Bartel Comp Materials

Short-range order in DRX cathodes significantly deviates from perfect randomness and affects local bond lengths and site occupancies. Accurate structural models must account for SRO to correctly interpret diffraction and PDF data from these materials.

Multiple synthesis routes targeting highly fluorinated DRX cathode Li1.2Mn0.4Ti0.4O1.6F0.4 were rationally designed using thermochemical analysis. MnF2 as a reactive fluorine source and alternative Li precursors (Li6MnO4, LiMnO2) were tested to avoid LiF formation that inhibits fluorination.

LiF formation during synthesis is the primary barrier to achieving high fluorination in DRX cathodes. Raising the F chemical potential through reactive fluoride precursors (MnF2) is more effective than using LiF directly as a fluorine source.

r2SCAN meta-GGA with rVV10 van der Waals correction provides improved prediction of formation and decomposition enthalpies for solids compared to PBE-GGA. For 1000+ compounds tested, r2SCAN+rVV10 reduces MAE of formation enthalpies and improves phase stability predictions.

r2SCAN restores numerical stability that was problematic in the original SCAN functional while maintaining similar accuracy for solid-state thermochemistry, making it suitable for high-throughput computational screening of materials stability.

Automated high-throughput comparison of r2SCAN and SCAN for 6,307 solid materials reveals that r2SCAN achieves comparable accuracy to SCAN for formation energies and band gaps while being significantly more numerically stable, enabling reliable high-throughput workflows.

Computational screening of pyrochlore and spinel crystal structures identifies materials with predicted giant anomalous Hall effect, demonstrating the application of high-throughput DFT for functional property discovery beyond thermodynamic stability.

Ca1.5Ba0.5Si5O3N6 is identified as a potential calcium solid-state conductor with partially occupied Ca sites enabling ion migration. Ab initio NEB calculations combined with neutron diffraction reveal the Ca migration mechanism through a 3D percolating pathway.

Development of Ca-ion batteries is largely limited by the low diffusion rate of divalent Ca2+ ions in solid-state materials. Understanding design principles for Ca2+ mobility requires identifying host structures with appropriate channel sizes and partial site occupancies.

Optimal solid-state synthesis heating temperatures correlate strongly with precursor material stability (melting points, formation energies) but show no direct correlation with thermodynamic features of the synthesis reaction itself, extending Tamman's rule from intermetallics to oxide systems.

Heating times in solid-state synthesis correlate with experimental procedures and instrument setups rather than material properties, suggesting potential human bias in reported synthesis conditions.

A data-driven precursor recommendation system trained on 29,900 text-mined solid-state synthesis procedures achieves at least 82% success rate when providing top-5 precursor recommendations for 2,654 unseen target materials.

The ARROWS3 algorithm for autonomous precursor selection identifies effective precursor sets by learning which precursors form highly stable intermediates that prevent target formation, requiring substantially fewer iterations than black-box optimization across 3 experimental datasets with 200+ synthesis procedures.

Kinetically controlled solid-state metathesis between MnCl2 and nitrogen-containing precursors (Mg2NCl, Mg3N2) produces Mn3N2 at low temperatures via a solid-solution intermediate mechanism, enabling synthesis of metal nitrides inaccessible by conventional ceramic methods.

Rate-limiting barriers for ion exchange reactions in solid-state synthesis are associated with the salt precursor (LiCl, LiBr) rather than the ceramic target material. Defect formation energy in LiBr is substantially lower than DFT predictions, also influencing rates.

Scaling relationships govern concentration-dependent reaction kinetics in lithium halide ion exchange, enabling predictions of conditions that could substantially accelerate ion exchange reactions through control of vacancy concentrations.

NaSICON-structured NaV2(PO4)3 demonstrates reversible dual Ca2+-Na+ ion electrochemistry, with topotactic (de)intercalation of 0.6 mol Ca2+ alongside Na+ exchange. This establishes NaSICON as a viable cathode framework for Ca-ion batteries.

Ca-ion batteries are plagued by a paucity of suitable cathode materials. NaV2(PO4)3 with NASICON structure overcomes this by leveraging the rigid polyanion framework that provides channels for Ca2+ transport.

CaB12H12 has a percolating Ca migration path with a low activation barrier of 650 meV, and can be doped with Al, Bi, or trivalent rare-earth cations to create vacancies and potentially improve conductivity for solid-state Ca batteries.

Large expansion of the CaFe2O4-type sodium postspinel phase space at ambient pressure achieved through systematic synthesis of NaCrTiO4, NaRhTiO4, NaCrSnO4, NaInSnO4, NaMgAlO4, and other new compositions. These tunnel-structured materials are prospective battery electrodes with Na+ in tunnel sites.

Computational approaches are emerging to accelerate battery materials synthesis, including thermodynamic selectivity analysis, precursor recommendation via ML, and autonomous experimental platforms. These techniques can predict synthesis conditions and identify optimal precursor sets for target battery phases.

Synthesis prediction for battery materials requires understanding both thermodynamic driving forces (which phases form) and kinetic barriers (reaction rates and temperatures). Current computational methods are strongest at predicting thermodynamic outcomes but kinetic prediction remains challenging.

Double-ion exchange (anion cometathesis) reactions enable perovskite LaMnO3 synthesis with reaction onset at 450-480C, compared to >1000C for traditional equilibrium synthesis. Lanthanum oxyhalide precursors achieve the lowest onset temperatures.

DFT accuracy for predicting solid stability depends critically on reaction type: PBE achieves MAE=70 meV/atom and SCAN achieves MAE=59 meV/atom for 1,012 compounds vs experiment, but for 231 compound-only reactions (no elemental phases), both functionals achieve ~35 meV/atom, matching experimental uncertainty.

ML models using 25 electronic structure features can correct PBE-calculated formation enthalpies from MAE=195 meV/atom to MAE=80 meV/atom across 1,011 solid-state compounds. For highly ionic compounds (ionicity>0.22), PBE errors are systematically larger.

Metathesis synthesis enables rapid and selective formation of MgCr2S4 thiospinel cathode material by altering the thermodynamic landscape. Traditional ceramic synthesis of this Mg cathode material is laborious, but metathesis bypasses intermediate phases.

Review establishes that machine learning methods including neural networks, Gaussian process regression, and random forests can accelerate heterogeneous catalyst design by learning structure-property relationships from DFT datasets, reducing the need for expensive quantum-chemical calculations.

Application of the Gibbs energy descriptor to ~30,000 ICSD materials reveals the temperature-dependent scale of metastability and provides insights into how composition and temperature jointly affect materials synthesizability.

For 231 Type 2 reactions (involving only compounds, no elements), SCAN, PBE, and experiment agree within ~35 meV/atom, comparable to experimental uncertainty, suggesting DFT is highly accurate for compound-compound reaction thermodynamics.

DFT and ML potentials predict negative hydrogen insertion energies for La0.5Sr0.5CoO3-delta, but convex hull analysis reveals protonated phases are thermodynamically unstable, decomposing into hydroxides and other products, explaining experimentally observed acid-etching during electrochemical cycling.

Thermodynamic selectivity analysis can be used prospectively to identify optimal precursor sets for synthesizing novel target materials, reducing the trial-and-error typically required for solid-state synthesis.

In situ synchrotron studies reveal that rate-limiting barriers in Li-ion exchange reactions are commonly associated with the LiCl/LiBr salt reactants rather than the ceramic target, quantifying both thermodynamic activation energies and kinetic barriers for the salt processes.

New ternary zinc molybdenum nitride compounds Zn3MoN4 and ZnMoN2 were predicted by theory and experimentally realized. These alloys form in a broad composition range in the wurtzite-derived structure, with redox-mediated stabilization enabling large off-stoichiometry accommodation.

A novel chemical looping process generating pure SO2 from raw sulfur and air is identified computationally, with 12 viable redox material combinations (2 supported by prior experimental evidence), potentially offering a more efficient, lower-emission route to sulfuric acid.

High-throughput computation of 3,881 potential NASICON phases identified stability rules based on a 2D descriptor combining cation properties. Five of six predicted NASICONs were successfully synthesized experimentally, validating the ab initio predictive capability.

Thermochemical modeling identifies a narrow temperature window for fluorinated disordered rocksalt cathode synthesis: too low produces LiF intermediates that restrict fluorine availability, while temperatures above 848C (LiF melting point) cause LiF volatility limiting F solubility.

Approximately two-thirds of decomposition reactions relevant to solid stability assessment contain no elemental phases. Correction schemes using fitted elemental reference energies provide negligible improvement for these majority reactions.

Diversifying training data composition improves ML predictions of thermodynamic properties for inorganic crystals more than simply increasing dataset size. A data-centric approach that balances chemical space representation yields more robust and generalizable formation energy models.

Partial dependence plot and GAM analysis reveals that compounds with high ionicity (I>0.22) have systematically larger DFT formation enthalpy errors, providing interpretable insight into where density functional approximations fail most significantly.

Li2MnP2S6, a new lithium metal thiophosphate synthesized via novel iodide-assisted route, shows electrochemical Li extraction at ~3 V, significantly higher than other sulfide cathodes. Mn and Fe analogs operate at similar voltages but through different redox mechanisms.

Computational screening of ternary metal nitrides identified 209 stable and 847 metastable (within 100 meV/atom of convex hull) previously unreported nitride compounds, creating a comprehensive stability map of the ternary metal nitride chemical space.

SISSO-identified physical descriptor predicts Gibbs free energy of stoichiometric inorganic compounds with ~50 meV/atom resolution across 300-1800K temperature range, enabling generation of thousands of temperature-dependent phase diagrams for ~30,000 known materials from the ICSD.

For 646 non-trivial decomposition reactions assessed against 1012 experimental formation enthalpies, PBE achieves MAD of 70 meV/atom and SCAN achieves MAD of 59 meV/atom, with commonly employed elemental reference energy corrections providing only ~2 meV/atom improvement.

Computational precursor selection methods that incorporate thermodynamic reaction analysis are broadly applicable across battery material classes (Li-ion cathodes, solid electrolytes, all-solid-state batteries), though effectiveness varies and current methods have limitations requiring further development for synthesis-by-design.

Four published ML synthesizability scores generally overestimate synthesis likelihood for hypothetical materials, but certain scoring approaches correlate with thermodynamic heuristics, assigning lower scores to materials lacking stability or available synthesis routes.

Ab initio thermodynamics correctly predicts which pair of precursors has the most reactive interface in multicomponent ceramic synthesis. Solid-state synthesis proceeds through sequential pairwise reactions at interfaces, with the most exothermic pair reacting first.

Although halide salts LiCl and LiBr are routinely used as Li sources for ion exchange reactions, the processes controlling their reaction rates were poorly understood. This work demonstrates that salt melting, dissolution, and diffusion through salt layers are quantifiable kinetic barriers.

Computational screening of 3,881 potential NASICON phases combined with a ML tolerance factor achieves 5 out of 6 successful synthesis attempts, demonstrating effective integration of stability prediction with experimental validation.

Comprehensive review establishes that computational prediction of inorganic solid stability has matured significantly with DFT functionals, correction schemes, and ML approaches, but systematic errors remain for reactions involving elemental phases and strongly correlated systems.

Imbalanced training data leads to biased ML predictions that perform well on common chemistries but poorly on underrepresented compositions, highlighting the need for active learning and targeted data collection strategies.

A comprehensive computational map of inorganic ternary metal nitrides identifies stable and metastable compositions across the full periodic table, revealing that nitride chemical space is far less explored than oxide space with many predicted-stable but unsynthesized compounds.

Li2FeP2S6 and Li2MnP2S6 operate at similar ~3 V voltages but via fundamentally different redox mechanisms: Fe undergoes cation redox while Mn involves anion (sulfur) redox, as revealed by DFT electronic structure analysis.

Comprehensive review identifies that predicting thermodynamic stability of inorganic solids requires computing energy above the convex hull (decomposition enthalpy) rather than formation enthalpy alone, and that DFT accuracy at the meta-GGA level (SCAN, r2SCAN) provides ~50-70 meV/atom MAE for formation enthalpies.

Analysis of 56,791 compounds shows that decomposition into elemental forms is rarely the competing reaction determining stability; approximately two-thirds of decomposition reactions involve no elemental phases, making formation enthalpy an incomplete metric for stability assessment.

MBE growth of rutile Sn1-xGexO2 achieves maximum 34% Ge incorporation at 600C, with DFT phase diagram analysis predicting spinodal decomposition beyond this concentration. Ge-rich rutile phase formation is suppressed by amorphization and GeO volatility.

Limited negative training examples (materials that failed to synthesize) represent a fundamental challenge for ML synthesis prediction, as models cannot learn failure modes from databases that predominantly report successful syntheses.

Two selectivity metrics (primary and secondary competition) quantify the thermodynamic favorability of target vs impurity phase formation in 3,520 literature solid-state reactions. These metrics successfully rank synthesis approaches and predict which precursor combinations yield the purest products.

Heterogeneous solid-state synthesis evolves through a complicated series of reaction intermediates determined by local interface thermodynamics, not bulk equilibrium. Modeling the most reactive interface pair enables understanding and prediction of the reaction pathway.

The thermodynamics of proton insertion across the perovskite-brownmillerite structural transition in La0.5Sr0.5CoO3-delta are characterized, revealing how oxygen vacancy ordering and proton incorporation energetics govern the structural phase transition relevant to protonic ceramic applications.

A systematic thermodynamic methodology correctly classifies all 17 experimentally tested redox materials for chemical looping and identifies over 1,300 previously unstudied promising candidates, demonstrating scalable materials discovery for industrial redox processes.

High-throughput thermodynamic screening of 1,148 nitride/metal oxide pairs for solar thermochemical ammonia synthesis identifies promising materials based on boron, vanadium, iron, and cerium through equilibrium analysis of four key reactions: hydrolysis, oxide reduction, nitrogen fixation, and nitride reformation.

Microkinetic modeling predicts programmable oxide catalysts can boost OER current density by 100-600x at fixed overpotentials, or reduce the overpotential needed to achieve 10 mA/cm2 by 45-140% compared to optimal static catalysts.

Interpretable ML models can forecast the behavior of programmable catalytic loops, advancing the ability to predict and design dynamic catalytic systems beyond what microkinetic simulations alone can achieve.

Ensemble convolutional neural network trained on physics-informed augmented simulated diffraction spectra achieves exceptional accuracy for multi-phase mixture identification, exceeding previously reported methods based on profile matching and deep learning.

Adaptive XRD enables in situ identification of short-lived intermediate phases formed during solid-state reactions using a standard in-house diffractometer, a capability previously requiring synchrotron facilities.

Heating times in solid-state synthesis are strongly correlated with experimental procedures and instrument setups rather than thermodynamic or kinetic properties, indicating significant human bias in reported synthesis durations.

A theoretical framework predicts and controls polymorph selectivity in solid-state reactions by using reaction energy as a handle for polymorph selection. Surface energy contributions at small particle sizes can tip the thermodynamic balance toward metastable polymorphs.

When reaction energies are large (strongly exothermic), thermodynamics primarily dictates the initial product formed regardless of stoichiometry. This principle enables predictive synthesis by selecting precursors that maximize thermodynamic driving force toward the target phase.

A physically interpretable SISSO-derived descriptor accurately predicts Gibbs energies of inorganic crystalline solids as a function of temperature, enabling temperature-dependent phase stability predictions without expensive phonon calculations for thousands of compounds.

In situ characterization of 37 reactant pairs reveals a thermodynamic threshold below which the first intermediate phase formed is thermodynamically controlled regardless of stoichiometry. Above this threshold, kinetics dominates.

Ion exchange of known compounds outperforms generative AI (diffusion models, VAEs, LLMs) for discovering novel inorganic crystals. Random enumeration of charge-balanced prototypes also performs competitively against generative methods.

A SISSO-derived descriptor accurately predicts Gibbs energies of inorganic solids as a function of temperature, enabling temperature-dependent phase stability predictions without expensive phonon calculations.

High-throughput thermodynamic screening using Gibbs energy models identified promising active materials for chemical looping from thousands of candidates, overcoming the scarcity of high-temperature thermochemical data.

The O*-to-OOH* and O*-to-OH* activation barriers emerge as the critical parameters governing OER catalytic performance under dynamic forcing conditions, determining the optimal oscillation parameters for programmable catalysts.

ML-driven adaptive XRD can accurately detect trace amounts of materials in multi-phase mixtures with significantly shorter measurement times than conventional full-range scans, by steering measurements toward features that improve phase identification confidence.

ML models trained on text-mined synthesis datasets reveal that optimal solid-state heating temperatures strongly correlate with precursor stability (melting points, formation energies), extending Tamman's rule from intermetallics to oxide systems and suggesting kinetics determine synthesis conditions.

Using an 18-element chemical reaction network with Materials Project thermodynamic data, two unconventional BaTiO3 synthesis reactions using BaS/BaCl2 and Na2TiO3 precursors were discovered that produce BaTiO3 faster and with fewer impurities than conventional methods.

In situ characterization of 37 reactant pairs reveals a thermodynamic threshold below which the first intermediate phase formed is controlled by thermodynamics regardless of reactant stoichiometry. Above this threshold, kinetics becomes dominant.

Generative AI for materials discovery has unclear advantages over traditional methods. Proper baseline comparisons are essential, and current generative models may be solving an easier problem than claimed when measured against simple composition/structure enumeration.

High-throughput thermodynamic screening using accurate Gibbs energy models identified promising active materials for chemical looping from thousands of candidates, demonstrating that computational materials screening can overcome the scarcity of high-temperature thermochemical data.

Catalytic resonance theory provides a framework for understanding circumfluence (directional flow patterns) in programmable catalytic loops, where forced oscillation of binding energies creates non-equilibrium steady states with enhanced catalytic throughput.

Physics-informed perturbations (peak shifting, broadening, intensity variations) and hypothetical solid solution augmentation of training data dramatically improve robustness of XRD phase identification models to experimental artifacts from sample preparation and synthesis.

Data-driven precursor recommendation system using 29,900 text-mined solid-state synthesis recipes achieves at least 82% success rate when proposing five precursor sets for each of 2,654 unseen test target materials, learning chemical similarity to mimic human synthesis design.

Two new selectivity metrics (primary and secondary competition) for solid-state reactions, applied to 3520 literature reactions, correlate with observed target/impurity formation in synchrotron XRD characterization of reaction pathways.

Despite advances in predictive synthesis for solution-based techniques, there remained no methods to design solid-state reactions targeting metastable materials. This framework fills that gap by linking precursor choice to polymorph selectivity through reaction thermodynamics.

Established methods such as ion exchange of known compounds outperform generative AI techniques (diffusion models, VAEs, LLMs) for discovering novel inorganic crystals when evaluated against proper baselines. Random enumeration of charge-balanced prototypes also performs competitively.

Temperature-dependent materials chemistry including high-temperature phase transitions and decomposition can be predicted using the SISSO Gibbs energy descriptor, extending stability predictions beyond the 0K DFT energies that dominate current high-throughput screening.

A theoretical framework predicts and controls polymorph selectivity in solid-state reactions by using reaction energy as a handle for polymorph selection, with surface energy contributions at small particle sizes tipping the balance toward metastable polymorphs.

ML approaches for heterogeneous catalyst design encompass descriptor-based methods, neural network potentials, and active learning, enabling systematic exploration of catalyst composition-structure-activity relationships at scales infeasible for first-principles calculations alone.

Aluminum antisite disorder in iron aluminate spinels lowers cation vacancy formation energy from >3 eV to 0.62 eV when one-third of sites are inverted, making cation vacancy-mediated water splitting thermodynamically accessible and supporting hydrogen yields up to 361 umol/g.

Cation vacancies rather than conventional oxygen vacancies mediate redox cycling in iron aluminate spinel water splitting, representing a fundamentally different mechanism that explains experimental hydrogen yields not accountable by oxygen vacancy models alone.

Betti curves derived from persistent homology of electron density improve ML model performance by over 33 percentage points compared to models trained on raw electron density data for tasks including crystal structure classification, thermodynamic stability prediction, and metallic/nonmetallic distinction.

Betti curve descriptors maintain comparable information content to full electron density representations while requiring substantially less computational data, offering an efficient low-dimensional encoding of bonding characteristics in crystalline solids.

Decision tree models trained on text-mined BiFeO3 synthesis data (177 articles, 331 procedures) identify experimental factors preventing impurity phases but have limited predictive capability because important features are frequently unreported in published literature.

r2SCAN predicts systematically larger lattice constants than SCAN across the ~6000 material benchmark set, a systematic shift that should be accounted for when comparing to experimental values or SCAN-computed databases.

Established baseline methods like ion exchange are better than generative AI models at generating novel materials that are thermodynamically stable, although many of these closely resemble known compounds. Generative models excel at proposing novel structural frameworks.

Machine learning interatomic potentials can accelerate the computation of surface phase diagrams for inorganic materials, enabling surface science studies at scales previously intractable with DFT alone, with applications in heterogeneous catalysis and thin film growth.

Systematic in situ synchrotron diffraction of Li+ ion exchange into Na2Mg2P3O9N using a novel high-throughput 2D detector setup enabled simultaneous study of many samples and precise quantification of kinetic rate constants for Li-ion exchange reactions.

Low-temperature anion cometathesis selectively synthesizes defect-rich LaMnO3 perovskite with controlled defect concentrations. The metathesis route provides a pathway to perovskite phases with tunable properties through defect engineering.

ML-driven adaptive X-ray diffraction, coupling an ML algorithm with a physical diffractometer, reliably detects minor phase amounts in complex mixtures within shortened timeframes and enables in situ detection of transient phases during solid-state reactions using conventional laboratory equipment.

ML-driven adaptive XRD reliably detects minor phases in complex mixtures and enables in situ detection of transient phases during solid-state reactions using conventional laboratory equipment.

Thermodynamic assessment of ML models for solid-state synthesis prediction reveals that current models have uneven performance across composition space and reaction types. Proper thermodynamic evaluation frameworks are needed to assess synthesis prediction reliability.

High-throughput computation of 3881 potential NASICON phases yielded a machine-learned 2D tolerance factor that classifies NASICON phases by synthetic accessibility. Predictive capability validated by successful synthesis of 5 out of 6 attempted materials.

Diversifying training data is important for making balanced predictions of thermodynamic properties for inorganic crystals; data-centric approaches improve ML model generalization across chemical space.

Electronic structure calculations of GdRu2X2 (X=Si,Ge,Sn) reveal antibonding instabilities that drive structural distortions, identifying a new pathway toward centrosymmetric altermagnets. This demonstrates how electronic structure analysis can predict structural instabilities in intermetallic phases.

High-throughput computational screening of pyrochlore and spinel structures identifies candidates with giant anomalous Hall effect, demonstrating the power of crystal-structure-targeted screening for discovering materials with specific electronic transport properties.

An ensemble CNN trained on simulated XRD spectra with physics-informed augmentations achieves automated identification of complex multi-phase mixtures from experimental X-ray diffraction data, critical for autonomous synthesis characterization workflows.

Reproduction of 9 literature syntheses for BiFeO3 with incomplete parameter information reveals that text-mined datasets can guide controlled experiments and identify unreported but critical synthesis variables.

r2SCAN predicts formation energies more accurately than both SCAN and PBEsol for ~6000 solid materials, including both strongly and weakly bound systems, while requiring modestly fewer computational resources and offering significantly more reliable convergence.

r2SCAN+rVV10 van der Waals correction predicts slightly more accurate cell volumes but marginally less accurate formation enthalpies compared to bare r2SCAN, whereas SCAN+rVV10 worsens SCAN formation enthalpies.

When sufficient training data is available, generative models can more effectively target properties such as electronic band gap and bulk modulus compared to baseline approaches, suggesting generative methods have unique value for property-targeted discovery.

Betti curves retain comparable information content to full electron density (measured by Shannon entropy) while requiring 2 orders of magnitude less data, making them efficient descriptors for materials property prediction.

High-throughput thermodynamic screening of 1,148 metal nitride/metal oxide pairs for solar thermochemical ammonia synthesis (STAS) identified promising redox candidates. The screening leveraged accurate Gibbs energy models to evaluate materials at high-temperature operating conditions.

Betti curves as topological descriptors compress electron densities into compact representations that capture bonding characteristics. These descriptors outperform traditional composition and structure features for predicting certain material properties, introducing a new class of electron-density-based features for materials ML.

A probabilistic neural network ensemble trained on simulated XRD spectra with physics-based perturbations outperforms profile matching and conventional ML approaches for automated multi-phase diffraction interpretation, with a branching algorithm that explores suspected phase combinations to strengthen prediction reliability.

A probabilistic neural network ensemble trained on simulated XRD spectra with physics-based perturbations outperforms profile matching and conventional ML for automated multi-phase diffraction interpretation.

Machine learning interatomic potentials (MLIPs) accelerate prediction of inorganic surface properties by orders of magnitude compared to DFT, enabling surface energy calculations and surface reconstruction studies for materials where surface properties dictate functionality.

Wide band gap alkaline-earth chalcogenides show complex alloying behavior with potential for band gap engineering. The study maps the composition-structure-property relationships relevant to optoelectronic applications.

Triple-alkali perovskites Cs2[Alk]+[TM]3+Cl6 identified as a class with remarkable optical properties including large and tunable exciton binding energies, with properties strongly influenced by sublattice mixing between Alk-Cl and TM-Cl sublattices.

No analogous ionicity-dependent error trend is observed for SCAN-calculated formation enthalpies, explaining why ML correction works better for PBE (systematic errors) than SCAN (more random errors).

Redox-mediated stabilization mechanisms in zinc molybdenum nitrides demonstrate that internal electron transfer between metal cations can thermodynamically stabilize otherwise metastable nitride phases, providing an alternative pathway to achieve phase stability beyond conventional energetic arguments.

Multi-modal integration of text mining, in situ characterization, and ab initio calculations provides complementary insights into BiFeO3 crystallization pathways that no single method achieves alone, establishing a template for literature-experiment-computation synthesis rationalization.

Average absolute errors in predicted formation enthalpies decrease by a factor of 1.5 to 2.5 from the GGA level to the meta-GGA level (r2SCAN/SCAN), with r2SCAN improving over SCAN specifically for intermetallic systems.

Post-generation screening through stability and property filters from pre-trained ML models (including universal interatomic potentials) leads to substantial improvement in success rates of ALL methods (both baseline and generative), providing a practical low-cost pathway to more effective materials discovery.

ML models trained on Betti curve topological descriptors outperform those trained on raw electron densities by an average of 33 percentage points in classifying structure prototypes, predicting thermodynamic stability, and distinguishing metals from nonmetals.

Kinetically controlled solid-state metathesis reactions achieve selective low-temperature synthesis of Mn3N2, overcoming the typical problem that high-temperature approaches cause N2 gas release. Controlling reaction exothermicity through precursor choice prevents activation of deleterious competing pathways.

Machine learning methods for heterogeneous catalyst design enable rapid screening of catalyst candidates by learning structure-activity relationships from computational and experimental data. The review establishes best practices for ML in materials discovery.

Electronic structure analysis of Li2FeS2 reveals distinct cation and anion redox mechanisms operating during Li extraction. Understanding these redox mechanisms is critical for designing sulfide cathodes with both high voltage and high capacity.

ML-derived tolerance factor tau screened 903 Cs2BB'Cl6 double perovskites down to 311 likely stable candidates; first-principles calculations then identified 261 as likely synthesizable (decomposition enthalpy <0.05 eV/atom) with 47 having direct band gaps between 1-3 eV.

ML model reduces PBE formation enthalpy errors from MAE of 195 meV/atom to 80 meV/atom. Interpretable analysis reveals compounds with high ionicity (I>0.22) have PBE errors twice as large as low-ionicity compounds (246 vs 113 meV/atom).

Phase evolution in multicomponent ceramic synthesis (demonstrated for YBCO) can be modeled as sequential pairwise interfacial reactions. Using BaO2 instead of traditional BaCO3 precursor enables YBCO synthesis in 30 minutes vs 12+ hours, rationalized by pairwise reaction thermodynamics.

ML models trained on text-mined solid-state synthesis data predict synthesis conditions (temperature, atmosphere, precursors) for inorganic materials. Feature importance analysis reveals that thermodynamic properties and ionic radii are the most predictive features for synthesis temperature.

A materials similarity metric learned from 29,900 text-mined synthesis recipes enables automatic precursor recommendation for novel target materials. The ML model learns which precursors are interchangeable and which are essential from the scientific literature.

Growth dynamics of rutile Sn1-xGexO2 films studied by MBE and DFT reveal composition-dependent growth mechanisms. Film composition and thickness are controlled by growth conditions rationalized through DFT surface energy and defect formation calculations.

The revised tolerance factor tau, a data-derived descriptor combining ionic radii and oxidation states, classifies perovskite stability with 92.2% accuracy across 576 ABX3 compounds, substantially outperforming the Goldschmidt tolerance factor (74%).

Computational mapping of the inorganic ternary metal nitride stability landscape using DFT identifies numerous previously unknown stable and metastable phases, providing a roadmap for experimental synthesis of novel nitride materials.

The ARROWS3 algorithm automates precursor selection for solid-state synthesis by actively learning from experimental outcomes to dynamically update precursor recommendations. The algorithm outperforms static computational predictions by incorporating feedback from synthesis attempts.

DFT reveals that AlN hydrolysis proceeds through hydroxyl-mediated surface proton hopping, with the proton hopping mechanism being the rate-determining step. This atomistic understanding explains AlN's degradation in water and informs protective coating strategies.

Conventional materials discovery approaches (random charge-balanced prototype enumeration, data-driven ion exchange) outperform generative AI models (diffusion, VAE, LLM) at producing novel stable materials, though generative models better identify novel structural designs when targeting specific properties.

In situ synchrotron studies of LiCl and LiBr ion exchange reactions precisely quantify thermodynamic activation energies for solid-state reactions. LiCl rate is limited by ion hopping barrier, while LiBr rate is also affected by defect formation energy substantially lower than DFT predictions.

When multiple phases have comparable thermodynamic driving force to form, initial product is determined by kinetic factors rather than thermodynamics. Analysis of Materials Project data shows 15% of possible reactions fall within the regime of thermodynamic control.

CHGNet foundation potential used to compute thermodynamic properties for thousands of hypothetical materials generated by the Chemeleon generative model, demonstrating the utility of MLIPs as rapid thermodynamic filters for generative materials discovery.

Sublattice mixing in halide double perovskites provides a thermodynamically accessible route to modulate optoelectronic properties, with DFT-computed mixing energies predicting experimental miscibility ranges for designing targeted compositions.

First-principles calculations of calcium ion diffusion in Ca1.5Ba0.5Si5O3N6 nitridosilicate characterize migration barriers and identify preferred diffusion pathways, informing design of solid-state calcium-ion conductors.

Decision tree models trained on 331 manually extracted sol-gel synthesis procedures for BiFeO3 reinforce important experimental heuristics for impurity avoidance but show limited predictive capability due to many important synthesis features being unreported in the literature.

Mn-rich disordered rocksalts exhibit similar phase transformation to spinel-like phase as ordered layered structures, leading to improved capacity and Li-ion transport kinetics, discovered through large-scale charge-informed MLIP simulations.

Domain knowledge incorporation (understanding that stable intermediates block target formation) substantially outperforms black-box optimization for precursor selection, highlighting the importance of physically-informed optimization in autonomous synthesis platforms.

The potential for programmable catalysts to fundamentally break scaling relations that limit static catalyst performance for OER, achieving performance regimes inaccessible to any fixed-composition catalyst.

DFT screening of MgLn2X4 spinels identifies 7 chalcogenide spinels with low Mg migration barriers (<380 meV) that are stable or nearly stable (within 50 meV/atom of hull). Increasing lanthanoid size improves Mg mobility but decreases spinel stability.

Machine learning-guided adaptive XRD measurements make on-the-fly decisions during diffraction experiments to achieve optimal measurement effectiveness for autonomous phase identification. This brings ML interpretation in-line with experiments for rapid learning.

Decision tree models trained on 331 text-mined BiFeO3 sol-gel synthesis procedures identify key experimental heuristics for avoiding phase impurities but show limited predictive capability, indicating that synthesis outcome prediction requires features beyond those typically reported in papers.

r2SCAN meta-GGA functional predicts formation energies of ~6,000 solid materials more accurately than PBEsol while achieving significantly more reliable convergence than SCAN, establishing it as preferred for large-scale computational materials screening.

Post-generation filtering using ML stability predictions and universal interatomic potentials substantially enhances generative materials discovery success rates while maintaining computational efficiency.

Varying relative amounts of reactants identifies rate-limiting reagent and elucidates a universal scaling relationship controlling concentration dependence of reaction rate. Global fits across doped/undoped salts probe both intrinsic and extrinsic vacancy concentrations.

Four recently published ML synthesizability prediction models generally overpredict the likelihood of synthesis when assessed against computed thermodynamics (CHGNet-computed Ehull and thermodynamic selectivity of synthesis reactions).

By altering the thermodynamic landscape through metathesis, rapid and selective synthesis of MgCr2S4 thiospinel (a Mg-cathode material) is achieved, replacing laborious traditional ceramic synthesis routes with a thermodynamically controlled approach.

A data-centric approach to ML for inorganic materials demonstrates that improving training data quality (cleaning, augmentation, better representations) yields larger performance gains than architectural changes to ML models, establishing data quality as the primary bottleneck.

Computational screening of chalcogenide spinel conductors identifies thermodynamically stable candidates for all-solid-state Mg batteries, with stability against decomposition being a critical filter that eliminates the majority of candidate compositions.

High-throughput computational screening using tolerance factor and thermochemical stability criteria, followed by high-throughput thin film growth, successfully discovered two entirely new nitride perovskites CeWN3 and CeMoN3, demonstrating the effectiveness of combined computational-experimental discovery pipelines.

Reproduction attempts of 9 literature syntheses with varying degrees of missing synthesis parameters demonstrate how text-mined datasets can inform controlled experiments and improve understanding of impurity phase formation in complex oxide systems.

Thermodynamic analysis reveals that chalcogenide perovskites are substantially less stable than their oxide and halide counterparts, with tolerance factor predictions showing poor agreement with first-principles stability calculations for this materials class.

Aluminum nitride hydrolysis proceeds through hydroxyl-mediated surface proton hopping rather than direct water dissociation, with DFT-computed activation barriers explaining the observed temperature-dependent degradation kinetics.

Combined computational stability prediction and experimental synthesis expands the known ambient-pressure phase space of CaFe2O4-type sodium postspinel compounds, demonstrating predictive synthesis guided by DFT energy calculations.

Formation enthalpy prediction errors decrease by factors of 1.5 to 2.5 when advancing from GGA (PBE) to meta-GGA (r2SCAN/SCAN) density functionals for over 1,000 solid materials, with r2SCAN balancing numerical stability with high accuracy.

Computational approaches for guiding precursor selection in solid-state battery materials synthesis are reviewed, covering Li-ion cathodes and solid electrolytes. Methods show effectiveness but limitations remain in predicting optimal synthesis routes for complex multicomponent materials.

In situ characterization of 37 reactant pairs reveals a threshold for thermodynamic control in solid-state reactions: initial product formation can be predicted when its driving force exceeds all competing phases by at least 60 meV/atom.

Some ML synthesizability model scores do trend with thermodynamic heuristics, assigning lower scores to materials that are less stable or lack thermodynamically selective synthesis routes, suggesting partial but incomplete alignment with physical principles.

High-throughput DFT screening successfully identifies two new Ce-based nitride perovskites (CeMoN3 and CeWN3) that are subsequently experimentally synthesized, demonstrating that computational stability predictions can directly guide discovery of novel inorganic phases.

Chalcogenide perovskites exhibit fundamental thermodynamic instability arising from the mismatch between chalcogenide anion size/polarizability and perovskite structural requirements, explaining why many computationally predicted phases remain experimentally inaccessible.

A competing fluorite-family phase was identified for both CeWN3 and CeMoN3 systems, hypothesized to be a transient intermediate phase during crystallization from amorphous precursor. Different processing routes demonstrated to overcome this competing phase.

CHGNet charge-informed MLIP enables modeling of Mn valence-dependent migration and charge disproportionation during orthorhombic LixMnO2 to spinel transformation, providing atomic-level insights into cathode electrochemistry that require charge-aware interatomic potentials.

Novel iodide-assisted synthesis route enables synthesis of Li2MP2S6 thiophosphates including the new compound Li2MnP2S6, which operates at ~3V (significantly higher than typical sulfide cathodes) via a redox mechanism involving significant anionic redox participation.

Chemical similarity patterns learned from 29,900 literature synthesis procedures capture decades of experimental knowledge mathematically, enabling automated precursor recommendation that mimics expert chemist reasoning for synthesis design.

A general methodology is established to quantify conditions under which metastable polymorphs become experimentally accessible through solid-state synthesis, based on critical nucleus size ratios determined by reaction energies and surface energy differences.

DFT calculations reveal that phase separation into Na-rich and Ca-rich NASICON phases limits Ca2+ electrochemistry capacity, and Na+ ions in host materials assist migration of neighboring Ca2+ ions, providing design principles for Ca-ion battery cathodes.

Novel high-throughput in situ synchrotron studies using 2D area detector enable simultaneous monitoring of many ion exchange reactions. Kinetic rate constants extracted from time-dependent lattice parameter evolution reveal ion exchange rates are limited by ion transport in the salt rather than the ceramic host.

Ab initio and machine learned models enable rapid computational screening of materials for solar thermal water splitting, combining thermodynamic and kinetic assessments to identify promising candidates from large materials databases.

The thermodynamic driving force from salt byproduct formation in metathesis reactions can be engineered to selectively target specific products, with the magnitude of the driving force directly controlling which competing phases form preferentially.

Gibbs energy descriptors computed from a statistically-learned model combined with Materials Project DFT data enable high-throughput equilibrium prediction for multi-step thermochemical cycles, identifying promising redox pairs for solar ammonia synthesis based on B, V, Fe, and Ce.

Systematic first-principles evaluation of layered transition metal oxides as calcium intercalation cathodes reveals trade-offs between voltage, thermodynamic stability, and Ca2+ mobility, with few candidates satisfying all requirements simultaneously.

Subgroup discovery and compressed sensing methods (SISSO) can identify interpretable patterns, correlations, and descriptors in materials data, providing physically meaningful feature selection for property prediction that outperforms black-box ML approaches for certain materials properties.

Large expansion of known Na-CaFe2O4 postspinel phase space demonstrated through systematic synthesis of 17 new compositions at ambient pressure. Stability trends explained by crystal chemistry and DFT, showing even strongly Jahn-Teller active cations can form Na-CFs when combined with larger Sn4+.

Integration of text mining, in situ XRD characterization, and ab initio calculations rationalizes BiFeO3 crystallization pathways, demonstrating how multi-modal data fusion can guide understanding of complex synthesis processes.

First-principles study of alloying behavior in wide band gap alkaline-earth chalcogenides reveals phase stability trends and band gap engineering opportunities in this semiconductor materials class.

The A-Lab's closed-loop system integrates computational predictions, robotic powder handling, automated furnace operation, and ML-based XRD analysis. Failed synthesis attempts trigger adaptive recipe modification, demonstrating genuine autonomous learning from experimental outcomes.

First-principles calculations identify CaB12H12 as a potential solid-state Ca conductor with low migration barriers, demonstrating how computational screening can discover new solid-state ionic conductors for beyond-lithium battery technologies.

The ternary nitride stability map reveals that transition metal nitrides dominate the stable compositions, with alkaline earth and rare earth elements providing the most stable ternary nitride host lattices, providing chemical design rules for nitride materials discovery.

Feature importance analysis reveals optimal heating temperatures correlate with both melting points and formation energies (delta-Gf, delta-Hf) of precursor materials, extending Tamman's rule to oxide systems and suggesting reaction kinetics are governed by precursor decomposition rather than product formation thermodynamics.

Combined theoretical and experimental study unravels growth dynamics of rutile Sn1-xGexO2, demonstrating how DFT phase diagrams can guide understanding of thin film growth thermodynamics and kinetics for novel semiconductor alloys.

The importance of considering complex chemistries with additional elements during precursor selection is demonstrated: unconventional precursors containing elements not present in the target product (e.g., Na2TiO3 for BaTiO3) can enable superior synthesis routes via thermodynamic selectivity.

Review identifies key ML strategies for catalyst discovery: (1) learning adsorption energy scaling relations to avoid explicit DFT calculations, (2) transfer learning from bulk properties to surface reactivity, (3) Bayesian optimization for active learning in catalyst design space.

The A-Lab autonomous laboratory successfully synthesized 41 of 58 targeted inorganic compounds (71% success rate) over 17 days without human intervention, using ML-guided precursor selection, robotic synthesis, and automated XRD characterization.

Review identifies key components needed for autonomous materials synthesis platforms: (1) robotic synthesis and characterization automation, (2) ML for phase identification from XRD/characterization data, (3) active learning optimization algorithms for closed-loop experimental design.

Thermodynamics of proton insertion across the perovskite-brownmillerite structural transition in La0.5Sr0.5CoO3-delta reveals how vacancy ordering and proton incorporation compete thermodynamically, with implications for protonic ceramic electrochemical cells.

The NASICON machine-learned tolerance factor is based on just two descriptors combining Na content, elemental radii, electronegativities, and Madelung energy, demonstrating that synthetic accessibility can be captured by physically interpretable low-dimensional descriptors.

Random enumeration of charge-balanced prototypes serves as a surprisingly competitive baseline for crystal structure generation, highlighting that current generative models have not yet clearly surpassed simple combinatorial approaches for discovering stable novel materials.

CHGNet is pre-trained on the Materials Project Trajectory dataset (MPtrj) containing 1.58M structures with energies, forces, stresses, and magnetic moments from GGA/GGA+U DFT calculations, making it a universal potential for charge-informed atomistic modeling.

Direct observation via in situ XRD and TEM confirms that sequential pairwise interfacial reactions model correctly predicts non-equilibrium intermediate phases during YBCO synthesis, validating ab initio thermodynamics as a tool for understanding and optimizing complex ceramic synthesis.

Perspective identifies three key computational methods for guiding precursor selection: (1) thermodynamic selectivity metrics from reaction networks, (2) ML precursor recommendation from text-mined data, and (3) kinetic modeling of pairwise reactions, each with distinct applicability to Li-ion cathodes vs solid electrolytes.

The branching algorithm for multi-phase mixture identification exploits the probabilistic nature of the ensemble CNN to systematically explore suspected mixtures and identify the set of phases maximizing prediction confidence, achieving higher accuracy than single-pass classification.

Betti curves from persistent homology capture bonding characteristics by encoding connected components, cycles, and voids across electron density thresholds, providing a mathematically rigorous compression of the 3D electron density into a 1D representation.

The classical Goldschmidt tolerance factor t = (r_A + r_X) / sqrt(2)(r_B + r_X) achieves only 74% classification accuracy for perovskite vs non-perovskite ABX3 compositions across 576 experimental data points, significantly underperforming the revised tau (92% accuracy).

The octahedral factor mu shows a clear lower bound of ~0.41 for perovskite formability — compositions with mu below this threshold almost never form stable perovskite structures, providing a useful necessary-but-not-sufficient screening criterion.

CHGNet achieves energy MAE of 30 meV/atom, force MAE of 77 meV/A, and stress MAE of 0.462 GPa on the MPtrj dataset, while also predicting magnetic moments to enable charge-informed atomistic modeling across the periodic table.

Analysis of electronic instabilities in GdRu2X2 compounds reveals how antibonding interactions drive structural phase transitions, demonstrating the role of electronic structure in determining materials stability beyond simple energy-based metrics.

The review identifies that ML models trained on DFT formation energies (e.g., graph neural networks, random forests) can predict formation energy with MAE of 20-50 meV/atom depending on architecture and training data, approaching the intrinsic uncertainty of the DFT reference data itself.

Validation across 37 reactant pairs establishes that when no single phase has a driving force advantage exceeding 60 meV/atom over competitors, kinetic factors (diffusion, interface geometry) dominate initial product selection, making thermodynamic prediction unreliable in this regime.

The precursor recommendation system learns chemical similarity of materials from 29,900 text-mined recipes and refers synthesis of new targets to precedent procedures of similar materials, effectively encoding decades of heuristic synthesis knowledge in a mathematical form for use in recommendation engines.

New approach to assess ML synthesizability models introduced: using thermodynamic bounds from successful synthesis recipes to determine likely limits beyond which materials are unlikely to be synthesized, providing a principled evaluation framework in the absence of extensive negative examples.

For a few classes of systems -- transition metals, intermetallics, weakly bound solids, and decomposition reactions into compounds -- GGA-level functionals are comparable to meta-GGAs, suggesting the extra computational cost of meta-GGA may not always be justified.

The SISSO-derived Gibbs energy descriptor uses only formation enthalpy, volume per atom, and a temperature-dependent correction term, achieving remarkable accuracy with minimal computational cost compared to full phonon calculations or molecular dynamics simulations.

Varying relative reactant amounts identifies rate-limiting reagent and reveals a universal scaling relationship controlling concentration dependence of solid-state reaction rates, providing a framework to predict conditions that can accelerate ion exchange reactions by orders of magnitude.

For 231 compound-only decomposition reactions (Type 2), theory-experiment agreement within ~35 meV/atom is comparable to experimental uncertainty, establishing that DFT is effectively exact for predicting relative stability among competing compound phases.

The Goldschmidt tolerance factor t requires separate parameterizations for oxide vs halide perovskites and fails to capture the stability of many halide perovskites that fall outside the traditional 0.8 < t < 1.0 range.

Review establishes that learned interatomic potentials trained on DFT data for bulk phases can be applied to study surfaces with reasonable accuracy, but surface-specific training data and fine-tuning improve predictions for surface reconstructions and adsorbate interactions.

Current computational synthesis guidance methods have limitations: thermodynamic approaches assume equilibrium which may not hold during rapid reactions, ML models require sufficient text-mined training data which is sparse for novel material classes, and kinetic models are computationally expensive for multicomponent systems.

By coupling ML algorithm with physical diffractometer, adaptive XRD integrates diffraction and analysis such that early experimental information steers measurements toward features improving phase identification confidence, reducing total measurement time while maintaining or improving accuracy.

Decomposition enthalpy threshold of <0.05 eV/atom used as a practical criterion for likely synthesizability in the double perovskite screening, with compounds below this threshold considered thermodynamically accessible under typical synthesis conditions.

The octahedral factor mu = r_B/r_X has limited standalone predictive power for perovskite stability (values 0.44-0.90 for stable perovskites overlap substantially with non-perovskites), but improves classification accuracy when combined with the tolerance factor as a secondary descriptor.

The revised tolerance factor tau incorporates the oxidation state of the A-site cation (n_A), which captures the strength of A-O bonding and is the key physical insight that enables a single descriptor to work across both oxide and halide chemistries.

Sublattice mixing in inorganic halide double perovskites (A2BB'X6) enables continuous tuning of band gaps across the visible spectrum (1.0-3.5 eV), with DFT-HSE06 calculations predicting band gaps within 0.3 eV of experimental measurements for validated compositions.

Perovskite formability is necessary but not sufficient for desirable band gaps — many stable perovskites have band gaps outside the optimal 1.1-1.7 eV window for photovoltaics, requiring joint optimization of stability and electronic properties.

A SISSO-derived physical descriptor accurately predicts Gibbs energy of formation for inorganic crystalline solids as a function of temperature, enabling temperature-dependent stability predictions without expensive phonon calculations. The descriptor achieves MAE of 47 meV/atom across 2500+ compounds.

Temperature-dependent Gibbs energy predictions reveal that ~15% of materials stable at 0K become unstable at synthesis temperatures, and ~5% of 0K-unstable materials become accessible at elevated temperatures — demonstrating that static DFT stability predictions can be misleading for synthesis planning.

Decomposition reactions to competing phases are essential for correctly assessing DFT-predicted stability — comparing only to elemental references overestimates stability by 0.1-0.5 eV/atom for many ternary oxides, leading to false positive stability predictions.

Machine learning approaches to heterogeneous catalyst design can accelerate screening by identifying structure-activity relationships from DFT data, but performance is limited by the quality and diversity of training data — underrepresented catalyst compositions show significantly higher prediction errors.

DFT calculations reveal that aluminum nitride hydrolysis proceeds via a hydroxyl-mediated surface proton hopping mechanism with an activation barrier of 0.8 eV, establishing a molecular-level understanding of this important surface reaction for the first time.

High-throughput thermodynamic screening of 1,148 binary and ternary metal oxide/nitride pairs identifies the most promising materials for solar thermochemical ammonia synthesis, with Mn-based systems showing optimal redox properties at achievable solar concentrator temperatures.

MgxCr2S4 spinel cathode operates via the high-voltage Cr3+/4+ redox couple. DFT predicted it as a suitable high-voltage Mg cathode, but experimental electrochemical cycling showed limited reversibility.

First-principles evaluation of P-type layered CaTM2O4 demonstrates that several compositions have excellent battery properties: thermodynamic stability, average voltages of 2.2-4.2 V vs Ca/Ca2+, energy densities up to 600-800 Wh/kg.

Nanocrystalline layered MnOx with high defect concentration and lattice water demonstrates remarkable room-temperature Ca2+ electrochemical activity, achieving capacity of ~100-130 mAh/g.

Betti curves derived from persistent homology of electron density improve ML model performance by over 33 percentage points compared to models trained on raw electron density data.

Ensemble convolutional neural network trained on physics-informed augmented simulated diffraction spectra achieves exceptional accuracy for multi-phase mixture identification.