Density Functional Theory Data Research Articles

Few-shot learning for screening 2D Ga2CoS4−x supported single-atom catalysts for hydrogen production

Hydrogen generation and related energy applications heavily rely on the hydrogen evolution reaction (HER), which faces challenges of slow kinetics and high overpotential. Efficient electrocatalysts, particularly single-atom catalysts (SACs) on two-dimensional (2D) materials, are essential. This study presents a few-shot machine learning (ML) assisted high-throughput screening of 2D septuple-atomic-layer Ga2CoS4−x supported SACs to predict HER catalytic activity. Initially, density functional theory (DFT) calculations showed that 2D Ga2CoS4 is inactive for HER. However, defective Ga2CoS4−x (x = 0–0.25) monolayers exhibit excellent HER activity due to surface sulfur vacancies (SVs), with predicted overpotentials (0–60 mV) comparable to or lower than commercial Pt/C, which typically exhibits an overpotential of around 50 mV in the acidic electrolyte, when the concentration of surface SV is lower than 8.3%. SVs generate spin-polarized states near the Fermi level, making them effective HER sites. We demonstrate ML-accelerated HER overpotential predictions for all transition metal SACs on 2D Ga2CoS4−x. Using DFT data from 18 SACs, an ML model with high prediction accuracy and reduced computation time was developed. An intrinsic descriptor linking SAC atomic properties to HER overpotential was identified. This study thus provides a framework for screening SACs on 2D materials, enhancing catalyst design.

Exploring the Polymorphism of Dicalcium Silicates Using Transfer Learning Enhanced Machine Learning Atomic Potentials.

Belitic cements are a greener alternative to Ordinary Portland Cement due to the lower CO2 associated with their production. However, their low reactivity with water is currently a drawback, resulting in longer setting times. In this study, we utilize a combination of evolutionary algorithms and machine learning atomic potentials (MLPs) to identify previously unreported belite polymorphs that may exhibit higher hydraulic reactivity than the known phases. To address the high computational demand of this methodology, we propose a novel transfer learning approach for generating MLPs. First, the models are pretrained on a large set of classical data (ReaxFF) and then retrained with density functional theory (DFT) data. We demonstrate that the transfer learning enhanced potentials exhibit higher accuracy, require less training data, and are more transferable than those trained exclusively on DFT data. The generated machine learning potential enables a fast, exhaustive, and reliable exploration of the dicalcium silicate polymorphs. This includes studying their stability through phonon analysis and calculating their structural and elastic properties. Overall, we identify ten new belite polymorphs within the energy range of the existing ones, including a layered phase with potentially high reactivity.

Atomic cluster expansion potential for large scale simulations of hydrocarbons under shock compression.

We present an Atomic Cluster Expansion (ACE) machine learned potential developed for high-fidelity atomistic simulations of hydrocarbons, targeting pressures and temperatures near and above supercritical fluid regimes for molecular fluids. A diverse set of stoichiometries were covered in training, including 1:0 (pure carbon), 1:4 (methane), and 1:1 (benzene), and rich bonding environments sampled at supercritical temperatures, hydrogen rich, reactive mixtures where metastable stoichiometries arise, including 1:2 (ethylene) and 1:3 (ethane). A high-fidelity training database was constructed by performing large-scale quantum molecular dynamic simulations [density functional theory (DFT) MD] of diamond, graphite, methane, and benzene. A novel approach to selecting structures from DFT MD is also presented, which allows for the rapid selection of unique DFT MD frames from complex trajectories. Comparisons to DFT and experimental data demonstrate that the presented ACE potential accurately reproduces isotherms, carbon melting curves, radial distribution functions, and shock Hugoniots for carbon and hydrocarbon systems for pressures up to 100GPa and temperatures up to 6000K for hydrocarbon systems and up to 9000K for pure carbon systems. This work delivers a potential that can be used for accurate, large-scale simulations of shocked hydrocarbons and demonstrates a methodology for fitting and validating machine learning interatomic potentials to complex molecular environments, which can be applied to energetic materials in future works.

Deciphering Absolute O 2p Position as a Practical Descriptor for Strain‐Dependent Perovskite Nickelate Electrocatalysts

AbstractDeveloping effective descriptors for perovskite oxygen evolution reaction (OER) electrocatalysts is crucial for advancing clean energy technologies, which are often constrained by incomplete theoretical calculations. Herein, using step‐wise strained (−3–3%) NdNiO3 as a model, the density functional theory (DFT) is employed, considering both bulk superlattice model (B‐model) and surface slab models (S‐model) to garner a set of electronic descriptors. The S‐model‐derived metal─oxygen orbital hybridization is found to agree with the X‐ray absorption spectroscopy (XAS) results, which, however, are disentangled from the observed OER activity trend. Further detailed analysis discovers that the alignment of the calculated O 2p band position with the OER redox potential on an absolute scale can serve as a hitherto unexplored descriptor for comprehending the activity of lattice oxygen and its bonding strength to OH*. Combined DFT and XAS data reveals that excessive strain hinders surface oxygen exchange, slows down the rate‐determining step of OH* adsorption, and impairs the durability of the electrocatalyst by altering the random electron occupancy in frontier orbitals. The work sheds light on the rational design of cost‐effective, high‐performance perovskite‐based electrocatalysts.

Structural Disorder by Octahedral Tilting in Inorganic Halide Perovskites: New Insight with Bayesian Optimization

Structural disorder is common in metal‐halide perovskites and important for understanding the functional properties of these materials. First‐principles methods can address structure variation on the atomistic scale, but they are often limited by the lack of structure‐sampling schemes required to characterize the disorder. Herein, structural disorder in the benchmark inorganic halide perovskites CsPbI3 and CsPbBr3 is computationally studied in terms of the three octahedral‐tilting angles. The subsequent variations in energetics and properties are described by 3D potential‐energy surfaces (PESs) and property landscapes, delivered by Bayesian optimization as implemented in the Bayesian optimization structure search code sampling density functional theory (DFT) calculations. The rapid convergence of the PES with about 200 DFT data points in 3D searches demonstrates the power of active learning and strategic sampling with Bayesian optimization. Further analysis indicates that disorder grows with increasing temperature and reveals that the material bandgap at finite temperatures is a statistical mean over disordered structures.

Prediction of Frequency-Dependent Optical Spectrum for Solid Materials: A Multioutput and Multifidelity Machine Learning Approach.

The frequency-dependent optical spectrum is pivotal for a broad range of applications from material characterization to optoelectronics and energy harvesting. Data-driven surrogate models, trained on density functional theory (DFT) data, have effectively alleviated the scalability limitations of DFT while preserving its chemical accuracy, expediting material discovery. However, prevailing machine learning (ML) efforts often focus on scalar properties such as the band gap, overlooking the complexities of optical spectra. In this work, we employ deep graph neural networks (GNNs) to predict the frequency-dependent complex-valued dielectric function across the infrared, visible, and ultraviolet spectra directly from the crystal structures. We explore multiple architectures for the spectral multioutput representation of the dielectric function and utilize various multifidelity learning strategies, such as transfer learning and fidelity embedding, to address the challenges associated with the scarcity of high-fidelity DFT data. Additionally, we model key solar cell absorption efficiency metrics, demonstrating that learning these parameters is enhanced when integrated through a learning bias within the learning of the frequency-dependent absorption coefficient. This study demonstrates that leveraging multioutput and multifidelity ML techniques enables accurate predictions of optical spectra from crystal structures, providing a versatile tool for rapidly screening materials for optoelectronics, optical sensing, and solar energy applications across an extensive frequency spectrum.

Multitask methods for predicting molecular properties from heterogeneous data.

Data generation remains a bottleneck in training surrogate models to predict molecular properties. We demonstrate that multitask Gaussian process regression overcomes this limitation by leveraging both expensive and cheap data sources. In particular, we consider training sets constructed from coupled-cluster (CC) and density functional theory (DFT) data. We report that multitask surrogates can predict at CC-level accuracy with a reduction in data generation cost by over an order of magnitude. Of note, our approach allows the training set to include DFT data generated by a heterogeneous mix of exchange-correlation functionals without imposing any artificial hierarchy on functional accuracy. More generally, the multitask framework can accommodate a wider range of training set structures-including the full disparity between the different levels of fidelity-than existing kernel approaches based on Δ-learning although we show that the accuracy of the two approaches can be similar. Consequently, multitask regression can be a tool for reducing data generation costs even further by opportunistically exploiting existing data sources.

Synthesis and characterization of bis-isoxazoline-thiosemicarbazone as a corrosion inhibitor for carbon steel: Experimental study, and molecular simulation

In this study, a novel compound, bis-isoxazoline-thiosemicarbazone (BIT), was synthesized and tested for its ability to protect carbon steel from corrosion in a 1 M HCl solution. The compound's structure was confirmed using spectroscopic techniques such as 1H NMR, 13C NMR, and HRMS. Various methods, including mass loss analysis, potentiodynamic polarization (PDP), electrochemical impedance spectroscopy (EIS), SEM, and EDX, were employed to assess BIT's effectiveness in inhibiting corrosion and its adsorption behavior on carbon steel in acidic conditions. The findings demonstrated the high effectiveness of BIT as a corrosion inhibitor, with a 95 % inhibition rate observed at a concentration of 0.77 mM and a temperature of 298°K. Analysis through PDP revealed that BIT operates as a mixed-type inhibitor, conforming to the Langmuir adsorption isotherm during its attachment to the carbon steel surface, indicating a mixed (chemical-physical) adsorption process.SEM-EDX analysis demonstrated even distribution of the inhibitor on the steel surface. Notably, BIT contains several functional groups (methoxy, carbonyl, isoxazolyl,..) and heteroatoms (N, O, and S) that play a crucial role, as supported by Density Functional Theory (DFT) studies. Experimental findings were validated by DFT and molecular dynamics (MD) simulation data.

Double half Heusler (DHH) alloys with a high absorption coefficient shed light on the optoelectronic, thermoelectric, and structural, properties of X4Fe2Hf4Pt2 (X = Bi, Sb), derived from DFT data analysis

In this investigation, we employ density functional theory (DFT) with the generalised gradient approximation of Perdew–Burke–Ernzerhof (GGA-PBE) and the modified Beck-Johnson approach (mBJ-GGA). Our focus is on examining the optoelectronic and thermoelectric properties of Bi4Fe2Hf4Pt2 and Sb4Fe2Hf4Pt2 compounds. The findings indicate that both compounds are stable in non-magnetic phase (NM). Regarding electronic characteristics, it is revealed that these compounds are direct bandgap semiconductors. The optical analysis suggests that these compounds demonstrate high reflectivity within the visible and ultraviolet spectra, rendering them promising candidates for solar cells and photovoltaic applications. Furthermore, these materials present as highly promising candidates for thermoelectric purposes due to their remarkable figure of merit (ZT) and notable Seebeck coefficient (S) values. However, there is a lack of theoretical or experimental data available for these compounds for conducting comparison with our results. We think future experimenters attempting to synthesise these chemicals experimentally will benefit from the current study.

Predicting Quantum-Mechanical Partial Charges in Arbitrarily Long Boron Nitride Nanotubes to Accurately Simulate Nanoscale Water Transport.

Single-walled boron nitride nanotubes (BNNTs) have been explored for various applications, ranging from water desalination to osmotic power harvesting. However, no simulation work so far has modeled the changes in the partial charge distribution when a flat sheet is rolled into a tube, hindering the ability to perform accurate molecular dynamics (MD) simulations of water flow through BNNTs. To address this knowledge gap, we employ electronic density functional theory (DFT) calculations to precisely estimate quantum-mechanically derived partial charges on boron (B) and nitrogen (N) atoms in BNNTs of varying lengths and diameters. We observe a spatially varying charge distribution inside both armchair and zigzag nanotubes of finite lengths. Performing DFT calculations for longer BNNTs is computationally intractable, even with state-of-the-art computing resources. To solve this issue, we devise a charge assignment scheme to predict partial charges for longer BNNTs using DFT data for shorter nanotubes, thus overcoming the need to perform more expensive DFT calculations. We show that these charges reproduce the electrostatic potential predicted from first-principles simulations. Subsequently, we carried out MD simulations to predict the effect of the charge distribution inside BNNTs on water flow enhancement via them. We find that using uniform charges leads to an underprediction in flow enhancement, as compared to using quantum-mechanical charges for both armchair and zigzag BNNTs. We also incorporate atomic vibrations into our simulations and show that these vibrations lead to a reduction in the water flow through aperiodic BNNTs. Our work demonstrates the requirement of a quantum-mechanical charge assignment scheme for BNNTs and evolves a framework to assign charges to nanotubes of arbitrary length, thus allowing realistic MD simulations of long BNNTs using accurate partial charges.

Towards atomistic modelling of solid Pb-O formation and dissolution in liquid lead coolant: Interatomic potential development

Microscopic description of solid Pb-O formation and dissolution in liquid lead coolant is important for the modelling of fast neutron reactors. For this purpose, in this work we develop interatomic potential models for lead melt with dissolved oxygen. Potentials fitting is based on ab initio density functional theory (DFT) calculations and quantum molecular dynamics. Two different potential models are trained on the DFT data using the force-matching method: a computationally cheap embedded atom method (EAM) potential and a computationally expensive machine-learned moment tensor potential (MTP) that allows reproducing DFT data very accurately. The potentials are further validated to reproduce the properties of lead melt and the coordination of oxygen atoms in molten lead. The diffusion coefficient for oxygen in molten lead is calculated and compared with the available experimental data. We analyse how these interatomic potential models describe the crystal structure of the solid lead oxide to consider PbO nucleation and dissolution in molten lead. We extensively discuss the arising problems with the description of long-range electrostatic interactions. The possibility of matching the experimental data on the oxygen solubility limit by tuning the EAM potential is shown.

A reactive molecular dynamics model for uranium/hydrogen containing systems.

Uranium-based materials are valuable assets in the energy, medical, and military industries. However, understanding their sensitivity to hydrogen embrittlement is particularly challenging due to the toxicity of uranium and the computationally expensive nature of quantum-based methods generally required to study such processes. In this regard, we have developed a Chebyshev Interaction Model for Efficient Simulation (ChIMES) that can be employed to compute energies and forces of U and UH3 bulk structures with vacancies and hydrogen interstitials with accuracy similar to that of Density Functional Theory (DFT) while yielding linear scaling and orders of magnitude improvement in computational efficiency. We show that the bulk structural parameters, uranium and hydrogen vacancy formation energies, and diffusion barriers predicted by the ChIMES potential are in strong agreement with the reference DFT data. We then use ChIMES to conduct molecular dynamics simulations of the temperature-dependent diffusion of a hydrogen interstitial and determine the corresponding diffusion activation energy. Our model has particular significance in studies of actinides and other high-Z materials, where there is a strong need for computationally efficient methods to bridge length and time scales between experiments and quantum theory.

Development and validation of a general-purpose ReaxFF reactive force field for earth material modeling.

ReaxFF reactive force field bridges the gap between nonreactive molecular simulations and quantum mechanical calculations and has been widely applied during the past two decades. However, its application to earth materials, especially those under high T-P conditions relevant to Earth's interior, is still limited due to the lack of available parameters. Here, we present the development and validation of a ReaxFF force field containing several of the most common elements in Earth's crust, i.e., Si/Al/O/H/Na/K. The force field was trained against a large data set obtained from density functional theory (DFT) calculations, including charges, bond/angle distortion curves, equation of states, ion migration energy profiles, and condensation reaction energies. Different coordination environments were considered in the training set. The fitting results showed that the current force field can well reproduce the DFT data (the Pearson correlation coefficient, Rp, is 0.95). We validated the force field on mineral-water interfaces, hydrous melts/supercritical geofluids, and bulk crystals. It was found that the current force field performed excellently in predicting the structural, thermodynamic, and transport properties of various systems (Rp = 0.95). Moreover, possible applications and future development have been discussed. The results obtained in this study suggest that the current force field holds good promise to model a wide range of processes and thus open opportunities to advance the application of ReaxFF in earth material modeling.

A strategy of combining energy‐resolved mass spectrometry and density functional theory for the recognition of forsythoside isomers

RationaleIsomerism is very common in nature, and the mass spectrometry (MS) behavior of isomers is highly similar, making their spectral analysis and identification difficult. Computational chemistry can effectively assist the study of MS spectra and molecule structures. Herein, three isomers of forsythoside (For‐A, For‐H, and For‐I) were successfully identified using energy‐resolved MS combined with density functional theory (DFT).MethodsThe MS scan spectra of the isomers were collected in electrospray ionization MS. The possible positive/negative ion adduct sites were investigated using the DFT method, and the molecular formation energy of each precursor ion was obtained. In the multireaction monitoring mode, the spectra of the common precursor ion → product ion pair of the three isomers were acquired under different collision energy, and the optimal collision energy (OCE) and maximum relative intensity (MRI) of each product ion were obtained. The energy of the broken chemical bond was calculated.ResultsThe experimental data showed that the product ions of [M + H]+ of the three isomers have the greatest differences in OCE and MRI and were the most suitable to distinguish the three isomers. The DFT data showed that there was a good positive correlation between the bond fragmentation energy calculated by the quasi‐molecular ion model and OCE. The results indicated that the energy of precursor ions and corresponding fragments could be correlated with their experimental spectra.ConclusionsA strategy of combining energy‐resolved MS and DFT was implemented on the recognition of forsythoside isomers, and this strategy was expected to be applied to the identification of more isomers.

First principle based rate equation (1pRE) for reduction kinetics of Fe2O3 with syngas in chemical looping

Chemical looping combustion (CLC) of solid fuel is a promising technology with inherent CO2 separation and low energy penalty for CO2 capture. Syngas is main intermediate species of solid fuel conversion in CLC, the reduction kinetics of oxygen carriers with syngas play a crucial role in CLC systems. However, the current research obtains the reaction rate constants by fitting the apparent models with the experimental data, and cannot explain the reduction kinetics behavior from a microscopic level. It remains a challenge to compute the reduction kinetics of oxygen carriers with syngas directly from first-principles density functional theory (DFT) without fitting experimental data. This study proposes a first-principle-based rate equation (1pRE) theory and integrates it into the random pore model (RPM) to predict the kinetics of Fe2O3 reduction by syngas in CLC. The developed 1pRE theory utilizes DFT calculations to search for reaction pathways and energy barriers of elementary reactions. Then the DFT data are introduced into the statistical mechanics partition function and transition state theory (TST) to calculate the reaction rate constants. Microkinetic rate equations of elementary reactions occurring at the surface scale are developed to describe the change of surface coverage of different surface species. The 1pRE theory is integrated into the RPM to account for the influence of particle-scale structural changes on the overall conversion rate during the reduction process. The theory can predict the reduction kinetics of oxygen carriers without fitting experimental data and establishes a connection between microscopic insights and macroscopic phenomena. The accuracy was validated by experimental data of Fe2O3 oxygen carriers obtained from the thermogravimetric analyzer (TGA) in the atmosphere of syngas. The developed 1pRE predicts the reduction kinetics of oxygen carriers accurately and can be used to optimize the design of oxygen carrier materials and the scale up of CLC reactors.

Composition-based phase stability model for multicomponent metal alloys

The vastness of the space of possible multicomponent metal alloys is hoped to provide improved structural materials but also challenges traditional, low-throughput materials design efforts. Computational screening could narrow this search space if models for materials stability and desired properties exist that are sufficiently inexpensive and accurate to efficiently guide experiments. Toward this effort, here we develop a method to rapidly assess the thermodynamic stability of a metal alloy composition of an arbitrary number of elements, stoichiometry, and temperature based on density functional theory (DFT) data. In our model, the Gibbs free energy of the solid solution contains binary enthalpy contributions and ideal configurational entropy, whereas only enthalpy is considered for intermetallic competing phases. Compared to a past model for predicting the formation of single-phase high-entropy alloys [M. C. Troparevsky et al., Phys. Rev. X 5, 011041 (2015)], our method is similarly inexpensive, since it assesses enthalpies based on existing DFT data, but less heuristic, more broadly applicable, and more accurate (70%–75%) compared to experiment.

Machine learning of kinetic energy densities with target and feature smoothing: Better results with fewer training data.

Machine learning (ML) of kinetic energy functionals (KEFs), in particular kinetic energy density (KED) functionals, is a promising way to construct KEFs for orbital-free density functional theory (DFT). Neural networks and kernel methods including Gaussian process regression (GPR) have been used to learn Kohn-Sham (KS) KED from density-based descriptors derived from KS DFT calculations. The descriptors are typically expressed as functions of different powers and derivatives of the electron density. This can generate large and extremely unevenly distributed datasets, which complicates effective application of ML techniques. Very uneven data distributions require many training datapoints, can cause overfitting, and can ultimately lower the quality of an ML KED model. We show that one can produce more accurate ML models from fewer data by working with smoothed density-dependent variables and KED. Smoothing palliates the issue of very uneven data distributions and associated difficulties of sampling while retaining enough spatial structure necessary for working within the paradigm of KEDF. We use GPR as a function of smoothed terms of the fourth order gradient expansion and KS effective potential and obtain accurate and stable (with respect to different random choices of training points) kinetic energy models for Al, Mg, and Si simultaneously from as few as 2000 samples (about 0.3% of the total KS DFT data). In particular, accuracies on the order of 1% in a measure of the quality of energy-volume dependence B'=EV0-ΔV-2EV0+E(V0+ΔV)ΔV/V02 (where V0 is the equilibrium volume and ΔV is a deviation from it) are obtained simultaneously for all three materials.

Ortho-Substituent Effects on Halogen Bond Geometry for N-Haloimide⋯2-Substituted Pyridine Complexes.

The nature of (imide)N-X⋯N(pyridine) halogen-bonded complexes formed by six N-haloimides and sixteen 2-substituted pyridines are studied using X-ray crystallography (68 crystal structures), Density Functional Theory (DFT)(86 complexation energies), and NMR spectroscopy (90 association constants). Strong halogen bond (XB) donors such as N-iodosuccinimide form only 1:1 haloimide:pyridine crystalline complexes, but even stronger N-iodosaccharin forms 1:1 haloimide:pyridine and three other distinct complexes. In 1:1 haloimide:pyridine crystalline complexes, the haloimide's N─X bond exhibits an unusual bond bending feature that is larger for stronger N-haloimides. DFT complexation energies (ΔEXB ) for iodoimide-pyridine complexes range from -44 to -99kJmol-1 , while for N-bromoimide-pyridine, they are between -31 and -77kJmol-1 . The ΔEXB of I⋯N XBs in 1:1 iodosaccharin:pyridine complexes are the largest of their kind, but they are substantially smaller than those in [bis(saccharinato)iodine(I)]pyridinium salts (-576kJmol-1 ), formed by N-iodosaccharin and pyridines. The NMR association constants and ΔEXB energies of 1:1 haloimide:pyridine complexes do not correlate as these complexes in solution are heavily influenced by secondary interactions, which DFT studies do not account for. Association constants follow the σ-hole strengths of N-haloimides, which agree with DFT and crystallography data. The haloimide:2-(N,N-dimethylamino)pyridine complex undergoes a halogenation reaction resulting in 5-iodo-2-dimethylaminopyridine.

VASP2KP: k⋅p Models and Landé g-Factors from ab initio Calculations

The k⋅p method is significant in condensed matter physics for the compact and analytical Hamiltonian. In the presence of magnetic field, it is described by the effective Zeeman’s coupling Hamiltonian with Landé g-factors. Here, we develop an open-source package VASP2KP (including two parts: vasp2mat and mat2kp) to compute k⋅p parameters and Landé g-factors directly from the wavefunctions provided by the density functional theory (DFT) as implemented in Vienna ab initio Simulation Package (VASP). First, we develop a VASP patch vasp2mat to compute matrix representations of the generalized momentum operator , spin operator , time reversal operator , and crystalline symmetry operators on the DFT wavefunctions. Second, we develop a python code mat2kp to obtain the unitary transformation U that rotates the degenerate DFT basis towards the standard basis, and then automatically compute the k⋅p parameters and g-factors. The theory and the methodology behind VASP2KP are described in detail. The matrix elements of the operators are derived comprehensively and computed correctly within the projector augmented wave method. We apply this package to some materials, e.g., Bi2Se3, Na3Bi, Te, InAs and 1H-TMD monolayers. The obtained effective model’s dispersions are in good agreement with the DFT data around the specific wave vector, and the g-factors are consistent with experimental data. The VASP2KP package is available at https://github.com/zjwang11/VASP2KP.

Graph neural networks for predicting structural stability of Cd- and Zn-doped [formula omitted]-CsPbI3

Computational modeling of disordered crystal structures is essential for the study of composition–structure–property relations for many families of functional materials. Efficient and reliable solutions are of great interest due to the promising reduction of labor-intensive calculations through data-driven prediction of target properties. One of the modern problems directly related to this topic is the enhancement of phase stability of functional materials with the required properties. In this work, we address the effects of Cd and Zn substitutions on the Pb sites on the structural stability of inorganic lead halide perovskite CsPbI3 — a semiconductor promising for optoelectronic devices. At room temperature, this material undergoes an undesirable phase transition from the direct band gap “black” to indirect “yellow” (ca. 15 meV/atom more favored) phase. For evaluations of effects caused by the substitutions, we combined the density functional theory (DFT) calculations and graph neural network (GNN) models. At low Cd and Zn contents, complete composition/configuration spaces comprising ca. 200 structures were set and studied using DFT. The GNN models trained on this data were used to predict energetics of ca. 74k structures with higher contents of substitutions and evaluate the energy differences between the phases. Because of the small amount of DFT data, an impact of data subsampling on the quality of predictions was comprehensively studied. Additionally, transfer learning routines were tested using available collections of DFT-derived properties. For the examined models, the root-mean-square errors of ca. 0.7 meV/atom were achieved in predicting formation energies. The data treatment workflows proposed and tested in this study open up perspectives for data-driven research of disordered materials of particular interest with low computational costs.