Hydrogen bonds' flexible distances and moderate strength entitle compounds to dynamic properties under external stimuli. Here we report multiple phase transitions and counter-intuitive CO2 adsorption behavior of dynamic guanidinium sulfate (GS) salt assembled via hydrogen-bonds. Exploration based on the energy landscape generated by crystal structure prediction (CSP) reveals three porous GS phases with stability of α > β > γ and the inverse order of porosity, agreeing with experimental results. Transformations among polymorphs via heating or compressing involve ion rearrangement. Adsorption isotherms of β-GS indicate that CO2 firstly enters the isolated cavities at a low gating pressure, and further increasing CO2 pressure leads to the continuous gas uptake but reduced pressure at a critical point and thus an unexpected negative pressure inflexion (NPI), followed by the final adsorption saturation. Theoretical calculations demonstrate that the NPI behavior stemmed from the GS structural transition from β to more porous γ-phase, with the γ-GS phase becoming more energy-favorable as CO2 uptake increases. Specific supramolecular interactions ensure CO2 selectivity and easy regeneration. With a CO2 uptake of 4.2 mmol g-1 (273 K, 100 kPa), GS salt exhibits great promise for CO2 capture and transport, demonstrating the potential of simple hydrogen-bonded salts as adaptive materials.

In this work, we explore the possibility of applying automated crystal structure prediction to reproduce the experimentally identified metastable porous polymorphs. Using our recently developed High-Throughput Organic Crystal Structure Prediction () framework, we conducted a systematic study on five representative organic crystalline systems including hydrogen-bonded frameworks (HOFs), featured by the presence of significant porosity, in conjunction with different choices of energy models from classical, machine learning force fields, tight binding to density functional theory. Our results suggest that the current structure generation framework, with careful selection of symmetry conditions, is likely to generate rather complex and abundant metastable crystal candidates for porous crystals. In conjunction with the recent advance in universal machine learning force fields, it becomes possible to identify experimental structures as the energetically favorable candidates from a simple energy versus density analysis, thus paving the way for computational design of complex porous materials with the target systems prior to the experimental synthesis and characterization.

Reinforcement learning with formation energy feedback for material diffusion models.

Crystal structure prediction in multifunctional materials such as Ni–Mn-based Heusler alloys is paramount to increasing applications in spintronics magnetic refrigeration and smart systems. This work provides a machine learning (ML) model for classifying these alloys’ austenitic and martensitic phases based on structural compositional and thermal properties. Several supervised learning models such as Decision Tree Naive Bayes k-Nearest Neighbors (kNN) and Artificial Neural Network (ANN) were trained and tested using a dataset compiled from experimental and first-principles literature. In order to enhance prediction accuracy, an ensemble classifier using majority voting was also proposed. The performance of the models was enhanced by feature selection and hyperparameter optimization. With 97.2% accuracy, the ensemble model outperformed any individual classifier. The results demonstrate the ability of machine learning (ML) to understand complex phase behaviors and offer a reliable route to accelerating the design and discovery of high-performing Heusler alloys.

The ability to understandcrystallization and predict the resultingsolid form of a system is not always easily achieved, but it is critical,particularly in the field of materials science. Intriguing (and previouslyunreported) crystallization behavior is observed with terephthalicdihydrazide (TeDi) as it rapidly forms two concomitant crystallinepolymorphs upon cooling in solution. The crystal morphology of FormI (FI) has not been seen before in organic systems and involves impressive,accordion-like stacks, composed of numerous twin domains and remainsstable in solution for years. Form II (FII) exists as large needlesthat disappear in solution after 20 h. All experimental methods employedreveal that FI is the most stable polymorph. Conversely, all computationalmethods utilized (conformational analyses, lattice energy calculations,and crystal structure prediction) suggest that FII is the most stablepolymorph. Isolation of FII was achieved by the crystallization ofTeDi powder with a supramolecular mimetic gelator, as the gel fibersact as a template for the preferential crystallization of FII, dueto the comparable crystal packing of FII and the gelator. This workhighlights the impact of crystallization behavior in a real laboratoryand the defects, disorder, and twinning that lead to remarkable crystalmorphologies that may not be accounted for with idealized calculations,and also explores approaches for controlling and directing crystallizationoutcomes.

Abstract Machine learning interatomic potentials have revolutionized complex materials design by enabling rapid exploration of material configurational spaces via crystal structure prediction with ab initio accuracy. However, critical challenges persist in ensuring robust generalization to unknown structures and minimizing the requirement for substantial expert knowledge and time-consuming manual interventions. Here, we propose an automated crystal structure prediction framework built upon the attention-coupled neural network potential to address these limitations. The generalizability of the potential is achieved by sampling regions across the local minima of the potential energy surface, where the self-evolving pipeline autonomously refines the potential iteratively while minimizing human intervention. The workflow is validated on Mg-Ca-H ternary and Be-P-N-O quaternary systems by exploring nearly 10 million configurations, demonstrating substantial speedup compared to first-principles calculations. These results underscore the effectiveness of our approach in accelerating the exploration and discovery of complex multi-component functional materials.

Exploring the ground-state structures of cluster molecules formed by hydrogen-bonding interactions remains a significant challenge for nanoscience. We propose a novel method called the hydrogen atom ratio-based docking evolution algorithm (HARDEA) for exploring the ground-state structures of molecules formed by hydrogen-bond interactions. This method converges faster than traditional evolutionary algorithms and has a higher probability of discovering new ground-state structures. Eighteen new ground-state structures were discovered using the HARDEA method for sulfuric acid (SA)-dimethylamine (DMA) ((SA)n(DMA)m (n = 1-4, m = 1-4)) and sulfuric acid (SA)-3-methyl-1,2,3-butane-tricarboxylic acid (MBTCA) ((SA)n(MBTCA)m (n = 1-3, m = 1-3)) systems and account for 67% of the total structures. The lowest-energy structures are 5.69 and 4.97 kcal/mol lower than those reported in the literature for (SA)4(DMA)4 and (SA)1(MBTCA)2, respectively. Compared to the field and experimental measurements of new particle formation rates, the simulated formation rates based on these new ground-state structures show improved accuracy by 20 and 3.7% than literature values for the (SA)n(DMA)m (n = 1-4, m = 1-4) and (SA)n(MBTCA)m (n = 1-3, m = 1-3) systems, respectively. The HARDEA method is general and flexible and can be applied to different kinds of problems with hydrogen-bond interaction, such as molecular crystal structure prediction, new particle formation, and protein-drug interactions.

Ionic liquids (ILs) represent an extensively studied class of materials. Nevertheless, their solid state has often been overlooked, leading to frequent knowledge gaps about their phase behavior or crystal structures that such materials may form. This work focuses on the development of a crystal structure prediction (CSP) scheme suitable for aprotic ILs, relying on quasi-random crystal structure generation, dispersion-corrected density functional theory (DFT-D)-based energy reranking, and quasi-harmonic phonon treatment. The interpretation of peculiar differences in the crystallizability of very similar ILs upon cooling of their melts is presented. The versatility of the computational protocol is validated for [emIm]-[MeSO3], an IL known to be polymorphic. The current CSP identifies the [emIm]-[MeSO3] polymorph that is thermodynamically stable in reality at the top of the stability ranking, both in terms of DFT-D refined lattice energies and quasi-harmonic Gibbs free energies. Several low-energy, high-entropy crystal structures are also proposed for [emIm]-[MeSO3] as candidates for the remaining known polymorphs with yet unresolved crystal structures. Our CSP modeling explains the extraordinary reluctance of [emIm]-[EtSO4] to crystallize due to its glassy shape of the polymorph landscape with no distinct global energy minimum crystal structure.

Abstract Lithium-based compounds with interstitial anionic electrons (IAEs) exhibit unique electronic properties, including superconductivity, and superionic behavior. The intrinsic connection between these properties offers valuable insights and potential applications in the materials science. In this study, we employed machine learning accelerated crystal structure prediction and first-principles calculations to investigate the phase stability of various Li−B systems under high pressures. Our results indicate that the known R-3m Li 6 B compound is an electride. At 150 GPa, R-3m Li6B exhibits a superconducting transition temperature of around 51 K and enters a superionic state at high temperature. Additionally, a monoclinic compound, C2/m LiB 6 , which is metastable at ambient pressure was found. More interestingly, an unpredicted caged-like metallic boron allotropes termed C2/m -B 12 can be obtained by removing Li from LiB 6 . These findings open avenues for interdisciplinary research and highlight the potential of exotic boron allotropes in advanced device applications.

Deep-ultraviolet (deep-UV) nonlinear optical (NLO) crystals are crucial for generating deep-UV lasers, and their performance is determined by the type, ratio, and arrangement of microscopic NLO functional units. Currently, there are no suitable materials capable of achieving deep-UV phase-matching (PM) laser output via direct second harmonic generation (SHG) at around 148.3nm - a key requirement for the 2 2 9Th nuclear clock. Here, we proposed a functional-units-ratio design principle to address this bottleneck. Applying this strategy to the Li-B-O-F system, we designed two novel compositions, LiB3O4F2 and Li2B4O5F4. Subsequent crystal structure prediction identified C2-LiB3O4F2 as an exceptional candidate, exhibiting a record-short PM wavelength of 145.2nm and a strong SHG response of 3.4×KH2PO4. The prediction also revealed several other metastable phases with outstanding performance, including Cc-LiB3O4F2 (149.7nm), P21-Li2B4O5F4 (151.6nm), P21-LiB3O4F2-5 (156.1nm), P21-LiB3O4F2-9 (156.8nm), and Cm-LiB3O4F2-7 (158.2nm), all of which surpass the previous record and have a high synthesis probability. Crucially, the combination of [BO3] and[BO2F2] functional units enables deep-UV PM with a moderate birefringence (∼0.05 @1064nm), effectively circumventing the traditional performance trade-off. This work provides a generalizable design strategy for next-generation deep-UV NLO materials and paves the way for the practical development of the 229Th nuclear clock.

Due to the strong electronegativity of oxygen ions, the valence band maximum (VBM) that is derived from the O 2p orbital leads to strong localization, as well as further heavy hole mass and low hole mobility, which makes it extremely difficult to obtain high-conductivity p-type transparent conductive materials. Herein, we propose the strategy of multiple anions through the introduction of weaker electronegative nitrogen, in consideration of the delocalization on VBM, as well as the stability of octahedral anion cages. As such, first-principles calculations in the framework of density functional theory (DFT) are used for this work. Crystal structure prediction software USPEX (version 2023.0) was adopted to investigate the N-O appropriate ratio in CaTiO3-xNx (0 ≤ x ≤ 1) to balance the high transmission of light and highly favorable dispersion at the VBM. Furthermore, the p-type TCO performance of CaTiO3-xNx was evaluated based on the hole effective mass, hole mobility, and conductivity. The effectiveness of modulating p-type TCO through N-O multiple anions was also evaluated through defect formation energy and ionization energy. Ultimately, the construction of a CaTiO3-xNx/Si heterojunction and band alignment were considered for practical application. This approach attempts to boost the diversity of p-type perovskite-based TCOs and opens a new perspective for engineering and innovative material design for sustainable TCOs demand.

Crystal structure prediction has developed into a valuable tool for anticipating the likely crystalline arrangement that a molecule will adopt, with applications in materials discovery and polymorph screening. Although powerful, crystal structure prediction is usually limited to locating the local minima of the crystal energy surface. We demonstrate how, by mapping the energy barriers between structures, applying the Monte Carlo threshold algorithm provides a richer description of the crystal energy landscape which allows us to rationalize the differences in experimental conditions under which different crystal polymorphs are observed. As a demonstration, we apply the method to three polymorphic polycyclic aromatic hydrocarbons, phenanthrene, pyrene, and perylene.

Today, machine learning (ML) and crystal structure prediction (CSP) guided by crystal engineering (CE) are principal tools in computational materials discovery.

Rn is a radioactive gas naturally produced in the Earth's crust that occurs in soils of different rock types. Cracking of rocks under pressure can affect the release of Rn, leading to an increase in atmospheric Rn concentrations. Hence, monitoring changes in Rn concentrations is important for predicting the geological activity. Silica is a major mineral in the Earth and super-Earth, and exploring the reaction between silica and radon has important implications for understanding the storage of radon in the mantle. Here, using crystal structure predictions and first-principles calculations, we have explored the reaction of radon gas with silica under the conditions of the mantle. The calculation results show that Rn and silica can form stable structures including Fd3̄m and P41 phases. These two phases exist in three regions including solid, partially diffused and liquid at high temperature and pressure. Under conditions of the Earth's mantle, the presence of RnSiO2 systems means that silica can act as a reservoir for radon gas.

Negative thermal expansion (NTE) is a counterintuitive phenomenon, in which materials undergo contraction as they are heated. ScF3, a well-known NTE material, has been reported to show NTE coefficients up to 1000 K. Under ambient conditions, ScF3 crystallizes in a cubic symmetry (Pm3̄m space group), the same as that in the ReO3-type structures. Crystal structure predictions (CSPs) show that at P = 1 GPa, a phase transition occurs in cubic ScF3 to form the rhombohedral phase (R3̄C space group). Quasi-harmonic approximation (QHA) calculations under high pressure conditions show that this new phase can show anisotropic NTE coefficients. The stability of this phase persists until 4 GPa. Beyond 4 GPa, the rhombohedral phase further undergoes a phase transition into an orthorhombic phase (Immm space group) with a non-corner-shared polyhedron network. This phase exhibits NTE along only one crystallographic axis, while the other two axes show no NTE response. On further increasing the pressure to 6 GPa, a trigonal-prismatic arrangement of ScF3 is obtained (R32 space group), which shows a reasonably better NTE than the previous phase due to the corner shared framework and remains stable until 9 GPa. All the phases show mechanical stability. Ab initio molecular dynamics (AIMD) simulations show that both the cubic and the rhombohedral phases show bond-length elongation as well as deviation in dihedral angle confirming their NTE.

TCSP 2.0: Template based crystal structure prediction with improved oxidation state prediction and chemistry heuristics

Crystal structure prediction for systems governed by weak non-covalent interactions remain a significant challenge due to the complex energy landscapes involved. Herein, we have experimentally investigated the impact of systematic halogen substitution in fluorinated aromatic co-formers on the formation, structure, and phase behaviour of donor–acceptor adducts and co-crystals with p-xylene (p-C6H4Me2). Using a combined approach of differential scanning calorimetry (DSC), variable-temperature powder X-ray diffraction (VT-PXRD), and single-crystal X-ray diffraction (SXD), we have characterized a series of co-crystals formed by p-C6H4Me2 with C6F5X (X = Cl, Br, I) and p-C6F4X2 derivatives. Our results revealed a clear evolution from columnar π-stacked adducts in the Cl-substituted systems to halogen-bonded structures with the heavier halogens (Br, I). The columnar 1 : 1 adducts exhibit complex solid-state phase behaviour linked to molecular dipole and steric effects, whereas co-crystals involving Br and I show simpler behaviour, with discrete η2 and η6 halogen–π interactions both being observed. In one instance, a 1 : 2 co-crystal was formed with antiferroelectric ordering requiring halogen bonding to p-C6H4Me2 from two C6F5I molecules. The results underscore the tunability of solid-state architectures through targeted halogen substitution to probe subtle non-covalent interactions. In summary, this work advances our understanding of weak intermolecular forces in crystalline materials and provides data for the predictive design of functional co-crystals.

Metal-carbon compounds are significant for their diverse carbon units, which exhibit distinctive electronic and bonding properties, and broad application potential. In this article, we present a detailed C-K phase diagram, constructed using the crystal structure prediction method, MAGUS, based on machine-learning potentials fitted from first-principles calculations, revealing diverse carbon units, from allylenide ions to graphene-like two-dimensional layers. We found that the Pnma C12K16 phase contains allylenide ions, which contribute to its insulating behavior. Meanwhile, the ambient-pressure stable Cmcm C12K4 phase contains potassium-intercalated carbon layers with unique pentagonal-hexagonal-heptagonal (5-6-7) carbon rings, which we term "σ-graphene." This σ-graphene monolayer can be synthesized either by exfoliating bulk C12K4 using an electrochemical method or by removing potassium atoms via evaporating. Furthermore, Boltzmann transport calculations show that pristine σ-graphene exhibits a high electrical conductivity (∼5.5 × 107 S/m at 300K), comparable with silver and copper, making it a promising material for electrical transport applications. In addition, σ-graphene demonstrates excellent adsorption capabilities for O2 and NO2, with adsorption energies of -0.503 and -0.528eV, respectively, suggesting potential applications in catalysis and environmental monitoring. Our work highlights the C-K system as a versatile platform for synthesizing and applying novel carbon-based materials.

Abstract Electrides, characterized by interstitial quasi-atoms (ISQs) where electrons occupy lattice voids instead of atomic orbitals, provide a unique platform for discovering novel superconductors and mixed-conduction materials. Here, using crystal structure prediction combined with first-principles calculations, we systematically explore lithium-rich Li-Bi compounds under high pressure. Several new Li rich stoichiometries, LiBi, Li 11 Bi 2 , Li 9 Bi, and Li 10 Bi, are identified as thermodynamically stable. Among them, the C 2/ m phase of Li 10 Bi features one-dimensional ISQ networks, exhibiting both metallic and electride characteristics. Electron-phonon coupling analysis reveals a dome-shaped evolution of superconducting transition temperature ( T c), reaching a maximum value of 9.9 K at 35 GPa, where the superconductivity is primarily driven by strong Li-derived phonon modes. Ab initio molecular dynamics simulations further reveal a temperature-induced superionic transition above 700 K, where Li + ions diffuse freely while Bi atoms remain fixed within the lattice. This coexistence of superconductivity and superionicity within a single crystalline framework highlights Li 10 Bi as a prototype dual-functional electride, bridging the gap between quantum superconductors and solid-state lithium-ion conductors. These findings open a new route for designing multifunctional materials that integrate electronic and ionic transport for next-generation energy and quantum applications.

The transformation of ritonavir form 1 into a less soluble form 2 is the most notorious example of the risks associated with crystal polymorphism in pharmaceuticals. Since then, significant advancements have occurred in the field of theoretical crystal structure prediction, which forecasts the potential polymorphs of a molecule and their stability ranking. However, a question remains whether in silico modeling would have predicted the ritonavir disaster and informed appropriate action. Furthermore, the experimental landscape of ritonavir remains incomplete as no solution of form 4 has been deposited. Here, we show that CSP would have foreseen the existence of more stable then-unfound form 2 of ritonavir at room temperature. From a risk standpoint, the threat posed by this polymorph would have been considered severe due to its unique conformational and structural characteristics, combined with the formulation’s low tolerance for solubility reduction. This would have prompted additional work that could have averted the crisis. Furthermore, we determined the crystal structure of form 4 of ritonavir by three-dimensional electron diffraction, combined with in silico modeling and experimental powder X-ray diffraction, revealing a disordered motif and proving it is thermodynamically unstable.