Aug 2026· Journal of Physical Chemistry A· Vol 130 35, pp.
7020-7026
· 0 citations· 65 references
Medicine
TL;DR
A global active learning framework is demonstrated to map these landscapes efficiently by coupling genetic algorithms with deep neural networks trained on density functional theory data, which provides a scalable and resource-efficient strategy for high-throughput materials discovery in applications such as hydrogen storage and ammonia synthesis.
Abstract
Optimizing catalysis requires the efficient exploration of immense chemical spaces, particularly for high-entropy (multielement) systems where properties depend on complex variables like geometry, composition, and site-specific interactions. In this work, we demonstrate a global active learning framework to map these landscapes efficiently. By coupling genetic algorithms with deep neural networks trained on density functional theory data, our approach learns the potential energy surface while optimizing chemical structures simultaneously, bypassing costly density functional theory relaxations. We apply this framework to predict H2, N2, and NH3 adsorption energies on 10-50 atom clusters composed of Ag, Au, Cu, Ni, Pd, and Pt. The model achieves a mean absolute error of less than 0.10 eV against density functional theory validation. Results identify regions with moderate H2 physisorption (-0.1 to -0.5 eV) and other regions with strong chemisorption (-1.0 to -2.0 eV). Physisorption mainly consists of adsorption on Ag, Au, Ni, and Cu, while chemisorption involves Pd and Pt sites. High-entropy environments can provide a diverse distribution of local chemical conditions. This site variance is critical for multistep reactions where intermediate steps possess conflicting optimal binding energies that cannot be simultaneously satisfied by low-entropy surfaces, thereby offering entropy as a tunable parameter for catalyst design. This framework provides a scalable and resource-efficient strategy for high-throughput materials discovery in applications such as hydrogen storage and ammonia synthesis.
The application of density functional theory to heterogeneous catalysis is hindered by the shortcomings of conventional density functional approximations. We combine machine learning with explicitly non-local physically informed descriptors and introduce an exchange-correlation functional (CIDER26SS) framework regularized for wide transferability. CIDER26SS is size-extensive, highly efficient, provides a balanced and accurate description of both molecular and solid-state systems, and is specifically well-optimized for transition metal surface chemistry. Surpassing existing conventional functionals, CIDER26SS resolves the CO/Pt puzzle, identifying the correct binding site for CO adsorption on the Pt(111) surface, along with an accurate adsorption energy, Pt lattice constant, and surface energy. Predictions agree well with the experimental values, even when all bulk and surface data for Pt are excluded from the training set. Remarkably, CIDER26SS exceeds the accuracy of semilocal approximations even for systems far outside the training domain.
M. S. Abdallah, Zhuotao Jin, Boris Kozinsky et al.· 0 citations
Metal-organic frameworks (MOFs) are promising materials for adsorption and separation, but accurately predicting water adsorption remains a major challenge in molecular simulation. Classical force fields, such as UFF, often fail to capture the strong, directional hydrogen-bonding interactions between water and MOFs, while high-throughput density functional theory (DFT) calculations are computationally prohibitive. Here, we develop a transferable machine-learning potential (MLP) for water adsorption in Al-based MOFs by fine-tuning the pretrained MACE foundation model. Using a data set of 420 Al-MOFs with diverse chemical environments, we show that accurate prediction of water adsorption thermodynamics requires not only an improved description of water-framework interactions but also explicit treatment of framework flexibility. The resulting model reproduces experimental heats of adsorption and Henry coefficients for a series of benchmark Al-MOFs and achieves DFT-level accuracy across more than 400 Al-MOFs. Analysis of MIL-160 establishes practical accuracy requirements of approximately 10 kJ mol-1 in total energies and 40 meV Å-1 in atomic forces for reliable adsorption thermodynamics. Compared with DFT-level predictions, UFF underestimates water adsorption enthalpies by more than 10 kJ mol-1 for 74% of the materials studied. In addition, the MLP identifies lower-energy framework configurations than previously reported DFT-optimized structures, highlighting the importance of enhanced configurational sampling. These results demonstrate that foundation-model-based machine-learning potentials can enable DFT-accurate, high-throughput screening of water adsorption in flexible MOFs.
Yu-Tao Li, Xiaoqi Zhang, Xin Jin et al.· Journal of Chemical Theory a...· 0 citations
Reliable prediction of hydrogen adsorption free energy (ΔGH) is essential for accelerating electrocatalyst discovery for the alkaline hydrogen evolution reaction (HER), yet practical machine-learning workflows remain limited by inconsistent energetic definitions, heterogeneous density functional theory (DFT) protocols, and poor out-of-distribution (OOD) generalization. Here, we present an energetically anchored machine-learning framework that integrates machine-learning interatomic potential (MLIP)-derived energetic descriptors with pretrained Crystal Hamiltonian Graph Neural Network (CHGNet) latent embeddings to predict DFT-defined hydrogen adsorption energetics across chemically diverse catalyst surfaces. The framework employs protocol-consistent single-point MLIP energy evaluation on reference geometries to construct thermodynamically aligned energetic descriptors while combining them with structural representations through gradient-boosting regression. Using curated data sets from Catalysis Hub and AQCat25 containing single-, bi-, and trimetallic adsorption systems, the embedding-augmented energetic model achieved high predictive accuracy (R2 = 0.976, MAE = 0.054 eV, RMSE = 0.105 eV) under site-level data splitting, substantially outperforming embedding-only and physicochemical-descriptor-only models. Shapley additive explanations analysis revealed a hierarchical learning mechanism in which the protocol-consistent energetic descriptor serves as the dominant thermodynamic anchor, whereas structural descriptors provide secondary refinements that improve adsorption-energy discrimination, particularly near the thermoneutral regime relevant to catalyst screening. Explicit evaluation of MLIP-only geometry workflows further demonstrated that the framework retains meaningful adsorption-energy ranking capability despite degradation introduced by MLIP-relaxed geometries. Additional OOD and transfer-learning analyses showed that predictive robustness depends strongly on the balance between compositional diversity and reference-protocol consistency. These results establish energetically anchored MLIP embeddings as an effective strategy for scalable post-DFT adsorption-energy refinement and data-efficient electrocatalyst screening while clarifying the practical limitations of MLIP-driven workflows for heterogeneous catalysis.
Ching-En Lin, Po-Wen Chen, Tien-Hsiang Hsueh· Journal of Chemical Informat...· 0 citations
Adequate characterization of two-dimensional (2D) materials with low energy barriers for impurity adsorption is key for advancing applications based on catalysis, sensing, and surface functionalization. However, first-principles methods, such as density functional theory, are often computationally extremely expensive for feasible large-scale screenings. Given such a scenario, we address a data-driven approach, which integrates the semiempirical extended Hückel method (EHM) with machine learning (ML) techniques to estimate adsorption energy barriers in the case of three relevant chalcogen impurities, sulfur (S), selenium (Se), and tellurium (Te). With this aim, we consider the 4036 2D materials found in the Computational 2D Materials Database (C2DB). The scheme employs the EHM to compute energy profiles along three in-plane migration paths, from which average barriers can be derived. The equilibrium distance between the impurity and the 2D surface is not calculated from a time-consuming geometry optimization. Instead, it is estimated from a simple effective phenomenological expression. Physicochemical descriptors are then obtained from the Matminer (Materials Data Mining) library for curated features. Four different ML models are tested, with XGBoost (considering hyperparameter optimization via Optuna) leading to the highest performance. We further use SHAP to verify the resulting predictions, focusing on the ∼1500 materials displaying the lowest barrier values. As could be anticipated, we establish that the average valence electron count, electronegativity, and atomic number are typically the most relevant attributes to validate the ML model. However, we are also able to determine, for the different chalcogen atoms, which other few descriptors likewise considerably influence the adsorption properties. Our results show that when combined with interpretable ML protocols, EHM (and potentially semiempirical calculations in general) can produce a scalable framework for choosing 2D structures that exhibit the desired capture/release dynamics pertinent in a variety of utilization.
M. L. Pereira, M. G. E. da Luz, P. Cesana et al.· Journal of Chemical Theory a...· 0 citations
The adsorption and desorption of gas-phase molecules on solid surfaces are elementary steps in heterogeneous catalysis. However, capturing these dynamics under finite-temperature and coverage-dependent conditions remains challenging because static models cannot fully describe the coupled motion of adsorbates and surface atoms. In this study, we investigate the adsorption/desorption dynamics of CO on Ru(0001) by combining machine learning potential energy surfaces (ML-PESs) with umbrella sampling molecular dynamics (US-MD). A 2 × 2 × 5 Ru(0001) model containing one CO molecule (0.250 monolayer, ML) is used to analyze temperature-dependent potential of mean force (PMF) profiles, local free-energy landscapes, and molecular orientation distributions. To clarify coverage and finite-size effects, enlarged 4 × 4 × 5 Ru(0001) models containing 1, 4, and 8 CO molecules are used to cover 0.063, 0.250, and 0.500 ML, respectively, while 20 × 20 × 5 Ru(0001) models containing 100 and 200 CO molecules are further examined at 0.250 and 0.500 ML using a fine-tuned DPA-2 potential. The PMF profiles show that CO adsorption is nearly barrierless at low coverage, whereas an adsorption-side free-energy bottleneck becomes most clearly resolved under crowded high-coverage conditions. This bottleneck arises from the cooperative effect of configurational/rotational entropy loss and lateral CO–CO repulsion, with finite-size periodicity and surface coverage modulating the strength of entropic confinement in the PMF profile. Comparison between rigid and relaxed 2 × 2 × 5 surface models further shows that neglecting lattice motion qualitatively preserves the overall temperature-dependent PMF trend but suppresses the small adsorption-side bottleneck relative to the relaxed-surface model. These results highlight the coupled roles of coverage, entropy, lateral adsorbate interactions, and surface lattice degrees of freedom in CO/Ru(0001) gas–surface dynamics.
Jin He, Mingjun Yang, Zhe-Ning Chen et al.· JACS Au· 0 citations
This study demonstrates an example of how machine-learning-driven materials discovery can accelerate catalyst design while simultaneously offering an experimental solution to the persistent challenge of high ORR overpotential.
Darik A. Rosser, Anto Felix Sotvik GS, Kevin C. Leonard· ACS Applied Energy Materials· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
