Experiment-driven atomistic materials modeling: A case study combining X-ray photoelectron spectroscopy and machine learning potentials to infer the structure of oxygen-rich amorphous carbon (2402.03219v2)

Abstract: An important yet challenging aspect of atomistic materials modeling is reconciling experimental and computational results. Conventional approaches involve generating numerous configurations through molecular dynamics or Monte Carlo structure optimization and selecting the one with the closest match to experiment. However, this inefficient process is not guaranteed to succeed. We introduce a general method to combine atomistic ML with experimental observables that produces atomistic structures compatible with experiment by design. We use this approach in combination with grand-canonical Monte Carlo within a modified Hamiltonian formalism, to generate configurations that agree with experimental data and are chemically sound (low in energy). We apply our approach to understand the atomistic structure of oxygenated amorphous carbon (a-CO${x}$), an intriguing carbon-based material, to answer the question of how much oxygen can be added to carbon before it fully decomposes into CO and CO$_2$. Utilizing an ML-based X-ray photoelectron spectroscopy (XPS) model trained from $GW$ and density functional theory (DFT) data, in conjunction with an ML interatomic potential, we identify a-CO${x}$ structures compliant with experimental XPS predictions that are also energetically favorable with respect to DFT. Employing a network analysis, we accurately deconvolve the XPS spectrum into motif contributions, both revealing the inaccuracies inherent to experimental XPS interpretation and granting us atomistic insight into the structure of a-CO$_{x}$. This method generalizes to multiple experimental observables and allows for the elucidation of the atomistic structure of materials directly from experimental data, thereby enabling experiment-driven materials modeling with a degree of realism previously out of reach.

• The paper introduces a novel framework that integrates experimental XPS with ML potentials to predict atomic structures of oxygen-rich amorphous carbon.
• It employs modified GCMC simulations with ML components trained on DFT/GW data to achieve precise spectral matching and realistic oxygen-to-carbon ratios.
• Compared to traditional melt-quench methods, the approach yields configurations that closely replicate experimental XPS spectra and inform improved materials design.

Integration of XPS and Machine Learning Potentials in Amorphous Carbon Modeling

This paper presents a methodological advancement in atomistic materials modeling, emphasizing the integration of machine learning potentials with experimental data to infer atomic structures, particularly focusing on oxygen-rich amorphous carbon (a-COx). The authors propose a novel framework that leverages both X-ray photoelectron spectroscopy (XPS) and ML, aiming to address the fundamental question of structural responses of amorphous carbon to varying oxygen content, which poses significant implications in materials science.

Methodology

The authors introduce a comprehensive method combining computational and experimental insights. This integration employs grand-canonical Monte Carlo (GCMC) simulations augmented by ML potentials. The ML components are trained using a synergy of density functional theory (DFT) and GW data, crafting interatomic potentials that enable an efficient exploration of potential energy surfaces (PES) while maintaining computational feasibility.

This "experiment-driven atomistic materials modeling" framework essentially serves a dual purpose: generating low-energy configurations and matching these with experimental XPS data through structural optimization. The paper utilizes a modified Hamiltonian formalism to scrutinize structures within a chemical context, favoring stability and experimental relevance.

Strong Numerical Results and Claims

The numerical results affirm that this hybrid approach successfully reproduces the XPS characteristics of amorphous carbon better than purely computational models. For low- and high-oxygen content a-COx configurations, the predicted structures and their XPS spectra converge significantly with experimental observations, which is a testament to the efficacy of the ML-augmented simulations.

Comparative Analysis with Traditional Methods

The paper critically compares this approach with traditional melt-quench methods, which often result in structures failing to reflect accurate chemical environments due to disproportionate formation of thermodynamically stable by-products like CO2. In contrast, the modified GCMC strategy yields structures with comprehensive XPS alignment and realistic oxygen-to-carbon ratios.

Implications and Future Directions

Practically, this methodology holds potential for simulating deposition processes in material synthesis, offering a pathway to model and understand materials at a granularity previously constrained by computational resources. Theoretically, it paves the way for enhanced ML applications in computational chemistry, transcending the predictive limitations of classical potentials.

The results encourage further exploration of this integrated modeling framework, suggesting its applicability beyond XPS to other spectroscopic techniques. Such expansions could streamline experimental validation processes across a variety of material systems and applications, notably in electronics and energy storage domains.

Conclusion

This paper underscores a pivotal step towards bridging computational predictions with experimental realities in materials science. By marrying the precision of machine-learned potentials with the robustness of experimental data, the authors introduce a paradigm capable of providing both fundamental insights and practical modeling solutions. Future work might extend this integration to consider more diverse environmental conditions and other amorphous systems, thereby enhancing predictive reliability and material design.