The membranes’ role is of supreme importance in the separation of compounds under different phases of matter. The topic addressed here is based on the use of membranes on the gases separation, specifically the advantages of mixed-matrix membranes (MMMs) when using carbon nanotubes as fillers to separate carbon dioxide (CO₂) from other carrier gas. MMMs consist of a polymer support with additive fillers to improve their efficiency by increasing both selectivity and permeability. The most promising fillers in the MMM development are nanostructured molecules. Due to the good prospects of carbon nanotubes (CNTs) as MMM fillers, this article aims to concentrate the advances and developments of CNT–MMM to separate gases, such as CO₂. The influence of functionalized CNT or mixtures of CNT with additional materials such as zeolites, hydrogel and, graphene sheets on membranes performance is highlighted in the present work.

Geometric optimization of carbon nanotubes (CNTs) is a fundamental step in computational simulations, enabling precise studies of their properties for various applications. However, this process becomes computationally expensive as the molecular structure grows in complexity and size. To address this challenge, this study utilized three deep-learning-based neural network architectures: Multi-Layer Perceptron (MLP), Bidirectional Long Short-Term Memory (BiLSTM), and 1D Convolutional Neural Networks (1D-CNNs). Simulations were performed using the CASTEP module in Material Studio to generate datasets for training the neural networks. While the final geometric optimization calculations were completed within Material Studio, the neural networks effectively generated preoptimized CNT structures that served as starting points, significantly reducing computational time. The results showed that the 1D-CNN architecture performed best for CNTs with 28, 52, 76, and 156 atoms, while the MLP outperformed others for CNTs with 84, 124, 148, and 196 atoms. Across all cases, computational time was reduced by 39.68% to 90.62%. Although the BiLSTM also achieved reductions, its performance was less effective than the other two architectures. This work highlights the potential of integrating deep learning techniques into materials science; it also offers a transformative approach to reducing computational costs in optimizing CNTs and presents a way for accelerated research in molecular systems.

Accurate estimation of molecule–nanotube interaction energies is critical for the computational screening of carbon-based materials for hydrogen storage; however, density functional theory (DFT) calculations remain computationally expensive for extensive configurational sampling. In this work, we develop a multilayer perceptron (MLP) surrogate model to predict H₂–CNT interaction energies, represented by $E_{ads}$, for H₂ interactions with a chiral (2,1) carbon nanotube. A curated dataset comprising 696 configurations was generated using DMol³ (BIOVIA Materials Studio), varying intermolecular distance, molecular orientation, and interaction site across three regions: internal cavity, edges, and external surface. The proposed MLP architecture (64–32–1) incorporates GELU activation functions, L2 regularization, and dropout to improve generalization. The model achieves coefficients of determination in the range R² = 0.90–0.96 across all interaction regions, with particularly strong performance at the nanotube edges (R² = 0.9358, MSE = 0.046 eV²), as well as on the external surface (R² = 0.9625, MSE = 0.574 eV²) and within the internal cavity (R² = 0.9051, MSE = 1.506 eV²). The original $E_{ads}$ distribution had a mean of 4.0955 eV and a sample standard deviation of 4.3189 eV. The elevated energy values observed in the internal cavity (up to 12 eV) are consistent with steric repulsion induced by geometric confinement rather than predictive artifacts. The trained MLP showed close agreement with DFT-derived trends, enabling exploration of interaction-energy landscapes spanning both attractive and repulsive regimes. These results indicate that MLP-based models trained on diverse configurational datasets provide a computationally efficient alternative for screening carbon nanostructures in hydrogen storage applications, without substantially compromising accuracy relative to first-principles methods.