Abstract
Unraveling the atomistic and the electronic structure of solid-liquid interfaces is the key to the design of new materials for many important applications, from heterogeneous catalysis to battery technology. Density functional theory (DFT) calculations can, in principle, provide a reliable description of such interfaces, but the high computational costs severely restrict the accessible time and length scales. Here, we report machine learning-driven simulations of various interfaces between water and lithium manganese oxide ($Li_xMn_2O_4$), an important electrode material in lithium ion batteries and a catalyst for the oxygen evolution reaction. We employ a high-dimensional neural network potential to compute the energies and forces several orders of magnitude faster than DFT without loss in accuracy. In addition, a high-dimensional neural network for spin prediction is utilized to analyze the electronic structure of the manganese ions. Combining these methods, a series of interfaces is investigated by large-scale molecular dynamics. The simulations allow us to gain insights into a variety of properties, such as the dissociation of water molecules, proton transfer processes, and hydrogen bonds, as well as the geometric and electronic structure of the solid surfaces, including the manganese oxidation state distribution, Jahn-Teller distortions, and electron hopping.
Introduction: The Challenge of the Battery Frontier
Understanding the boundary between a solid electrode and a liquid electrolyte is a primary goal in modern energy research. These interfaces dictate the efficiency of lithium-ion batteries and the performance of catalysts for the Oxygen Evolution Reaction (OER). However, modeling these regions with quantum-level precision is computationally prohibitive. While Density Functional Theory (DFT) offers the necessary accuracy, it is limited to small systems and short timescales, often missing critical rare events like proton transfer or long-term structural equilibration.
To overcome these limitations, researchers have developed a machine learning framework that bridges the gap between quantum accuracy and large-scale simulation. By utilizing high-dimensional neural networks, they have successfully modeled the $Li_xMn_2O_4$–water interface, revealing complex dynamics that were previously inaccessible.
Methodology: Machine Learning Framework
The study employs two distinct machine learning architectures to achieve a comprehensive description of the interface:
- High-Dimensional Neural Network Potential (HDNNP): This model predicts total system energy and atomic forces by summing local atomic contributions. It utilizes Atom-Centered Symmetry Functions (ACSFs) to describe local environments, enabling the simulation of chemical reactions such as bond breaking and formation with DFT-level accuracy.
- High-Dimensional Neural Network for Spin (HDNNS): A specialized network designed to predict the oxidation states of Manganese ions ($Mn^{II}$, $Mn^{III}$, $Mn^{IV}$) based on local geometric distortions. This allows for the tracking of electronic states, which are critical for understanding material stability and catalytic activity.
The models were trained on a massive dataset comprising over 37,000 structures, including bulk $Li_xMn_2O_4$, bulk water, and various interface configurations calculated using the PBE0r functional. This robust training set allows the Molecular Dynamics (MD) simulations to scale to systems of approximately 6,000 atoms over nanosecond timescales.
Interfacial Dynamics and Electronic Structure
The simulations revealed several key insights into the behavior of the $Li_xMn_2O_4$ surface in aqueous environments:
- Spontaneous Reactivity and Proton Highways: Water molecules were observed to spontaneously dissociate on most surfaces, forming dense hydroxide ($OH^-$) layers. On {110} surfaces, the researchers identified 2D zigzag networks that facilitate rapid proton hopping, suggesting the surface structure acts as a template for transport.
- Surface Polarization: The {100} surface terminated with Lithium exhibits positive polarization, which prevents the accumulation of $OH^-$ ions, resulting in a uniquely "dry" and less reactive interface compared to other orientations.
- The $Mn^{II}$ Distribution: A critical finding for battery longevity concerns "capacity fading" caused by $Mn^{II}$ dissolution. The simulations showed that $Mn^{II}$ ions are extremely rare and localized strictly at the topmost surface layer, providing a structural explanation for why dissolution is a surface-limited process.
Key Results and Accuracy
The HDNNP demonstrated exceptional fidelity, reproducing DFT energies with a mean error of only ~2.4 meV/atom and predicting Mn spin states with 99% accuracy. MD results confirmed that oxygen atoms at the interface remain bound for 31–37 ns—significantly longer than in bulk water (~9 ns)—indicating the formation of a rigid, strongly bound interfacial layer. Furthermore, the model successfully tracked electron hopping between Mn ions, validating the physical consistency of the charge distribution over time.
Conclusion and Future Outlook
While this work represents a significant advancement in simulating complex electrochemical interfaces, several avenues for future research remain. Incorporating nuclear quantum effects could further refine proton transfer rates, and extending simulation timescales might eventually allow for the direct observation of Manganese dissolution. Nevertheless, the combination of HDNNPs and HDNNSs provides a powerful toolkit for the rational design of next-generation battery materials and catalysts.
Summary updated