workflow automatically executes multi-configurational calculations (CASPT2) to validate the results.
The entire stack is deployed on Microsoft Azure, utilizing specialized hardware and low-priority "Spot" VMs to maximize cost-efficiency.
Case Study: Iron-Catalyzed Ketone Reduction
The workflow was demonstrated on the asymmetric transfer hydrogenation of ketones using a homogeneous iron catalyst. This system represents a significant challenge due to the presence of a transition metal and the potential for numerous competing pathways.
Starting from four initial compounds, the AutoRXN workflow generated a reaction network comprising 540 reactions and 2,227 elementary steps. This massive exploration identified hundreds of side reactions and decomposition paths that would likely be missed in a manual study.
Key Insights and Results
A critical finding of the study was the discrepancy between DFT and Coupled Cluster results. While DFT performed reasonably well for final reaction energies, it systematically underestimated reaction barriers by an average of 34.4 kJ/mol. Such errors can lead to incorrect conclusions regarding rate-limiting steps and overall catalytic efficiency.
Furthermore, the study highlighted the limitations of standard diagnostics. The commonly used T1 diagnostic for Coupled Cluster reliability was found to be unreliable, whereas orbital entropy-based diagnostics ($Z_{s(1)}$) provided a much clearer indication of when multi-reference methods were necessary.
From an infrastructure perspective, the workflow achieved a 10x reduction in cost and a 30x speedup in runtime by utilizing optimized software (NWChemEx) on low-priority cloud VMs. This demonstrates that high-level ab initio calculations can be feasibly integrated into high-throughput pipelines.
Conclusion and Future Outlook
The AutoRXN workflow represents a significant step toward fully autonomous chemical discovery. By removing the expertise and infrastructure barriers to high-level quantum chemistry, it enables the mapping of complex reaction spaces with unprecedented confidence. Future developments will focus on integrating solvation models to better simulate real-world conditions and refining conformer search strategies to ensure no critical reaction paths are missed.
Summary updated